freebsd-skq/sys/kern/uipc_socket.c

2105 lines
53 KiB
C
Raw Normal View History

1994-05-24 10:09:53 +00:00
/*
* Copyright (c) 1982, 1986, 1988, 1990, 1993
* The Regents of the University of California. All rights reserved.
*
* 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.
* 4. Neither the name of the University nor the names of its contributors
* 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.
*
* @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
*/
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_inet.h"
#include "opt_mac.h"
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
#include "opt_zero.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/fcntl.h>
#include <sys/limits.h>
#include <sys/lock.h>
#include <sys/mac.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/domain.h>
#include <sys/file.h> /* for struct knote */
1994-05-24 10:09:53 +00:00
#include <sys/kernel.h>
#include <sys/event.h>
#include <sys/poll.h>
#include <sys/proc.h>
1994-05-24 10:09:53 +00:00
#include <sys/protosw.h>
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/resourcevar.h>
#include <sys/signalvar.h>
#include <sys/sysctl.h>
#include <sys/uio.h>
#include <sys/jail.h>
#include <vm/uma.h>
static int soreceive_rcvoob(struct socket *so, struct uio *uio,
int flags);
#ifdef INET
static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
#endif
static void filt_sordetach(struct knote *kn);
static int filt_soread(struct knote *kn, long hint);
static void filt_sowdetach(struct knote *kn);
static int filt_sowrite(struct knote *kn, long hint);
static int filt_solisten(struct knote *kn, long hint);
static struct filterops solisten_filtops =
{ 1, NULL, filt_sordetach, filt_solisten };
static struct filterops soread_filtops =
{ 1, NULL, filt_sordetach, filt_soread };
static struct filterops sowrite_filtops =
{ 1, NULL, filt_sowdetach, filt_sowrite };
uma_zone_t socket_zone;
so_gen_t so_gencnt; /* generation count for sockets */
MALLOC_DEFINE(M_SONAME, "soname", "socket name");
MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
SYSCTL_DECL(_kern_ipc);
static int somaxconn = SOMAXCONN;
SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
&somaxconn, 0, "Maximum pending socket connection queue size");
static int numopensockets;
SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
&numopensockets, 0, "Number of open sockets");
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
#ifdef ZERO_COPY_SOCKETS
/* These aren't static because they're used in other files. */
int so_zero_copy_send = 1;
int so_zero_copy_receive = 1;
SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
"Zero copy controls");
SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
&so_zero_copy_receive, 0, "Enable zero copy receive");
SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
&so_zero_copy_send, 0, "Enable zero copy send");
#endif /* ZERO_COPY_SOCKETS */
/*
* accept_mtx locks down per-socket fields relating to accept queues. See
* socketvar.h for an annotation of the protected fields of struct socket.
*/
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
struct mtx accept_mtx;
MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
/*
* so_global_mtx protects so_gencnt, numopensockets, and the per-socket
* so_gencnt field.
*
* XXXRW: These variables might be better manipulated using atomic operations
* for improved efficiency.
*/
static struct mtx so_global_mtx;
MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
1994-05-24 10:09:53 +00:00
/*
* Socket operation routines.
* These routines are called by the routines in
* sys_socket.c or from a system process, and
* implement the semantics of socket operations by
* switching out to the protocol specific routines.
*/
/*
* Get a socket structure from our zone, and initialize it.
* Note that it would probably be better to allocate socket
* and PCB at the same time, but I'm not convinced that all
* the protocols can be easily modified to do this.
*
* soalloc() returns a socket with a ref count of 0.
*/
struct socket *
soalloc(int mflags)
{
struct socket *so;
#ifdef MAC
int error;
#endif
so = uma_zalloc(socket_zone, mflags | M_ZERO);
if (so != NULL) {
#ifdef MAC
error = mac_init_socket(so, mflags);
if (error != 0) {
uma_zfree(socket_zone, so);
so = NULL;
return so;
}
#endif
SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
/* sx_init(&so->so_sxlock, "socket sxlock"); */
TAILQ_INIT(&so->so_aiojobq);
mtx_lock(&so_global_mtx);
so->so_gencnt = ++so_gencnt;
++numopensockets;
mtx_unlock(&so_global_mtx);
}
return so;
}
/*
* socreate returns a socket with a ref count of 1. The socket should be
* closed with soclose().
*/
int
socreate(dom, aso, type, proto, cred, td)
1994-05-24 10:09:53 +00:00
int dom;
struct socket **aso;
int type;
1994-05-24 10:09:53 +00:00
int proto;
struct ucred *cred;
struct thread *td;
1994-05-24 10:09:53 +00:00
{
struct protosw *prp;
struct socket *so;
int error;
1994-05-24 10:09:53 +00:00
if (proto)
prp = pffindproto(dom, proto, type);
else
prp = pffindtype(dom, type);
if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL)
return (EPROTONOSUPPORT);
if (jailed(cred) && jail_socket_unixiproute_only &&
prp->pr_domain->dom_family != PF_LOCAL &&
prp->pr_domain->dom_family != PF_INET &&
prp->pr_domain->dom_family != PF_ROUTE) {
return (EPROTONOSUPPORT);
}
1994-05-24 10:09:53 +00:00
if (prp->pr_type != type)
return (EPROTOTYPE);
so = soalloc(M_WAITOK);
if (so == NULL)
return (ENOBUFS);
TAILQ_INIT(&so->so_incomp);
TAILQ_INIT(&so->so_comp);
1994-05-24 10:09:53 +00:00
so->so_type = type;
so->so_cred = crhold(cred);
1994-05-24 10:09:53 +00:00
so->so_proto = prp;
#ifdef MAC
mac_create_socket(cred, so);
#endif
SOCK_LOCK(so);
soref(so);
SOCK_UNLOCK(so);
error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
1994-05-24 10:09:53 +00:00
if (error) {
SOCK_LOCK(so);
1994-05-24 10:09:53 +00:00
so->so_state |= SS_NOFDREF;
sorele(so);
1994-05-24 10:09:53 +00:00
return (error);
}
*aso = so;
return (0);
}
int
sobind(so, nam, td)
1994-05-24 10:09:53 +00:00
struct socket *so;
struct sockaddr *nam;
struct thread *td;
1994-05-24 10:09:53 +00:00
{
return ((*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td));
1994-05-24 10:09:53 +00:00
}
void
sodealloc(struct socket *so)
{
KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
mtx_lock(&so_global_mtx);
so->so_gencnt = ++so_gencnt;
mtx_unlock(&so_global_mtx);
if (so->so_rcv.sb_hiwat)
(void)chgsbsize(so->so_cred->cr_uidinfo,
&so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
if (so->so_snd.sb_hiwat)
(void)chgsbsize(so->so_cred->cr_uidinfo,
&so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
#ifdef INET
/* remove acccept filter if one is present. */
if (so->so_accf != NULL)
do_setopt_accept_filter(so, NULL);
#endif
#ifdef MAC
mac_destroy_socket(so);
#endif
crfree(so->so_cred);
SOCKBUF_LOCK_DESTROY(&so->so_snd);
SOCKBUF_LOCK_DESTROY(&so->so_rcv);
/* sx_destroy(&so->so_sxlock); */
uma_zfree(socket_zone, so);
/*
* XXXRW: Seems like a shame to grab the mutex again down here, but
* we don't want to decrement the socket count until after we free
* the socket, and we can't increment the gencnt on the socket after
* we free, it so...
*/
mtx_lock(&so_global_mtx);
--numopensockets;
mtx_unlock(&so_global_mtx);
}
int
solisten(so, backlog, td)
struct socket *so;
1994-05-24 10:09:53 +00:00
int backlog;
struct thread *td;
1994-05-24 10:09:53 +00:00
{
int error;
1994-05-24 10:09:53 +00:00
/*
* XXXRW: Ordering issue here -- perhaps we need to set
* SO_ACCEPTCONN before the call to pru_listen()?
* XXXRW: General atomic test-and-set concerns here also.
*/
if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
SS_ISDISCONNECTING))
return (EINVAL);
error = (*so->so_proto->pr_usrreqs->pru_listen)(so, td);
if (error)
1994-05-24 10:09:53 +00:00
return (error);
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
ACCEPT_LOCK();
if (TAILQ_EMPTY(&so->so_comp)) {
SOCK_LOCK(so);
1994-05-24 10:09:53 +00:00
so->so_options |= SO_ACCEPTCONN;
SOCK_UNLOCK(so);
}
if (backlog < 0 || backlog > somaxconn)
backlog = somaxconn;
so->so_qlimit = backlog;
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
ACCEPT_UNLOCK();
1994-05-24 10:09:53 +00:00
return (0);
}
void
1994-05-24 10:09:53 +00:00
sofree(so)
struct socket *so;
1994-05-24 10:09:53 +00:00
{
struct socket *head;
1994-05-24 10:09:53 +00:00
KASSERT(so->so_count == 0, ("socket %p so_count not 0", so));
SOCK_LOCK_ASSERT(so);
if (so->so_pcb != NULL || (so->so_state & SS_NOFDREF) == 0) {
SOCK_UNLOCK(so);
1994-05-24 10:09:53 +00:00
return;
}
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
SOCK_UNLOCK(so);
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
ACCEPT_LOCK();
head = so->so_head;
if (head != NULL) {
KASSERT((so->so_qstate & SQ_COMP) != 0 ||
(so->so_qstate & SQ_INCOMP) != 0,
("sofree: so_head != NULL, but neither SQ_COMP nor "
"SQ_INCOMP"));
KASSERT((so->so_qstate & SQ_COMP) == 0 ||
(so->so_qstate & SQ_INCOMP) == 0,
("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
/*
* accept(2) is responsible draining the completed
* connection queue and freeing those sockets, so
* we just return here if this socket is currently
* on the completed connection queue. Otherwise,
* accept(2) may hang after select(2) has indicating
* that a listening socket was ready. If it's an
* incomplete connection, we remove it from the queue
* and free it; otherwise, it won't be released until
* the listening socket is closed.
*/
if ((so->so_qstate & SQ_COMP) != 0) {
ACCEPT_UNLOCK();
return;
}
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
TAILQ_REMOVE(&head->so_incomp, so, so_list);
head->so_incqlen--;
so->so_qstate &= ~SQ_INCOMP;
so->so_head = NULL;
}
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
KASSERT((so->so_qstate & SQ_COMP) == 0 &&
(so->so_qstate & SQ_INCOMP) == 0,
("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
ACCEPT_UNLOCK();
SOCKBUF_LOCK(&so->so_snd);
so->so_snd.sb_flags |= SB_NOINTR;
(void)sblock(&so->so_snd, M_WAITOK);
Merge next step in socket buffer locking: - sowakeup() now asserts the socket buffer lock on entry. Move the call to KNOTE higher in sowakeup() so that it is made with the socket buffer lock held for consistency with other calls. Release the socket buffer lock prior to calling into pgsigio(), so_upcall(), or aio_swake(). Locking for this event management will need revisiting in the future, but this model avoids lock order reversals when upcalls into other subsystems result in socket/socket buffer operations. Assert that the socket buffer lock is not held at the end of the function. - Wrapper macros for sowakeup(), sorwakeup() and sowwakeup(), now have _locked versions which assert the socket buffer lock on entry. If a wakeup is required by sb_notify(), invoke sowakeup(); otherwise, unconditionally release the socket buffer lock. This results in the socket buffer lock being released whether a wakeup is required or not. - Break out socantsendmore() into socantsendmore_locked() that asserts the socket buffer lock. socantsendmore() unconditionally locks the socket buffer before calling socantsendmore_locked(). Note that both functions return with the socket buffer unlocked as socantsendmore_locked() calls sowwakeup_locked() which has the same properties. Assert that the socket buffer is unlocked on return. - Break out socantrcvmore() into socantrcvmore_locked() that asserts the socket buffer lock. socantrcvmore() unconditionally locks the socket buffer before calling socantrcvmore_locked(). Note that both functions return with the socket buffer unlocked as socantrcvmore_locked() calls sorwakeup_locked() which has similar properties. Assert that the socket buffer is unlocked on return. - Break out sbrelease() into a sbrelease_locked() that asserts the socket buffer lock. sbrelease() unconditionally locks the socket buffer before calling sbrelease_locked(). sbrelease_locked() now invokes sbflush_locked() instead of sbflush(). - Assert the socket buffer lock in socket buffer sanity check functions sblastrecordchk(), sblastmbufchk(). - Assert the socket buffer lock in SBLINKRECORD(). - Break out various sbappend() functions into sbappend_locked() (and variations on that name) that assert the socket buffer lock. The !_locked() variations unconditionally lock the socket buffer before calling their _locked counterparts. Internally, make sure to call _locked() support routines, etc, if already holding the socket buffer lock. - Break out sbinsertoob() into sbinsertoob_locked() that asserts the socket buffer lock. sbinsertoob() unconditionally locks the socket buffer before calling sbinsertoob_locked(). - Break out sbflush() into sbflush_locked() that asserts the socket buffer lock. sbflush() unconditionally locks the socket buffer before calling sbflush_locked(). Update panic strings for new function names. - Break out sbdrop() into sbdrop_locked() that asserts the socket buffer lock. sbdrop() unconditionally locks the socket buffer before calling sbdrop_locked(). - Break out sbdroprecord() into sbdroprecord_locked() that asserts the socket buffer lock. sbdroprecord() unconditionally locks the socket buffer before calling sbdroprecord_locked(). - sofree() now calls socantsendmore_locked() and re-acquires the socket buffer lock on return. It also now calls sbrelease_locked(). - sorflush() now calls socantrcvmore_locked() and re-acquires the socket buffer lock on return. Clean up/mess up other behavior in sorflush() relating to the temporary stack copy of the socket buffer used with dom_dispose by more properly initializing the temporary copy, and selectively bzeroing/copying more carefully to prevent WITNESS from getting confused by improperly initialized mutexes. Annotate why that's necessary, or at least, needed. - soisconnected() now calls sbdrop_locked() before unlocking the socket buffer to avoid locking overhead. Some parts of this change were: Submitted by: sam Sponsored by: FreeBSD Foundation Obtained from: BSD/OS
2004-06-21 00:20:43 +00:00
/*
* socantsendmore_locked() drops the socket buffer mutex so that it
* can safely perform wakeups. Re-acquire the mutex before
* continuing.
