2 * Copyright (c) 2013-2015 Gleb Smirnoff <glebius@FreeBSD.org>
3 * Copyright (c) 1998, David Greenman. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
33 #include "opt_compat.h"
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/capsicum.h>
38 #include <sys/kernel.h>
40 #include <sys/mutex.h>
41 #include <sys/sysproto.h>
42 #include <sys/malloc.h>
45 #include <sys/mount.h>
47 #include <sys/protosw.h>
48 #include <sys/rwlock.h>
49 #include <sys/sf_buf.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/syscallsubr.h>
53 #include <sys/sysctl.h>
54 #include <sys/vnode.h>
58 #include <security/audit/audit.h>
59 #include <security/mac/mac_framework.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_pager.h>
66 * Structure describing a single sendfile(2) I/O, which may consist of
67 * several underlying pager I/Os.
69 * The syscall context allocates the structure and initializes 'nios'
70 * to 1. As sendfile_swapin() runs through pages and starts asynchronous
71 * paging operations, it increments 'nios'.
73 * Every I/O completion calls sendfile_iodone(), which decrements the 'nios',
74 * and the syscall also calls sendfile_iodone() after allocating all mbufs,
75 * linking them and sending to socket. Whoever reaches zero 'nios' is
76 * responsible to * call pru_ready on the socket, to notify it of readyness
89 * Structure used to track requests with SF_SYNC flag.
91 struct sendfile_sync
{
97 counter_u64_t sfstat
[sizeof(struct sfstat
) / sizeof(uint64_t)];
100 sfstat_init(const void *unused
)
103 COUNTER_ARRAY_ALLOC(sfstat
, sizeof(struct sfstat
) / sizeof(uint64_t),
106 SYSINIT(sfstat
, SI_SUB_MBUF
, SI_ORDER_FIRST
, sfstat_init
, NULL
);
109 sfstat_sysctl(SYSCTL_HANDLER_ARGS
)
113 COUNTER_ARRAY_COPY(sfstat
, &s
, sizeof(s
) / sizeof(uint64_t));
115 COUNTER_ARRAY_ZERO(sfstat
, sizeof(s
) / sizeof(uint64_t));
116 return (SYSCTL_OUT(req
, &s
, sizeof(s
)));
118 SYSCTL_PROC(_kern_ipc
, OID_AUTO
, sfstat
, CTLTYPE_OPAQUE
| CTLFLAG_RW
,
119 NULL
, 0, sfstat_sysctl
, "I", "sendfile statistics");
122 * Detach mapped page and release resources back to the system. Called
123 * by mbuf(9) code when last reference to a page is freed.
126 sf_ext_free(void *arg1
, void *arg2
)
128 struct sf_buf
*sf
= arg1
;
129 struct sendfile_sync
*sfs
= arg2
;
130 vm_page_t pg
= sf_buf_page(sf
);
136 * Check for the object going away on us. This can
137 * happen since we don't hold a reference to it.
138 * If so, we're responsible for freeing the page.
140 if (vm_page_unwire(pg
, PQ_INACTIVE
) && pg
->object
== NULL
)
146 KASSERT(sfs
->count
> 0, ("Sendfile sync botchup count == 0"));
147 if (--sfs
->count
== 0)
149 mtx_unlock(&sfs
->mtx
);
154 * Same as above, but forces the page to be detached from the object
155 * and go into free pool.
158 sf_ext_free_nocache(void *arg1
, void *arg2
)
160 struct sf_buf
*sf
= arg1
;
161 struct sendfile_sync
*sfs
= arg2
;
162 vm_page_t pg
= sf_buf_page(sf
);
167 if (vm_page_unwire(pg
, PQ_NONE
)) {
170 /* Try to free the page, but only if it is cheap to. */
171 if ((obj
= pg
->object
) == NULL
)
173 else if (!vm_page_xbusied(pg
) && VM_OBJECT_TRYWLOCK(obj
)) {
175 VM_OBJECT_WUNLOCK(obj
);
177 vm_page_deactivate(pg
);
183 KASSERT(sfs
->count
> 0, ("Sendfile sync botchup count == 0"));
184 if (--sfs
->count
== 0)
186 mtx_unlock(&sfs
->mtx
);
191 * Helper function to calculate how much data to put into page i of n.
