freebsd-dev/sys/net/bpf.c

3092 lines
70 KiB
C
Raw Normal View History

/*-
* SPDX-License-Identifier: BSD-3-Clause
*
1994-05-24 10:09:53 +00:00
* Copyright (c) 1990, 1991, 1993
* The Regents of the University of California. All rights reserved.
* Copyright (c) 2019 Andrey V. Elsukov <ae@FreeBSD.org>
1994-05-24 10:09:53 +00:00
*
* This code is derived from the Stanford/CMU enet packet filter,
* (net/enet.c) distributed as part of 4.3BSD, and code contributed
1995-05-30 08:16:23 +00:00
* to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
1994-05-24 10:09:53 +00:00
* Berkeley Laboratory.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
1994-05-24 10:09:53 +00:00
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)bpf.c 8.4 (Berkeley) 1/9/95
1994-05-24 10:09:53 +00:00
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_bpf.h"
#include "opt_ddb.h"
#include "opt_netgraph.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/conf.h>
#include <sys/eventhandler.h>
#include <sys/fcntl.h>
#include <sys/jail.h>
#include <sys/ktr.h>
#include <sys/lock.h>
#include <sys/malloc.h>
1994-05-24 10:09:53 +00:00
#include <sys/mbuf.h>
#include <sys/mutex.h>
1994-05-24 10:09:53 +00:00
#include <sys/time.h>
#include <sys/priv.h>
1994-05-24 10:09:53 +00:00
#include <sys/proc.h>
#include <sys/signalvar.h>
#include <sys/filio.h>
#include <sys/sockio.h>
#include <sys/ttycom.h>
#include <sys/uio.h>
#include <sys/sysent.h>
#include <sys/systm.h>
1994-05-24 10:09:53 +00:00
#include <sys/event.h>
#include <sys/file.h>
#include <sys/poll.h>
#include <sys/proc.h>
1994-05-24 10:09:53 +00:00
#include <sys/socket.h>
#ifdef DDB
#include <ddb/ddb.h>
#endif
#include <net/if.h>
#include <net/if_var.h>
Implement interface link header precomputation API. Add if_requestencap() interface method which is capable of calculating various link headers for given interface. Right now there is support for INET/INET6/ARP llheader calculation (IFENCAP_LL type request). Other types are planned to support more complex calculation (L2 multipath lagg nexthops, tunnel encap nexthops, etc..). Reshape 'struct route' to be able to pass additional data (with is length) to prepend to mbuf. These two changes permits routing code to pass pre-calculated nexthop data (like L2 header for route w/gateway) down to the stack eliminating the need for other lookups. It also brings us closer to more complex scenarios like transparently handling MPLS nexthops and tunnel interfaces. Last, but not least, it removes layering violation introduced by flowtable code (ro_lle) and simplifies handling of existing if_output consumers. ARP/ND changes: Make arp/ndp stack pre-calculate link header upon installing/updating lle record. Interface link address change are handled by re-calculating headers for all lles based on if_lladdr event. After these changes, arpresolve()/nd6_resolve() returns full pre-calculated header for supported interfaces thus simplifying if_output(). Move these lookups to separate ether_resolve_addr() function which ether returs error or fully-prepared link header. Add <arp|nd6_>resolve_addr() compat versions to return link addresses instead of pre-calculated data. BPF changes: Raw bpf writes occupied _two_ cases: AF_UNSPEC and pseudo_AF_HDRCMPLT. Despite the naming, both of there have ther header "complete". The only difference is that interface source mac has to be filled by OS for AF_UNSPEC (controlled via BIOCGHDRCMPLT). This logic has to stay inside BPF and not pollute if_output() routines. Convert BPF to pass prepend data via new 'struct route' mechanism. Note that it does not change non-optimized if_output(): ro_prepend handling is purely optional. Side note: hackish pseudo_AF_HDRCMPLT is supported for ethernet and FDDI. It is not needed for ethernet anymore. The only remaining FDDI user is dev/pdq mostly untouched since 2007. FDDI support was eliminated from OpenBSD in 2013 (sys/net/if_fddisubr.c rev 1.65). Flowtable changes: Flowtable violates layering by saving (and not correctly managing) rtes/lles. Instead of passing lle pointer, pass pointer to pre-calculated header data from that lle. Differential Revision: https://reviews.freebsd.org/D4102
2015-12-31 05:03:27 +00:00
#include <net/if_dl.h>
1994-05-24 10:09:53 +00:00
#include <net/bpf.h>
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
#include <net/bpf_buffer.h>
#ifdef BPF_JITTER
#include <net/bpf_jitter.h>
#endif
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
#include <net/bpf_zerocopy.h>
1994-05-24 10:09:53 +00:00
#include <net/bpfdesc.h>
Implement interface link header precomputation API. Add if_requestencap() interface method which is capable of calculating various link headers for given interface. Right now there is support for INET/INET6/ARP llheader calculation (IFENCAP_LL type request). Other types are planned to support more complex calculation (L2 multipath lagg nexthops, tunnel encap nexthops, etc..). Reshape 'struct route' to be able to pass additional data (with is length) to prepend to mbuf. These two changes permits routing code to pass pre-calculated nexthop data (like L2 header for route w/gateway) down to the stack eliminating the need for other lookups. It also brings us closer to more complex scenarios like transparently handling MPLS nexthops and tunnel interfaces. Last, but not least, it removes layering violation introduced by flowtable code (ro_lle) and simplifies handling of existing if_output consumers. ARP/ND changes: Make arp/ndp stack pre-calculate link header upon installing/updating lle record. Interface link address change are handled by re-calculating headers for all lles based on if_lladdr event. After these changes, arpresolve()/nd6_resolve() returns full pre-calculated header for supported interfaces thus simplifying if_output(). Move these lookups to separate ether_resolve_addr() function which ether returs error or fully-prepared link header. Add <arp|nd6_>resolve_addr() compat versions to return link addresses instead of pre-calculated data. BPF changes: Raw bpf writes occupied _two_ cases: AF_UNSPEC and pseudo_AF_HDRCMPLT. Despite the naming, both of there have ther header "complete". The only difference is that interface source mac has to be filled by OS for AF_UNSPEC (controlled via BIOCGHDRCMPLT). This logic has to stay inside BPF and not pollute if_output() routines. Convert BPF to pass prepend data via new 'struct route' mechanism. Note that it does not change non-optimized if_output(): ro_prepend handling is purely optional. Side note: hackish pseudo_AF_HDRCMPLT is supported for ethernet and FDDI. It is not needed for ethernet anymore. The only remaining FDDI user is dev/pdq mostly untouched since 2007. FDDI support was eliminated from OpenBSD in 2013 (sys/net/if_fddisubr.c rev 1.65). Flowtable changes: Flowtable violates layering by saving (and not correctly managing) rtes/lles. Instead of passing lle pointer, pass pointer to pre-calculated header data from that lle. Differential Revision: https://reviews.freebsd.org/D4102
2015-12-31 05:03:27 +00:00
#include <net/route.h>
#include <net/vnet.h>
1994-05-24 10:09:53 +00:00
#include <netinet/in.h>
#include <netinet/if_ether.h>
#include <sys/kernel.h>
#include <sys/sysctl.h>
#include <net80211/ieee80211_freebsd.h>
#include <security/mac/mac_framework.h>
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
static struct bpf_if_ext dead_bpf_if = {
.bif_dlist = CK_LIST_HEAD_INITIALIZER()
};
struct bpf_if {
#define bif_next bif_ext.bif_next
#define bif_dlist bif_ext.bif_dlist
struct bpf_if_ext bif_ext; /* public members */
u_int bif_dlt; /* link layer type */
u_int bif_hdrlen; /* length of link header */
struct bpfd_list bif_wlist; /* writer-only list */
struct ifnet *bif_ifp; /* corresponding interface */
struct bpf_if **bif_bpf; /* Pointer to pointer to us */
volatile u_int bif_refcnt;
struct epoch_context epoch_ctx;
};
CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
struct bpf_program_buffer {
struct epoch_context epoch_ctx;
#ifdef BPF_JITTER
bpf_jit_filter *func;
#endif
void *buffer[0];
};
#if defined(DEV_BPF) || defined(NETGRAPH_BPF)
1994-05-24 10:09:53 +00:00
#define PRINET 26 /* interruptible */
#define SIZEOF_BPF_HDR(type) \
(offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
#ifdef COMPAT_FREEBSD32
#include <sys/mount.h>
#include <compat/freebsd32/freebsd32.h>
#define BPF_ALIGNMENT32 sizeof(int32_t)
#define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
#ifndef BURN_BRIDGES
/*
* 32-bit version of structure prepended to each packet. We use this header
* instead of the standard one for 32-bit streams. We mark the a stream as
* 32-bit the first time we see a 32-bit compat ioctl request.
*/
struct bpf_hdr32 {
struct timeval32 bh_tstamp; /* time stamp */
uint32_t bh_caplen; /* length of captured portion */
uint32_t bh_datalen; /* original length of packet */
uint16_t bh_hdrlen; /* length of bpf header (this struct
plus alignment padding) */
};
#endif
struct bpf_program32 {
u_int bf_len;
uint32_t bf_insns;
};
struct bpf_dltlist32 {
u_int bfl_len;
u_int bfl_list;
};
#define BIOCSETF32 _IOW('B', 103, struct bpf_program32)
#define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
#define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
#define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32)
#define BIOCSETWF32 _IOW('B', 123, struct bpf_program32)
#define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32)
#endif
#define BPF_LOCK() sx_xlock(&bpf_sx)
#define BPF_UNLOCK() sx_xunlock(&bpf_sx)
#define BPF_LOCK_ASSERT() sx_assert(&bpf_sx, SA_XLOCKED)
1994-05-24 10:09:53 +00:00
/*
* bpf_iflist is a list of BPF interface structures, each corresponding to a
* specific DLT. The same network interface might have several BPF interface
* structures registered by different layers in the stack (i.e., 802.11
* frames, ethernet frames, etc).
1994-05-24 10:09:53 +00:00
*/
CK_LIST_HEAD(bpf_iflist, bpf_if);
static struct bpf_iflist bpf_iflist;
static struct sx bpf_sx; /* bpf global lock */
static int bpf_bpfd_cnt;
1994-05-24 10:09:53 +00:00
static void bpfif_ref(struct bpf_if *);
static void bpfif_rele(struct bpf_if *);
static void bpfd_ref(struct bpf_d *);
static void bpfd_rele(struct bpf_d *);
static void bpf_attachd(struct bpf_d *, struct bpf_if *);
static void bpf_detachd(struct bpf_d *);
static void bpf_detachd_locked(struct bpf_d *, bool);
static void bpfd_free(epoch_context_t);
static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
Implement interface link header precomputation API. Add if_requestencap() interface method which is capable of calculating various link headers for given interface. Right now there is support for INET/INET6/ARP llheader calculation (IFENCAP_LL type request). Other types are planned to support more complex calculation (L2 multipath lagg nexthops, tunnel encap nexthops, etc..). Reshape 'struct route' to be able to pass additional data (with is length) to prepend to mbuf. These two changes permits routing code to pass pre-calculated nexthop data (like L2 header for route w/gateway) down to the stack eliminating the need for other lookups. It also brings us closer to more complex scenarios like transparently handling MPLS nexthops and tunnel interfaces. Last, but not least, it removes layering violation introduced by flowtable code (ro_lle) and simplifies handling of existing if_output consumers. ARP/ND changes: Make arp/ndp stack pre-calculate link header upon installing/updating lle record. Interface link address change are handled by re-calculating headers for all lles based on if_lladdr event. After these changes, arpresolve()/nd6_resolve() returns full pre-calculated header for supported interfaces thus simplifying if_output(). Move these lookups to separate ether_resolve_addr() function which ether returs error or fully-prepared link header. Add <arp|nd6_>resolve_addr() compat versions to return link addresses instead of pre-calculated data. BPF changes: Raw bpf writes occupied _two_ cases: AF_UNSPEC and pseudo_AF_HDRCMPLT. Despite the naming, both of there have ther header "complete". The only difference is that interface source mac has to be filled by OS for AF_UNSPEC (controlled via BIOCGHDRCMPLT). This logic has to stay inside BPF and not pollute if_output() routines. Convert BPF to pass prepend data via new 'struct route' mechanism. Note that it does not change non-optimized if_output(): ro_prepend handling is purely optional. Side note: hackish pseudo_AF_HDRCMPLT is supported for ethernet and FDDI. It is not needed for ethernet anymore. The only remaining FDDI user is dev/pdq mostly untouched since 2007. FDDI support was eliminated from OpenBSD in 2013 (sys/net/if_fddisubr.c rev 1.65). Flowtable changes: Flowtable violates layering by saving (and not correctly managing) rtes/lles. Instead of passing lle pointer, pass pointer to pre-calculated header data from that lle. Differential Revision: https://reviews.freebsd.org/D4102
2015-12-31 05:03:27 +00:00
struct sockaddr *, int *, struct bpf_d *);
2002-03-19 21:54:18 +00:00
static int bpf_setif(struct bpf_d *, struct ifreq *);
static void bpf_timed_out(void *);
static __inline void
2002-03-19 21:54:18 +00:00
bpf_wakeup(struct bpf_d *);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
static void catchpacket(struct bpf_d *, u_char *, u_int, u_int,
void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
struct bintime *);
2002-03-19 21:54:18 +00:00
static void reset_d(struct bpf_d *);
static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
static int bpf_setdlt(struct bpf_d *, u_int);
static void filt_bpfdetach(struct knote *);
static int filt_bpfread(struct knote *, long);
static void bpf_drvinit(void *);
static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
int bpf_maxinsns = BPF_MAXINSNS;
SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
&bpf_maxinsns, 0, "Maximum bpf program instructions");
static int bpf_zerocopy_enable = 0;
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
&bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
bpf_stats_sysctl, "bpf statistics portal");
1994-05-24 10:09:53 +00:00
VNET_DEFINE_STATIC(int, bpf_optimize_writers) = 0;
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
#define V_bpf_optimize_writers VNET(bpf_optimize_writers)
SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RW,
&VNET_NAME(bpf_optimize_writers), 0,
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
"Do not send packets until BPF program is set");
static d_open_t bpfopen;
static d_read_t bpfread;
static d_write_t bpfwrite;
static d_ioctl_t bpfioctl;
static d_poll_t bpfpoll;
static d_kqfilter_t bpfkqfilter;
static struct cdevsw bpf_cdevsw = {
.d_version = D_VERSION,
.d_open = bpfopen,
.d_read = bpfread,
.d_write = bpfwrite,
.d_ioctl = bpfioctl,
.d_poll = bpfpoll,
.d_name = "bpf",
.d_kqfilter = bpfkqfilter,
};
static struct filterops bpfread_filtops = {
.f_isfd = 1,
.f_detach = filt_bpfdetach,
.f_event = filt_bpfread,
};
/*
* LOCKING MODEL USED BY BPF
*
* Locks:
* 1) global lock (BPF_LOCK). Sx, used to protect some global counters,
* every bpf_iflist changes, serializes ioctl access to bpf descriptors.
* 2) Descriptor lock. Mutex, used to protect BPF buffers and various
* structure fields used by bpf_*tap* code.
*
* Lock order: global lock, then descriptor lock.
*
* There are several possible consumers:
*
* 1. The kernel registers interface pointer with bpfattach().
* Each call allocates new bpf_if structure, references ifnet pointer
* and links bpf_if into bpf_iflist chain. This is protected with global
* lock.
*
* 2. An userland application uses ioctl() call to bpf_d descriptor.
* All such call are serialized with global lock. BPF filters can be
* changed, but pointer to old filter will be freed using epoch_call().
* Thus it should be safe for bpf_tap/bpf_mtap* code to do access to
* filter pointers, even if change will happen during bpf_tap execution.
* Destroying of bpf_d descriptor also is doing using epoch_call().
*
* 3. An userland application can write packets into bpf_d descriptor.
* There we need to be sure, that ifnet won't disappear during bpfwrite().
*
* 4. The kernel invokes bpf_tap/bpf_mtap* functions. The access to
* bif_dlist is protected with net_epoch_preempt section. So, it should
* be safe to make access to bpf_d descriptor inside the section.
*
* 5. The kernel invokes bpfdetach() on interface destroying. All lists
* are modified with global lock held and actual free() is done using
* epoch_call().
