d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
freebsd-skq/sys/sys/sf_buf.h
glebius aaa09777e1 New sendfile(2) syscall. A joint effort of NGINX and Netflix from 2013 and 
up to now.

The new sendfile is the code that Netflix uses to send their multiple tens
of gigabits of data per second. The new implementation features asynchronous
I/O, when I/O operations are launched, but not awaited to be complete. An
explanation of why such behavior is beneficial compared to old one is
going to be too long for a commit message, so we will skip it here.

Additional features of new syscall are extra flags, which provide an
application more control over data sent. The SF_NOCACHE flag tells
kernel that data shouldn't be cached after it was sent. The SF_READAHEAD()
macro allows to specify readahead size in pages.

The new syscalls is a drop in replacement. No modifications are required
to applications. One can take nginx binary for stable/10 and run it
successfully on head. Although SF_NODISKIO lost its original sense, as now
sendfile doesn't block, and now means something completely different (tm),
using the new sendfile the old way is absolutely safe.

Celebrates:	Netflix global launch!
Sponsored by:	Nginx, Inc.
Sponsored by:	Netflix
Relnotes:	yes
2016-01-08 20:34:57 +00:00

204 lines
6.1 KiB
C
Raw Blame History

/*-
* Copyright (c) 2014 Gleb Smirnoff <glebius@FreeBSD.org>
* Copyright (c) 2003-2004 Alan L. Cox <alc@cs.rice.edu>
* 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.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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.
*
* $FreeBSD$
*/
#ifndef _SYS_SF_BUF_H_
#define _SYS_SF_BUF_H_
struct sfstat { /* sendfile statistics */
uint64_t sf_syscalls; /* times sendfile was called */
uint64_t sf_noiocnt; /* times sendfile didn't require I/O */
uint64_t sf_iocnt; /* times sendfile had to do disk I/O */
uint64_t sf_pages_read; /* pages read as part of a request */
uint64_t sf_pages_valid; /* pages were valid for a request */
uint64_t sf_rhpages_requested; /* readahead pages requested */
uint64_t sf_rhpages_read; /* readahead pages read */
uint64_t sf_busy; /* times aborted on a busy page */
uint64_t sf_allocfail; /* times sfbuf allocation failed */
uint64_t sf_allocwait; /* times sfbuf allocation had to wait */
};
#ifdef _KERNEL
#include <sys/types.h>
#include <sys/systm.h>
#include <sys/counter.h>
#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/vm_page.h>
/*
* Sf_bufs, or sendfile(2) buffers provide a vm_page that is mapped
* into kernel address space. Note, that they aren't used only
* by sendfile(2)!
*
* Sf_bufs could be implemented as a feature of vm_page_t, but that
* would require growth of the structure. That's why they are implemented
* as a separate hash indexed by vm_page address. Implementation lives in
* kern/subr_sfbuf.c. Meanwhile, most 64-bit machines have a physical map,
* so they don't require this hash at all, thus ignore subr_sfbuf.c.
*
* Different 32-bit architectures demand different requirements on sf_buf
* hash and functions. They request features in machine/vmparam.h, which
* enable parts of this file. They can also optionally provide helpers in
* machine/sf_buf.h
*
* Defines are:
* SFBUF This machine requires sf_buf hash.
* subr_sfbuf.c should be compiled.
* SFBUF_CPUSET This machine can perform SFB_CPUPRIVATE mappings,
* that do no invalidate cache on the rest of CPUs.
* SFBUF_NOMD This machine doesn't have machine/sf_buf.h
*
* SFBUF_OPTIONAL_DIRECT_MAP Value of this define is used as boolean
* variable that tells whether machine is
* capable of direct map or not at runtime.
* SFBUF_MAP This machine provides its own sf_buf_map() and
* sf_buf_unmap().
* SFBUF_PROCESS_PAGE This machine provides sf_buf_process_page()
* function.
*/
#ifdef SFBUF
#if defined(SMP) && defined(SFBUF_CPUSET)
#include <sys/_cpuset.h>
#endif
#include <sys/queue.h>
struct sf_buf {
LIST_ENTRY(sf_buf) list_entry; /* list of buffers */
TAILQ_ENTRY(sf_buf) free_entry; /* list of buffers */
vm_page_t m; /* currently mapped page */
vm_offset_t kva; /* va of mapping */
int ref_count; /* usage of this mapping */
#if defined(SMP) && defined(SFBUF_CPUSET)
cpuset_t cpumask; /* where mapping is valid */
#endif
};
#else /* ! SFBUF */
struct sf_buf;
#endif /* SFBUF */
#ifndef SFBUF_NOMD
#include <machine/sf_buf.h>
#endif
#ifdef SFBUF_OPTIONAL_DIRECT_MAP
#include <machine/md_var.h>
#endif
#ifdef SFBUF
struct sf_buf *sf_buf_alloc(struct vm_page *, int);
void sf_buf_free(struct sf_buf *);
void sf_buf_ref(struct sf_buf *);
static inline vm_offset_t
sf_buf_kva(struct sf_buf *sf)
{
#ifdef SFBUF_OPTIONAL_DIRECT_MAP
if (SFBUF_OPTIONAL_DIRECT_MAP)
return (SFBUF_PHYS_DMAP(VM_PAGE_TO_PHYS((vm_page_t)sf)));
#endif
return (sf->kva);
}
static inline vm_page_t
sf_buf_page(struct sf_buf *sf)
{
#ifdef SFBUF_OPTIONAL_DIRECT_MAP
if (SFBUF_OPTIONAL_DIRECT_MAP)
return ((vm_page_t)sf);
#endif
return (sf->m);
}
#ifndef SFBUF_MAP
#include <vm/pmap.h>
static inline void
sf_buf_map(struct sf_buf *sf, int flags)
{
pmap_qenter(sf->kva, &sf->m, 1);
}
static inline int
sf_buf_unmap(struct sf_buf *sf)
{
return (0);
}
#endif /* SFBUF_MAP */
#if defined(SMP) && defined(SFBUF_CPUSET)
void sf_buf_shootdown(struct sf_buf *, int);
#endif
#ifdef SFBUF_PROCESS_PAGE
boolean_t sf_buf_process_page(vm_page_t, void (*)(struct sf_buf *));
#endif
#else /* ! SFBUF */
static inline struct sf_buf *
sf_buf_alloc(struct vm_page *m, int pri)
{
return ((struct sf_buf *)m);
}
static inline void
sf_buf_free(struct sf_buf *sf)
{
}
static inline void
sf_buf_ref(struct sf_buf *sf)
{
}
#endif /* SFBUF */
/*
* Options to sf_buf_alloc() are specified through its flags argument. This
* argument's value should be the result of a bitwise or'ing of one or more
* of the following values.
*/
#define SFB_CATCH 1 /* Check signals if the allocation
sleeps. */
#define SFB_CPUPRIVATE 2 /* Create a CPU private mapping. */
#define SFB_DEFAULT 0
#define SFB_NOWAIT 4 /* Return NULL if all bufs are used. */
extern counter_u64_t sfstat[sizeof(struct sfstat) / sizeof(uint64_t)];
#define SFSTAT_ADD(name, val) \
counter_u64_add(sfstat[offsetof(struct sfstat, name) / sizeof(uint64_t)],\
(val))
#define SFSTAT_INC(name) SFSTAT_ADD(name, 1)
#endif /* _KERNEL */
#endif /* !_SYS_SF_BUF_H_ */
Reference in New Issue View Git Blame Copy Permalink