192 * Only first and last pages are special.
195 xfsize(int i
, int n
, off_t off
, off_t len
)
199 return (omin(PAGE_SIZE
- (off
& PAGE_MASK
), len
));
201 if (i
== n
- 1 && ((off
+ len
) & PAGE_MASK
) > 0)
202 return ((off
+ len
) & PAGE_MASK
);
208 * Helper function to get offset within object for i page.
210 static inline vm_offset_t
211 vmoff(int i
, off_t off
)
215 return ((vm_offset_t
)off
);
217 return (trunc_page(off
+ i
* PAGE_SIZE
));
221 * Helper function used when allocation of a page or sf_buf failed.
222 * Pretend as if we don't have enough space, subtract xfsize() of
223 * all pages that failed.
226 fixspace(int old
, int new, off_t off
, int *space
)
229 KASSERT(old
> new, ("%s: old %d new %d", __func__
, old
, new));
231 /* Subtract last one. */
232 *space
-= xfsize(old
- 1, old
, off
, *space
);
236 /* There was only one page. */
239 /* Subtract first one. */
241 *space
-= xfsize(0, old
, off
, *space
);
245 /* Rest of pages are full sized. */
246 *space
-= (old
- new) * PAGE_SIZE
;
248 KASSERT(*space
>= 0, ("%s: space went backwards", __func__
));
252 * I/O completion callback.
255 sendfile_iodone(void *arg
, vm_page_t
*pg
, int count
, int error
)
257 struct sf_io
*sfio
= arg
;
260 for (int i
= 0; i
< count
; i
++)
261 vm_page_xunbusy(pg
[i
]);
266 if (!refcount_release(&sfio
->nios
))
269 so
= sfio
->sock_fp
->f_data
;
275 * I/O operation failed. The state of data in the socket
276 * is now inconsistent, and all what we can do is to tear
277 * it down. Protocol abort method would tear down protocol
278 * state, free all ready mbufs and detach not ready ones.
279 * We will free the mbufs corresponding to this I/O manually.
281 * The socket would be marked with EIO and made available
282 * for read, so that application receives EIO on next
283 * syscall and eventually closes the socket.
285 so
->so_proto
->pr_usrreqs
->pru_abort(so
);
289 for (int i
= 0; i
< sfio
->npages
; i
++)
292 CURVNET_SET(so
->so_vnet
);
293 (void )(so
->so_proto
->pr_usrreqs
->pru_ready
)(so
, sfio
->m
,
298 /* XXXGL: curthread */
299 fdrop(sfio
->sock_fp
, curthread
);
304 * Iterate through pages vector and request paging for non-valid pages.
307 sendfile_swapin(vm_object_t obj
, struct sf_io
*sfio
, off_t off
, off_t len
,
308 int npages
, int rhpages
, int flags
)
310 vm_page_t
*pa
= sfio
->pa
;
314 flags
= (flags
& SF_NODISKIO
) ? VM_ALLOC_NOWAIT
: 0;
317 * First grab all the pages and wire them. Note that we grab
318 * only required pages. Readahead pages are dealt with later.