*/
static void
bpfif_free(epoch_context_t ctx)
{
struct bpf_if *bp;
bp = __containerof(ctx, struct bpf_if, epoch_ctx);
if_rele(bp->bif_ifp);
free(bp, M_BPF);
}
static void
bpfif_ref(struct bpf_if *bp)
{
refcount_acquire(&bp->bif_refcnt);
}
static void
bpfif_rele(struct bpf_if *bp)
{
if (!refcount_release(&bp->bif_refcnt))
return;
epoch_call(net_epoch_preempt, &bp->epoch_ctx, bpfif_free);
}
static void
bpfd_ref(struct bpf_d *d)
{
refcount_acquire(&d->bd_refcnt);
}
static void
bpfd_rele(struct bpf_d *d)
{
if (!refcount_release(&d->bd_refcnt))
return;
epoch_call(net_epoch_preempt, &d->epoch_ctx, bpfd_free);
}
static struct bpf_program_buffer*
bpf_program_buffer_alloc(size_t size, int flags)
{
return (malloc(sizeof(struct bpf_program_buffer) + size,
M_BPF, flags));
}
static void
bpf_program_buffer_free(epoch_context_t ctx)
{
struct bpf_program_buffer *ptr;
ptr = __containerof(ctx, struct bpf_program_buffer, epoch_ctx);
#ifdef BPF_JITTER
if (ptr->func != NULL)
bpf_destroy_jit_filter(ptr->func);
#endif
free(ptr, M_BPF);
}
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
/*
* Wrapper functions for various buffering methods. If the set of buffer
* modes expands, we will probably want to introduce a switch data structure
* similar to protosw, et.
*/
static void
bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
u_int len)
{
BPFD_LOCK_ASSERT(d);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
switch (d->bd_bufmode) {
case BPF_BUFMODE_BUFFER:
return (bpf_buffer_append_bytes(d, buf, offset, src, len));
case BPF_BUFMODE_ZBUF:
counter_u64_add(d->bd_zcopy, 1);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
default:
panic("bpf_buf_append_bytes");
}
}
static void
bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
u_int len)
{
BPFD_LOCK_ASSERT(d);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
switch (d->bd_bufmode) {
case BPF_BUFMODE_BUFFER:
return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
case BPF_BUFMODE_ZBUF:
counter_u64_add(d->bd_zcopy, 1);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
default:
panic("bpf_buf_append_mbuf");
}
}
/*
* This function gets called when the free buffer is re-assigned.
*/
static void
bpf_buf_reclaimed(struct bpf_d *d)
{
BPFD_LOCK_ASSERT(d);
switch (d->bd_bufmode) {
case BPF_BUFMODE_BUFFER:
return;
case BPF_BUFMODE_ZBUF:
bpf_zerocopy_buf_reclaimed(d);
return;
default:
panic("bpf_buf_reclaimed");
}
}
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
/*
* If the buffer mechanism has a way to decide that a held buffer can be made
* free, then it is exposed via the bpf_canfreebuf() interface. (1) is
* returned if the buffer can be discarded, (0) is returned if it cannot.
*/
static int
bpf_canfreebuf(struct bpf_d *d)
{
BPFD_LOCK_ASSERT(d);
switch (d->bd_bufmode) {
case BPF_BUFMODE_ZBUF:
return (bpf_zerocopy_canfreebuf(d));
}
return (0);
}
/*
* Allow the buffer model to indicate that the current store buffer is
* immutable, regardless of the appearance of space. Return (1) if the
* buffer is writable, and (0) if not.
*/
static int
bpf_canwritebuf(struct bpf_d *d)
{
BPFD_LOCK_ASSERT(d);
switch (d->bd_bufmode) {
case BPF_BUFMODE_ZBUF:
return (bpf_zerocopy_canwritebuf(d));
}
return (1);
}
/*
* Notify buffer model that an attempt to write to the store buffer has
* resulted in a dropped packet, in which case the buffer may be considered
* full.
*/
static void
bpf_buffull(struct bpf_d *d)
{
BPFD_LOCK_ASSERT(d);
switch (d->bd_bufmode) {
case BPF_BUFMODE_ZBUF:
bpf_zerocopy_buffull(d);
break;
}
}
/*
* Notify the buffer model that a buffer has moved into the hold position.
*/
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
void
bpf_bufheld(struct bpf_d *d)
{
BPFD_LOCK_ASSERT(d);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
switch (d->bd_bufmode) {
case BPF_BUFMODE_ZBUF:
bpf_zerocopy_bufheld(d);
break;
}
}
static void
bpf_free(struct bpf_d *d)
{
switch (d->bd_bufmode) {
case BPF_BUFMODE_BUFFER:
return (bpf_buffer_free(d));
case BPF_BUFMODE_ZBUF:
return (bpf_zerocopy_free(d));
default:
panic("bpf_buf_free");
}
}
static int
bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
{
if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
return (EOPNOTSUPP);
return (bpf_buffer_uiomove(d, buf, len, uio));
}
static int
bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
{
if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
return (EOPNOTSUPP);
return (bpf_buffer_ioctl_sblen(d, i));
}
static int
bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
{
if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
return (EOPNOTSUPP);
return (bpf_zerocopy_ioctl_getzmax(td, d, i));
}
static int
bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
{
if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
return (EOPNOTSUPP);
return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
}
static int
bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
{
if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
return (EOPNOTSUPP);
return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
}
/*
* General BPF functions.
*/
1994-05-24 10:09:53 +00:00
static int
bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
Implement interface link header precomputation API. Add if_requestencap() interface method which is capable of calculating various link headers for given interface. Right now there is support for INET/INET6/ARP llheader calculation (IFENCAP_LL type request). Other types are planned to support more complex calculation (L2 multipath lagg nexthops, tunnel encap nexthops, etc..). Reshape 'struct route' to be able to pass additional data (with is length) to prepend to mbuf. These two changes permits routing code to pass pre-calculated nexthop data (like L2 header for route w/gateway) down to the stack eliminating the need for other lookups. It also brings us closer to more complex scenarios like transparently handling MPLS nexthops and tunnel interfaces. Last, but not least, it removes layering violation introduced by flowtable code (ro_lle) and simplifies handling of existing if_output consumers. ARP/ND changes: Make arp/ndp stack pre-calculate link header upon installing/updating lle record. Interface link address change are handled by re-calculating headers for all lles based on if_lladdr event. After these changes, arpresolve()/nd6_resolve() returns full pre-calculated header for supported interfaces thus simplifying if_output(). Move these lookups to separate ether_resolve_addr() function which ether returs error or fully-prepared link header. Add <arp|nd6_>resolve_addr() compat versions to return link addresses instead of pre-calculated data. BPF changes: Raw bpf writes occupied _two_ cases: AF_UNSPEC and pseudo_AF_HDRCMPLT. Despite the naming, both of there have ther header "complete". The only difference is that interface source mac has to be filled by OS for AF_UNSPEC (controlled via BIOCGHDRCMPLT). This logic has to stay inside BPF and not pollute if_output() routines. Convert BPF to pass prepend data via new 'struct route' mechanism. Note that it does not change non-optimized if_output(): ro_prepend handling is purely optional. Side note: hackish pseudo_AF_HDRCMPLT is supported for ethernet and FDDI. It is not needed for ethernet anymore. The only remaining FDDI user is dev/pdq mostly untouched since 2007. FDDI support was eliminated from OpenBSD in 2013 (sys/net/if_fddisubr.c rev 1.65). Flowtable changes: Flowtable violates layering by saving (and not correctly managing) rtes/lles. Instead of passing lle pointer, pass pointer to pre-calculated header data from that lle. Differential Revision: https://reviews.freebsd.org/D4102
2015-12-31 05:03:27 +00:00
struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
1994-05-24 10:09:53 +00:00
{
const struct ieee80211_bpf_params *p;
struct ether_header *eh;
1994-05-24 10:09:53 +00:00
struct mbuf *m;
int error;
int len;
int hlen;
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
int slen;
1994-05-24 10:09:53 +00:00
/*
* Build a sockaddr based on the data link layer type.
* We do this at this level because the ethernet header
* is copied directly into the data field of the sockaddr.
* In the case of SLIP, there is no header and the packet
* is forwarded as is.
* Also, we are careful to leave room at the front of the mbuf
* for the link level header.
*/
switch (linktype) {
case DLT_SLIP:
sockp->sa_family = AF_INET;
hlen = 0;
break;
case DLT_EN10MB:
sockp->sa_family = AF_UNSPEC;
/* XXX Would MAXLINKHDR be better? */
hlen = ETHER_HDR_LEN;
1994-05-24 10:09:53 +00:00
break;
case DLT_FDDI:
sockp->sa_family = AF_IMPLINK;
hlen = 0;
1994-05-24 10:09:53 +00:00
break;
1998-08-18 10:13:11 +00:00
case DLT_RAW:
1994-05-24 10:09:53 +00:00
sockp->sa_family = AF_UNSPEC;
hlen = 0;
break;
case DLT_NULL:
/*
* null interface types require a 4 byte pseudo header which
* corresponds to the address family of the packet.
*/
sockp->sa_family = AF_UNSPEC;
hlen = 4;
break;
case DLT_ATM_RFC1483:
/*
* en atm driver requires 4-byte atm pseudo header.
* though it isn't standard, vpi:vci needs to be
* specified anyway.
*/
sockp->sa_family = AF_UNSPEC;
hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
break;
case DLT_PPP:
sockp->sa_family = AF_UNSPEC;
hlen = 4; /* This should match PPP_HDRLEN */
break;
case DLT_IEEE802_11: /* IEEE 802.11 wireless */
sockp->sa_family = AF_IEEE80211;
hlen = 0;
break;
case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
sockp->sa_family = AF_IEEE80211;
sockp->sa_len = 12; /* XXX != 0 */
hlen = sizeof(struct ieee80211_bpf_params);
break;
1994-05-24 10:09:53 +00:00
default:
return (EIO);
}
len = uio->uio_resid;
if (len < hlen || len - hlen > ifp->if_mtu)
return (EMSGSIZE);
m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR);
if (m == NULL)
1994-05-24 10:09:53 +00:00
return (EIO);
m->m_pkthdr.len = m->m_len = len;
1994-05-24 10:09:53 +00:00
*mp = m;
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
error = uiomove(mtod(m, u_char *), len, uio);
if (error)
goto bad;
Implement interface link header precomputation API. Add if_requestencap() interface method which is capable of calculating various link headers for given interface. Right now there is support for INET/INET6/ARP llheader calculation (IFENCAP_LL type request). Other types are planned to support more complex calculation (L2 multipath lagg nexthops, tunnel encap nexthops, etc..). Reshape 'struct route' to be able to pass additional data (with is length) to prepend to mbuf. These two changes permits routing code to pass pre-calculated nexthop data (like L2 header for route w/gateway) down to the stack eliminating the need for other lookups. It also brings us closer to more complex scenarios like transparently handling MPLS nexthops and tunnel interfaces. Last, but not least, it removes layering violation introduced by flowtable code (ro_lle) and simplifies handling of existing if_output consumers. ARP/ND changes: Make arp/ndp stack pre-calculate link header upon installing/updating lle record. Interface link address change are handled by re-calculating headers for all lles based on if_lladdr event. After these changes, arpresolve()/nd6_resolve() returns full pre-calculated header for supported interfaces thus simplifying if_output(). Move these lookups to separate ether_resolve_addr() function which ether returs error or fully-prepared link header. Add <arp|nd6_>resolve_addr() compat versions to return link addresses instead of pre-calculated data. BPF changes: Raw bpf writes occupied _two_ cases: AF_UNSPEC and pseudo_AF_HDRCMPLT. Despite the naming, both of there have ther header "complete". The only difference is that interface source mac has to be filled by OS for AF_UNSPEC (controlled via BIOCGHDRCMPLT). This logic has to stay inside BPF and not pollute if_output() routines. Convert BPF to pass prepend data via new 'struct route' mechanism. Note that it does not change non-optimized if_output(): ro_prepend handling is purely optional. Side note: hackish pseudo_AF_HDRCMPLT is supported for ethernet and FDDI. It is not needed for ethernet anymore. The only remaining FDDI user is dev/pdq mostly untouched since 2007. FDDI support was eliminated from OpenBSD in 2013 (sys/net/if_fddisubr.c rev 1.65). Flowtable changes: Flowtable violates layering by saving (and not correctly managing) rtes/lles. Instead of passing lle pointer, pass pointer to pre-calculated header data from that lle. Differential Revision: https://reviews.freebsd.org/D4102
2015-12-31 05:03:27 +00:00
slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
if (slen == 0) {
error = EPERM;
goto bad;
}
/* Check for multicast destination */
switch (linktype) {
case DLT_EN10MB:
eh = mtod(m, struct ether_header *);
if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
ETHER_ADDR_LEN) == 0)
m->m_flags |= M_BCAST;
else
m->m_flags |= M_MCAST;
}
Implement interface link header precomputation API. Add if_requestencap() interface method which is capable of calculating various link headers for given interface. Right now there is support for INET/INET6/ARP llheader calculation (IFENCAP_LL type request). Other types are planned to support more complex calculation (L2 multipath lagg nexthops, tunnel encap nexthops, etc..). Reshape 'struct route' to be able to pass additional data (with is length) to prepend to mbuf. These two changes permits routing code to pass pre-calculated nexthop data (like L2 header for route w/gateway) down to the stack eliminating the need for other lookups. It also brings us closer to more complex scenarios like transparently handling MPLS nexthops and tunnel interfaces. Last, but not least, it removes layering violation introduced by flowtable code (ro_lle) and simplifies handling of existing if_output consumers. ARP/ND changes: Make arp/ndp stack pre-calculate link header upon installing/updating lle record. Interface link address change are handled by re-calculating headers for all lles based on if_lladdr event. After these changes, arpresolve()/nd6_resolve() returns full pre-calculated header for supported interfaces thus simplifying if_output(). Move these lookups to separate ether_resolve_addr() function which ether returs error or fully-prepared link header. Add <arp|nd6_>resolve_addr() compat versions to return link addresses instead of pre-calculated data. BPF changes: Raw bpf writes occupied _two_ cases: AF_UNSPEC and pseudo_AF_HDRCMPLT. Despite the naming, both of there have ther header "complete". The only difference is that interface source mac has to be filled by OS for AF_UNSPEC (controlled via BIOCGHDRCMPLT). This logic has to stay inside BPF and not pollute if_output() routines. Convert BPF to pass prepend data via new 'struct route' mechanism. Note that it does not change non-optimized if_output(): ro_prepend handling is purely optional. Side note: hackish pseudo_AF_HDRCMPLT is supported for ethernet and FDDI. It is not needed for ethernet anymore. The only remaining FDDI user is dev/pdq mostly untouched since 2007. FDDI support was eliminated from OpenBSD in 2013 (sys/net/if_fddisubr.c rev 1.65). Flowtable changes: Flowtable violates layering by saving (and not correctly managing) rtes/lles. Instead of passing lle pointer, pass pointer to pre-calculated header data from that lle. Differential Revision: https://reviews.freebsd.org/D4102
2015-12-31 05:03:27 +00:00
if (d->bd_hdrcmplt == 0) {
memcpy(eh->ether_shost, IF_LLADDR(ifp),
sizeof(eh->ether_shost));
}
break;
}
1994-05-24 10:09:53 +00:00
/*
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
* Make room for link header, and copy it to sockaddr
1994-05-24 10:09:53 +00:00
*/
if (hlen != 0) {
if (sockp->sa_family == AF_IEEE80211) {
/*
* Collect true length from the parameter header
* NB: sockp is known to be zero'd so if we do a
* short copy unspecified parameters will be
* zero.
* NB: packet may not be aligned after stripping
* bpf params
* XXX check ibp_vers
*/
p = mtod(m, const struct ieee80211_bpf_params *);
hlen = p->ibp_len;
if (hlen > sizeof(sockp->sa_data)) {
error = EINVAL;
goto bad;
}
}
bcopy(mtod(m, const void *), sockp->sa_data, hlen);
1994-05-24 10:09:53 +00:00
}
*hdrlen = hlen;
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
return (0);
bad:
1994-05-24 10:09:53 +00:00
m_freem(m);
return (error);
}
/*
* Attach descriptor to the bpf interface, i.e. make d listen on bp,
* then reset its buffers and counters with reset_d().
1994-05-24 10:09:53 +00:00
*/
static void
bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
1994-05-24 10:09:53 +00:00
{
int op_w;
BPF_LOCK_ASSERT();
/*
* Save sysctl value to protect from sysctl change
* between reads
*/
op_w = V_bpf_optimize_writers || d->bd_writer;
if (d->bd_bif != NULL)
bpf_detachd_locked(d, false);
1994-05-24 10:09:53 +00:00
/*
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
* Point d at bp, and add d to the interface's list.
* Since there are many applications using BPF for
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
* sending raw packets only (dhcpd, cdpd are good examples)
* we can delay adding d to the list of active listeners until
* some filter is configured.
1994-05-24 10:09:53 +00:00
*/
BPFD_LOCK(d);
/*
* Hold reference to bpif while descriptor uses this interface.
*/
bpfif_ref(bp);
d->bd_bif = bp;
if (op_w != 0) {
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
/* Add to writers-only list */
CK_LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
/*
* We decrement bd_writer on every filter set operation.