*/
socantsendmore_locked(so);
SOCKBUF_LOCK(&so->so_snd);
sbunlock(&so->so_snd);
Merge next step in socket buffer locking: - sowakeup() now asserts the socket buffer lock on entry. Move the call to KNOTE higher in sowakeup() so that it is made with the socket buffer lock held for consistency with other calls. Release the socket buffer lock prior to calling into pgsigio(), so_upcall(), or aio_swake(). Locking for this event management will need revisiting in the future, but this model avoids lock order reversals when upcalls into other subsystems result in socket/socket buffer operations. Assert that the socket buffer lock is not held at the end of the function. - Wrapper macros for sowakeup(), sorwakeup() and sowwakeup(), now have _locked versions which assert the socket buffer lock on entry. If a wakeup is required by sb_notify(), invoke sowakeup(); otherwise, unconditionally release the socket buffer lock. This results in the socket buffer lock being released whether a wakeup is required or not. - Break out socantsendmore() into socantsendmore_locked() that asserts the socket buffer lock. socantsendmore() unconditionally locks the socket buffer before calling socantsendmore_locked(). Note that both functions return with the socket buffer unlocked as socantsendmore_locked() calls sowwakeup_locked() which has the same properties. Assert that the socket buffer is unlocked on return. - Break out socantrcvmore() into socantrcvmore_locked() that asserts the socket buffer lock. socantrcvmore() unconditionally locks the socket buffer before calling socantrcvmore_locked(). Note that both functions return with the socket buffer unlocked as socantrcvmore_locked() calls sorwakeup_locked() which has similar properties. Assert that the socket buffer is unlocked on return. - Break out sbrelease() into a sbrelease_locked() that asserts the socket buffer lock. sbrelease() unconditionally locks the socket buffer before calling sbrelease_locked(). sbrelease_locked() now invokes sbflush_locked() instead of sbflush(). - Assert the socket buffer lock in socket buffer sanity check functions sblastrecordchk(), sblastmbufchk(). - Assert the socket buffer lock in SBLINKRECORD(). - Break out various sbappend() functions into sbappend_locked() (and variations on that name) that assert the socket buffer lock. The !_locked() variations unconditionally lock the socket buffer before calling their _locked counterparts. Internally, make sure to call _locked() support routines, etc, if already holding the socket buffer lock. - Break out sbinsertoob() into sbinsertoob_locked() that asserts the socket buffer lock. sbinsertoob() unconditionally locks the socket buffer before calling sbinsertoob_locked(). - Break out sbflush() into sbflush_locked() that asserts the socket buffer lock. sbflush() unconditionally locks the socket buffer before calling sbflush_locked(). Update panic strings for new function names. - Break out sbdrop() into sbdrop_locked() that asserts the socket buffer lock. sbdrop() unconditionally locks the socket buffer before calling sbdrop_locked(). - Break out sbdroprecord() into sbdroprecord_locked() that asserts the socket buffer lock. sbdroprecord() unconditionally locks the socket buffer before calling sbdroprecord_locked(). - sofree() now calls socantsendmore_locked() and re-acquires the socket buffer lock on return. It also now calls sbrelease_locked(). - sorflush() now calls socantrcvmore_locked() and re-acquires the socket buffer lock on return. Clean up/mess up other behavior in sorflush() relating to the temporary stack copy of the socket buffer used with dom_dispose by more properly initializing the temporary copy, and selectively bzeroing/copying more carefully to prevent WITNESS from getting confused by improperly initialized mutexes. Annotate why that's necessary, or at least, needed. - soisconnected() now calls sbdrop_locked() before unlocking the socket buffer to avoid locking overhead. Some parts of this change were: Submitted by: sam Sponsored by: FreeBSD Foundation Obtained from: BSD/OS
2004-06-21 00:20:43 +00:00
sbrelease_locked(&so->so_snd, so);
SOCKBUF_UNLOCK(&so->so_snd);
1994-05-24 10:09:53 +00:00
sorflush(so);
sodealloc(so);
1994-05-24 10:09:53 +00:00
}
/*
* Close a socket on last file table reference removal.
* Initiate disconnect if connected.
* Free socket when disconnect complete.
*
* This function will sorele() the socket. Note that soclose() may be
* called prior to the ref count reaching zero. The actual socket
* structure will not be freed until the ref count reaches zero.
1994-05-24 10:09:53 +00:00
*/
int
1994-05-24 10:09:53 +00:00
soclose(so)
struct socket *so;
1994-05-24 10:09:53 +00:00
{
int error = 0;
funsetown(&so->so_sigio);
1994-05-24 10:09:53 +00:00
if (so->so_options & SO_ACCEPTCONN) {
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
struct socket *sp;
ACCEPT_LOCK();
while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
TAILQ_REMOVE(&so->so_incomp, sp, so_list);
so->so_incqlen--;
sp->so_qstate &= ~SQ_INCOMP;
sp->so_head = NULL;
ACCEPT_UNLOCK();
(void) soabort(sp);
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
ACCEPT_LOCK();
}
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
TAILQ_REMOVE(&so->so_comp, sp, so_list);
so->so_qlen--;
sp->so_qstate &= ~SQ_COMP;
sp->so_head = NULL;
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
ACCEPT_UNLOCK();
(void) soabort(sp);
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
ACCEPT_LOCK();
}
Integrate accept locking from rwatson_netperf, introducing a new global mutex, accept_mtx, which serializes access to the following fields across all sockets: so_qlen so_incqlen so_qstate so_comp so_incomp so_list so_head While providing only coarse granularity, this approach avoids lock order issues between sockets by avoiding ownership of the fields by a specific socket and its per-socket mutexes. While here, rewrite soclose(), sofree(), soaccept(), and sonewconn() to add assertions, close additional races and address lock order concerns. In particular: - Reorganize the optimistic concurrency behavior in accept1() to always allocate a file descriptor with falloc() so that if we do find a socket, we don't have to encounter the "Oh, there wasn't a socket" race that can occur if falloc() sleeps in the current code, which broke inbound accept() ordering, not to mention requiring backing out socket state changes in a way that raced with the protocol level. We may want to add a lockless read of the queue state if polling of empty queues proves to be important to optimize. - In accept1(), soref() the socket while holding the accept lock so that the socket cannot be free'd in a race with the protocol layer. Likewise in netgraph equivilents of the accept1() code. - In sonewconn(), loop waiting for the queue to be small enough to insert our new socket once we've committed to inserting it, or races can occur that cause the incomplete socket queue to overfill. In the previously implementation, it was sufficient to simply tested once since calling soabort() didn't release synchronization permitting another thread to insert a socket as we discard a previous one. - In soclose()/sofree()/et al, it is the responsibility of the caller to remove a socket from the incomplete connection queue before calling soabort(), which prevents soabort() from having to walk into the accept socket to release the socket from its queue, and avoids races when releasing the accept mutex to enter soabort(), permitting soabort() to avoid lock ordering issues with the caller. - Generally cluster accept queue related operations together throughout these functions in order to facilitate locking. Annotate new locking in socketvar.h.
2004-06-02 04:15:39 +00:00
ACCEPT_UNLOCK();
1994-05-24 10:09:53 +00:00
}
if (so->so_pcb == NULL)
1994-05-24 10:09:53 +00:00
goto discard;
if (so->so_state & SS_ISCONNECTED) {
if ((so->so_state & SS_ISDISCONNECTING) == 0) {
error = sodisconnect(so);
if (error)
1994-05-24 10:09:53 +00:00
goto drop;
}
if (so->so_options & SO_LINGER) {
if ((so->so_state & SS_ISDISCONNECTING) &&
(so->so_state & SS_NBIO))
1994-05-24 10:09:53 +00:00
goto drop;
while (so->so_state & SS_ISCONNECTED) {
error = tsleep(&so->so_timeo,
PSOCK | PCATCH, "soclos", so->so_linger * hz);
if (error)
1994-05-24 10:09:53 +00:00
break;
}
1994-05-24 10:09:53 +00:00
}
}
drop:
if (so->so_pcb != NULL) {
int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so);
1994-05-24 10:09:53 +00:00
if (error == 0)
error = error2;
}
discard:
SOCK_LOCK(so);
KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1994-05-24 10:09:53 +00:00
so->so_state |= SS_NOFDREF;
sorele(so);
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* soabort() must not be called with any socket locks held, as it calls
* into the protocol, which will call back into the socket code causing
* it to acquire additional socket locks that may cause recursion or lock
* order reversals.
1994-05-24 10:09:53 +00:00
*/
int
1994-05-24 10:09:53 +00:00
soabort(so)
struct socket *so;
{
int error;
1994-05-24 10:09:53 +00:00
error = (*so->so_proto->pr_usrreqs->pru_abort)(so);
if (error) {
SOCK_LOCK(so);
sotryfree(so); /* note: does not decrement the ref count */
return error;
}
return (0);
1994-05-24 10:09:53 +00:00
}
int
1994-05-24 10:09:53 +00:00
soaccept(so, nam)
struct socket *so;
struct sockaddr **nam;
1994-05-24 10:09:53 +00:00
{
int error;
SOCK_LOCK(so);
KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1994-05-24 10:09:53 +00:00
so->so_state &= ~SS_NOFDREF;
SOCK_UNLOCK(so);
error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1994-05-24 10:09:53 +00:00
return (error);
}
int
soconnect(so, nam, td)
struct socket *so;
struct sockaddr *nam;
struct thread *td;
1994-05-24 10:09:53 +00:00
{
int error;
if (so->so_options & SO_ACCEPTCONN)
1994-05-24 10:09:53 +00:00
return (EOPNOTSUPP);
/*
* If protocol is connection-based, can only connect once.
* Otherwise, if connected, try to disconnect first.
* This allows user to disconnect by connecting to, e.g.,
* a null address.
*/
if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
(error = sodisconnect(so))))
error = EISCONN;
else
error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
1994-05-24 10:09:53 +00:00
return (error);
}
int
1994-05-24 10:09:53 +00:00
soconnect2(so1, so2)
struct socket *so1;
1994-05-24 10:09:53 +00:00
struct socket *so2;
{
return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
1994-05-24 10:09:53 +00:00
}
int
1994-05-24 10:09:53 +00:00
sodisconnect(so)
struct socket *so;
1994-05-24 10:09:53 +00:00
{
int error;
if ((so->so_state & SS_ISCONNECTED) == 0)
return (ENOTCONN);
if (so->so_state & SS_ISDISCONNECTING)
return (EALREADY);
error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1994-05-24 10:09:53 +00:00
return (error);
}
#define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
1994-05-24 10:09:53 +00:00
/*
* Send on a socket.
* If send must go all at once and message is larger than
* send buffering, then hard error.
* Lock against other senders.
* If must go all at once and not enough room now, then
* inform user that this would block and do nothing.
* Otherwise, if nonblocking, send as much as possible.
* The data to be sent is described by "uio" if nonzero,
* otherwise by the mbuf chain "top" (which must be null
* if uio is not). Data provided in mbuf chain must be small
* enough to send all at once.
*
* Returns nonzero on error, timeout or signal; callers
* must check for short counts if EINTR/ERESTART are returned.
* Data and control buffers are freed on return.
*/
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
#ifdef ZERO_COPY_SOCKETS
struct so_zerocopy_stats{
int size_ok;
int align_ok;
int found_ifp;
};
struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
#include <netinet/in.h>
#include <net/route.h>
#include <netinet/in_pcb.h>
#include <vm/vm.h>
#include <vm/vm_page.h>
#include <vm/vm_object.h>
#endif /*ZERO_COPY_SOCKETS*/
int
sosend(so, addr, uio, top, control, flags, td)
struct socket *so;
struct sockaddr *addr;
1994-05-24 10:09:53 +00:00
struct uio *uio;
struct mbuf *top;
struct mbuf *control;
int flags;
struct thread *td;
1994-05-24 10:09:53 +00:00
{
struct mbuf **mp;
struct mbuf *m;
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
long space, len = 0, resid;
int clen = 0, error, dontroute;
1994-05-24 10:09:53 +00:00
int atomic = sosendallatonce(so) || top;
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
#ifdef ZERO_COPY_SOCKETS
int cow_send;
#endif /* ZERO_COPY_SOCKETS */
1994-05-24 10:09:53 +00:00
if (uio != NULL)
1994-05-24 10:09:53 +00:00
resid = uio->uio_resid;
else
resid = top->m_pkthdr.len;
/*
* In theory resid should be unsigned.