* First BIOCSETF is done by pcap_open_live() to set up
* snap length. After that appliation usually sets its own
* filter.
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
*/
d->bd_writer = 2;
} else
CK_LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
reset_d(d);
BPFD_UNLOCK(d);
bpf_bpfd_cnt++;
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
__func__, d->bd_pid, d->bd_writer ? "writer" : "active");
if (op_w == 0)
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
}
/*
* Check if we need to upgrade our descriptor @d from write-only mode.
*/
static int
bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode,
int flen)
{
int is_snap, need_upgrade;
/*
* Check if we've already upgraded or new filter is empty.
*/
if (d->bd_writer == 0 || fcode == NULL)
return (0);
need_upgrade = 0;
/*
* Check if cmd looks like snaplen setting from
* pcap_bpf.c:pcap_open_live().
* Note we're not checking .k value here:
* while pcap_open_live() definitely sets to non-zero value,
* we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
* do not consider upgrading immediately
*/
if (cmd == BIOCSETF && flen == 1 &&
fcode[0].code == (BPF_RET | BPF_K))
is_snap = 1;
else
is_snap = 0;
if (is_snap == 0) {
/*
* We're setting first filter and it doesn't look like
* setting snaplen. We're probably using bpf directly.
* Upgrade immediately.
*/
need_upgrade = 1;
} else {
/*
* Do not require upgrade by first BIOCSETF
* (used to set snaplen) by pcap_open_live().
*/
if (--d->bd_writer == 0) {
/*
* First snaplen filter has already
* been set. This is probably catch-all
* filter
*/
need_upgrade = 1;
}
}
CTR5(KTR_NET,
"%s: filter function set by pid %d, "
"bd_writer counter %d, snap %d upgrade %d",
__func__, d->bd_pid, d->bd_writer,
is_snap, need_upgrade);
return (need_upgrade);
}
1994-05-24 10:09:53 +00:00
/*
* Detach a file from its interface.
*/
static void
bpf_detachd(struct bpf_d *d)
{
BPF_LOCK();
bpf_detachd_locked(d, false);
BPF_UNLOCK();
}
static void
bpf_detachd_locked(struct bpf_d *d, bool detached_ifp)
1994-05-24 10:09:53 +00:00
{
struct bpf_if *bp;
struct ifnet *ifp;
int error;
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
BPF_LOCK_ASSERT();
CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
/* Check if descriptor is attached */
if ((bp = d->bd_bif) == NULL)
return;
BPFD_LOCK(d);
Rework r348303 to reduce the time of holding global BPF lock. It appeared that using NET_EPOCH_WAIT() while holding global BPF lock can lead to another panic: spin lock 0xfffff800183c9840 (turnstile lock) held by 0xfffff80018e2c5a0 (tid 100325) too long panic: spin lock held too long ... #0 sched_switch (td=0xfffff80018e2c5a0, newtd=0xfffff8000389e000, flags=<optimized out>) at /usr/src/sys/kern/sched_ule.c:2133 #1 0xffffffff80bf9912 in mi_switch (flags=256, newtd=0x0) at /usr/src/sys/kern/kern_synch.c:439 #2 0xffffffff80c21db7 in sched_bind (td=<optimized out>, cpu=<optimized out>) at /usr/src/sys/kern/sched_ule.c:2704 #3 0xffffffff80c34c33 in epoch_block_handler_preempt (global=<optimized out>, cr=0xfffffe00005a1a00, arg=<optimized out>) at /usr/src/sys/kern/subr_epoch.c:394 #4 0xffffffff803c741b in epoch_block (global=<optimized out>, cr=<optimized out>, cb=<optimized out>, ct=<optimized out>) at /usr/src/sys/contrib/ck/src/ck_epoch.c:416 #5 ck_epoch_synchronize_wait (global=0xfffff8000380cd80, cb=<optimized out>, ct=<optimized out>) at /usr/src/sys/contrib/ck/src/ck_epoch.c:465 #6 0xffffffff80c3475e in epoch_wait_preempt (epoch=0xfffff8000380cd80) at /usr/src/sys/kern/subr_epoch.c:513 #7 0xffffffff80ce970b in bpf_detachd_locked (d=0xfffff801d309cc00, detached_ifp=<optimized out>) at /usr/src/sys/net/bpf.c:856 #8 0xffffffff80ced166 in bpf_detachd (d=<optimized out>) at /usr/src/sys/net/bpf.c:836 #9 bpf_dtor (data=0xfffff801d309cc00) at /usr/src/sys/net/bpf.c:914 To fix this add the check to the catchpacket() that BPF descriptor was not detached just before we acquired BPFD_LOCK(). Reported by: slavash Tested by: slavash MFC after: 1 week
2019-05-28 11:45:00 +00:00
/* Remove d from the interface's descriptor list. */
CK_LIST_REMOVE(d, bd_next);
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
/* Save bd_writer value */
error = d->bd_writer;
ifp = bp->bif_ifp;
d->bd_bif = NULL;
if (detached_ifp) {
/*
* Notify descriptor as it's detached, so that any
* sleepers wake up and get ENXIO.
*/
bpf_wakeup(d);
}
BPFD_UNLOCK(d);
bpf_bpfd_cnt--;
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
/* Call event handler iff d is attached */
if (error == 0)
EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
1994-05-24 10:09:53 +00:00
/*
* Check if this descriptor had requested promiscuous mode.
* If so and ifnet is not detached, turn it off.
1994-05-24 10:09:53 +00:00
*/
if (d->bd_promisc && !detached_ifp) {
1994-05-24 10:09:53 +00:00
d->bd_promisc = 0;
CURVNET_SET(ifp->if_vnet);
error = ifpromisc(ifp, 0);
CURVNET_RESTORE();
if (error != 0 && error != ENXIO) {
1994-05-24 10:09:53 +00:00
/*
* ENXIO can happen if a pccard is unplugged
1994-05-24 10:09:53 +00:00
* Something is really wrong if we were able to put
* the driver into promiscuous mode, but can't
* take it out.
*/
if_printf(bp->bif_ifp,
"bpf_detach: ifpromisc failed (%d)\n", error);
}
1994-05-24 10:09:53 +00:00
}
bpfif_rele(bp);
1994-05-24 10:09:53 +00:00
}
/*
* Close the descriptor by detaching it from its interface,
* deallocating its buffers, and marking it free.
*/
static void
bpf_dtor(void *data)
{
struct bpf_d *d = data;
BPFD_LOCK(d);
if (d->bd_state == BPF_WAITING)
callout_stop(&d->bd_callout);
d->bd_state = BPF_IDLE;
BPFD_UNLOCK(d);
funsetown(&d->bd_sigio);
bpf_detachd(d);
#ifdef MAC
mac_bpfdesc_destroy(d);
#endif /* MAC */
seldrain(&d->bd_sel);
knlist_destroy(&d->bd_sel.si_note);
callout_drain(&d->bd_callout);
bpfd_rele(d);
}
1994-05-24 10:09:53 +00:00
/*
* Open ethernet device. Returns ENXIO for illegal minor device number,
* EBUSY if file is open by another process.
*/
/* ARGSUSED */
static int
bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
1994-05-24 10:09:53 +00:00
{
struct bpf_d *d;
int error;
1994-05-24 10:09:53 +00:00
d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
error = devfs_set_cdevpriv(d, bpf_dtor);
if (error != 0) {
free(d, M_BPF);
return (error);
}
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
/* Setup counters */
d->bd_rcount = counter_u64_alloc(M_WAITOK);
d->bd_dcount = counter_u64_alloc(M_WAITOK);
d->bd_fcount = counter_u64_alloc(M_WAITOK);
d->bd_wcount = counter_u64_alloc(M_WAITOK);
d->bd_wfcount = counter_u64_alloc(M_WAITOK);
d->bd_wdcount = counter_u64_alloc(M_WAITOK);
d->bd_zcopy = counter_u64_alloc(M_WAITOK);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
/*
* For historical reasons, perform a one-time initialization call to
* the buffer routines, even though we're not yet committed to a
* particular buffer method.
*/
bpf_buffer_init(d);
if ((flags & FREAD) == 0)
d->bd_writer = 2;
d->bd_hbuf_in_use = 0;
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
d->bd_bufmode = BPF_BUFMODE_BUFFER;
d->bd_sig = SIGIO;
d->bd_direction = BPF_D_INOUT;
d->bd_refcnt = 1;
BPF_PID_REFRESH(d, td);
#ifdef MAC
mac_bpfdesc_init(d);
mac_bpfdesc_create(td->td_ucred, d);
#endif
mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
1994-05-24 10:09:53 +00:00
return (0);
}
/*
* bpfread - read next chunk of packets from buffers
*/
static int
bpfread(struct cdev *dev, struct uio *uio, int ioflag)
1994-05-24 10:09:53 +00:00
{
struct bpf_d *d;
1994-05-24 10:09:53 +00:00
int error;
int non_block;
int timed_out;
1994-05-24 10:09:53 +00:00
error = devfs_get_cdevpriv((void **)&d);
if (error != 0)
return (error);
1994-05-24 10:09:53 +00:00
/*
* Restrict application to use a buffer the same size as
* as kernel buffers.
*/
if (uio->uio_resid != d->bd_bufsize)
return (EINVAL);
non_block = ((ioflag & O_NONBLOCK) != 0);
BPFD_LOCK(d);
BPF_PID_REFRESH_CUR(d);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
BPFD_UNLOCK(d);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
return (EOPNOTSUPP);
}
if (d->bd_state == BPF_WAITING)
callout_stop(&d->bd_callout);
timed_out = (d->bd_state == BPF_TIMED_OUT);
d->bd_state = BPF_IDLE;
while (d->bd_hbuf_in_use) {
error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
PRINET|PCATCH, "bd_hbuf", 0);
if (error != 0) {
BPFD_UNLOCK(d);
return (error);
}
}
1994-05-24 10:09:53 +00:00
/*
* If the hold buffer is empty, then do a timed sleep, which
* ends when the timeout expires or when enough packets
* have arrived to fill the store buffer.
*/
while (d->bd_hbuf == NULL) {
if (d->bd_slen != 0) {
1994-05-24 10:09:53 +00:00
/*
* A packet(s) either arrived since the previous
* read or arrived while we were asleep.
*/
if (d->bd_immediate || non_block || timed_out) {
/*
* Rotate the buffers and return what's here
* if we are in immediate mode, non-blocking
* flag is set, or this descriptor timed out.
*/
ROTATE_BUFFERS(d);
break;
}
1994-05-24 10:09:53 +00:00
}
/*
* No data is available, check to see if the bpf device
* is still pointed at a real interface. If not, return
* ENXIO so that the userland process knows to rebind
* it before using it again.
*/
if (d->bd_bif == NULL) {
BPFD_UNLOCK(d);
return (ENXIO);
}
if (non_block) {
BPFD_UNLOCK(d);
return (EWOULDBLOCK);
}
error = msleep(d, &d->bd_lock, PRINET|PCATCH,
"bpf", d->bd_rtout);
1994-05-24 10:09:53 +00:00
if (error == EINTR || error == ERESTART) {
BPFD_UNLOCK(d);
1994-05-24 10:09:53 +00:00
return (error);
}
if (error == EWOULDBLOCK) {
/*
* On a timeout, return what's in the buffer,
* which may be nothing. If there is something
* in the store buffer, we can rotate the buffers.
*/
if (d->bd_hbuf)
/*
* We filled up the buffer in between
* getting the timeout and arriving
* here, so we don't need to rotate.
*/
break;
if (d->bd_slen == 0) {
BPFD_UNLOCK(d);
1994-05-24 10:09:53 +00:00
return (0);
}
ROTATE_BUFFERS(d);
break;
}
}
/*
* At this point, we know we have something in the hold slot.
*/
d->bd_hbuf_in_use = 1;
BPFD_UNLOCK(d);
1994-05-24 10:09:53 +00:00
/*
* Move data from hold buffer into user space.
* We know the entire buffer is transferred since
* we checked above that the read buffer is bpf_bufsize bytes.
*
* We do not have to worry about simultaneous reads because
* we waited for sole access to the hold buffer above.
1994-05-24 10:09:53 +00:00
*/
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1994-05-24 10:09:53 +00:00
BPFD_LOCK(d);
KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
d->bd_fbuf = d->bd_hbuf;
d->bd_hbuf = NULL;
d->bd_hlen = 0;
bpf_buf_reclaimed(d);
d->bd_hbuf_in_use = 0;
wakeup(&d->bd_hbuf_in_use);
BPFD_UNLOCK(d);
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* If there are processes sleeping on this descriptor, wake them up.
*/
static __inline void
bpf_wakeup(struct bpf_d *d)
1994-05-24 10:09:53 +00:00
{
BPFD_LOCK_ASSERT(d);
if (d->bd_state == BPF_WAITING) {
callout_stop(&d->bd_callout);
d->bd_state = BPF_IDLE;
}
wakeup(d);
if (d->bd_async && d->bd_sig && d->bd_sigio)
pgsigio(&d->bd_sigio, d->bd_sig, 0);
selwakeuppri(&d->bd_sel, PRINET);
KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1994-05-24 10:09:53 +00:00
}
static void
bpf_timed_out(void *arg)
{
struct bpf_d *d = (struct bpf_d *)arg;
BPFD_LOCK_ASSERT(d);
if (callout_pending(&d->bd_callout) ||
!callout_active(&d->bd_callout))
return;
if (d->bd_state == BPF_WAITING) {
d->bd_state = BPF_TIMED_OUT;
if (d->bd_slen != 0)
bpf_wakeup(d);
}
}
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
static int
bpf_ready(struct bpf_d *d)
{
BPFD_LOCK_ASSERT(d);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
return (1);
if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
d->bd_slen != 0)
return (1);
return (0);
}
static int
bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1994-05-24 10:09:53 +00:00
{
struct route ro;
struct sockaddr dst;
struct epoch_tracker et;
struct bpf_if *bp;
struct bpf_d *d;
1994-05-24 10:09:53 +00:00
struct ifnet *ifp;
struct mbuf *m, *mc;
int error, hlen;
1994-05-24 10:09:53 +00:00
error = devfs_get_cdevpriv((void **)&d);
if (error != 0)
return (error);
NET_EPOCH_ENTER(et);
BPFD_LOCK(d);
BPF_PID_REFRESH_CUR(d);
counter_u64_add(d->bd_wcount, 1);
if ((bp = d->bd_bif) == NULL) {
error = ENXIO;
goto out_locked;
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
}
1994-05-24 10:09:53 +00:00
ifp = bp->bif_ifp;
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
if ((ifp->if_flags & IFF_UP) == 0) {
error = ENETDOWN;
goto out_locked;
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
}
if (uio->uio_resid == 0)
goto out_locked;
1994-05-24 10:09:53 +00:00
bzero(&dst, sizeof(dst));
m = NULL;
hlen = 0;
/*
* Take extra reference, unlock d and exit from epoch section,
* since bpf_movein() can sleep.
*/
bpfd_ref(d);
NET_EPOCH_EXIT(et);
BPFD_UNLOCK(d);
error = bpf_movein(uio, (int)bp->bif_dlt, ifp,
Implement interface link header precomputation API. Add if_requestencap() interface method which is capable of calculating various link headers for given interface. Right now there is support for INET/INET6/ARP llheader calculation (IFENCAP_LL type request). Other types are planned to support more complex calculation (L2 multipath lagg nexthops, tunnel encap nexthops, etc..). Reshape 'struct route' to be able to pass additional data (with is length) to prepend to mbuf. These two changes permits routing code to pass pre-calculated nexthop data (like L2 header for route w/gateway) down to the stack eliminating the need for other lookups. It also brings us closer to more complex scenarios like transparently handling MPLS nexthops and tunnel interfaces. Last, but not least, it removes layering violation introduced by flowtable code (ro_lle) and simplifies handling of existing if_output consumers. ARP/ND changes: Make arp/ndp stack pre-calculate link header upon installing/updating lle record. Interface link address change are handled by re-calculating headers for all lles based on if_lladdr event. After these changes, arpresolve()/nd6_resolve() returns full pre-calculated header for supported interfaces thus simplifying if_output(). Move these lookups to separate ether_resolve_addr() function which ether returs error or fully-prepared link header. Add <arp|nd6_>resolve_addr() compat versions to return link addresses instead of pre-calculated data. BPF changes: Raw bpf writes occupied _two_ cases: AF_UNSPEC and pseudo_AF_HDRCMPLT. Despite the naming, both of there have ther header "complete". The only difference is that interface source mac has to be filled by OS for AF_UNSPEC (controlled via BIOCGHDRCMPLT). This logic has to stay inside BPF and not pollute if_output() routines. Convert BPF to pass prepend data via new 'struct route' mechanism. Note that it does not change non-optimized if_output(): ro_prepend handling is purely optional. Side note: hackish pseudo_AF_HDRCMPLT is supported for ethernet and FDDI. It is not needed for ethernet anymore. The only remaining FDDI user is dev/pdq mostly untouched since 2007. FDDI support was eliminated from OpenBSD in 2013 (sys/net/if_fddisubr.c rev 1.65). Flowtable changes: Flowtable violates layering by saving (and not correctly managing) rtes/lles. Instead of passing lle pointer, pass pointer to pre-calculated header data from that lle. Differential Revision: https://reviews.freebsd.org/D4102
2015-12-31 05:03:27 +00:00
&m, &dst, &hlen, d);
if (error != 0) {
counter_u64_add(d->bd_wdcount, 1);
bpfd_rele(d);
1994-05-24 10:09:53 +00:00
return (error);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
}
BPFD_LOCK(d);
/*
* Check that descriptor is still attached to the interface.