* However, space must be signed, as it might be less than 0
* if we over-committed, and we must use a signed comparison
* of space and resid. On the other hand, a negative resid
* causes us to loop sending 0-length segments to the protocol.
*
* Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
* type sockets since that's an error.
1994-05-24 10:09:53 +00:00
*/
if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
error = EINVAL;
goto out;
}
1994-05-24 10:09:53 +00:00
dontroute =
(flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
(so->so_proto->pr_flags & PR_ATOMIC);
if (td != NULL)
td->td_proc->p_stats->p_ru.ru_msgsnd++;
if (control != NULL)
1994-05-24 10:09:53 +00:00
clen = control->m_len;
#define snderr(errno) { error = (errno); goto release; }
1994-05-24 10:09:53 +00:00
SOCKBUF_LOCK(&so->so_snd);
1994-05-24 10:09:53 +00:00
restart:
SOCKBUF_LOCK_ASSERT(&so->so_snd);
error = sblock(&so->so_snd, SBLOCKWAIT(flags));
if (error)
goto out_locked;
1994-05-24 10:09:53 +00:00
do {
SOCKBUF_LOCK_ASSERT(&so->so_snd);
if (so->so_snd.sb_state & SBS_CANTSENDMORE)
1994-05-24 10:09:53 +00:00
snderr(EPIPE);
if (so->so_error) {
error = so->so_error;
so->so_error = 0;
goto release;
}
1994-05-24 10:09:53 +00:00
if ((so->so_state & SS_ISCONNECTED) == 0) {
/*
* `sendto' and `sendmsg' is allowed on a connection-
* based socket if it supports implied connect.
* Return ENOTCONN if not connected and no address is
* supplied.
*/
if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
(so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1994-05-24 10:09:53 +00:00
if ((so->so_state & SS_ISCONFIRMING) == 0 &&
!(resid == 0 && clen != 0))
snderr(ENOTCONN);
} else if (addr == NULL)
snderr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
ENOTCONN : EDESTADDRREQ);
1994-05-24 10:09:53 +00:00
}
space = sbspace(&so->so_snd);
if (flags & MSG_OOB)
space += 1024;
if ((atomic && resid > so->so_snd.sb_hiwat) ||
clen > so->so_snd.sb_hiwat)
1994-05-24 10:09:53 +00:00
snderr(EMSGSIZE);
if (space < resid + clen &&
1994-05-24 10:09:53 +00:00
(atomic || space < so->so_snd.sb_lowat || space < clen)) {
if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO))
1994-05-24 10:09:53 +00:00
snderr(EWOULDBLOCK);
sbunlock(&so->so_snd);
error = sbwait(&so->so_snd);
if (error)
goto out_locked;
1994-05-24 10:09:53 +00:00
goto restart;
}
SOCKBUF_UNLOCK(&so->so_snd);
1994-05-24 10:09:53 +00:00
mp = &top;
space -= clen;
do {
if (uio == NULL) {
/*
* Data is prepackaged in "top".
*/
resid = 0;
if (flags & MSG_EOR)
top->m_flags |= M_EOR;
} else do {
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
#ifdef ZERO_COPY_SOCKETS
cow_send = 0;
#endif /* ZERO_COPY_SOCKETS */
if (resid >= MINCLSIZE) {
#ifdef ZERO_COPY_SOCKETS
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
if (top == NULL) {
MGETHDR(m, M_TRYWAIT, MT_DATA);
if (m == NULL) {
error = ENOBUFS;
SOCKBUF_LOCK(&so->so_snd);
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
goto release;
}
m->m_pkthdr.len = 0;
m->m_pkthdr.rcvif = (struct ifnet *)0;
} else {
MGET(m, M_TRYWAIT, MT_DATA);
if (m == NULL) {
error = ENOBUFS;
SOCKBUF_LOCK(&so->so_snd);
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
goto release;
}
}
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
if (so_zero_copy_send &&
resid>=PAGE_SIZE &&
space>=PAGE_SIZE &&
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
uio->uio_iov->iov_len>=PAGE_SIZE) {
so_zerocp_stats.size_ok++;
if (!((vm_offset_t)
uio->uio_iov->iov_base & PAGE_MASK)){
so_zerocp_stats.align_ok++;
cow_send = socow_setup(m, uio);
}
}
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
if (!cow_send) {
MCLGET(m, M_TRYWAIT);
if ((m->m_flags & M_EXT) == 0) {
m_free(m);
m = NULL;
} else {
len = min(min(MCLBYTES, resid), space);
}
} else
len = PAGE_SIZE;
#else /* ZERO_COPY_SOCKETS */
if (top == NULL) {
m = m_getcl(M_TRYWAIT, MT_DATA, M_PKTHDR);
m->m_pkthdr.len = 0;
m->m_pkthdr.rcvif = (struct ifnet *)0;
} else
m = m_getcl(M_TRYWAIT, MT_DATA, 0);
len = min(min(MCLBYTES, resid), space);
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
#endif /* ZERO_COPY_SOCKETS */
1994-05-24 10:09:53 +00:00
} else {
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
if (top == NULL) {
m = m_gethdr(M_TRYWAIT, MT_DATA);
m->m_pkthdr.len = 0;
m->m_pkthdr.rcvif = (struct ifnet *)0;
len = min(min(MHLEN, resid), space);
/*
* For datagram protocols, leave room
* for protocol headers in first mbuf.
*/
if (atomic && m && len < MHLEN)
MH_ALIGN(m, len);
} else {
m = m_get(M_TRYWAIT, MT_DATA);
len = min(min(MLEN, resid), space);
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
}
1994-05-24 10:09:53 +00:00
}
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
if (m == NULL) {
error = ENOBUFS;
SOCKBUF_LOCK(&so->so_snd);
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
goto release;
}
space -= len;
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
#ifdef ZERO_COPY_SOCKETS
if (cow_send)
error = 0;
else
#endif /* ZERO_COPY_SOCKETS */
error = uiomove(mtod(m, void *), (int)len, uio);
1994-05-24 10:09:53 +00:00
resid = uio->uio_resid;
m->m_len = len;
*mp = m;
top->m_pkthdr.len += len;
if (error) {
SOCKBUF_LOCK(&so->so_snd);
1994-05-24 10:09:53 +00:00
goto release;
}
1994-05-24 10:09:53 +00:00
mp = &m->m_next;
if (resid <= 0) {
if (flags & MSG_EOR)
top->m_flags |= M_EOR;
break;
}
} while (space > 0 && atomic);
if (dontroute) {
SOCK_LOCK(so);
1994-05-24 10:09:53 +00:00
so->so_options |= SO_DONTROUTE;
SOCK_UNLOCK(so);
}
/*
* XXX all the SBS_CANTSENDMORE checks previously
* done could be out of date. We could have recieved
* a reset packet in an interrupt or maybe we slept
* while doing page faults in uiomove() etc. We could
* probably recheck again inside the splnet() protection
* here, but there are probably other places that this
* also happens. We must rethink this.
*/
error = (*so->so_proto->pr_usrreqs->pru_send)(so,
(flags & MSG_OOB) ? PRUS_OOB :
/*
* If the user set MSG_EOF, the protocol
* understands this flag and nothing left to
* send then use PRU_SEND_EOF instead of PRU_SEND.
*/
((flags & MSG_EOF) &&
(so->so_proto->pr_flags & PR_IMPLOPCL) &&
(resid <= 0)) ?
PRUS_EOF :
/* If there is more to send set PRUS_MORETOCOME */
(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
top, addr, control, td);
if (dontroute) {
SOCK_LOCK(so);
1994-05-24 10:09:53 +00:00
so->so_options &= ~SO_DONTROUTE;
SOCK_UNLOCK(so);
}
1994-05-24 10:09:53 +00:00
clen = 0;
control = NULL;
top = NULL;
1994-05-24 10:09:53 +00:00
mp = &top;
if (error) {
SOCKBUF_LOCK(&so->so_snd);
1994-05-24 10:09:53 +00:00
goto release;
}
1994-05-24 10:09:53 +00:00
} while (resid && space > 0);
SOCKBUF_LOCK(&so->so_snd);
1994-05-24 10:09:53 +00:00
} while (resid);
release:
SOCKBUF_LOCK_ASSERT(&so->so_snd);
1994-05-24 10:09:53 +00:00
sbunlock(&so->so_snd);
out_locked:
SOCKBUF_LOCK_ASSERT(&so->so_snd);
SOCKBUF_UNLOCK(&so->so_snd);
1994-05-24 10:09:53 +00:00
out:
if (top != NULL)
1994-05-24 10:09:53 +00:00
m_freem(top);
if (control != NULL)
1994-05-24 10:09:53 +00:00
m_freem(control);
return (error);
}
/*
* The part of soreceive() that implements reading non-inline out-of-band
* data from a socket. For more complete comments, see soreceive(), from
* which this code originated.
*
* XXXRW: Note that soreceive_rcvoob(), unlike the remainder of soreceve(),
* is unable to return an mbuf chain to the caller.
*/
static int
soreceive_rcvoob(so, uio, flags)
struct socket *so;
struct uio *uio;
int flags;
{
struct protosw *pr = so->so_proto;
struct mbuf *m;
int error;
KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
m = m_get(M_TRYWAIT, MT_DATA);
if (m == NULL)
return (ENOBUFS);
error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
if (error)
goto bad;
do {
#ifdef ZERO_COPY_SOCKETS
if (so_zero_copy_receive) {
vm_page_t pg;
int disposable;
if ((m->m_flags & M_EXT)
&& (m->m_ext.ext_type == EXT_DISPOSABLE))
disposable = 1;
else
disposable = 0;
pg = PHYS_TO_VM_PAGE(vtophys(mtod(m, caddr_t)));
if (uio->uio_offset == -1)
uio->uio_offset =IDX_TO_OFF(pg->pindex);
error = uiomoveco(mtod(m, void *),
min(uio->uio_resid, m->m_len),
uio, pg->object,
disposable);
} else
#endif /* ZERO_COPY_SOCKETS */
error = uiomove(mtod(m, void *),
(int) min(uio->uio_resid, m->m_len), uio);
m = m_free(m);
} while (uio->uio_resid && error == 0 && m);
bad:
if (m != NULL)
m_freem(m);
return (error);
}
1994-05-24 10:09:53 +00:00
/*
* Implement receive operations on a socket.
* We depend on the way that records are added to the sockbuf
* by sbappend*. In particular, each record (mbufs linked through m_next)
* must begin with an address if the protocol so specifies,
* followed by an optional mbuf or mbufs containing ancillary data,
* and then zero or more mbufs of data.
* In order to avoid blocking network interrupts for the entire time here,
* we splx() while doing the actual copy to user space.
* Although the sockbuf is locked, new data may still be appended,
* and thus we must maintain consistency of the sockbuf during that time.
*
* The caller may receive the data as a single mbuf chain by supplying
* an mbuf **mp0 for use in returning the chain. The uio is then used
* only for the count in uio_resid.
*/
int
soreceive(so, psa, uio, mp0, controlp, flagsp)
struct socket *so;
struct sockaddr **psa;
1994-05-24 10:09:53 +00:00
struct uio *uio;
struct mbuf **mp0;
struct mbuf **controlp;
int *flagsp;
{
struct mbuf *m, **mp;
int flags, len, error, offset;
1994-05-24 10:09:53 +00:00
struct protosw *pr = so->so_proto;
struct mbuf *nextrecord;
int moff, type = 0;
1994-05-24 10:09:53 +00:00
int orig_resid = uio->uio_resid;
mp = mp0;
if (psa != NULL)
*psa = NULL;
if (controlp != NULL)
*controlp = NULL;
if (flagsp != NULL)
1994-05-24 10:09:53 +00:00
flags = *flagsp &~ MSG_EOR;
else
flags = 0;
if (flags & MSG_OOB)
return (soreceive_rcvoob(so, uio, flags));
if (mp != NULL)
*mp = NULL;
if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1994-05-24 10:09:53 +00:00
SOCKBUF_LOCK(&so->so_rcv);
1994-05-24 10:09:53 +00:00
restart:
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
if (error)
goto out;
1994-05-24 10:09:53 +00:00
m = so->so_rcv.sb_mb;
/*
* If we have less data than requested, block awaiting more
* (subject to any timeout) if:
* 1. the current count is less than the low water mark, or
* 2. MSG_WAITALL is set, and it is possible to do the entire
* receive operation at once if we block (resid <= hiwat).
* 3. MSG_DONTWAIT is not set
* If MSG_WAITALL is set but resid is larger than the receive buffer,
* we have to do the receive in sections, and thus risk returning
* a short count if a timeout or signal occurs after we start.