* This can happen on bpfdetach(). To avoid access to detached
* ifnet, free mbuf and return ENXIO.
*/
if (d->bd_bif == NULL) {
counter_u64_add(d->bd_wdcount, 1);
BPFD_UNLOCK(d);
bpfd_rele(d);
m_freem(m);
return (ENXIO);
}
counter_u64_add(d->bd_wfcount, 1);
if (d->bd_hdrcmplt)
dst.sa_family = pseudo_AF_HDRCMPLT;
if (d->bd_feedback) {
mc = m_dup(m, M_NOWAIT);
if (mc != NULL)
mc->m_pkthdr.rcvif = ifp;
/* Set M_PROMISC for outgoing packets to be discarded. */
if (d->bd_direction == BPF_D_INOUT)
m->m_flags |= M_PROMISC;
} else
mc = NULL;
m->m_pkthdr.len -= hlen;
m->m_len -= hlen;
m->m_data += hlen; /* XXX */
Change the curvnet variable from a global const struct vnet *, previously always pointing to the default vnet context, to a dynamically changing thread-local one. The currvnet context should be set on entry to networking code via CURVNET_SET() macros, and reverted to previous state via CURVNET_RESTORE(). Recursions on curvnet are permitted, though strongly discuouraged. This change should have no functional impact on nooptions VIMAGE kernel builds, where CURVNET_* macros expand to whitespace. The curthread->td_vnet (aka curvnet) variable's purpose is to be an indicator of the vnet context in which the current network-related operation takes place, in case we cannot deduce the current vnet context from any other source, such as by looking at mbuf's m->m_pkthdr.rcvif->if_vnet, sockets's so->so_vnet etc. Moreover, so far curvnet has turned out to be an invaluable consistency checking aid: it helps to catch cases when sockets, ifnets or any other vnet-aware structures may have leaked from one vnet to another. The exact placement of the CURVNET_SET() / CURVNET_RESTORE() macros was a result of an empirical iterative process, whith an aim to reduce recursions on CURVNET_SET() to a minimum, while still reducing the scope of CURVNET_SET() to networking only operations - the alternative would be calling CURVNET_SET() on each system call entry. In general, curvnet has to be set in three typicall cases: when processing socket-related requests from userspace or from within the kernel; when processing inbound traffic flowing from device drivers to upper layers of the networking stack, and when executing timer-driven networking functions. This change also introduces a DDB subcommand to show the list of all vnet instances. Approved by: julian (mentor)
2009-05-05 10:56:12 +00:00
CURVNET_SET(ifp->if_vnet);
#ifdef MAC
mac_bpfdesc_create_mbuf(d, m);
if (mc != NULL)
mac_bpfdesc_create_mbuf(d, mc);
#endif
Implement interface link header precomputation API. Add if_requestencap() interface method which is capable of calculating various link headers for given interface. Right now there is support for INET/INET6/ARP llheader calculation (IFENCAP_LL type request). Other types are planned to support more complex calculation (L2 multipath lagg nexthops, tunnel encap nexthops, etc..). Reshape 'struct route' to be able to pass additional data (with is length) to prepend to mbuf. These two changes permits routing code to pass pre-calculated nexthop data (like L2 header for route w/gateway) down to the stack eliminating the need for other lookups. It also brings us closer to more complex scenarios like transparently handling MPLS nexthops and tunnel interfaces. Last, but not least, it removes layering violation introduced by flowtable code (ro_lle) and simplifies handling of existing if_output consumers. ARP/ND changes: Make arp/ndp stack pre-calculate link header upon installing/updating lle record. Interface link address change are handled by re-calculating headers for all lles based on if_lladdr event. After these changes, arpresolve()/nd6_resolve() returns full pre-calculated header for supported interfaces thus simplifying if_output(). Move these lookups to separate ether_resolve_addr() function which ether returs error or fully-prepared link header. Add <arp|nd6_>resolve_addr() compat versions to return link addresses instead of pre-calculated data. BPF changes: Raw bpf writes occupied _two_ cases: AF_UNSPEC and pseudo_AF_HDRCMPLT. Despite the naming, both of there have ther header "complete". The only difference is that interface source mac has to be filled by OS for AF_UNSPEC (controlled via BIOCGHDRCMPLT). This logic has to stay inside BPF and not pollute if_output() routines. Convert BPF to pass prepend data via new 'struct route' mechanism. Note that it does not change non-optimized if_output(): ro_prepend handling is purely optional. Side note: hackish pseudo_AF_HDRCMPLT is supported for ethernet and FDDI. It is not needed for ethernet anymore. The only remaining FDDI user is dev/pdq mostly untouched since 2007. FDDI support was eliminated from OpenBSD in 2013 (sys/net/if_fddisubr.c rev 1.65). Flowtable changes: Flowtable violates layering by saving (and not correctly managing) rtes/lles. Instead of passing lle pointer, pass pointer to pre-calculated header data from that lle. Differential Revision: https://reviews.freebsd.org/D4102
2015-12-31 05:03:27 +00:00
bzero(&ro, sizeof(ro));
if (hlen != 0) {
ro.ro_prepend = (u_char *)&dst.sa_data;
ro.ro_plen = hlen;
ro.ro_flags = RT_HAS_HEADER;
}
/* Avoid possible recursion on BPFD_LOCK(). */
NET_EPOCH_ENTER(et);
BPFD_UNLOCK(d);
Implement interface link header precomputation API. Add if_requestencap() interface method which is capable of calculating various link headers for given interface. Right now there is support for INET/INET6/ARP llheader calculation (IFENCAP_LL type request). Other types are planned to support more complex calculation (L2 multipath lagg nexthops, tunnel encap nexthops, etc..). Reshape 'struct route' to be able to pass additional data (with is length) to prepend to mbuf. These two changes permits routing code to pass pre-calculated nexthop data (like L2 header for route w/gateway) down to the stack eliminating the need for other lookups. It also brings us closer to more complex scenarios like transparently handling MPLS nexthops and tunnel interfaces. Last, but not least, it removes layering violation introduced by flowtable code (ro_lle) and simplifies handling of existing if_output consumers. ARP/ND changes: Make arp/ndp stack pre-calculate link header upon installing/updating lle record. Interface link address change are handled by re-calculating headers for all lles based on if_lladdr event. After these changes, arpresolve()/nd6_resolve() returns full pre-calculated header for supported interfaces thus simplifying if_output(). Move these lookups to separate ether_resolve_addr() function which ether returs error or fully-prepared link header. Add <arp|nd6_>resolve_addr() compat versions to return link addresses instead of pre-calculated data. BPF changes: Raw bpf writes occupied _two_ cases: AF_UNSPEC and pseudo_AF_HDRCMPLT. Despite the naming, both of there have ther header "complete". The only difference is that interface source mac has to be filled by OS for AF_UNSPEC (controlled via BIOCGHDRCMPLT). This logic has to stay inside BPF and not pollute if_output() routines. Convert BPF to pass prepend data via new 'struct route' mechanism. Note that it does not change non-optimized if_output(): ro_prepend handling is purely optional. Side note: hackish pseudo_AF_HDRCMPLT is supported for ethernet and FDDI. It is not needed for ethernet anymore. The only remaining FDDI user is dev/pdq mostly untouched since 2007. FDDI support was eliminated from OpenBSD in 2013 (sys/net/if_fddisubr.c rev 1.65). Flowtable changes: Flowtable violates layering by saving (and not correctly managing) rtes/lles. Instead of passing lle pointer, pass pointer to pre-calculated header data from that lle. Differential Revision: https://reviews.freebsd.org/D4102
2015-12-31 05:03:27 +00:00
error = (*ifp->if_output)(ifp, m, &dst, &ro);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
if (error)
counter_u64_add(d->bd_wdcount, 1);
if (mc != NULL) {
if (error == 0)
(*ifp->if_input)(ifp, mc);
else
m_freem(mc);
}
NET_EPOCH_EXIT(et);
Change the curvnet variable from a global const struct vnet *, previously always pointing to the default vnet context, to a dynamically changing thread-local one. The currvnet context should be set on entry to networking code via CURVNET_SET() macros, and reverted to previous state via CURVNET_RESTORE(). Recursions on curvnet are permitted, though strongly discuouraged. This change should have no functional impact on nooptions VIMAGE kernel builds, where CURVNET_* macros expand to whitespace. The curthread->td_vnet (aka curvnet) variable's purpose is to be an indicator of the vnet context in which the current network-related operation takes place, in case we cannot deduce the current vnet context from any other source, such as by looking at mbuf's m->m_pkthdr.rcvif->if_vnet, sockets's so->so_vnet etc. Moreover, so far curvnet has turned out to be an invaluable consistency checking aid: it helps to catch cases when sockets, ifnets or any other vnet-aware structures may have leaked from one vnet to another. The exact placement of the CURVNET_SET() / CURVNET_RESTORE() macros was a result of an empirical iterative process, whith an aim to reduce recursions on CURVNET_SET() to a minimum, while still reducing the scope of CURVNET_SET() to networking only operations - the alternative would be calling CURVNET_SET() on each system call entry. In general, curvnet has to be set in three typicall cases: when processing socket-related requests from userspace or from within the kernel; when processing inbound traffic flowing from device drivers to upper layers of the networking stack, and when executing timer-driven networking functions. This change also introduces a DDB subcommand to show the list of all vnet instances. Approved by: julian (mentor)
2009-05-05 10:56:12 +00:00
CURVNET_RESTORE();
bpfd_rele(d);
return (error);
out_locked:
counter_u64_add(d->bd_wdcount, 1);
NET_EPOCH_EXIT(et);
BPFD_UNLOCK(d);
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* Reset a descriptor by flushing its packet buffer and clearing the receive
* and drop counts. This is doable for kernel-only buffers, but with
* zero-copy buffers, we can't write to (or rotate) buffers that are
* currently owned by userspace. It would be nice if we could encapsulate
* this logic in the buffer code rather than here.
1994-05-24 10:09:53 +00:00
*/
static void
reset_d(struct bpf_d *d)
1994-05-24 10:09:53 +00:00
{
BPFD_LOCK_ASSERT(d);
while (d->bd_hbuf_in_use)
mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
"bd_hbuf", 0);
if ((d->bd_hbuf != NULL) &&
(d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1994-05-24 10:09:53 +00:00
/* Free the hold buffer. */
d->bd_fbuf = d->bd_hbuf;
d->bd_hbuf = NULL;
d->bd_hlen = 0;
bpf_buf_reclaimed(d);
1994-05-24 10:09:53 +00:00
}
if (bpf_canwritebuf(d))
d->bd_slen = 0;
counter_u64_zero(d->bd_rcount);
counter_u64_zero(d->bd_dcount);
counter_u64_zero(d->bd_fcount);
counter_u64_zero(d->bd_wcount);
counter_u64_zero(d->bd_wfcount);
counter_u64_zero(d->bd_wdcount);
counter_u64_zero(d->bd_zcopy);
1994-05-24 10:09:53 +00:00
}
/*
* FIONREAD Check for read packet available.
* BIOCGBLEN Get buffer len [for read()].
* BIOCSETF Set read filter.
* BIOCSETFNR Set read filter without resetting descriptor.
* BIOCSETWF Set write filter.
1994-05-24 10:09:53 +00:00
* BIOCFLUSH Flush read packet buffer.
* BIOCPROMISC Put interface into promiscuous mode.
* BIOCGDLT Get link layer type.
* BIOCGETIF Get interface name.
* BIOCSETIF Set interface.
* BIOCSRTIMEOUT Set read timeout.
* BIOCGRTIMEOUT Get read timeout.
* BIOCGSTATS Get packet stats.
* BIOCIMMEDIATE Set immediate mode.
* BIOCVERSION Get filter language version.
* BIOCGHDRCMPLT Get "header already complete" flag
* BIOCSHDRCMPLT Set "header already complete" flag
* BIOCGDIRECTION Get packet direction flag
* BIOCSDIRECTION Set packet direction flag
* BIOCGTSTAMP Get time stamp format and resolution.
* BIOCSTSTAMP Set time stamp format and resolution.
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
* BIOCLOCK Set "locked" flag
* BIOCFEEDBACK Set packet feedback mode.
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
* BIOCSETZBUF Set current zero-copy buffer locations.
* BIOCGETZMAX Get maximum zero-copy buffer size.
* BIOCROTZBUF Force rotation of zero-copy buffer
* BIOCSETBUFMODE Set buffer mode.
* BIOCGETBUFMODE Get current buffer mode.
1994-05-24 10:09:53 +00:00
*/
/* ARGSUSED */
static int
bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
struct thread *td)
1994-05-24 10:09:53 +00:00
{
struct bpf_d *d;
int error;
error = devfs_get_cdevpriv((void **)&d);
if (error != 0)
return (error);
1994-05-24 10:09:53 +00:00
/*
* Refresh PID associated with this descriptor.
*/
BPFD_LOCK(d);
BPF_PID_REFRESH(d, td);
if (d->bd_state == BPF_WAITING)
callout_stop(&d->bd_callout);
d->bd_state = BPF_IDLE;
BPFD_UNLOCK(d);
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
if (d->bd_locked == 1) {
switch (cmd) {
case BIOCGBLEN:
case BIOCFLUSH:
case BIOCGDLT:
case BIOCGDLTLIST:
#ifdef COMPAT_FREEBSD32
case BIOCGDLTLIST32:
#endif
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
case BIOCGETIF:
case BIOCGRTIMEOUT:
#if defined(COMPAT_FREEBSD32) && defined(__amd64__)
case BIOCGRTIMEOUT32:
#endif
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
case BIOCGSTATS:
case BIOCVERSION:
case BIOCGRSIG:
case BIOCGHDRCMPLT:
case BIOCSTSTAMP:
case BIOCFEEDBACK:
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
case FIONREAD:
case BIOCLOCK:
case BIOCSRTIMEOUT:
#if defined(COMPAT_FREEBSD32) && defined(__amd64__)
case BIOCSRTIMEOUT32:
#endif
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
case BIOCIMMEDIATE:
case TIOCGPGRP:
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
case BIOCROTZBUF:
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
break;
default:
return (EPERM);
}
}
#ifdef COMPAT_FREEBSD32
/*
* If we see a 32-bit compat ioctl, mark the stream as 32-bit so
* that it will get 32-bit packet headers.
*/
switch (cmd) {
case BIOCSETF32:
case BIOCSETFNR32:
case BIOCSETWF32:
case BIOCGDLTLIST32:
case BIOCGRTIMEOUT32:
case BIOCSRTIMEOUT32:
if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
BPFD_LOCK(d);
d->bd_compat32 = 1;
BPFD_UNLOCK(d);
}
}
#endif
CURVNET_SET(TD_TO_VNET(td));
1994-05-24 10:09:53 +00:00
switch (cmd) {
default:
error = EINVAL;
break;
/*
* Check for read packet available.
*/
case FIONREAD:
{
int n;
BPFD_LOCK(d);
1994-05-24 10:09:53 +00:00
n = d->bd_slen;
while (d->bd_hbuf_in_use)
mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
PRINET, "bd_hbuf", 0);
1994-05-24 10:09:53 +00:00
if (d->bd_hbuf)
n += d->bd_hlen;
BPFD_UNLOCK(d);
1994-05-24 10:09:53 +00:00
*(int *)addr = n;
break;
}
/*
* Get buffer len [for read()].
*/
case BIOCGBLEN:
BPFD_LOCK(d);
1994-05-24 10:09:53 +00:00
*(u_int *)addr = d->bd_bufsize;
BPFD_UNLOCK(d);
1994-05-24 10:09:53 +00:00
break;
/*
* Set buffer length.
*/
case BIOCSBLEN:
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
error = bpf_ioctl_sblen(d, (u_int *)addr);
1994-05-24 10:09:53 +00:00
break;
/*
* Set link layer read filter.
*/
case BIOCSETF:
case BIOCSETFNR:
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
case BIOCSETWF:
#ifdef COMPAT_FREEBSD32
case BIOCSETF32:
case BIOCSETFNR32:
case BIOCSETWF32:
#endif
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1994-05-24 10:09:53 +00:00
break;
/*
* Flush read packet buffer.
*/
case BIOCFLUSH:
BPFD_LOCK(d);
1994-05-24 10:09:53 +00:00
reset_d(d);
BPFD_UNLOCK(d);
1994-05-24 10:09:53 +00:00
break;
/*
* Put interface into promiscuous mode.
*/
case BIOCPROMISC:
if (d->bd_bif == NULL) {
1994-05-24 10:09:53 +00:00
/*
* No interface attached yet.
*/
error = EINVAL;
break;
}
if (d->bd_promisc == 0) {
error = ifpromisc(d->bd_bif->bif_ifp, 1);
if (error == 0)
d->bd_promisc = 1;
}
break;
/*
* Get current data link type.
1994-05-24 10:09:53 +00:00
*/
case BIOCGDLT:
BPF_LOCK();
if (d->bd_bif == NULL)
1994-05-24 10:09:53 +00:00
error = EINVAL;
else
*(u_int *)addr = d->bd_bif->bif_dlt;
BPF_UNLOCK();
1994-05-24 10:09:53 +00:00
break;
/*
* Get a list of supported data link types.
*/
#ifdef COMPAT_FREEBSD32
case BIOCGDLTLIST32:
{
struct bpf_dltlist32 *list32;
struct bpf_dltlist dltlist;
list32 = (struct bpf_dltlist32 *)addr;
dltlist.bfl_len = list32->bfl_len;
dltlist.bfl_list = PTRIN(list32->bfl_list);
BPF_LOCK();
if (d->bd_bif == NULL)
error = EINVAL;
else {
error = bpf_getdltlist(d, &dltlist);
if (error == 0)
list32->bfl_len = dltlist.bfl_len;
}
BPF_UNLOCK();
break;
}
#endif
case BIOCGDLTLIST:
BPF_LOCK();
if (d->bd_bif == NULL)
error = EINVAL;
else
error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
BPF_UNLOCK();
break;
/*
* Set data link type.
*/
case BIOCSDLT:
BPF_LOCK();
if (d->bd_bif == NULL)
error = EINVAL;
else
error = bpf_setdlt(d, *(u_int *)addr);
BPF_UNLOCK();
break;
1994-05-24 10:09:53 +00:00
/*
* Get interface name.
1994-05-24 10:09:53 +00:00
*/
case BIOCGETIF:
BPF_LOCK();
if (d->bd_bif == NULL)
1994-05-24 10:09:53 +00:00
error = EINVAL;
else {
struct ifnet *const ifp = d->bd_bif->bif_ifp;
struct ifreq *const ifr = (struct ifreq *)addr;
strlcpy(ifr->ifr_name, ifp->if_xname,
sizeof(ifr->ifr_name));
}
BPF_UNLOCK();
1994-05-24 10:09:53 +00:00
break;
/*
* Set interface.
*/
case BIOCSETIF:
{
int alloc_buf, size;
/*
* Behavior here depends on the buffering model. If
* we're using kernel memory buffers, then we can
* allocate them here. If we're using zero-copy,
* then the user process must have registered buffers
* by the time we get here.
*/
alloc_buf = 0;
BPFD_LOCK(d);
if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
d->bd_sbuf == NULL)
alloc_buf = 1;
BPFD_UNLOCK(d);
if (alloc_buf) {
size = d->bd_bufsize;
error = bpf_buffer_ioctl_sblen(d, &size);
if (error != 0)
break;
}
BPF_LOCK();
error = bpf_setif(d, (struct ifreq *)addr);
BPF_UNLOCK();
break;
}
1994-05-24 10:09:53 +00:00
/*
* Set read timeout.
*/
case BIOCSRTIMEOUT:
#if defined(COMPAT_FREEBSD32) && defined(__amd64__)
case BIOCSRTIMEOUT32:
#endif
1994-05-24 10:09:53 +00:00
{
struct timeval *tv = (struct timeval *)addr;
#if defined(COMPAT_FREEBSD32) && !defined(__mips__)
struct timeval32 *tv32;
struct timeval tv64;
if (cmd == BIOCSRTIMEOUT32) {
tv32 = (struct timeval32 *)addr;
tv = &tv64;
tv->tv_sec = tv32->tv_sec;
tv->tv_usec = tv32->tv_usec;
} else
#endif
tv = (struct timeval *)addr;
/*
* Subtract 1 tick from tvtohz() since this isn't
* a one-shot timer.
*/
if ((error = itimerfix(tv)) == 0)
d->bd_rtout = tvtohz(tv) - 1;
1994-05-24 10:09:53 +00:00
break;
}
/*
* Get read timeout.
*/
case BIOCGRTIMEOUT:
#if defined(COMPAT_FREEBSD32) && defined(__amd64__)
case BIOCGRTIMEOUT32:
#endif
1994-05-24 10:09:53 +00:00
{
struct timeval *tv;
#if defined(COMPAT_FREEBSD32) && defined(__amd64__)
struct timeval32 *tv32;
struct timeval tv64;
if (cmd == BIOCGRTIMEOUT32)
tv = &tv64;
else
#endif
tv = (struct timeval *)addr;
1994-05-24 10:09:53 +00:00
tv->tv_sec = d->bd_rtout / hz;
tv->tv_usec = (d->bd_rtout % hz) * tick;
#if defined(COMPAT_FREEBSD32) && defined(__amd64__)
if (cmd == BIOCGRTIMEOUT32) {
tv32 = (struct timeval32 *)addr;
tv32->tv_sec = tv->tv_sec;
tv32->tv_usec = tv->tv_usec;
}
#endif
1994-05-24 10:09:53 +00:00
break;
}
/*
* Get packet stats.
*/
case BIOCGSTATS:
{
struct bpf_stat *bs = (struct bpf_stat *)addr;
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
/* XXXCSJP overflow */
bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount);
bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount);
1994-05-24 10:09:53 +00:00
break;
}
/*
* Set immediate mode.
*/
case BIOCIMMEDIATE:
BPFD_LOCK(d);
1994-05-24 10:09:53 +00:00
d->bd_immediate = *(u_int *)addr;
BPFD_UNLOCK(d);
1994-05-24 10:09:53 +00:00
break;
case BIOCVERSION:
{
struct bpf_version *bv = (struct bpf_version *)addr;
bv->bv_major = BPF_MAJOR_VERSION;
bv->bv_minor = BPF_MINOR_VERSION;
break;
}
/*
* Get "header already complete" flag
*/
case BIOCGHDRCMPLT:
BPFD_LOCK(d);
*(u_int *)addr = d->bd_hdrcmplt;
BPFD_UNLOCK(d);
break;
/*
* Set "header already complete" flag
*/
case BIOCSHDRCMPLT:
BPFD_LOCK(d);
d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
BPFD_UNLOCK(d);
break;
/*
* Get packet direction flag
*/
case BIOCGDIRECTION:
BPFD_LOCK(d);
*(u_int *)addr = d->bd_direction;
BPFD_UNLOCK(d);
break;
/*
* Set packet direction flag
*/
case BIOCSDIRECTION:
{
u_int direction;
direction = *(u_int *)addr;
switch (direction) {
case BPF_D_IN:
case BPF_D_INOUT:
case BPF_D_OUT:
BPFD_LOCK(d);
d->bd_direction = direction;
BPFD_UNLOCK(d);
break;
default:
error = EINVAL;
}
}
break;
/*
* Get packet timestamp format and resolution.
*/
case BIOCGTSTAMP:
BPFD_LOCK(d);
*(u_int *)addr = d->bd_tstamp;
BPFD_UNLOCK(d);
break;
/*
* Set packet timestamp format and resolution.
*/
case BIOCSTSTAMP:
{
u_int func;
func = *(u_int *)addr;
if (BPF_T_VALID(func))
d->bd_tstamp = func;
else
error = EINVAL;
}
break;
case BIOCFEEDBACK:
BPFD_LOCK(d);
d->bd_feedback = *(u_int *)addr;
BPFD_UNLOCK(d);
break;
case BIOCLOCK:
BPFD_LOCK(d);
d->bd_locked = 1;
BPFD_UNLOCK(d);
break;
case FIONBIO: /* Non-blocking I/O */
break;
case FIOASYNC: /* Send signal on receive packets */
BPFD_LOCK(d);
d->bd_async = *(int *)addr;
BPFD_UNLOCK(d);
break;
case FIOSETOWN:
/*
* XXX: Add some sort of locking here?
* fsetown() can sleep.
*/
error = fsetown(*(int *)addr, &d->bd_sigio);
break;
case FIOGETOWN:
BPFD_LOCK(d);
*(int *)addr = fgetown(&d->bd_sigio);
BPFD_UNLOCK(d);
break;
/* This is deprecated, FIOSETOWN should be used instead. */
case TIOCSPGRP:
error = fsetown(-(*(int *)addr), &d->bd_sigio);
break;
/* This is deprecated, FIOGETOWN should be used instead. */
case TIOCGPGRP:
*(int *)addr = -fgetown(&d->bd_sigio);
break;
case BIOCSRSIG: /* Set receive signal */
{
u_int sig;
sig = *(u_int *)addr;
if (sig >= NSIG)
error = EINVAL;
else {
BPFD_LOCK(d);
d->bd_sig = sig;
BPFD_UNLOCK(d);
}
break;
}
case BIOCGRSIG:
BPFD_LOCK(d);
*(u_int *)addr = d->bd_sig;
BPFD_UNLOCK(d);
break;
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
case BIOCGETBUFMODE:
BPFD_LOCK(d);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
*(u_int *)addr = d->bd_bufmode;
BPFD_UNLOCK(d);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
break;
case BIOCSETBUFMODE:
/*
* Allow the buffering mode to be changed as long as we
* haven't yet committed to a particular mode. Our
* definition of commitment, for now, is whether or not a
* buffer has been allocated or an interface attached, since
* that's the point where things get tricky.
*/
switch (*(u_int *)addr) {
case BPF_BUFMODE_BUFFER:
break;
case BPF_BUFMODE_ZBUF:
if (bpf_zerocopy_enable)
break;
/* FALLSTHROUGH */
default:
CURVNET_RESTORE();
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
return (EINVAL);
}
BPFD_LOCK(d);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
d->bd_fbuf != NULL || d->bd_bif != NULL) {
BPFD_UNLOCK(d);
CURVNET_RESTORE();
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
return (EBUSY);
}
d->bd_bufmode = *(u_int *)addr;
BPFD_UNLOCK(d);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
break;
case BIOCGETZMAX:
error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
break;
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
case BIOCSETZBUF:
error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
break;
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
case BIOCROTZBUF:
error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
break;
1994-05-24 10:09:53 +00:00
}
CURVNET_RESTORE();
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* Set d's packet filter program to fp. If this file already has a filter,
* free it and replace it. Returns EINVAL for bogus requests.
*
* Note we use global lock here to serialize bpf_setf() and bpf_setif()
* calls.
1994-05-24 10:09:53 +00:00
*/
static int
bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1994-05-24 10:09:53 +00:00
{
#ifdef COMPAT_FREEBSD32
struct bpf_program fp_swab;
struct bpf_program32 *fp32;
#endif
struct bpf_program_buffer *fcode;
struct bpf_insn *filter;
2006-05-30 19:24:01 +00:00
#ifdef BPF_JITTER
bpf_jit_filter *jfunc;
#endif
size_t size;
u_int flen;
bool track_event;
#ifdef COMPAT_FREEBSD32
switch (cmd) {
case BIOCSETF32:
case BIOCSETWF32:
case BIOCSETFNR32:
fp32 = (struct bpf_program32 *)fp;
fp_swab.bf_len = fp32->bf_len;
fp_swab.bf_insns =
(struct bpf_insn *)(uintptr_t)fp32->bf_insns;
fp = &fp_swab;
switch (cmd) {
case BIOCSETF32:
cmd = BIOCSETF;
break;
case BIOCSETWF32:
cmd = BIOCSETWF;
break;
}
break;
}
#endif
filter = NULL;
#ifdef BPF_JITTER
jfunc = NULL;
#endif
/*
* Check new filter validness before acquiring any locks.
* Allocate memory for new filter, if needed.
*/
flen = fp->bf_len;
if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
return (EINVAL);
size = flen * sizeof(*fp->bf_insns);
if (size > 0) {
/* We're setting up new filter. Copy and check actual data. */
fcode = bpf_program_buffer_alloc(size, M_WAITOK);
filter = (struct bpf_insn *)fcode->buffer;
if (copyin(fp->bf_insns, filter, size) != 0 ||
!bpf_validate(filter, flen)) {
free(fcode, M_BPF);
return (EINVAL);
}
#ifdef BPF_JITTER
if (cmd != BIOCSETWF) {
/*
* Filter is copied inside fcode and is
* perfectly valid.
*/
jfunc = bpf_jitter(filter, flen);
}
#endif
}
track_event = false;
fcode = NULL;
BPF_LOCK();
BPFD_LOCK(d);
/* Set up new filter. */
if (cmd == BIOCSETWF) {
if (d->bd_wfilter != NULL) {
fcode = __containerof((void *)d->bd_wfilter,
struct bpf_program_buffer, buffer);
#ifdef BPF_JITTER
fcode->func = NULL;
#endif
}
d->bd_wfilter = filter;
} else {
if (d->bd_rfilter != NULL) {
fcode = __containerof((void *)d->bd_rfilter,
struct bpf_program_buffer, buffer);
#ifdef BPF_JITTER
fcode->func = d->bd_bfilter;
#endif
}
d->bd_rfilter = filter;
2006-05-30 19:24:01 +00:00
#ifdef BPF_JITTER
d->bd_bfilter = jfunc;
#endif
if (cmd == BIOCSETF)
reset_d(d);
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
if (bpf_check_upgrade(cmd, d, filter, flen) != 0) {
/*
* Filter can be set several times without
* specifying interface. In this case just mark d
* as reader.
*/
d->bd_writer = 0;
if (d->bd_bif != NULL) {
/*
* Remove descriptor from writers-only list
* and add it to active readers list.
*/
CK_LIST_REMOVE(d, bd_next);
CK_LIST_INSERT_HEAD(&d->bd_bif->bif_dlist,
d, bd_next);
CTR2(KTR_NET,
"%s: upgrade required by pid %d",
__func__, d->bd_pid);
track_event = true;
}
}
}
BPFD_UNLOCK(d);
1994-05-24 10:09:53 +00:00
if (fcode != NULL)
epoch_call(net_epoch_preempt, &fcode->epoch_ctx,
bpf_program_buffer_free);
if (track_event)
EVENTHANDLER_INVOKE(bpf_track,
d->bd_bif->bif_ifp, d->bd_bif->bif_dlt, 1);
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
BPF_UNLOCK();
return (0);
1994-05-24 10:09:53 +00:00
}
/*
* Detach a file from its current interface (if attached at all) and attach
* to the interface indicated by the name stored in ifr.
* Return an errno or 0.