*/
if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1994-05-24 10:09:53 +00:00
so->so_rcv.sb_cc < uio->uio_resid) &&
(so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
KASSERT(m != NULL || !so->so_rcv.sb_cc,
("receive: m == %p so->so_rcv.sb_cc == %u",
m, so->so_rcv.sb_cc));
1994-05-24 10:09:53 +00:00
if (so->so_error) {
if (m != NULL)
1994-05-24 10:09:53 +00:00
goto dontblock;
error = so->so_error;
if ((flags & MSG_PEEK) == 0)
so->so_error = 0;
goto release;
}
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1994-05-24 10:09:53 +00:00
if (m)
goto dontblock;
else
goto release;
}
for (; m != NULL; m = m->m_next)
1994-05-24 10:09:53 +00:00
if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
m = so->so_rcv.sb_mb;
goto dontblock;
}
if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
(so->so_proto->pr_flags & PR_CONNREQUIRED)) {
error = ENOTCONN;
goto release;
}
if (uio->uio_resid == 0)
1994-05-24 10:09:53 +00:00
goto release;
if ((so->so_state & SS_NBIO) ||
(flags & (MSG_DONTWAIT|MSG_NBIO))) {
1994-05-24 10:09:53 +00:00
error = EWOULDBLOCK;
goto release;
}
SBLASTRECORDCHK(&so->so_rcv);
SBLASTMBUFCHK(&so->so_rcv);
1994-05-24 10:09:53 +00:00
sbunlock(&so->so_rcv);
error = sbwait(&so->so_rcv);
if (error)
goto out;
1994-05-24 10:09:53 +00:00
goto restart;
}
dontblock:
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
if (uio->uio_td)
uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++;
SBLASTRECORDCHK(&so->so_rcv);
SBLASTMBUFCHK(&so->so_rcv);
1994-05-24 10:09:53 +00:00
nextrecord = m->m_nextpkt;
if (pr->pr_flags & PR_ADDR) {
KASSERT(m->m_type == MT_SONAME,
("m->m_type == %d", m->m_type));
1994-05-24 10:09:53 +00:00
orig_resid = 0;
if (psa != NULL)
*psa = sodupsockaddr(mtod(m, struct sockaddr *),
M_NOWAIT);
1994-05-24 10:09:53 +00:00
if (flags & MSG_PEEK) {
m = m->m_next;
} else {
sbfree(&so->so_rcv, m);
so->so_rcv.sb_mb = m_free(m);
m = so->so_rcv.sb_mb;
if (m != NULL)
m->m_nextpkt = nextrecord;
1994-05-24 10:09:53 +00:00
}
}
while (m != NULL && m->m_type == MT_CONTROL && error == 0) {
1994-05-24 10:09:53 +00:00
if (flags & MSG_PEEK) {
if (controlp != NULL)
1994-05-24 10:09:53 +00:00
*controlp = m_copy(m, 0, m->m_len);
m = m->m_next;
} else {
sbfree(&so->so_rcv, m);
so->so_rcv.sb_mb = m->m_next;
m->m_next = NULL;
if (pr->pr_domain->dom_externalize) {
SOCKBUF_UNLOCK(&so->so_rcv);
error = (*pr->pr_domain->dom_externalize)
(m, controlp);
SOCKBUF_LOCK(&so->so_rcv);
} else if (controlp != NULL)
1994-05-24 10:09:53 +00:00
*controlp = m;
else
m_freem(m);
m = so->so_rcv.sb_mb;
1994-05-24 10:09:53 +00:00
}
if (controlp != NULL) {
1994-05-24 10:09:53 +00:00
orig_resid = 0;
while (*controlp != NULL)
controlp = &(*controlp)->m_next;
1994-05-24 10:09:53 +00:00
}
}
if (m != NULL) {
if ((flags & MSG_PEEK) == 0) {
1994-05-24 10:09:53 +00:00
m->m_nextpkt = nextrecord;
/*
* If nextrecord == NULL (this is a single chain),
* then sb_lastrecord may not be valid here if m
* was changed earlier.
*/
if (nextrecord == NULL) {
KASSERT(so->so_rcv.sb_mb == m,
("receive tailq 1"));
so->so_rcv.sb_lastrecord = m;
}
}
1994-05-24 10:09:53 +00:00
type = m->m_type;
if (type == MT_OOBDATA)
flags |= MSG_OOB;
} else {
if ((flags & MSG_PEEK) == 0) {
KASSERT(so->so_rcv.sb_mb == m,("receive tailq 2"));
so->so_rcv.sb_mb = nextrecord;
SB_EMPTY_FIXUP(&so->so_rcv);
}
1994-05-24 10:09:53 +00:00
}
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
SBLASTRECORDCHK(&so->so_rcv);
SBLASTMBUFCHK(&so->so_rcv);
1994-05-24 10:09:53 +00:00
moff = 0;
offset = 0;
while (m != NULL && uio->uio_resid > 0 && error == 0) {
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1994-05-24 10:09:53 +00:00
if (m->m_type == MT_OOBDATA) {
if (type != MT_OOBDATA)
break;
} else if (type == MT_OOBDATA)
break;
else
KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
("m->m_type == %d", m->m_type));
so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1994-05-24 10:09:53 +00:00
len = uio->uio_resid;
if (so->so_oobmark && len > so->so_oobmark - offset)
len = so->so_oobmark - offset;
if (len > m->m_len - moff)
len = m->m_len - moff;
/*
* If mp is set, just pass back the mbufs.
* Otherwise copy them out via the uio, then free.
* Sockbuf must be consistent here (points to current mbuf,
* it points to next record) when we drop priority;
* we must note any additions to the sockbuf when we
* block interrupts again.
*/
if (mp == NULL) {
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
SBLASTRECORDCHK(&so->so_rcv);
SBLASTMBUFCHK(&so->so_rcv);
SOCKBUF_UNLOCK(&so->so_rcv);
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
#ifdef ZERO_COPY_SOCKETS
if (so_zero_copy_receive) {
vm_page_t pg;
int disposable;
if ((m->m_flags & M_EXT)
&& (m->m_ext.ext_type == EXT_DISPOSABLE))
disposable = 1;
else
disposable = 0;
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
pg = PHYS_TO_VM_PAGE(vtophys(mtod(m, caddr_t) +
moff));
if (uio->uio_offset == -1)
uio->uio_offset =IDX_TO_OFF(pg->pindex);
error = uiomoveco(mtod(m, char *) + moff,
At long last, commit the zero copy sockets code. MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes. ti.4: Update the ti(4) man page to include information on the TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options, and also include information about the new character device interface and the associated ioctls. man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated links. jumbo.9: New man page describing the jumbo buffer allocator interface and operation. zero_copy.9: New man page describing the general characteristics of the zero copy send and receive code, and what an application author should do to take advantage of the zero copy functionality. NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS, TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT. conf/files: Add uipc_jumbo.c and uipc_cow.c. conf/options: Add the 5 options mentioned above. kern_subr.c: Receive side zero copy implementation. This takes "disposable" pages attached to an mbuf, gives them to a user process, and then recycles the user's page. This is only active when ZERO_COPY_SOCKETS is turned on and the kern.ipc.zero_copy.receive sysctl variable is set to 1. uipc_cow.c: Send side zero copy functions. Takes a page written by the user and maps it copy on write and assigns it kernel virtual address space. Removes copy on write mapping once the buffer has been freed by the network stack. uipc_jumbo.c: Jumbo disposable page allocator code. This allocates (optionally) disposable pages for network drivers that want to give the user the option of doing zero copy receive. uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are enabled if ZERO_COPY_SOCKETS is turned on. Add zero copy send support to sosend() -- pages get mapped into the kernel instead of getting copied if they meet size and alignment restrictions. uipc_syscalls.c:Un-staticize some of the sf* functions so that they can be used elsewhere. (uipc_cow.c) if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid calling malloc() with M_WAITOK. Return an error if the M_NOWAIT malloc fails. The ti(4) driver and the wi(4) driver, at least, call this with a mutex held. This causes witness warnings for 'ifconfig -a' with a wi(4) or ti(4) board in the system. (I've only verified for ti(4)). ip_output.c: Fragment large datagrams so that each segment contains a multiple of PAGE_SIZE amount of data plus headers. This allows the receiver to potentially do page flipping on receives. if_ti.c: Add zero copy receive support to the ti(4) driver. If TI_PRIVATE_JUMBOS is not defined, it now uses the jumbo(9) buffer allocator for jumbo receive buffers. Add a new character device interface for the ti(4) driver for the new debugging interface. This allows (a patched version of) gdb to talk to the Tigon board and debug the firmware. There are also a few additional debugging ioctls available through this interface. Add header splitting support to the ti(4) driver. Tweak some of the default interrupt coalescing parameters to more useful defaults. Add hooks for supporting transmit flow control, but leave it turned off with a comment describing why it is turned off. if_tireg.h: Change the firmware rev to 12.4.11, since we're really at 12.4.11 plus fixes from 12.4.13. Add defines needed for debugging. Remove the ti_stats structure, it is now defined in sys/tiio.h. ti_fw.h: 12.4.11 firmware. ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13, and my header splitting patches. Revision 12.4.13 doesn't handle 10/100 negotiation properly. (This firmware is the same as what was in the tree previously, with the addition of header splitting support.) sys/jumbo.h: Jumbo buffer allocator interface. sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to indicate that the payload buffer can be thrown away / flipped to a userland process. socketvar.h: Add prototype for socow_setup. tiio.h: ioctl interface to the character portion of the ti(4) driver, plus associated structure/type definitions. uio.h: Change prototype for uiomoveco() so that we'll know whether the source page is disposable. ufs_readwrite.c:Update for new prototype of uiomoveco(). vm_fault.c: In vm_fault(), check to see whether we need to do a page based copy on write fault. vm_object.c: Add a new function, vm_object_allocate_wait(). This does the same thing that vm_object allocate does, except that it gives the caller the opportunity to specify whether it should wait on the uma_zalloc() of the object structre. This allows vm objects to be allocated while holding a mutex. (Without generating WITNESS warnings.) vm_object_allocate() is implemented as a call to vm_object_allocate_wait() with the malloc flag set to M_WAITOK. vm_object.h: Add prototype for vm_object_allocate_wait(). vm_page.c: Add page-based copy on write setup, clear and fault routines. vm_page.h: Add page based COW function prototypes and variable in the vm_page structure. Many thanks to Drew Gallatin, who wrote the zero copy send and receive code, and to all the other folks who have tested and reviewed this code over the years.
2002-06-26 03:37:47 +00:00
(int)len, uio,pg->object,
disposable);
} else
#endif /* ZERO_COPY_SOCKETS */
error = uiomove(mtod(m, char *) + moff, (int)len, uio);
SOCKBUF_LOCK(&so->so_rcv);
if (error)
goto release;
1994-05-24 10:09:53 +00:00
} else
uio->uio_resid -= len;
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1994-05-24 10:09:53 +00:00
if (len == m->m_len - moff) {
if (m->m_flags & M_EOR)
flags |= MSG_EOR;
if (flags & MSG_PEEK) {
m = m->m_next;
moff = 0;
} else {
nextrecord = m->m_nextpkt;
sbfree(&so->so_rcv, m);
if (mp != NULL) {
1994-05-24 10:09:53 +00:00
*mp = m;
mp = &m->m_next;
so->so_rcv.sb_mb = m = m->m_next;
*mp = NULL;
1994-05-24 10:09:53 +00:00
} else {
so->so_rcv.sb_mb = m_free(m);
1994-05-24 10:09:53 +00:00
m = so->so_rcv.sb_mb;
}
if (m != NULL) {
1994-05-24 10:09:53 +00:00
m->m_nextpkt = nextrecord;
if (nextrecord == NULL)
so->so_rcv.sb_lastrecord = m;
} else {
so->so_rcv.sb_mb = nextrecord;
SB_EMPTY_FIXUP(&so->so_rcv);
}
SBLASTRECORDCHK(&so->so_rcv);
SBLASTMBUFCHK(&so->so_rcv);
1994-05-24 10:09:53 +00:00
}
} else {
if (flags & MSG_PEEK)
moff += len;
else {
if (mp != NULL) {
SOCKBUF_UNLOCK(&so->so_rcv);
*mp = m_copym(m, 0, len, M_TRYWAIT);
SOCKBUF_LOCK(&so->so_rcv);
}
1994-05-24 10:09:53 +00:00
m->m_data += len;
m->m_len -= len;
so->so_rcv.sb_cc -= len;
}
}
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1994-05-24 10:09:53 +00:00
if (so->so_oobmark) {
if ((flags & MSG_PEEK) == 0) {
so->so_oobmark -= len;
if (so->so_oobmark == 0) {
so->so_rcv.sb_state |= SBS_RCVATMARK;
1994-05-24 10:09:53 +00:00
break;
}
} else {
offset += len;
if (offset == so->so_oobmark)
break;
}
}
if (flags & MSG_EOR)
break;
/*
* If the MSG_WAITALL flag is set (for non-atomic socket),
* we must not quit until "uio->uio_resid == 0" or an error
* termination. If a signal/timeout occurs, return
* with a short count but without error.
* Keep sockbuf locked against other readers.
*/
while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
!sosendallatonce(so) && nextrecord == NULL) {
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1994-05-24 10:09:53 +00:00
break;
/*
* Notify the protocol that some data has been
* drained before blocking.