*/
static int
bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1994-05-24 10:09:53 +00:00
{
struct bpf_if *bp;
struct ifnet *theywant;
BPF_LOCK_ASSERT();
theywant = ifunit(ifr->ifr_name);
Fix the following bpf(4) race condition which can result in a panic: (1) bpf peer attaches to interface netif0 (2) Packet is received by netif0 (3) ifp->if_bpf pointer is checked and handed off to bpf (4) bpf peer detaches from netif0 resulting in ifp->if_bpf being initialized to NULL. (5) ifp->if_bpf is dereferenced by bpf machinery (6) Kaboom This race condition likely explains the various different kernel panics reported around sending SIGINT to tcpdump or dhclient processes. But really this race can result in kernel panics anywhere you have frequent bpf attach and detach operations with high packet per second load. Summary of changes: - Remove the bpf interface's "driverp" member - When we attach bpf interfaces, we now set the ifp->if_bpf member to the bpf interface structure. Once this is done, ifp->if_bpf should never be NULL. [1] - Introduce bpf_peers_present function, an inline operation which will do a lockless read bpf peer list associated with the interface. It should be noted that the bpf code will pickup the bpf_interface lock before adding or removing bpf peers. This should serialize the access to the bpf descriptor list, removing the race. - Expose the bpf_if structure in bpf.h so that the bpf_peers_present function can use it. This also removes the struct bpf_if; hack that was there. - Adjust all consumers of the raw if_bpf structure to use bpf_peers_present Now what happens is: (1) Packet is received by netif0 (2) Check to see if bpf descriptor list is empty (3) Pickup the bpf interface lock (4) Hand packet off to process From the attach/detach side: (1) Pickup the bpf interface lock (2) Add/remove from bpf descriptor list Now that we are storing the bpf interface structure with the ifnet, there is is no need to walk the bpf interface list to locate the correct bpf interface. We now simply look up the interface, and initialize the pointer. This has a nice side effect of changing a bpf interface attach operation from O(N) (where N is the number of bpf interfaces), to O(1). [1] From now on, we can no longer check ifp->if_bpf to tell us whether or not we have any bpf peers that might be interested in receiving packets. In collaboration with: sam@ MFC after: 1 month
2006-06-02 19:59:33 +00:00
if (theywant == NULL || theywant->if_bpf == NULL)
return (ENXIO);
1994-05-24 10:09:53 +00:00
Fix the following bpf(4) race condition which can result in a panic: (1) bpf peer attaches to interface netif0 (2) Packet is received by netif0 (3) ifp->if_bpf pointer is checked and handed off to bpf (4) bpf peer detaches from netif0 resulting in ifp->if_bpf being initialized to NULL. (5) ifp->if_bpf is dereferenced by bpf machinery (6) Kaboom This race condition likely explains the various different kernel panics reported around sending SIGINT to tcpdump or dhclient processes. But really this race can result in kernel panics anywhere you have frequent bpf attach and detach operations with high packet per second load. Summary of changes: - Remove the bpf interface's "driverp" member - When we attach bpf interfaces, we now set the ifp->if_bpf member to the bpf interface structure. Once this is done, ifp->if_bpf should never be NULL. [1] - Introduce bpf_peers_present function, an inline operation which will do a lockless read bpf peer list associated with the interface. It should be noted that the bpf code will pickup the bpf_interface lock before adding or removing bpf peers. This should serialize the access to the bpf descriptor list, removing the race. - Expose the bpf_if structure in bpf.h so that the bpf_peers_present function can use it. This also removes the struct bpf_if; hack that was there. - Adjust all consumers of the raw if_bpf structure to use bpf_peers_present Now what happens is: (1) Packet is received by netif0 (2) Check to see if bpf descriptor list is empty (3) Pickup the bpf interface lock (4) Hand packet off to process From the attach/detach side: (1) Pickup the bpf interface lock (2) Add/remove from bpf descriptor list Now that we are storing the bpf interface structure with the ifnet, there is is no need to walk the bpf interface list to locate the correct bpf interface. We now simply look up the interface, and initialize the pointer. This has a nice side effect of changing a bpf interface attach operation from O(N) (where N is the number of bpf interfaces), to O(1). [1] From now on, we can no longer check ifp->if_bpf to tell us whether or not we have any bpf peers that might be interested in receiving packets. In collaboration with: sam@ MFC after: 1 month
2006-06-02 19:59:33 +00:00
bp = theywant->if_bpf;
1994-05-24 10:09:53 +00:00
/*
* At this point, we expect the buffer is already allocated. If not,
* return an error.
1994-05-24 10:09:53 +00:00
*/
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
switch (d->bd_bufmode) {
case BPF_BUFMODE_BUFFER:
case BPF_BUFMODE_ZBUF:
if (d->bd_sbuf == NULL)
return (EINVAL);
break;
default:
panic("bpf_setif: bufmode %d", d->bd_bufmode);
}
if (bp != d->bd_bif)
Fix the following bpf(4) race condition which can result in a panic: (1) bpf peer attaches to interface netif0 (2) Packet is received by netif0 (3) ifp->if_bpf pointer is checked and handed off to bpf (4) bpf peer detaches from netif0 resulting in ifp->if_bpf being initialized to NULL. (5) ifp->if_bpf is dereferenced by bpf machinery (6) Kaboom This race condition likely explains the various different kernel panics reported around sending SIGINT to tcpdump or dhclient processes. But really this race can result in kernel panics anywhere you have frequent bpf attach and detach operations with high packet per second load. Summary of changes: - Remove the bpf interface's "driverp" member - When we attach bpf interfaces, we now set the ifp->if_bpf member to the bpf interface structure. Once this is done, ifp->if_bpf should never be NULL. [1] - Introduce bpf_peers_present function, an inline operation which will do a lockless read bpf peer list associated with the interface. It should be noted that the bpf code will pickup the bpf_interface lock before adding or removing bpf peers. This should serialize the access to the bpf descriptor list, removing the race. - Expose the bpf_if structure in bpf.h so that the bpf_peers_present function can use it. This also removes the struct bpf_if; hack that was there. - Adjust all consumers of the raw if_bpf structure to use bpf_peers_present Now what happens is: (1) Packet is received by netif0 (2) Check to see if bpf descriptor list is empty (3) Pickup the bpf interface lock (4) Hand packet off to process From the attach/detach side: (1) Pickup the bpf interface lock (2) Add/remove from bpf descriptor list Now that we are storing the bpf interface structure with the ifnet, there is is no need to walk the bpf interface list to locate the correct bpf interface. We now simply look up the interface, and initialize the pointer. This has a nice side effect of changing a bpf interface attach operation from O(N) (where N is the number of bpf interfaces), to O(1). [1] From now on, we can no longer check ifp->if_bpf to tell us whether or not we have any bpf peers that might be interested in receiving packets. In collaboration with: sam@ MFC after: 1 month
2006-06-02 19:59:33 +00:00
bpf_attachd(d, bp);
else {
BPFD_LOCK(d);
reset_d(d);
BPFD_UNLOCK(d);
}
Fix the following bpf(4) race condition which can result in a panic: (1) bpf peer attaches to interface netif0 (2) Packet is received by netif0 (3) ifp->if_bpf pointer is checked and handed off to bpf (4) bpf peer detaches from netif0 resulting in ifp->if_bpf being initialized to NULL. (5) ifp->if_bpf is dereferenced by bpf machinery (6) Kaboom This race condition likely explains the various different kernel panics reported around sending SIGINT to tcpdump or dhclient processes. But really this race can result in kernel panics anywhere you have frequent bpf attach and detach operations with high packet per second load. Summary of changes: - Remove the bpf interface's "driverp" member - When we attach bpf interfaces, we now set the ifp->if_bpf member to the bpf interface structure. Once this is done, ifp->if_bpf should never be NULL. [1] - Introduce bpf_peers_present function, an inline operation which will do a lockless read bpf peer list associated with the interface. It should be noted that the bpf code will pickup the bpf_interface lock before adding or removing bpf peers. This should serialize the access to the bpf descriptor list, removing the race. - Expose the bpf_if structure in bpf.h so that the bpf_peers_present function can use it. This also removes the struct bpf_if; hack that was there. - Adjust all consumers of the raw if_bpf structure to use bpf_peers_present Now what happens is: (1) Packet is received by netif0 (2) Check to see if bpf descriptor list is empty (3) Pickup the bpf interface lock (4) Hand packet off to process From the attach/detach side: (1) Pickup the bpf interface lock (2) Add/remove from bpf descriptor list Now that we are storing the bpf interface structure with the ifnet, there is is no need to walk the bpf interface list to locate the correct bpf interface. We now simply look up the interface, and initialize the pointer. This has a nice side effect of changing a bpf interface attach operation from O(N) (where N is the number of bpf interfaces), to O(1). [1] From now on, we can no longer check ifp->if_bpf to tell us whether or not we have any bpf peers that might be interested in receiving packets. In collaboration with: sam@ MFC after: 1 month
2006-06-02 19:59:33 +00:00
return (0);
1994-05-24 10:09:53 +00:00
}
/*
* Support for select() and poll() system calls
1994-05-24 10:09:53 +00:00
*
* Return true iff the specific operation will not block indefinitely.
* Otherwise, return false but make a note that a selwakeup() must be done.
*/
static int
bpfpoll(struct cdev *dev, int events, struct thread *td)
1994-05-24 10:09:53 +00:00
{
struct bpf_d *d;
int revents;
1994-05-24 10:09:53 +00:00
if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
return (events &
(POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
/*
* Refresh PID associated with this descriptor.
*/
revents = events & (POLLOUT | POLLWRNORM);
BPFD_LOCK(d);
BPF_PID_REFRESH(d, td);
This Implements the mumbled about "Jail" feature. This is a seriously beefed up chroot kind of thing. The process is jailed along the same lines as a chroot does it, but with additional tough restrictions imposed on what the superuser can do. For all I know, it is safe to hand over the root bit inside a prison to the customer living in that prison, this is what it was developed for in fact: "real virtual servers". Each prison has an ip number associated with it, which all IP communications will be coerced to use and each prison has its own hostname. Needless to say, you need more RAM this way, but the advantage is that each customer can run their own particular version of apache and not stomp on the toes of their neighbors. It generally does what one would expect, but setting up a jail still takes a little knowledge. A few notes: I have no scripts for setting up a jail, don't ask me for them. The IP number should be an alias on one of the interfaces. mount a /proc in each jail, it will make ps more useable. /proc/<pid>/status tells the hostname of the prison for jailed processes. Quotas are only sensible if you have a mountpoint per prison. There are no privisions for stopping resource-hogging. Some "#ifdef INET" and similar may be missing (send patches!) If somebody wants to take it from here and develop it into more of a "virtual machine" they should be most welcome! Tools, comments, patches & documentation most welcome. Have fun... Sponsored by: http://www.rndassociates.com/ Run for almost a year by: http://www.servetheweb.com/
1999-04-28 11:38:52 +00:00
if (events & (POLLIN | POLLRDNORM)) {
if (bpf_ready(d))
revents |= events & (POLLIN | POLLRDNORM);
else {
selrecord(td, &d->bd_sel);
/* Start the read timeout if necessary. */
if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
callout_reset(&d->bd_callout, d->bd_rtout,
bpf_timed_out, d);
d->bd_state = BPF_WAITING;
}
}
This Implements the mumbled about "Jail" feature. This is a seriously beefed up chroot kind of thing. The process is jailed along the same lines as a chroot does it, but with additional tough restrictions imposed on what the superuser can do. For all I know, it is safe to hand over the root bit inside a prison to the customer living in that prison, this is what it was developed for in fact: "real virtual servers". Each prison has an ip number associated with it, which all IP communications will be coerced to use and each prison has its own hostname. Needless to say, you need more RAM this way, but the advantage is that each customer can run their own particular version of apache and not stomp on the toes of their neighbors. It generally does what one would expect, but setting up a jail still takes a little knowledge. A few notes: I have no scripts for setting up a jail, don't ask me for them. The IP number should be an alias on one of the interfaces. mount a /proc in each jail, it will make ps more useable. /proc/<pid>/status tells the hostname of the prison for jailed processes. Quotas are only sensible if you have a mountpoint per prison. There are no privisions for stopping resource-hogging. Some "#ifdef INET" and similar may be missing (send patches!) If somebody wants to take it from here and develop it into more of a "virtual machine" they should be most welcome! Tools, comments, patches & documentation most welcome. Have fun... Sponsored by: http://www.rndassociates.com/ Run for almost a year by: http://www.servetheweb.com/
1999-04-28 11:38:52 +00:00
}
BPFD_UNLOCK(d);
return (revents);
1994-05-24 10:09:53 +00:00
}
/*
* Support for kevent() system call. Register EVFILT_READ filters and
* reject all others.
*/
int
bpfkqfilter(struct cdev *dev, struct knote *kn)
{
struct bpf_d *d;
if (devfs_get_cdevpriv((void **)&d) != 0 ||
kn->kn_filter != EVFILT_READ)
return (1);
/*
* Refresh PID associated with this descriptor.
*/
BPFD_LOCK(d);
BPF_PID_REFRESH_CUR(d);
kn->kn_fop = &bpfread_filtops;
kn->kn_hook = d;
knlist_add(&d->bd_sel.si_note, kn, 1);
BPFD_UNLOCK(d);
return (0);
}
static void
filt_bpfdetach(struct knote *kn)
{
struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
knlist_remove(&d->bd_sel.si_note, kn, 0);
}
static int
filt_bpfread(struct knote *kn, long hint)
{
struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
int ready;
BPFD_LOCK_ASSERT(d);
ready = bpf_ready(d);
if (ready) {
kn->kn_data = d->bd_slen;
/*
* Ignore the hold buffer if it is being copied to user space.
*/
if (!d->bd_hbuf_in_use && d->bd_hbuf)
kn->kn_data += d->bd_hlen;
2010-03-12 19:42:42 +00:00
} else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
callout_reset(&d->bd_callout, d->bd_rtout,
bpf_timed_out, d);
d->bd_state = BPF_WAITING;
}
return (ready);
}
#define BPF_TSTAMP_NONE 0
#define BPF_TSTAMP_FAST 1
#define BPF_TSTAMP_NORMAL 2
#define BPF_TSTAMP_EXTERN 3
static int
bpf_ts_quality(int tstype)
{
if (tstype == BPF_T_NONE)
return (BPF_TSTAMP_NONE);
if ((tstype & BPF_T_FAST) != 0)
return (BPF_TSTAMP_FAST);
return (BPF_TSTAMP_NORMAL);
}
static int
bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
{
struct m_tag *tag;
int quality;
quality = bpf_ts_quality(tstype);
if (quality == BPF_TSTAMP_NONE)
return (quality);
if (m != NULL) {
tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
if (tag != NULL) {
*bt = *(struct bintime *)(tag + 1);
return (BPF_TSTAMP_EXTERN);
}
}
if (quality == BPF_TSTAMP_NORMAL)
binuptime(bt);
else
getbinuptime(bt);
return (quality);
}
1994-05-24 10:09:53 +00:00
/*
* Incoming linkage from device drivers. Process the packet pkt, of length
* pktlen, which is stored in a contiguous buffer. The packet is parsed
* by each process' filter, and if accepted, stashed into the corresponding
* buffer.
*/
void
bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1994-05-24 10:09:53 +00:00
{
struct epoch_tracker et;
struct bintime bt;
struct bpf_d *d;
#ifdef BPF_JITTER
bpf_jit_filter *bf;
#endif
u_int slen;
int gottime;
gottime = BPF_TSTAMP_NONE;
NET_EPOCH_ENTER(et);
CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
counter_u64_add(d->bd_rcount, 1);
/*
* NB: We dont call BPF_CHECK_DIRECTION() here since there
* is no way for the caller to indiciate to us whether this
* packet is inbound or outbound. In the bpf_mtap() routines,
* we use the interface pointers on the mbuf to figure it out.
*/
#ifdef BPF_JITTER
bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
if (bf != NULL)
slen = (*(bf->func))(pkt, pktlen, pktlen);
else
#endif
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
if (slen != 0) {
/*
* Filter matches. Let's to acquire write lock.
*/
BPFD_LOCK(d);
counter_u64_add(d->bd_fcount, 1);
if (gottime < bpf_ts_quality(d->bd_tstamp))
gottime = bpf_gettime(&bt, d->bd_tstamp,
NULL);
#ifdef MAC
if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
#endif
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
catchpacket(d, pkt, pktlen, slen,
bpf_append_bytes, &bt);
BPFD_UNLOCK(d);
}
1994-05-24 10:09:53 +00:00
}
NET_EPOCH_EXIT(et);
1994-05-24 10:09:53 +00:00
}
#define BPF_CHECK_DIRECTION(d, r, i) \
(((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
((d)->bd_direction == BPF_D_OUT && (r) == (i)))
1994-05-24 10:09:53 +00:00
/*
* Incoming linkage from device drivers, when packet is in an mbuf chain.
* Locking model is explained in bpf_tap().
1994-05-24 10:09:53 +00:00
*/
void
bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1994-05-24 10:09:53 +00:00
{
struct epoch_tracker et;
struct bintime bt;
1994-05-24 10:09:53 +00:00
struct bpf_d *d;
#ifdef BPF_JITTER
bpf_jit_filter *bf;
#endif
u_int pktlen, slen;
int gottime;
/* Skip outgoing duplicate packets. */
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
if ((m->m_flags & M_PROMISC) != 0 && m_rcvif(m) == NULL) {
m->m_flags &= ~M_PROMISC;
return;
}
pktlen = m_length(m, NULL);
gottime = BPF_TSTAMP_NONE;
NET_EPOCH_ENTER(et);
CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
Restructure mbuf send tags to provide stronger guarantees. - Perform ifp mismatch checks (to determine if a send tag is allocated for a different ifp than the one the packet is being output on), in ip_output() and ip6_output(). This avoids sending packets with send tags to ifnet drivers that don't support send tags. Since we are now checking for ifp mismatches before invoking if_output, we can now try to allocate a new tag before invoking if_output sending the original packet on the new tag if allocation succeeds. To avoid code duplication for the fragment and unfragmented cases, add ip_output_send() and ip6_output_send() as wrappers around if_output and nd6_output_ifp, respectively. All of the logic for setting send tags and dealing with send tag-related errors is done in these wrapper functions. For pseudo interfaces that wrap other network interfaces (vlan and lagg), wrapper send tags are now allocated so that ip*_output see the wrapper ifp as the ifp in the send tag. The if_transmit routines rewrite the send tags after performing an ifp mismatch check. If an ifp mismatch is detected, the transmit routines fail with EAGAIN. - To provide clearer life cycle management of send tags, especially in the presence of vlan and lagg wrapper tags, add a reference count to send tags managed via m_snd_tag_ref() and m_snd_tag_rele(). Provide a helper function (m_snd_tag_init()) for use by drivers supporting send tags. m_snd_tag_init() takes care of the if_ref on the ifp meaning that code alloating send tags via if_snd_tag_alloc no longer has to manage that manually. Similarly, m_snd_tag_rele drops the refcount on the ifp after invoking if_snd_tag_free when the last reference to a send tag is dropped. This also closes use after free races if there are pending packets in driver tx rings after the socket is closed (e.g. from tcpdrop). In order for m_free to work reliably, add a new CSUM_SND_TAG flag in csum_flags to indicate 'snd_tag' is set (rather than 'rcvif'). Drivers now also check this flag instead of checking snd_tag against NULL. This avoids false positive matches when a forwarded packet has a non-NULL rcvif that was treated as a send tag. - cxgbe was relying on snd_tag_free being called when the inp was detached so that it could kick the firmware to flush any pending work on the flow. This is because the driver doesn't require ACK messages from the firmware for every request, but instead does a kind of manual interrupt coalescing by only setting a flag to request a completion on a subset of requests. If all of the in-flight requests don't have the flag when the tag is detached from the inp, the flow might never return the credits. The current snd_tag_free command issues a flush command to force the credits to return. However, the credit return is what also frees the mbufs, and since those mbufs now hold references on the tag, this meant that snd_tag_free would never be called. To fix, explicitly drop the mbuf's reference on the snd tag when the mbuf is queued in the firmware work queue. This means that once the inp's reference on the tag goes away and all in-flight mbufs have been queued to the firmware, tag's refcount will drop to zero and snd_tag_free will kick in and send the flush request. Note that we need to avoid doing this in the middle of ethofld_tx(), so the driver grabs a temporary reference on the tag around that loop to defer the free to the end of the function in case it sends the last mbuf to the queue after the inp has dropped its reference on the tag. - mlx5 preallocates send tags and was using the ifp pointer even when the send tag wasn't in use. Explicitly use the ifp from other data structures instead. - Sprinkle some assertions in various places to assert that received packets don't have a send tag, and that other places that overwrite rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer. Reviewed by: gallatin, hselasky, rgrimes, ae Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20117
2019-05-24 22:30:40 +00:00
if (BPF_CHECK_DIRECTION(d, m_rcvif(m), bp->bif_ifp))
continue;
counter_u64_add(d->bd_rcount, 1);
#ifdef BPF_JITTER
bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
/* XXX We cannot handle multiple mbufs. */
if (bf != NULL && m->m_next == NULL)
slen = (*(bf->func))(mtod(m, u_char *), pktlen,
pktlen);
else
#endif
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
if (slen != 0) {
BPFD_LOCK(d);
counter_u64_add(d->bd_fcount, 1);
if (gottime < bpf_ts_quality(d->bd_tstamp))
gottime = bpf_gettime(&bt, d->bd_tstamp, m);
#ifdef MAC
if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
#endif
catchpacket(d, (u_char *)m, pktlen, slen,
bpf_append_mbuf, &bt);
BPFD_UNLOCK(d);
}
1994-05-24 10:09:53 +00:00
}
NET_EPOCH_EXIT(et);
1994-05-24 10:09:53 +00:00
}
/*
* Incoming linkage from device drivers, when packet is in
* an mbuf chain and to be prepended by a contiguous header.
*/
void
bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
{
struct epoch_tracker et;
struct bintime bt;
struct mbuf mb;
struct bpf_d *d;
u_int pktlen, slen;
int gottime;
/* Skip outgoing duplicate packets. */
if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
m->m_flags &= ~M_PROMISC;
return;
}
pktlen = m_length(m, NULL);
/*
* Craft on-stack mbuf suitable for passing to bpf_filter.
* Note that we cut corners here; we only setup what's
* absolutely needed--this mbuf should never go anywhere else.
*/
Add an external mbuf buffer type that holds multiple unmapped pages. Unmapped mbufs allow sendfile to carry multiple pages of data in a single mbuf, without mapping those pages. It is a requirement for Netflix's in-kernel TLS, and provides a 5-10% CPU savings on heavy web serving workloads when used by sendfile, due to effectively compressing socket buffers by an order of magnitude, and hence reducing cache misses. For this new external mbuf buffer type (EXT_PGS), the ext_buf pointer now points to a struct mbuf_ext_pgs structure instead of a data buffer. This structure contains an array of physical addresses (this reduces cache misses compared to an earlier version that stored an array of vm_page_t pointers). It also stores additional fields needed for in-kernel TLS such as the TLS header and trailer data that are currently unused. To more easily detect these mbufs, the M_NOMAP flag is set in m_flags in addition to M_EXT. Various functions like m_copydata() have been updated to safely access packet contents (using uiomove_fromphys()), to make things like BPF safe. NIC drivers advertise support for unmapped mbufs on transmit via a new IFCAP_NOMAP capability. This capability can be toggled via the new 'nomap' and '-nomap' ifconfig(8) commands. For NIC drivers that only transmit packet contents via DMA and use bus_dma, adding the capability to if_capabilities and if_capenable should be all that is required. If a NIC does not support unmapped mbufs, they are converted to a chain of mapped mbufs (using sf_bufs to provide the mapping) in ip_output or ip6_output. If an unmapped mbuf requires software checksums, it is also converted to a chain of mapped mbufs before computing the checksum. Submitted by: gallatin (earlier version) Reviewed by: gallatin, hselasky, rrs Discussed with: ae, kp (firewalls) Relnotes: yes Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D20616
2019-06-29 00:48:33 +00:00
mb.m_flags = 0;
mb.m_next = m;
mb.m_data = data;
mb.m_len = dlen;
pktlen += dlen;
gottime = BPF_TSTAMP_NONE;
NET_EPOCH_ENTER(et);
CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
continue;
counter_u64_add(d->bd_rcount, 1);
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
if (slen != 0) {
BPFD_LOCK(d);
counter_u64_add(d->bd_fcount, 1);
if (gottime < bpf_ts_quality(d->bd_tstamp))
gottime = bpf_gettime(&bt, d->bd_tstamp, m);
#ifdef MAC
if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
#endif
catchpacket(d, (u_char *)&mb, pktlen, slen,
bpf_append_mbuf, &bt);
BPFD_UNLOCK(d);
}
}
NET_EPOCH_EXIT(et);
}
#undef BPF_CHECK_DIRECTION
#undef BPF_TSTAMP_NONE
#undef BPF_TSTAMP_FAST
#undef BPF_TSTAMP_NORMAL
#undef BPF_TSTAMP_EXTERN
static int
bpf_hdrlen(struct bpf_d *d)
{
int hdrlen;
hdrlen = d->bd_bif->bif_hdrlen;
#ifndef BURN_BRIDGES
if (d->bd_tstamp == BPF_T_NONE ||
BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
#ifdef COMPAT_FREEBSD32
if (d->bd_compat32)
hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
else
#endif
hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
else
#endif
hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
#ifdef COMPAT_FREEBSD32
if (d->bd_compat32)
hdrlen = BPF_WORDALIGN32(hdrlen);
else
#endif
hdrlen = BPF_WORDALIGN(hdrlen);
return (hdrlen - d->bd_bif->bif_hdrlen);
}
static void
bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
{
struct bintime bt2, boottimebin;
struct timeval tsm;
struct timespec tsn;
if ((tstype & BPF_T_MONOTONIC) == 0) {
bt2 = *bt;
getboottimebin(&boottimebin);
bintime_add(&bt2, &boottimebin);
bt = &bt2;
}
switch (BPF_T_FORMAT(tstype)) {
case BPF_T_MICROTIME:
bintime2timeval(bt, &tsm);
ts->bt_sec = tsm.tv_sec;
ts->bt_frac = tsm.tv_usec;
break;
case BPF_T_NANOTIME:
bintime2timespec(bt, &tsn);
ts->bt_sec = tsn.tv_sec;
ts->bt_frac = tsn.tv_nsec;
break;
case BPF_T_BINTIME:
ts->bt_sec = bt->sec;
ts->bt_frac = bt->frac;
break;
}
}
1994-05-24 10:09:53 +00:00
/*
* Move the packet data from interface memory (pkt) into the
* store buffer. "cpfn" is the routine called to do the actual data
1994-05-24 10:09:53 +00:00
* transfer. bcopy is passed in to copy contiguous chunks, while
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
* bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
1994-05-24 10:09:53 +00:00
* pkt is really an mbuf.
*/
static void
catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
struct bintime *bt)
1994-05-24 10:09:53 +00:00
{
struct bpf_xhdr hdr;
#ifndef BURN_BRIDGES
struct bpf_hdr hdr_old;
#ifdef COMPAT_FREEBSD32
struct bpf_hdr32 hdr32_old;
#endif
#endif
int caplen, curlen, hdrlen, totlen;
int do_wakeup = 0;
int do_timestamp;
int tstype;
BPFD_LOCK_ASSERT(d);
Rework r348303 to reduce the time of holding global BPF lock. It appeared that using NET_EPOCH_WAIT() while holding global BPF lock can lead to another panic: spin lock 0xfffff800183c9840 (turnstile lock) held by 0xfffff80018e2c5a0 (tid 100325) too long panic: spin lock held too long ... #0 sched_switch (td=0xfffff80018e2c5a0, newtd=0xfffff8000389e000, flags=<optimized out>) at /usr/src/sys/kern/sched_ule.c:2133 #1 0xffffffff80bf9912 in mi_switch (flags=256, newtd=0x0) at /usr/src/sys/kern/kern_synch.c:439 #2 0xffffffff80c21db7 in sched_bind (td=<optimized out>, cpu=<optimized out>) at /usr/src/sys/kern/sched_ule.c:2704 #3 0xffffffff80c34c33 in epoch_block_handler_preempt (global=<optimized out>, cr=0xfffffe00005a1a00, arg=<optimized out>) at /usr/src/sys/kern/subr_epoch.c:394 #4 0xffffffff803c741b in epoch_block (global=<optimized out>, cr=<optimized out>, cb=<optimized out>, ct=<optimized out>) at /usr/src/sys/contrib/ck/src/ck_epoch.c:416 #5 ck_epoch_synchronize_wait (global=0xfffff8000380cd80, cb=<optimized out>, ct=<optimized out>) at /usr/src/sys/contrib/ck/src/ck_epoch.c:465 #6 0xffffffff80c3475e in epoch_wait_preempt (epoch=0xfffff8000380cd80) at /usr/src/sys/kern/subr_epoch.c:513 #7 0xffffffff80ce970b in bpf_detachd_locked (d=0xfffff801d309cc00, detached_ifp=<optimized out>) at /usr/src/sys/net/bpf.c:856 #8 0xffffffff80ced166 in bpf_detachd (d=<optimized out>) at /usr/src/sys/net/bpf.c:836 #9 bpf_dtor (data=0xfffff801d309cc00) at /usr/src/sys/net/bpf.c:914 To fix this add the check to the catchpacket() that BPF descriptor was not detached just before we acquired BPFD_LOCK(). Reported by: slavash Tested by: slavash MFC after: 1 week
2019-05-28 11:45:00 +00:00
if (d->bd_bif == NULL) {
/* Descriptor was detached in concurrent thread */
counter_u64_add(d->bd_dcount, 1);
return;
}
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
/*
* Detect whether user space has released a buffer back to us, and if
* so, move it from being a hold buffer to a free buffer. This may
* not be the best place to do it (for example, we might only want to
* run this check if we need the space), but for now it's a reliable
* spot to do it.
*/
if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
d->bd_fbuf = d->bd_hbuf;
d->bd_hbuf = NULL;
d->bd_hlen = 0;
bpf_buf_reclaimed(d);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
}
1994-05-24 10:09:53 +00:00
/*
* Figure out how many bytes to move. If the packet is
* greater or equal to the snapshot length, transfer that
* much. Otherwise, transfer the whole packet (unless
* we hit the buffer size limit).
*/
hdrlen = bpf_hdrlen(d);
1994-05-24 10:09:53 +00:00
totlen = hdrlen + min(snaplen, pktlen);
if (totlen > d->bd_bufsize)
totlen = d->bd_bufsize;
/*
* Round up the end of the previous packet to the next longword.
*
* Drop the packet if there's no room and no hope of room
* If the packet would overflow the storage buffer or the storage
* buffer is considered immutable by the buffer model, try to rotate
* the buffer and wakeup pending processes.
1994-05-24 10:09:53 +00:00
*/
#ifdef COMPAT_FREEBSD32
if (d->bd_compat32)
curlen = BPF_WORDALIGN32(d->bd_slen);
else
#endif
curlen = BPF_WORDALIGN(d->bd_slen);
if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
if (d->bd_fbuf == NULL) {
1994-05-24 10:09:53 +00:00
/*
* There's no room in the store buffer, and no
* prospect of room, so drop the packet. Notify the
* buffer model.
1994-05-24 10:09:53 +00:00
*/
bpf_buffull(d);
counter_u64_add(d->bd_dcount, 1);
1994-05-24 10:09:53 +00:00
return;
}
KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
1994-05-24 10:09:53 +00:00
ROTATE_BUFFERS(d);
do_wakeup = 1;
1994-05-24 10:09:53 +00:00
curlen = 0;
} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
1994-05-24 10:09:53 +00:00
/*
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
* Immediate mode is set, or the read timeout has already
* expired during a select call. A packet arrived, so the
* reader should be woken up.
1994-05-24 10:09:53 +00:00
*/
do_wakeup = 1;
caplen = totlen - hdrlen;
tstype = d->bd_tstamp;
do_timestamp = tstype != BPF_T_NONE;
#ifndef BURN_BRIDGES
if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
struct bpf_ts ts;
if (do_timestamp)
bpf_bintime2ts(bt, &ts, tstype);
#ifdef COMPAT_FREEBSD32
if (d->bd_compat32) {
bzero(&hdr32_old, sizeof(hdr32_old));
if (do_timestamp) {
hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
}
hdr32_old.bh_datalen = pktlen;
hdr32_old.bh_hdrlen = hdrlen;
hdr32_old.bh_caplen = caplen;
bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
sizeof(hdr32_old));
goto copy;
}
#endif
bzero(&hdr_old, sizeof(hdr_old));
if (do_timestamp) {
hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
}
hdr_old.bh_datalen = pktlen;
hdr_old.bh_hdrlen = hdrlen;
hdr_old.bh_caplen = caplen;
bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
sizeof(hdr_old));
goto copy;
}
#endif
1994-05-24 10:09:53 +00:00
/*
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
* Append the bpf header. Note we append the actual header size, but
* move forward the length of the header plus padding.
1994-05-24 10:09:53 +00:00
*/
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
bzero(&hdr, sizeof(hdr));
if (do_timestamp)
bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
hdr.bh_datalen = pktlen;
hdr.bh_hdrlen = hdrlen;
hdr.bh_caplen = caplen;
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
1994-05-24 10:09:53 +00:00
/*
* Copy the packet data into the store buffer and update its length.
*/
#ifndef BURN_BRIDGES
copy:
#endif
(*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
1994-05-24 10:09:53 +00:00
d->bd_slen = curlen + totlen;
if (do_wakeup)
bpf_wakeup(d);
1994-05-24 10:09:53 +00:00
}
/*
* Free buffers currently in use by a descriptor.
* Called on close.
*/
static void
bpfd_free(epoch_context_t ctx)
1994-05-24 10:09:53 +00:00
{
struct bpf_d *d;
struct bpf_program_buffer *p;
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
1994-05-24 10:09:53 +00:00
/*
* We don't need to lock out interrupts since this descriptor has
* been detached from its interface and it yet hasn't been marked
* free.
*/
d = __containerof(ctx, struct bpf_d, epoch_ctx);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
bpf_free(d);
if (d->bd_rfilter != NULL) {
p = __containerof((void *)d->bd_rfilter,
struct bpf_program_buffer, buffer);
#ifdef BPF_JITTER
p->func = d->bd_bfilter;
#endif
bpf_program_buffer_free(&p->epoch_ctx);
}
if (d->bd_wfilter != NULL) {
p = __containerof((void *)d->bd_wfilter,
struct bpf_program_buffer, buffer);
#ifdef BPF_JITTER
p->func = NULL;
#endif
bpf_program_buffer_free(&p->epoch_ctx);
}
mtx_destroy(&d->bd_lock);
counter_u64_free(d->bd_rcount);
counter_u64_free(d->bd_dcount);
counter_u64_free(d->bd_fcount);
counter_u64_free(d->bd_wcount);
counter_u64_free(d->bd_wfcount);
counter_u64_free(d->bd_wdcount);
counter_u64_free(d->bd_zcopy);
free(d, M_BPF);
1994-05-24 10:09:53 +00:00
}
/*
* Attach an interface to bpf. dlt is the link layer type; hdrlen is the
* fixed size of the link header (variable length headers not yet supported).