*/
if (pr->pr_flags & PR_WANTRCVD && so->so_pcb != NULL) {
SOCKBUF_UNLOCK(&so->so_rcv);
(*pr->pr_usrreqs->pru_rcvd)(so, flags);
SOCKBUF_LOCK(&so->so_rcv);
}
SBLASTRECORDCHK(&so->so_rcv);
SBLASTMBUFCHK(&so->so_rcv);
1994-05-24 10:09:53 +00:00
error = sbwait(&so->so_rcv);
if (error)
goto release;
m = so->so_rcv.sb_mb;
if (m != NULL)
1994-05-24 10:09:53 +00:00
nextrecord = m->m_nextpkt;
}
}
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1994-05-24 10:09:53 +00:00
flags |= MSG_TRUNC;
if ((flags & MSG_PEEK) == 0)
Merge next step in socket buffer locking: - sowakeup() now asserts the socket buffer lock on entry. Move the call to KNOTE higher in sowakeup() so that it is made with the socket buffer lock held for consistency with other calls. Release the socket buffer lock prior to calling into pgsigio(), so_upcall(), or aio_swake(). Locking for this event management will need revisiting in the future, but this model avoids lock order reversals when upcalls into other subsystems result in socket/socket buffer operations. Assert that the socket buffer lock is not held at the end of the function. - Wrapper macros for sowakeup(), sorwakeup() and sowwakeup(), now have _locked versions which assert the socket buffer lock on entry. If a wakeup is required by sb_notify(), invoke sowakeup(); otherwise, unconditionally release the socket buffer lock. This results in the socket buffer lock being released whether a wakeup is required or not. - Break out socantsendmore() into socantsendmore_locked() that asserts the socket buffer lock. socantsendmore() unconditionally locks the socket buffer before calling socantsendmore_locked(). Note that both functions return with the socket buffer unlocked as socantsendmore_locked() calls sowwakeup_locked() which has the same properties. Assert that the socket buffer is unlocked on return. - Break out socantrcvmore() into socantrcvmore_locked() that asserts the socket buffer lock. socantrcvmore() unconditionally locks the socket buffer before calling socantrcvmore_locked(). Note that both functions return with the socket buffer unlocked as socantrcvmore_locked() calls sorwakeup_locked() which has similar properties. Assert that the socket buffer is unlocked on return. - Break out sbrelease() into a sbrelease_locked() that asserts the socket buffer lock. sbrelease() unconditionally locks the socket buffer before calling sbrelease_locked(). sbrelease_locked() now invokes sbflush_locked() instead of sbflush(). - Assert the socket buffer lock in socket buffer sanity check functions sblastrecordchk(), sblastmbufchk(). - Assert the socket buffer lock in SBLINKRECORD(). - Break out various sbappend() functions into sbappend_locked() (and variations on that name) that assert the socket buffer lock. The !_locked() variations unconditionally lock the socket buffer before calling their _locked counterparts. Internally, make sure to call _locked() support routines, etc, if already holding the socket buffer lock. - Break out sbinsertoob() into sbinsertoob_locked() that asserts the socket buffer lock. sbinsertoob() unconditionally locks the socket buffer before calling sbinsertoob_locked(). - Break out sbflush() into sbflush_locked() that asserts the socket buffer lock. sbflush() unconditionally locks the socket buffer before calling sbflush_locked(). Update panic strings for new function names. - Break out sbdrop() into sbdrop_locked() that asserts the socket buffer lock. sbdrop() unconditionally locks the socket buffer before calling sbdrop_locked(). - Break out sbdroprecord() into sbdroprecord_locked() that asserts the socket buffer lock. sbdroprecord() unconditionally locks the socket buffer before calling sbdroprecord_locked(). - sofree() now calls socantsendmore_locked() and re-acquires the socket buffer lock on return. It also now calls sbrelease_locked(). - sorflush() now calls socantrcvmore_locked() and re-acquires the socket buffer lock on return. Clean up/mess up other behavior in sorflush() relating to the temporary stack copy of the socket buffer used with dom_dispose by more properly initializing the temporary copy, and selectively bzeroing/copying more carefully to prevent WITNESS from getting confused by improperly initialized mutexes. Annotate why that's necessary, or at least, needed. - soisconnected() now calls sbdrop_locked() before unlocking the socket buffer to avoid locking overhead. Some parts of this change were: Submitted by: sam Sponsored by: FreeBSD Foundation Obtained from: BSD/OS
2004-06-21 00:20:43 +00:00
(void) sbdroprecord_locked(&so->so_rcv);
1994-05-24 10:09:53 +00:00
}
if ((flags & MSG_PEEK) == 0) {
if (m == NULL) {
/*
* First part is an inline SB_EMPTY_FIXUP(). Second
* part makes sure sb_lastrecord is up-to-date if
* there is still data in the socket buffer.
*/
1994-05-24 10:09:53 +00:00
so->so_rcv.sb_mb = nextrecord;
if (so->so_rcv.sb_mb == NULL) {
so->so_rcv.sb_mbtail = NULL;
so->so_rcv.sb_lastrecord = NULL;
} else if (nextrecord->m_nextpkt == NULL)
so->so_rcv.sb_lastrecord = nextrecord;
}
SBLASTRECORDCHK(&so->so_rcv);
SBLASTMBUFCHK(&so->so_rcv);
if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) {
SOCKBUF_UNLOCK(&so->so_rcv);
(*pr->pr_usrreqs->pru_rcvd)(so, flags);
SOCKBUF_LOCK(&so->so_rcv);
}
1994-05-24 10:09:53 +00:00
}
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1994-05-24 10:09:53 +00:00
if (orig_resid == uio->uio_resid && orig_resid &&
(flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1994-05-24 10:09:53 +00:00
sbunlock(&so->so_rcv);
goto restart;
}
1995-05-30 08:16:23 +00:00
if (flagsp != NULL)
1994-05-24 10:09:53 +00:00
*flagsp |= flags;
release:
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1994-05-24 10:09:53 +00:00
sbunlock(&so->so_rcv);
out:
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
SOCKBUF_UNLOCK(&so->so_rcv);
1994-05-24 10:09:53 +00:00
return (error);
}
int
1994-05-24 10:09:53 +00:00
soshutdown(so, how)
struct socket *so;
int how;
1994-05-24 10:09:53 +00:00
{
struct protosw *pr = so->so_proto;
1994-05-24 10:09:53 +00:00
if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
return (EINVAL);
if (how != SHUT_WR)
1994-05-24 10:09:53 +00:00
sorflush(so);
if (how != SHUT_RD)
return ((*pr->pr_usrreqs->pru_shutdown)(so));
1994-05-24 10:09:53 +00:00
return (0);
}
void
1994-05-24 10:09:53 +00:00
sorflush(so)
struct socket *so;
1994-05-24 10:09:53 +00:00
{
struct sockbuf *sb = &so->so_rcv;
struct protosw *pr = so->so_proto;
1994-05-24 10:09:53 +00:00
struct sockbuf asb;
Merge next step in socket buffer locking: - sowakeup() now asserts the socket buffer lock on entry. Move the call to KNOTE higher in sowakeup() so that it is made with the socket buffer lock held for consistency with other calls. Release the socket buffer lock prior to calling into pgsigio(), so_upcall(), or aio_swake(). Locking for this event management will need revisiting in the future, but this model avoids lock order reversals when upcalls into other subsystems result in socket/socket buffer operations. Assert that the socket buffer lock is not held at the end of the function. - Wrapper macros for sowakeup(), sorwakeup() and sowwakeup(), now have _locked versions which assert the socket buffer lock on entry. If a wakeup is required by sb_notify(), invoke sowakeup(); otherwise, unconditionally release the socket buffer lock. This results in the socket buffer lock being released whether a wakeup is required or not. - Break out socantsendmore() into socantsendmore_locked() that asserts the socket buffer lock. socantsendmore() unconditionally locks the socket buffer before calling socantsendmore_locked(). Note that both functions return with the socket buffer unlocked as socantsendmore_locked() calls sowwakeup_locked() which has the same properties. Assert that the socket buffer is unlocked on return. - Break out socantrcvmore() into socantrcvmore_locked() that asserts the socket buffer lock. socantrcvmore() unconditionally locks the socket buffer before calling socantrcvmore_locked(). Note that both functions return with the socket buffer unlocked as socantrcvmore_locked() calls sorwakeup_locked() which has similar properties. Assert that the socket buffer is unlocked on return. - Break out sbrelease() into a sbrelease_locked() that asserts the socket buffer lock. sbrelease() unconditionally locks the socket buffer before calling sbrelease_locked(). sbrelease_locked() now invokes sbflush_locked() instead of sbflush(). - Assert the socket buffer lock in socket buffer sanity check functions sblastrecordchk(), sblastmbufchk(). - Assert the socket buffer lock in SBLINKRECORD(). - Break out various sbappend() functions into sbappend_locked() (and variations on that name) that assert the socket buffer lock. The !_locked() variations unconditionally lock the socket buffer before calling their _locked counterparts. Internally, make sure to call _locked() support routines, etc, if already holding the socket buffer lock. - Break out sbinsertoob() into sbinsertoob_locked() that asserts the socket buffer lock. sbinsertoob() unconditionally locks the socket buffer before calling sbinsertoob_locked(). - Break out sbflush() into sbflush_locked() that asserts the socket buffer lock. sbflush() unconditionally locks the socket buffer before calling sbflush_locked(). Update panic strings for new function names. - Break out sbdrop() into sbdrop_locked() that asserts the socket buffer lock. sbdrop() unconditionally locks the socket buffer before calling sbdrop_locked(). - Break out sbdroprecord() into sbdroprecord_locked() that asserts the socket buffer lock. sbdroprecord() unconditionally locks the socket buffer before calling sbdroprecord_locked(). - sofree() now calls socantsendmore_locked() and re-acquires the socket buffer lock on return. It also now calls sbrelease_locked(). - sorflush() now calls socantrcvmore_locked() and re-acquires the socket buffer lock on return. Clean up/mess up other behavior in sorflush() relating to the temporary stack copy of the socket buffer used with dom_dispose by more properly initializing the temporary copy, and selectively bzeroing/copying more carefully to prevent WITNESS from getting confused by improperly initialized mutexes. Annotate why that's necessary, or at least, needed. - soisconnected() now calls sbdrop_locked() before unlocking the socket buffer to avoid locking overhead. Some parts of this change were: Submitted by: sam Sponsored by: FreeBSD Foundation Obtained from: BSD/OS
2004-06-21 00:20:43 +00:00
/*
* XXXRW: This is quite ugly. The existing code made a copy of the
* socket buffer, then zero'd the original to clear the buffer
* fields. However, with mutexes in the socket buffer, this causes
* problems. We only clear the zeroable bits of the original;
* however, we have to initialize and destroy the mutex in the copy
* so that dom_dispose() and sbrelease() can lock t as needed.
*/
SOCKBUF_LOCK(sb);
1994-05-24 10:09:53 +00:00
sb->sb_flags |= SB_NOINTR;
(void) sblock(sb, M_WAITOK);
Merge next step in socket buffer locking: - sowakeup() now asserts the socket buffer lock on entry. Move the call to KNOTE higher in sowakeup() so that it is made with the socket buffer lock held for consistency with other calls. Release the socket buffer lock prior to calling into pgsigio(), so_upcall(), or aio_swake(). Locking for this event management will need revisiting in the future, but this model avoids lock order reversals when upcalls into other subsystems result in socket/socket buffer operations. Assert that the socket buffer lock is not held at the end of the function. - Wrapper macros for sowakeup(), sorwakeup() and sowwakeup(), now have _locked versions which assert the socket buffer lock on entry. If a wakeup is required by sb_notify(), invoke sowakeup(); otherwise, unconditionally release the socket buffer lock. This results in the socket buffer lock being released whether a wakeup is required or not. - Break out socantsendmore() into socantsendmore_locked() that asserts the socket buffer lock. socantsendmore() unconditionally locks the socket buffer before calling socantsendmore_locked(). Note that both functions return with the socket buffer unlocked as socantsendmore_locked() calls sowwakeup_locked() which has the same properties. Assert that the socket buffer is unlocked on return. - Break out socantrcvmore() into socantrcvmore_locked() that asserts the socket buffer lock. socantrcvmore() unconditionally locks the socket buffer before calling socantrcvmore_locked(). Note that both functions return with the socket buffer unlocked as socantrcvmore_locked() calls sorwakeup_locked() which has similar properties. Assert that the socket buffer is unlocked on return. - Break out sbrelease() into a sbrelease_locked() that asserts the socket buffer lock. sbrelease() unconditionally locks the socket buffer before calling sbrelease_locked(). sbrelease_locked() now invokes sbflush_locked() instead of sbflush(). - Assert the socket buffer lock in socket buffer sanity check functions sblastrecordchk(), sblastmbufchk(). - Assert the socket buffer lock in SBLINKRECORD(). - Break out various sbappend() functions into sbappend_locked() (and variations on that name) that assert the socket buffer lock. The !_locked() variations unconditionally lock the socket buffer before calling their _locked counterparts. Internally, make sure to call _locked() support routines, etc, if already holding the socket buffer lock. - Break out sbinsertoob() into sbinsertoob_locked() that asserts the socket buffer lock. sbinsertoob() unconditionally locks the socket buffer before calling sbinsertoob_locked(). - Break out sbflush() into sbflush_locked() that asserts the socket buffer lock. sbflush() unconditionally locks the socket buffer before calling sbflush_locked(). Update panic strings for new function names. - Break out sbdrop() into sbdrop_locked() that asserts the socket buffer lock. sbdrop() unconditionally locks the socket buffer before calling sbdrop_locked(). - Break out sbdroprecord() into sbdroprecord_locked() that asserts the socket buffer lock. sbdroprecord() unconditionally locks the socket buffer before calling sbdroprecord_locked(). - sofree() now calls socantsendmore_locked() and re-acquires the socket buffer lock on return. It also now calls sbrelease_locked(). - sorflush() now calls socantrcvmore_locked() and re-acquires the socket buffer lock on return. Clean up/mess up other behavior in sorflush() relating to the temporary stack copy of the socket buffer used with dom_dispose by more properly initializing the temporary copy, and selectively bzeroing/copying more carefully to prevent WITNESS from getting confused by improperly initialized mutexes. Annotate why that's necessary, or at least, needed. - soisconnected() now calls sbdrop_locked() before unlocking the socket buffer to avoid locking overhead. Some parts of this change were: Submitted by: sam Sponsored by: FreeBSD Foundation Obtained from: BSD/OS
2004-06-21 00:20:43 +00:00
/*
* socantrcvmore_locked() drops the socket buffer mutex so that it
* can safely perform wakeups. Re-acquire the mutex before
* continuing.
*/
socantrcvmore_locked(so);
SOCKBUF_LOCK(sb);
1994-05-24 10:09:53 +00:00
sbunlock(sb);
/*
Merge next step in socket buffer locking: - sowakeup() now asserts the socket buffer lock on entry. Move the call to KNOTE higher in sowakeup() so that it is made with the socket buffer lock held for consistency with other calls. Release the socket buffer lock prior to calling into pgsigio(), so_upcall(), or aio_swake(). Locking for this event management will need revisiting in the future, but this model avoids lock order reversals when upcalls into other subsystems result in socket/socket buffer operations. Assert that the socket buffer lock is not held at the end of the function. - Wrapper macros for sowakeup(), sorwakeup() and sowwakeup(), now have _locked versions which assert the socket buffer lock on entry. If a wakeup is required by sb_notify(), invoke sowakeup(); otherwise, unconditionally release the socket buffer lock. This results in the socket buffer lock being released whether a wakeup is required or not. - Break out socantsendmore() into socantsendmore_locked() that asserts the socket buffer lock. socantsendmore() unconditionally locks the socket buffer before calling socantsendmore_locked(). Note that both functions return with the socket buffer unlocked as socantsendmore_locked() calls sowwakeup_locked() which has the same properties. Assert that the socket buffer is unlocked on return. - Break out socantrcvmore() into socantrcvmore_locked() that asserts the socket buffer lock. socantrcvmore() unconditionally locks the socket buffer before calling socantrcvmore_locked(). Note that both functions return with the socket buffer unlocked as socantrcvmore_locked() calls sorwakeup_locked() which has similar properties. Assert that the socket buffer is unlocked on return. - Break out sbrelease() into a sbrelease_locked() that asserts the socket buffer lock. sbrelease() unconditionally locks the socket buffer before calling sbrelease_locked(). sbrelease_locked() now invokes sbflush_locked() instead of sbflush(). - Assert the socket buffer lock in socket buffer sanity check functions sblastrecordchk(), sblastmbufchk(). - Assert the socket buffer lock in SBLINKRECORD(). - Break out various sbappend() functions into sbappend_locked() (and variations on that name) that assert the socket buffer lock. The !_locked() variations unconditionally lock the socket buffer before calling their _locked counterparts. Internally, make sure to call _locked() support routines, etc, if already holding the socket buffer lock. - Break out sbinsertoob() into sbinsertoob_locked() that asserts the socket buffer lock. sbinsertoob() unconditionally locks the socket buffer before calling sbinsertoob_locked(). - Break out sbflush() into sbflush_locked() that asserts the socket buffer lock. sbflush() unconditionally locks the socket buffer before calling sbflush_locked(). Update panic strings for new function names. - Break out sbdrop() into sbdrop_locked() that asserts the socket buffer lock. sbdrop() unconditionally locks the socket buffer before calling sbdrop_locked(). - Break out sbdroprecord() into sbdroprecord_locked() that asserts the socket buffer lock. sbdroprecord() unconditionally locks the socket buffer before calling sbdroprecord_locked(). - sofree() now calls socantsendmore_locked() and re-acquires the socket buffer lock on return. It also now calls sbrelease_locked(). - sorflush() now calls socantrcvmore_locked() and re-acquires the socket buffer lock on return. Clean up/mess up other behavior in sorflush() relating to the temporary stack copy of the socket buffer used with dom_dispose by more properly initializing the temporary copy, and selectively bzeroing/copying more carefully to prevent WITNESS from getting confused by improperly initialized mutexes. Annotate why that's necessary, or at least, needed. - soisconnected() now calls sbdrop_locked() before unlocking the socket buffer to avoid locking overhead. Some parts of this change were: Submitted by: sam Sponsored by: FreeBSD Foundation Obtained from: BSD/OS
2004-06-21 00:20:43 +00:00
* Invalidate/clear most of the sockbuf structure, but leave
* selinfo and mutex data unchanged.
*/
Merge next step in socket buffer locking: - sowakeup() now asserts the socket buffer lock on entry. Move the call to KNOTE higher in sowakeup() so that it is made with the socket buffer lock held for consistency with other calls. Release the socket buffer lock prior to calling into pgsigio(), so_upcall(), or aio_swake(). Locking for this event management will need revisiting in the future, but this model avoids lock order reversals when upcalls into other subsystems result in socket/socket buffer operations. Assert that the socket buffer lock is not held at the end of the function. - Wrapper macros for sowakeup(), sorwakeup() and sowwakeup(), now have _locked versions which assert the socket buffer lock on entry. If a wakeup is required by sb_notify(), invoke sowakeup(); otherwise, unconditionally release the socket buffer lock. This results in the socket buffer lock being released whether a wakeup is required or not. - Break out socantsendmore() into socantsendmore_locked() that asserts the socket buffer lock. socantsendmore() unconditionally locks the socket buffer before calling socantsendmore_locked(). Note that both functions return with the socket buffer unlocked as socantsendmore_locked() calls sowwakeup_locked() which has the same properties. Assert that the socket buffer is unlocked on return. - Break out socantrcvmore() into socantrcvmore_locked() that asserts the socket buffer lock. socantrcvmore() unconditionally locks the socket buffer before calling socantrcvmore_locked(). Note that both functions return with the socket buffer unlocked as socantrcvmore_locked() calls sorwakeup_locked() which has similar properties. Assert that the socket buffer is unlocked on return. - Break out sbrelease() into a sbrelease_locked() that asserts the socket buffer lock. sbrelease() unconditionally locks the socket buffer before calling sbrelease_locked(). sbrelease_locked() now invokes sbflush_locked() instead of sbflush(). - Assert the socket buffer lock in socket buffer sanity check functions sblastrecordchk(), sblastmbufchk(). - Assert the socket buffer lock in SBLINKRECORD(). - Break out various sbappend() functions into sbappend_locked() (and variations on that name) that assert the socket buffer lock. The !_locked() variations unconditionally lock the socket buffer before calling their _locked counterparts. Internally, make sure to call _locked() support routines, etc, if already holding the socket buffer lock. - Break out sbinsertoob() into sbinsertoob_locked() that asserts the socket buffer lock. sbinsertoob() unconditionally locks the socket buffer before calling sbinsertoob_locked(). - Break out sbflush() into sbflush_locked() that asserts the socket buffer lock. sbflush() unconditionally locks the socket buffer before calling sbflush_locked(). Update panic strings for new function names. - Break out sbdrop() into sbdrop_locked() that asserts the socket buffer lock. sbdrop() unconditionally locks the socket buffer before calling sbdrop_locked(). - Break out sbdroprecord() into sbdroprecord_locked() that asserts the socket buffer lock. sbdroprecord() unconditionally locks the socket buffer before calling sbdroprecord_locked(). - sofree() now calls socantsendmore_locked() and re-acquires the socket buffer lock on return. It also now calls sbrelease_locked(). - sorflush() now calls socantrcvmore_locked() and re-acquires the socket buffer lock on return. Clean up/mess up other behavior in sorflush() relating to the temporary stack copy of the socket buffer used with dom_dispose by more properly initializing the temporary copy, and selectively bzeroing/copying more carefully to prevent WITNESS from getting confused by improperly initialized mutexes. Annotate why that's necessary, or at least, needed. - soisconnected() now calls sbdrop_locked() before unlocking the socket buffer to avoid locking overhead. Some parts of this change were: Submitted by: sam Sponsored by: FreeBSD Foundation Obtained from: BSD/OS
2004-06-21 00:20:43 +00:00
bzero(&asb, offsetof(struct sockbuf, sb_startzero));
bcopy(&sb->sb_startzero, &asb.sb_startzero,
sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
bzero(&sb->sb_startzero,
sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
SOCKBUF_UNLOCK(sb);
Merge next step in socket buffer locking: - sowakeup() now asserts the socket buffer lock on entry. Move the call to KNOTE higher in sowakeup() so that it is made with the socket buffer lock held for consistency with other calls. Release the socket buffer lock prior to calling into pgsigio(), so_upcall(), or aio_swake(). Locking for this event management will need revisiting in the future, but this model avoids lock order reversals when upcalls into other subsystems result in socket/socket buffer operations. Assert that the socket buffer lock is not held at the end of the function. - Wrapper macros for sowakeup(), sorwakeup() and sowwakeup(), now have _locked versions which assert the socket buffer lock on entry. If a wakeup is required by sb_notify(), invoke sowakeup(); otherwise, unconditionally release the socket buffer lock. This results in the socket buffer lock being released whether a wakeup is required or not. - Break out socantsendmore() into socantsendmore_locked() that asserts the socket buffer lock. socantsendmore() unconditionally locks the socket buffer before calling socantsendmore_locked(). Note that both functions return with the socket buffer unlocked as socantsendmore_locked() calls sowwakeup_locked() which has the same properties. Assert that the socket buffer is unlocked on return. - Break out socantrcvmore() into socantrcvmore_locked() that asserts the socket buffer lock. socantrcvmore() unconditionally locks the socket buffer before calling socantrcvmore_locked(). Note that both functions return with the socket buffer unlocked as socantrcvmore_locked() calls sorwakeup_locked() which has similar properties. Assert that the socket buffer is unlocked on return. - Break out sbrelease() into a sbrelease_locked() that asserts the socket buffer lock. sbrelease() unconditionally locks the socket buffer before calling sbrelease_locked(). sbrelease_locked() now invokes sbflush_locked() instead of sbflush(). - Assert the socket buffer lock in socket buffer sanity check functions sblastrecordchk(), sblastmbufchk(). - Assert the socket buffer lock in SBLINKRECORD(). - Break out various sbappend() functions into sbappend_locked() (and variations on that name) that assert the socket buffer lock. The !_locked() variations unconditionally lock the socket buffer before calling their _locked counterparts. Internally, make sure to call _locked() support routines, etc, if already holding the socket buffer lock. - Break out sbinsertoob() into sbinsertoob_locked() that asserts the socket buffer lock. sbinsertoob() unconditionally locks the socket buffer before calling sbinsertoob_locked(). - Break out sbflush() into sbflush_locked() that asserts the socket buffer lock. sbflush() unconditionally locks the socket buffer before calling sbflush_locked(). Update panic strings for new function names. - Break out sbdrop() into sbdrop_locked() that asserts the socket buffer lock. sbdrop() unconditionally locks the socket buffer before calling sbdrop_locked(). - Break out sbdroprecord() into sbdroprecord_locked() that asserts the socket buffer lock. sbdroprecord() unconditionally locks the socket buffer before calling sbdroprecord_locked(). - sofree() now calls socantsendmore_locked() and re-acquires the socket buffer lock on return. It also now calls sbrelease_locked(). - sorflush() now calls socantrcvmore_locked() and re-acquires the socket buffer lock on return. Clean up/mess up other behavior in sorflush() relating to the temporary stack copy of the socket buffer used with dom_dispose by more properly initializing the temporary copy, and selectively bzeroing/copying more carefully to prevent WITNESS from getting confused by improperly initialized mutexes. Annotate why that's necessary, or at least, needed. - soisconnected() now calls sbdrop_locked() before unlocking the socket buffer to avoid locking overhead. Some parts of this change were: Submitted by: sam Sponsored by: FreeBSD Foundation Obtained from: BSD/OS