*/
void
bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
{
bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
}
1994-05-24 10:09:53 +00:00
/*
* Attach an interface to bpf. ifp is a pointer to the structure
* defining the interface to be attached, dlt is the link layer type,
* and hdrlen is the fixed size of the link header (variable length
* headers are not yet supporrted).
1994-05-24 10:09:53 +00:00
*/
void
bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen,
struct bpf_if **driverp)
1994-05-24 10:09:53 +00:00
{
struct bpf_if *bp;
KASSERT(*driverp == NULL,
("bpfattach2: driverp already initialized"));
bp = malloc(sizeof(*bp), M_BPF, M_WAITOK | M_ZERO);
1994-05-24 10:09:53 +00:00
CK_LIST_INIT(&bp->bif_dlist);
CK_LIST_INIT(&bp->bif_wlist);
1994-05-24 10:09:53 +00:00
bp->bif_ifp = ifp;
bp->bif_dlt = dlt;
bp->bif_hdrlen = hdrlen;
bp->bif_bpf = driverp;
bp->bif_refcnt = 1;
Fix the following bpf(4) race condition which can result in a panic: (1) bpf peer attaches to interface netif0 (2) Packet is received by netif0 (3) ifp->if_bpf pointer is checked and handed off to bpf (4) bpf peer detaches from netif0 resulting in ifp->if_bpf being initialized to NULL. (5) ifp->if_bpf is dereferenced by bpf machinery (6) Kaboom This race condition likely explains the various different kernel panics reported around sending SIGINT to tcpdump or dhclient processes. But really this race can result in kernel panics anywhere you have frequent bpf attach and detach operations with high packet per second load. Summary of changes: - Remove the bpf interface's "driverp" member - When we attach bpf interfaces, we now set the ifp->if_bpf member to the bpf interface structure. Once this is done, ifp->if_bpf should never be NULL. [1] - Introduce bpf_peers_present function, an inline operation which will do a lockless read bpf peer list associated with the interface. It should be noted that the bpf code will pickup the bpf_interface lock before adding or removing bpf peers. This should serialize the access to the bpf descriptor list, removing the race. - Expose the bpf_if structure in bpf.h so that the bpf_peers_present function can use it. This also removes the struct bpf_if; hack that was there. - Adjust all consumers of the raw if_bpf structure to use bpf_peers_present Now what happens is: (1) Packet is received by netif0 (2) Check to see if bpf descriptor list is empty (3) Pickup the bpf interface lock (4) Hand packet off to process From the attach/detach side: (1) Pickup the bpf interface lock (2) Add/remove from bpf descriptor list Now that we are storing the bpf interface structure with the ifnet, there is is no need to walk the bpf interface list to locate the correct bpf interface. We now simply look up the interface, and initialize the pointer. This has a nice side effect of changing a bpf interface attach operation from O(N) (where N is the number of bpf interfaces), to O(1). [1] From now on, we can no longer check ifp->if_bpf to tell us whether or not we have any bpf peers that might be interested in receiving packets. In collaboration with: sam@ MFC after: 1 month
2006-06-02 19:59:33 +00:00
*driverp = bp;
/*
* Reference ifnet pointer, so it won't freed until
* we release it.
*/
if_ref(ifp);
BPF_LOCK();
CK_LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
BPF_UNLOCK();
1994-05-24 10:09:53 +00:00
if (bootverbose && IS_DEFAULT_VNET(curvnet))
if_printf(ifp, "bpf attached\n");
1994-05-24 10:09:53 +00:00
}
#ifdef VIMAGE
/*
* When moving interfaces between vnet instances we need a way to
* query the dlt and hdrlen before detach so we can re-attch the if_bpf
* after the vmove. We unfortunately have no device driver infrastructure
* to query the interface for these values after creation/attach, thus
* add this as a workaround.
*/
int
bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
{
if (bp == NULL)
return (ENXIO);
if (bif_dlt == NULL && bif_hdrlen == NULL)
return (0);
if (bif_dlt != NULL)
*bif_dlt = bp->bif_dlt;
if (bif_hdrlen != NULL)
*bif_hdrlen = bp->bif_hdrlen;
return (0);
}
#endif
/*
* Detach bpf from an interface. This involves detaching each descriptor
* associated with the interface. Notify each descriptor as it's detached
* so that any sleepers wake up and get ENXIO.
*/
void
bpfdetach(struct ifnet *ifp)
{
struct bpf_if *bp, *bp_temp;
struct bpf_d *d;
BPF_LOCK();
/* Find all bpf_if struct's which reference ifp and detach them. */
CK_LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
if (ifp != bp->bif_ifp)
continue;
CK_LIST_REMOVE(bp, bif_next);
*bp->bif_bpf = (struct bpf_if *)&dead_bpf_if;
CTR4(KTR_NET,
"%s: sheduling free for encap %d (%p) for if %p",
__func__, bp->bif_dlt, bp, ifp);
/* Detach common descriptors */
while ((d = CK_LIST_FIRST(&bp->bif_dlist)) != NULL) {
bpf_detachd_locked(d, true);
}
/* Detach writer-only descriptors */
while ((d = CK_LIST_FIRST(&bp->bif_wlist)) != NULL) {
bpf_detachd_locked(d, true);
}
bpfif_rele(bp);
}
BPF_UNLOCK();
}
/*
* Get a list of available data link type of the interface.
*/
static int
bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
{
struct ifnet *ifp;
struct bpf_if *bp;
u_int *lst;
int error, n, n1;
BPF_LOCK_ASSERT();
ifp = d->bd_bif->bif_ifp;
n1 = 0;
CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
if (bp->bif_ifp == ifp)
n1++;
}
if (bfl->bfl_list == NULL) {
bfl->bfl_len = n1;
return (0);
}
if (n1 > bfl->bfl_len)
return (ENOMEM);
lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
n = 0;
CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
if (bp->bif_ifp != ifp)
continue;
lst[n++] = bp->bif_dlt;
}
error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
free(lst, M_TEMP);
bfl->bfl_len = n;
return (error);
}
/*
* Set the data link type of a BPF instance.
*/
static int
bpf_setdlt(struct bpf_d *d, u_int dlt)
{
int error, opromisc;
struct ifnet *ifp;
struct bpf_if *bp;
BPF_LOCK_ASSERT();
MPASS(d->bd_bif != NULL);
/*
* It is safe to check bd_bif without BPFD_LOCK, it can not be
* changed while we hold global lock.
*/
if (d->bd_bif->bif_dlt == dlt)
return (0);
ifp = d->bd_bif->bif_ifp;
CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
break;
}
if (bp == NULL)
return (EINVAL);
opromisc = d->bd_promisc;
bpf_attachd(d, bp);
if (opromisc) {
error = ifpromisc(bp->bif_ifp, 1);
if (error)
if_printf(bp->bif_ifp, "%s: ifpromisc failed (%d)\n",
__func__, error);
else
d->bd_promisc = 1;
}
return (0);
}
1997-09-16 11:44:05 +00:00
static void
bpf_drvinit(void *unused)
{
struct cdev *dev;
sx_init(&bpf_sx, "bpf global lock");
CK_LIST_INIT(&bpf_iflist);
dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
/* For compatibility */
make_dev_alias(dev, "bpf0");
}
/*
* Zero out the various packet counters associated with all of the bpf
* descriptors. At some point, we will probably want to get a bit more
* granular and allow the user to specify descriptors to be zeroed.
*/
static void
bpf_zero_counters(void)
{
struct bpf_if *bp;
struct bpf_d *bd;
BPF_LOCK();
/*
* We are protected by global lock here, interfaces and
* descriptors can not be deleted while we hold it.
*/
CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
counter_u64_zero(bd->bd_rcount);
counter_u64_zero(bd->bd_dcount);
counter_u64_zero(bd->bd_fcount);
counter_u64_zero(bd->bd_wcount);
counter_u64_zero(bd->bd_wfcount);
counter_u64_zero(bd->bd_zcopy);
}
}
BPF_UNLOCK();
}
/*
* Fill filter statistics
*/
static void
bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
{
BPF_LOCK_ASSERT();
bzero(d, sizeof(*d));
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
d->bd_structsize = sizeof(*d);
d->bd_immediate = bd->bd_immediate;
d->bd_promisc = bd->bd_promisc;
d->bd_hdrcmplt = bd->bd_hdrcmplt;
d->bd_direction = bd->bd_direction;
d->bd_feedback = bd->bd_feedback;
d->bd_async = bd->bd_async;
d->bd_rcount = counter_u64_fetch(bd->bd_rcount);
d->bd_dcount = counter_u64_fetch(bd->bd_dcount);
d->bd_fcount = counter_u64_fetch(bd->bd_fcount);
d->bd_sig = bd->bd_sig;
d->bd_slen = bd->bd_slen;
d->bd_hlen = bd->bd_hlen;
d->bd_bufsize = bd->bd_bufsize;
d->bd_pid = bd->bd_pid;
strlcpy(d->bd_ifname,
bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
Introduce two new ioctl(2) commands, BIOCLOCK and BIOCSETWF. These commands enhance the security of bpf(4) by further relinquishing the privilege of the bpf(4) consumer (assuming the ioctl commands are being implemented). Once BIOCLOCK is executed, the device becomes locked which prevents the execution of ioctl(2) commands which can change the underly parameters of the bpf(4) device. An example might be the setting of bpf(4) filter programs or attaching to different network interfaces. BIOCSETWF can be used to set write filters for outgoing packets. Currently if a bpf(4) consumer is compromised, the bpf(4) descriptor can essentially be used as a raw socket, regardless of consumer's UID. Write filters give users the ability to constrain which packets can be sent through the bpf(4) descriptor. These features are currently implemented by a couple programs which came from OpenBSD, such as the new dhclient and pflogd. -Modify bpf_setf(9) to accept a "cmd" parameter. This will be used to specify whether a read or write filter is to be set. -Add a bpf(4) filter program as a parameter to bpf_movein(9) as we will run the filter program on the mbuf data once we move the packet in from user-space. -Rather than execute two uiomove operations, (one for the link header and the other for the packet data), execute one and manually copy the linker header into the sockaddr structure via bcopy. -Restructure bpf_setf to compensate for write filters, as well as read. -Adjust bpf(4) stats structures to include a bd_locked member. It should be noted that the FreeBSD and OpenBSD implementations differ a bit in the sense that we unconditionally enforce the lock, where OpenBSD enforces it only if the calling credential is not root. Idea from: OpenBSD Reviewed by: mlaier
2005-08-22 19:35:48 +00:00
d->bd_locked = bd->bd_locked;
d->bd_wcount = counter_u64_fetch(bd->bd_wcount);
d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount);
d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount);
d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy);
Introduce support for zero-copy BPF buffering, which reduces the overhead of packet capture by allowing a user process to directly "loan" buffer memory to the kernel rather than using read(2) to explicitly copy data from kernel address space. The user process will issue new BPF ioctls to set the shared memory buffer mode and provide pointers to buffers and their size. The kernel then wires and maps the pages into kernel address space using sf_buf(9), which on supporting architectures will use the direct map region. The current "buffered" access mode remains the default, and support for zero-copy buffers must, for the time being, be explicitly enabled using a sysctl for the kernel to accept requests to use it. The kernel and user process synchronize use of the buffers with atomic operations, avoiding the need for system calls under load; the user process may use select()/poll()/kqueue() to manage blocking while waiting for network data if the user process is able to consume data faster than the kernel generates it. Patchs to libpcap are available to allow libpcap applications to transparently take advantage of this support. Detailed information on the new API may be found in bpf(4), including specific atomic operations and memory barriers required to synchronize buffer use safely. These changes modify the base BPF implementation to (roughly) abstrac the current buffer model, allowing the new shared memory model to be added, and add new monitoring statistics for netstat to print. The implementation, with the exception of some monitoring hanges that break the netstat monitoring ABI for BPF, will be MFC'd. Zerocopy bpf buffers are still considered experimental are disabled by default. To experiment with this new facility, adjust the net.bpf.zerocopy_enable sysctl variable to 1. Changes to libpcap will be made available as a patch for the time being, and further refinements to the implementation are expected. Sponsored by: Seccuris Inc. In collaboration with: rwatson Tested by: pwood, gallatin MFC after: 4 months [1] [1] Certain portions will probably not be MFCed, specifically things that can break the monitoring ABI.
2008-03-24 13:49:17 +00:00
d->bd_bufmode = bd->bd_bufmode;
}
/*
* Handle `netstat -B' stats request
*/
static int
bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
{
static const struct xbpf_d zerostats;
struct xbpf_d *xbdbuf, *xbd, tempstats;
int index, error;
struct bpf_if *bp;
struct bpf_d *bd;
/*
* XXX This is not technically correct. It is possible for non
* privileged users to open bpf devices. It would make sense
* if the users who opened the devices were able to retrieve
* the statistics for them, too.
*/
error = priv_check(req->td, PRIV_NET_BPF);
if (error)
return (error);
/*
* Check to see if the user is requesting that the counters be
* zeroed out. Explicitly check that the supplied data is zeroed,
* as we aren't allowing the user to set the counters currently.
*/
if (req->newptr != NULL) {
if (req->newlen != sizeof(tempstats))
return (EINVAL);
memset(&tempstats, 0, sizeof(tempstats));
error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
if (error)
return (error);
if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
return (EINVAL);
bpf_zero_counters();
return (0);
}
if (req->oldptr == NULL)
return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
if (bpf_bpfd_cnt == 0)
return (SYSCTL_OUT(req, 0, 0));
xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
BPF_LOCK();
if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
BPF_UNLOCK();
free(xbdbuf, M_BPF);
return (ENOMEM);
}
index = 0;
CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
/* Send writers-only first */
CK_LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
- Improve performace for writer-only BPF users. Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send raw ethernet frames. The only FreeBSD interface that can be used to send raw frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff uses BPF only to send data. This leads us to the situation when software like cdpd, being run on high-traffic-volume interface significantly reduces overall performance since we have to acquire additional locks for every packet. Here we add sysctl that changes BPF behavior in the following way: If program came and opens BPF socket without explicitly specifyin read filter we assume it to be write-only and add it to special writer-only per-interface list. This makes bpf_peers_present() return 0, so no additional overhead is introduced. After filter is supplied, descriptor is added to original per-interface list permitting packets to be captured. Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose of setting snap length. Fortunately, most programs explicitly sets (event catch-all) filter after that. tcpdump(1) is a good example. So a bit hackis approach is taken: we upgrade description only after second BIOCSETF is received. Sysctl is named net.bpf.optimize_writers and is turned off by default. - While here, document all sysctl variables in bpf.4 Sponsored by Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 4 weeks
2012-04-06 06:55:21 +00:00
xbd = &xbdbuf[index++];
bpfstats_fill_xbpf(xbd, bd);
}
CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
xbd = &xbdbuf[index++];
bpfstats_fill_xbpf(xbd, bd);
}
}
BPF_UNLOCK();
error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
free(xbdbuf, M_BPF);
return (error);
}
SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
#else /* !DEV_BPF && !NETGRAPH_BPF */
/*
* NOP stubs to allow bpf-using drivers to load and function.
*
* A 'better' implementation would allow the core bpf functionality
* to be loaded at runtime.
*/
void
bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
{
}
void
bpf_mtap(struct bpf_if *bp, struct mbuf *m)
{
}
void
bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
{
}
void
bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
{
bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
}
void
bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
{
*driverp = (struct bpf_if *)&dead_bpf_if;
}
void
bpfdetach(struct ifnet *ifp)
{
}
u_int
bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
{
return -1; /* "no filter" behaviour */
}
int
bpf_validate(const struct bpf_insn *f, int len)
{
return 0; /* false */
}
#endif /* !DEV_BPF && !NETGRAPH_BPF */
#ifdef DDB
static void
bpf_show_bpf_if(struct bpf_if *bpf_if)
{
if (bpf_if == NULL)
return;
db_printf("%p:\n", bpf_if);
#define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e);
/* bif_ext.bif_next */
/* bif_ext.bif_dlist */
BPF_DB_PRINTF("%#x", bif_dlt);
BPF_DB_PRINTF("%u", bif_hdrlen);
/* bif_wlist */
BPF_DB_PRINTF("%p", bif_ifp);
BPF_DB_PRINTF("%p", bif_bpf);
BPF_DB_PRINTF("%u", bif_refcnt);
}
DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
{
if (!have_addr) {
db_printf("usage: show bpf_if <struct bpf_if *>\n");
return;
}
bpf_show_bpf_if((struct bpf_if *)addr);
}
#endif