2004-06-21 00:20:43 +00:00
SOCKBUF_LOCK_INIT(&asb, "so_rcv");
if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1994-05-24 10:09:53 +00:00
(*pr->pr_domain->dom_dispose)(asb.sb_mb);
sbrelease(&asb, so);
Merge next step in socket buffer locking: - sowakeup() now asserts the socket buffer lock on entry. Move the call to KNOTE higher in sowakeup() so that it is made with the socket buffer lock held for consistency with other calls. Release the socket buffer lock prior to calling into pgsigio(), so_upcall(), or aio_swake(). Locking for this event management will need revisiting in the future, but this model avoids lock order reversals when upcalls into other subsystems result in socket/socket buffer operations. Assert that the socket buffer lock is not held at the end of the function. - Wrapper macros for sowakeup(), sorwakeup() and sowwakeup(), now have _locked versions which assert the socket buffer lock on entry. If a wakeup is required by sb_notify(), invoke sowakeup(); otherwise, unconditionally release the socket buffer lock. This results in the socket buffer lock being released whether a wakeup is required or not. - Break out socantsendmore() into socantsendmore_locked() that asserts the socket buffer lock. socantsendmore() unconditionally locks the socket buffer before calling socantsendmore_locked(). Note that both functions return with the socket buffer unlocked as socantsendmore_locked() calls sowwakeup_locked() which has the same properties. Assert that the socket buffer is unlocked on return. - Break out socantrcvmore() into socantrcvmore_locked() that asserts the socket buffer lock. socantrcvmore() unconditionally locks the socket buffer before calling socantrcvmore_locked(). Note that both functions return with the socket buffer unlocked as socantrcvmore_locked() calls sorwakeup_locked() which has similar properties. Assert that the socket buffer is unlocked on return. - Break out sbrelease() into a sbrelease_locked() that asserts the socket buffer lock. sbrelease() unconditionally locks the socket buffer before calling sbrelease_locked(). sbrelease_locked() now invokes sbflush_locked() instead of sbflush(). - Assert the socket buffer lock in socket buffer sanity check functions sblastrecordchk(), sblastmbufchk(). - Assert the socket buffer lock in SBLINKRECORD(). - Break out various sbappend() functions into sbappend_locked() (and variations on that name) that assert the socket buffer lock. The !_locked() variations unconditionally lock the socket buffer before calling their _locked counterparts. Internally, make sure to call _locked() support routines, etc, if already holding the socket buffer lock. - Break out sbinsertoob() into sbinsertoob_locked() that asserts the socket buffer lock. sbinsertoob() unconditionally locks the socket buffer before calling sbinsertoob_locked(). - Break out sbflush() into sbflush_locked() that asserts the socket buffer lock. sbflush() unconditionally locks the socket buffer before calling sbflush_locked(). Update panic strings for new function names. - Break out sbdrop() into sbdrop_locked() that asserts the socket buffer lock. sbdrop() unconditionally locks the socket buffer before calling sbdrop_locked(). - Break out sbdroprecord() into sbdroprecord_locked() that asserts the socket buffer lock. sbdroprecord() unconditionally locks the socket buffer before calling sbdroprecord_locked(). - sofree() now calls socantsendmore_locked() and re-acquires the socket buffer lock on return. It also now calls sbrelease_locked(). - sorflush() now calls socantrcvmore_locked() and re-acquires the socket buffer lock on return. Clean up/mess up other behavior in sorflush() relating to the temporary stack copy of the socket buffer used with dom_dispose by more properly initializing the temporary copy, and selectively bzeroing/copying more carefully to prevent WITNESS from getting confused by improperly initialized mutexes. Annotate why that's necessary, or at least, needed. - soisconnected() now calls sbdrop_locked() before unlocking the socket buffer to avoid locking overhead. Some parts of this change were: Submitted by: sam Sponsored by: FreeBSD Foundation Obtained from: BSD/OS
2004-06-21 00:20:43 +00:00
SOCKBUF_LOCK_DESTROY(&asb);
1994-05-24 10:09:53 +00:00
}
#ifdef INET
static int
do_setopt_accept_filter(so, sopt)
struct socket *so;
struct sockopt *sopt;
{
struct accept_filter_arg *afap = NULL;
struct accept_filter *afp;
struct so_accf *af = so->so_accf;
int error = 0;
/* do not set/remove accept filters on non listen sockets */
if ((so->so_options & SO_ACCEPTCONN) == 0) {
error = EINVAL;
goto out;
}
/* removing the filter */
if (sopt == NULL) {
if (af != NULL) {
if (af->so_accept_filter != NULL &&
af->so_accept_filter->accf_destroy != NULL) {
af->so_accept_filter->accf_destroy(so);
}
if (af->so_accept_filter_str != NULL) {
FREE(af->so_accept_filter_str, M_ACCF);
}
FREE(af, M_ACCF);
so->so_accf = NULL;
}
so->so_options &= ~SO_ACCEPTFILTER;
return (0);
}
/* adding a filter */
/* must remove previous filter first */
if (af != NULL) {
error = EINVAL;
goto out;
}
/* don't put large objects on the kernel stack */
MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
afap->af_name[sizeof(afap->af_name)-1] = '\0';
afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
if (error)
goto out;
afp = accept_filt_get(afap->af_name);
if (afp == NULL) {
error = ENOENT;
goto out;
}
MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
if (afp->accf_create != NULL) {
if (afap->af_name[0] != '\0') {
int len = strlen(afap->af_name) + 1;
MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
strcpy(af->so_accept_filter_str, afap->af_name);
}
af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
if (af->so_accept_filter_arg == NULL) {
FREE(af->so_accept_filter_str, M_ACCF);
FREE(af, M_ACCF);
so->so_accf = NULL;
error = EINVAL;
goto out;
}
}
af->so_accept_filter = afp;
so->so_accf = af;
so->so_options |= SO_ACCEPTFILTER;
out:
if (afap != NULL)
FREE(afap, M_TEMP);
return (error);
}
#endif /* INET */
/*
* Perhaps this routine, and sooptcopyout(), below, ought to come in
* an additional variant to handle the case where the option value needs
* to be some kind of integer, but not a specific size.
* In addition to their use here, these functions are also called by the
* protocol-level pr_ctloutput() routines.
*/
int
sooptcopyin(sopt, buf, len, minlen)
struct sockopt *sopt;
void *buf;
size_t len;
size_t minlen;
1994-05-24 10:09:53 +00:00
{
size_t valsize;
/*
* If the user gives us more than we wanted, we ignore it,
* but if we don't get the minimum length the caller
* wants, we return EINVAL. On success, sopt->sopt_valsize
* is set to however much we actually retrieved.
*/
if ((valsize = sopt->sopt_valsize) < minlen)
return EINVAL;
if (valsize > len)
sopt->sopt_valsize = valsize = len;
if (sopt->sopt_td != NULL)
return (copyin(sopt->sopt_val, buf, valsize));
1994-05-24 10:09:53 +00:00
bcopy(sopt->sopt_val, buf, valsize);
return 0;
}
int
sosetopt(so, sopt)
struct socket *so;
struct sockopt *sopt;
{
int error, optval;
struct linger l;
struct timeval tv;
u_long val;
#ifdef MAC
struct mac extmac;
#endif
error = 0;
if (sopt->sopt_level != SOL_SOCKET) {
1994-05-24 10:09:53 +00:00
if (so->so_proto && so->so_proto->pr_ctloutput)
return ((*so->so_proto->pr_ctloutput)
(so, sopt));
1994-05-24 10:09:53 +00:00
error = ENOPROTOOPT;
} else {
switch (sopt->sopt_name) {
#ifdef INET
case SO_ACCEPTFILTER:
error = do_setopt_accept_filter(so, sopt);
if (error)
goto bad;
break;
#endif
1994-05-24 10:09:53 +00:00
case SO_LINGER:
error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
if (error)
1994-05-24 10:09:53 +00:00
goto bad;
SOCK_LOCK(so);
so->so_linger = l.l_linger;
if (l.l_onoff)
so->so_options |= SO_LINGER;
else
so->so_options &= ~SO_LINGER;
SOCK_UNLOCK(so);
break;
1994-05-24 10:09:53 +00:00
case SO_DEBUG:
case SO_KEEPALIVE:
case SO_DONTROUTE:
case SO_USELOOPBACK:
case SO_BROADCAST:
case SO_REUSEADDR:
case SO_REUSEPORT:
case SO_OOBINLINE:
case SO_TIMESTAMP:
case SO_BINTIME:
case SO_NOSIGPIPE:
error = sooptcopyin(sopt, &optval, sizeof optval,
sizeof optval);
if (error)
1994-05-24 10:09:53 +00:00
goto bad;
SOCK_LOCK(so);
if (optval)
so->so_options |= sopt->sopt_name;
1994-05-24 10:09:53 +00:00
else
so->so_options &= ~sopt->sopt_name;
SOCK_UNLOCK(so);
1994-05-24 10:09:53 +00:00
break;
case SO_SNDBUF:
case SO_RCVBUF:
case SO_SNDLOWAT:
case SO_RCVLOWAT:
error = sooptcopyin(sopt, &optval, sizeof optval,
sizeof optval);
if (error)
1994-05-24 10:09:53 +00:00
goto bad;
/*
* Values < 1 make no sense for any of these
* options, so disallow them.
*/
if (optval < 1) {
error = EINVAL;
goto bad;
}
switch (sopt->sopt_name) {
1994-05-24 10:09:53 +00:00
case SO_SNDBUF:
case SO_RCVBUF:
if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
&so->so_snd : &so->so_rcv, (u_long)optval,
so, curthread) == 0) {
1994-05-24 10:09:53 +00:00
error = ENOBUFS;
goto bad;
}
break;
/*
* Make sure the low-water is never greater than
* the high-water.
*/
1994-05-24 10:09:53 +00:00
case SO_SNDLOWAT:
SOCKBUF_LOCK(&so->so_snd);
so->so_snd.sb_lowat =
(optval > so->so_snd.sb_hiwat) ?
so->so_snd.sb_hiwat : optval;
SOCKBUF_UNLOCK(&so->so_snd);
1994-05-24 10:09:53 +00:00
break;
case SO_RCVLOWAT:
SOCKBUF_LOCK(&so->so_rcv);
so->so_rcv.sb_lowat =
(optval > so->so_rcv.sb_hiwat) ?
so->so_rcv.sb_hiwat : optval;
SOCKBUF_UNLOCK(&so->so_rcv);
1994-05-24 10:09:53 +00:00
break;
}
break;
case SO_SNDTIMEO:
case SO_RCVTIMEO:
error = sooptcopyin(sopt, &tv, sizeof tv,
sizeof tv);
if (error)
1994-05-24 10:09:53 +00:00
goto bad;
/* assert(hz > 0); */
if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
error = EDOM;
goto bad;
}
/* assert(tick > 0); */
/* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
if (val > SHRT_MAX) {
1994-05-24 10:09:53 +00:00
error = EDOM;
goto bad;
}
if (val == 0 && tv.tv_usec != 0)
val = 1;
1994-05-24 10:09:53 +00:00
switch (sopt->sopt_name) {
1994-05-24 10:09:53 +00:00
case SO_SNDTIMEO:
so->so_snd.sb_timeo = val;
break;
case SO_RCVTIMEO:
so->so_rcv.sb_timeo = val;
break;
}
break;
case SO_LABEL:
#ifdef MAC
error = sooptcopyin(sopt, &extmac, sizeof extmac,
sizeof extmac);
if (error)
goto bad;
error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
so, &extmac);
#else
error = EOPNOTSUPP;
#endif
break;
1994-05-24 10:09:53 +00:00
default:
error = ENOPROTOOPT;
break;
}
if (error == 0 && so->so_proto != NULL &&
so->so_proto->pr_ctloutput != NULL) {
1994-05-24 10:09:53 +00:00
(void) ((*so->so_proto->pr_ctloutput)
(so, sopt));
1994-05-24 10:09:53 +00:00
}
}
bad:
return (error);
}
/* Helper routine for getsockopt */
int
sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1994-05-24 10:09:53 +00:00
{
int error;
size_t valsize;
error = 0;
/*
* Documented get behavior is that we always return a value,
* possibly truncated to fit in the user's buffer.
* Traditional behavior is that we always tell the user
* precisely how much we copied, rather than something useful
* like the total amount we had available for her.
* Note that this interface is not idempotent; the entire answer must
* generated ahead of time.
*/
valsize = min(len, sopt->sopt_valsize);
sopt->sopt_valsize = valsize;
if (sopt->sopt_val != NULL) {
if (sopt->sopt_td != NULL)
error = copyout(buf, sopt->sopt_val, valsize);
else
bcopy(buf, sopt->sopt_val, valsize);
}
return error;
}
int
sogetopt(so, sopt)
struct socket *so;
struct sockopt *sopt;
{
int error, optval;
struct linger l;
struct timeval tv;
#ifdef INET
struct accept_filter_arg *afap;
#endif
#ifdef MAC
struct mac extmac;
#endif
1994-05-24 10:09:53 +00:00
error = 0;
if (sopt->sopt_level != SOL_SOCKET) {
1994-05-24 10:09:53 +00:00
if (so->so_proto && so->so_proto->pr_ctloutput) {
return ((*so->so_proto->pr_ctloutput)
(so, sopt));
1994-05-24 10:09:53 +00:00
} else
return (ENOPROTOOPT);
} else {
switch (sopt->sopt_name) {
#ifdef INET
case SO_ACCEPTFILTER:
if ((so->so_options & SO_ACCEPTCONN) == 0)
return (EINVAL);
MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
M_TEMP, M_WAITOK | M_ZERO);
if ((so->so_options & SO_ACCEPTFILTER) != 0) {
strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
if (so->so_accf->so_accept_filter_str != NULL)
strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
}
error = sooptcopyout(sopt, afap, sizeof(*afap));
FREE(afap, M_TEMP);
break;
#endif
1994-05-24 10:09:53 +00:00
case SO_LINGER:
/*
* XXXRW: We grab the lock here to get a consistent
* snapshot of both fields. This may not really
* be necessary.
*/
SOCK_LOCK(so);
l.l_onoff = so->so_options & SO_LINGER;
l.l_linger = so->so_linger;
SOCK_UNLOCK(so);
error = sooptcopyout(sopt, &l, sizeof l);
1994-05-24 10:09:53 +00:00
break;
case SO_USELOOPBACK:
case SO_DONTROUTE:
case SO_DEBUG:
case SO_KEEPALIVE:
case SO_REUSEADDR:
case SO_REUSEPORT:
case SO_BROADCAST:
case SO_OOBINLINE:
case SO_TIMESTAMP:
case SO_BINTIME:
case SO_NOSIGPIPE:
optval = so->so_options & sopt->sopt_name;
integer:
error = sooptcopyout(sopt, &optval, sizeof optval);
1994-05-24 10:09:53 +00:00
break;
case SO_TYPE:
optval = so->so_type;
goto integer;
1994-05-24 10:09:53 +00:00
case SO_ERROR:
optval = so->so_error;
1994-05-24 10:09:53 +00:00
so->so_error = 0;
goto integer;
1994-05-24 10:09:53 +00:00
case SO_SNDBUF:
optval = so->so_snd.sb_hiwat;
goto integer;
1994-05-24 10:09:53 +00:00
case SO_RCVBUF:
optval = so->so_rcv.sb_hiwat;
goto integer;
1994-05-24 10:09:53 +00:00
case SO_SNDLOWAT:
optval = so->so_snd.sb_lowat;
goto integer;
1994-05-24 10:09:53 +00:00
case SO_RCVLOWAT:
optval = so->so_rcv.sb_lowat;
goto integer;
1994-05-24 10:09:53 +00:00
case SO_SNDTIMEO:
case SO_RCVTIMEO:
optval = (sopt->sopt_name == SO_SNDTIMEO ?
so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
tv.tv_sec = optval / hz;
tv.tv_usec = (optval % hz) * tick;
error = sooptcopyout(sopt, &tv, sizeof tv);
break;
case SO_LABEL:
#ifdef MAC
error = sooptcopyin(sopt, &extmac, sizeof(extmac),
sizeof(extmac));
if (error)
return (error);
error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
so, &extmac);
if (error)
return (error);
error = sooptcopyout(sopt, &extmac, sizeof extmac);
#else
error = EOPNOTSUPP;
#endif
break;
case SO_PEERLABEL:
#ifdef MAC
error = sooptcopyin(sopt, &extmac, sizeof(extmac),
sizeof(extmac));
if (error)
return (error);
error = mac_getsockopt_peerlabel(
sopt->sopt_td->td_ucred, so, &extmac);
if (error)
return (error);
error = sooptcopyout(sopt, &extmac, sizeof extmac);
#else
error = EOPNOTSUPP;
#endif
break;
1994-05-24 10:09:53 +00:00
default:
error = ENOPROTOOPT;
break;
1994-05-24 10:09:53 +00:00
}
return (error);
1994-05-24 10:09:53 +00:00
}
}
/* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
int
soopt_getm(struct sockopt *sopt, struct mbuf **mp)
{
struct mbuf *m, *m_prev;
int sopt_size = sopt->sopt_valsize;
MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
if (m == NULL)
return ENOBUFS;
if (sopt_size > MLEN) {
MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT);
if ((m->m_flags & M_EXT) == 0) {
m_free(m);
return ENOBUFS;
}
m->m_len = min(MCLBYTES, sopt_size);
} else {
m->m_len = min(MLEN, sopt_size);
}
sopt_size -= m->m_len;
*mp = m;
m_prev = m;
while (sopt_size) {
MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
if (m == NULL) {
m_freem(*mp);
return ENOBUFS;
}
if (sopt_size > MLEN) {
MCLGET(m, sopt->sopt_td != NULL ? M_TRYWAIT :
M_DONTWAIT);
if ((m->m_flags & M_EXT) == 0) {
m_freem(m);
m_freem(*mp);
return ENOBUFS;
}
m->m_len = min(MCLBYTES, sopt_size);
} else {
m->m_len = min(MLEN, sopt_size);
}
sopt_size -= m->m_len;
m_prev->m_next = m;
m_prev = m;
}
return 0;
}
/* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
int
soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
{
struct mbuf *m0 = m;
if (sopt->sopt_val == NULL)
return 0;
while (m != NULL && sopt->sopt_valsize >= m->m_len) {
if (sopt->sopt_td != NULL) {
int error;
error = copyin(sopt->sopt_val, mtod(m, char *),
m->m_len);
if (error != 0) {
m_freem(m0);
return(error);
}
} else
bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
sopt->sopt_valsize -= m->m_len;
sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
m = m->m_next;
}
if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
panic("ip6_sooptmcopyin");
return 0;
}
/* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
int
soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
{
struct mbuf *m0 = m;
size_t valsize = 0;
if (sopt->sopt_val == NULL)
return 0;
while (m != NULL && sopt->sopt_valsize >= m->m_len) {
if (sopt->sopt_td != NULL) {
int error;
error = copyout(mtod(m, char *), sopt->sopt_val,
m->m_len);
if (error != 0) {
m_freem(m0);
return(error);
}
} else
bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
sopt->sopt_valsize -= m->m_len;
sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
valsize += m->m_len;
m = m->m_next;
}
if (m != NULL) {
/* enough soopt buffer should be given from user-land */
m_freem(m0);
return(EINVAL);
}
sopt->sopt_valsize = valsize;
return 0;
}
void
1994-05-24 10:09:53 +00:00
sohasoutofband(so)
struct socket *so;
1994-05-24 10:09:53 +00:00
{
if (so->so_sigio != NULL)
pgsigio(&so->so_sigio, SIGURG, 0);
selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
1994-05-24 10:09:53 +00:00
}
int
Make similar changes to fo_stat() and fo_poll() as made earlier to fo_read() and fo_write(): explicitly use the cred argument to fo_poll() as "active_cred" using the passed file descriptor's f_cred reference to provide access to the file credential. Add an active_cred argument to fo_stat() so that implementers have access to the active credential as well as the file credential. Generally modify callers of fo_stat() to pass in td->td_ucred rather than fp->f_cred, which was redundantly provided via the fp argument. This set of modifications also permits threads to perform these operations on behalf of another thread without modifying their credential. Trickle this change down into fo_stat/poll() implementations: - badfo_poll(), badfo_stat(): modify/add arguments. - kqueue_poll(), kqueue_stat(): modify arguments. - pipe_poll(), pipe_stat(): modify/add arguments, pass active_cred to MAC checks rather than td->td_ucred. - soo_poll(), soo_stat(): modify/add arguments, pass fp->f_cred rather than cred to pru_sopoll() to maintain current semantics. - sopoll(): moidfy arguments. - vn_poll(), vn_statfile(): modify/add arguments, pass new arguments to vn_stat(). Pass active_cred to MAC and fp->f_cred to VOP_POLL() to maintian current semantics. - vn_close(): rename cred to file_cred to reflect reality while I'm here. - vn_stat(): Add active_cred and file_cred arguments to vn_stat() and consumers so that this distinction is maintained at the VFS as well as 'struct file' layer. Pass active_cred instead of td->td_ucred to MAC and to VOP_GETATTR() to maintain current semantics. - fifofs: modify the creation of a "filetemp" so that the file credential is properly initialized and can be used in the socket code if desired. Pass ap->a_td->td_ucred as the active credential to soo_poll(). If we teach the vnop interface about the distinction between file and active credentials, we would use the active credential here. Note that current inconsistent passing of active_cred vs. file_cred to VOP's is maintained. It's not clear why GETATTR would be authorized using active_cred while POLL would be authorized using file_cred at the file system level. Obtained from: TrustedBSD Project Sponsored by: DARPA, NAI Labs
2002-08-16 12:52:03 +00:00
sopoll(struct socket *so, int events, struct ucred *active_cred,
struct thread *td)
{
int revents = 0;
if (events & (POLLIN | POLLRDNORM))
if (soreadable(so))
revents |= events & (POLLIN | POLLRDNORM);
if (events & POLLINIGNEOF)
if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
!TAILQ_EMPTY(&so->so_comp) || so->so_error)
revents |= POLLINIGNEOF;
if (events & (POLLOUT | POLLWRNORM))
if (sowriteable(so))
revents |= events & (POLLOUT | POLLWRNORM);
if (events & (POLLPRI | POLLRDBAND))
if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
revents |= events & (POLLPRI | POLLRDBAND);
if (revents == 0) {
if (events &
(POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
POLLRDBAND)) {
SOCKBUF_LOCK(&so->so_rcv);
selrecord(td, &so->so_rcv.sb_sel);
so->so_rcv.sb_flags |= SB_SEL;
SOCKBUF_UNLOCK(&so->so_rcv);
}
if (events & (POLLOUT | POLLWRNORM)) {
SOCKBUF_LOCK(&so->so_snd);
selrecord(td, &so->so_snd.sb_sel);
so->so_snd.sb_flags |= SB_SEL;
SOCKBUF_UNLOCK(&so->so_snd);
}
}
return (revents);
}
int
soo_kqfilter(struct file *fp, struct knote *kn)
{
struct socket *so = kn->kn_fp->f_data;
struct sockbuf *sb;
switch (kn->kn_filter) {
case EVFILT_READ:
if (so->so_options & SO_ACCEPTCONN)
kn->kn_fop = &solisten_filtops;
else
kn->kn_fop = &soread_filtops;
sb = &so->so_rcv;
break;
case EVFILT_WRITE:
kn->kn_fop = &sowrite_filtops;
sb = &so->so_snd;
break;
default:
return (1);
}
SOCKBUF_LOCK(sb);
SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext);
sb->sb_flags |= SB_KNOTE;
SOCKBUF_UNLOCK(sb);
return (0);
}
static void
filt_sordetach(struct knote *kn)
{
struct socket *so = kn->kn_fp->f_data;
SOCKBUF_LOCK(&so->so_rcv);
SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext);
if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note))
so->so_rcv.sb_flags &= ~SB_KNOTE;
SOCKBUF_UNLOCK(&so->so_rcv);
}
/*ARGSUSED*/
static int
filt_soread(struct knote *kn, long hint)
{
struct socket *so = kn->kn_fp->f_data;
int need_lock, result;
/*
* XXXRW: Conditional locking because filt_soread() can be called
* either from KNOTE() in the socket context where the socket buffer
* lock is already held, or from kqueue() itself.
*/
need_lock = !SOCKBUF_OWNED(&so->so_rcv);
if (need_lock)
SOCKBUF_LOCK(&so->so_rcv);
kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
kn->kn_flags |= EV_EOF;
kn->kn_fflags = so->so_error;
result = 1;
} else if (so->so_error) /* temporary udp error */
result = 1;
else if (kn->kn_sfflags & NOTE_LOWAT)
result = (kn->kn_data >= kn->kn_sdata);
else
result = (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
if (need_lock)
SOCKBUF_UNLOCK(&so->so_rcv);
return (result);
}
static void
filt_sowdetach(struct knote *kn)
{
struct socket *so = kn->kn_fp->f_data;
SOCKBUF_LOCK(&so->so_snd);
SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext);
if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note))
so->so_snd.sb_flags &= ~SB_KNOTE;
SOCKBUF_UNLOCK(&so->so_snd);
}
/*ARGSUSED*/
static int
filt_sowrite(struct knote *kn, long hint)
{
struct socket *so = kn->kn_fp->f_data;
int need_lock, result;
/*
* XXXRW: Conditional locking because filt_soread() can be called
* either from KNOTE() in the socket context where the socket buffer
* lock is already held, or from kqueue() itself.
*/
need_lock = !SOCKBUF_OWNED(&so->so_snd);
if (need_lock)
SOCKBUF_LOCK(&so->so_snd);
kn->kn_data = sbspace(&so->so_snd);
if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
kn->kn_flags |= EV_EOF;
kn->kn_fflags = so->so_error;
result = 1;
} else if (so->so_error) /* temporary udp error */
result = 1;
else if (((so->so_state & SS_ISCONNECTED) == 0) &&
(so->so_proto->pr_flags & PR_CONNREQUIRED))
result = 0;
else if (kn->kn_sfflags & NOTE_LOWAT)
result = (kn->kn_data >= kn->kn_sdata);
else
result = (kn->kn_data >= so->so_snd.sb_lowat);
if (need_lock)
SOCKBUF_UNLOCK(&so->so_snd);
return (result);
}
/*ARGSUSED*/
static int
filt_solisten(struct knote *kn, long hint)
{
struct socket *so = kn->kn_fp->f_data;
kn->kn_data = so->so_qlen;
return (! TAILQ_EMPTY(&so->so_comp));
}
int
socheckuid(struct socket *so, uid_t uid)
{
if (so == NULL)
return (EPERM);
if (so->so_cred->cr_uid == uid)
return (0);
return (EPERM);
}