320 VM_OBJECT_WLOCK(obj
);
321 for (int i
= 0; i
< npages
; i
++) {
322 pa
[i
] = vm_page_grab(obj
, OFF_TO_IDX(vmoff(i
, off
)),
323 VM_ALLOC_WIRED
| VM_ALLOC_NORMAL
| flags
);
331 for (int i
= 0; i
< npages
;) {
334 /* Skip valid pages. */
335 if (vm_page_is_valid(pa
[i
], vmoff(i
, off
) & PAGE_MASK
,
336 xfsize(i
, npages
, off
, len
))) {
337 vm_page_xunbusy(pa
[i
]);
338 SFSTAT_INC(sf_pages_valid
);
344 * Now 'i' points to first invalid page, iterate further
345 * to make 'j' point at first valid after a bunch of
348 for (j
= i
+ 1; j
< npages
; j
++)
349 if (vm_page_is_valid(pa
[j
], vmoff(j
, off
) & PAGE_MASK
,
350 xfsize(j
, npages
, off
, len
))) {
351 SFSTAT_INC(sf_pages_valid
);
356 * Now we got region of invalid pages between 'i' and 'j'.
357 * Check that they belong to pager. They may not be there,
358 * which is a regular situation for shmem pager. For vnode
359 * pager this happens only in case of sparse file.
361 * Important feature of vm_pager_has_page() is the hint
362 * stored in 'a', about how many pages we can pagein after
363 * this page in a single I/O.
365 while (!vm_pager_has_page(obj
, OFF_TO_IDX(vmoff(i
, off
)),
366 NULL
, &a
) && i
< j
) {
367 pmap_zero_page(pa
[i
]);
368 pa
[i
]->valid
= VM_PAGE_BITS_ALL
;
370 vm_page_xunbusy(pa
[i
]);
377 * We want to pagein as many pages as possible, limited only
378 * by the 'a' hint and actual request.
380 * We should not pagein into already valid page, thus if
381 * 'j' didn't reach last page, trim by that page.
383 * When the pagein fulfils the request, also specify readahead.
386 a
= min(a
, j
- i
- 1);
387 count
= min(a
+ 1, npages
- i
);
389 refcount_acquire(&sfio
->nios
);
390 rv
= vm_pager_get_pages_async(obj
, pa
+ i
, count
, NULL
,
391 i
+ count
== npages
? &rhpages
: NULL
,
392 &sendfile_iodone
, sfio
);
393 KASSERT(rv
== VM_PAGER_OK
, ("%s: pager fail obj %p page %p",
394 __func__
, obj
, pa
[i
]));
396 SFSTAT_INC(sf_iocnt
);
397 SFSTAT_ADD(sf_pages_read
, count
);
398 if (i
+ count
== npages
)
399 SFSTAT_ADD(sf_rhpages_read
, rhpages
);
402 for (j
= i
; j
< i
+ count
&& j
< npages
; j
++)
403 KASSERT(pa
[j
] == vm_page_lookup(obj
,
404 OFF_TO_IDX(vmoff(j
, off
))),
405 ("pa[j] %p lookup %p\n", pa
[j
],
406 vm_page_lookup(obj
, OFF_TO_IDX(vmoff(j
, off
)))));
412 VM_OBJECT_WUNLOCK(obj
);
414 if (nios
== 0 && npages
!= 0)
415 SFSTAT_INC(sf_noiocnt
);
421 sendfile_getobj(struct thread
*td
, struct file
*fp
, vm_object_t
*obj_res
,
422 struct vnode
**vp_res
, struct shmfd
**shmfd_res
, off_t
*obj_size
,
433 shmfd
= *shmfd_res
= NULL
;
437 * The file descriptor must be a regular file and have a
440 if (fp
->f_type
== DTYPE_VNODE
) {
442 vn_lock(vp
, LK_SHARED
| LK_RETRY
);
443 if (vp
->v_type
!= VREG
) {
447 *bsize
= vp
->v_mount
->mnt_stat
.f_iosize
;
448 error
= VOP_GETATTR(vp
, &va
, td
->td_ucred
);
451 *obj_size
= va
.va_size
;
457 } else if (fp
->f_type
== DTYPE_SHM
) {
460 obj
= shmfd
->shm_object
;
461 *obj_size
= shmfd
->shm_size
;
467 VM_OBJECT_WLOCK(obj
);
468 if ((obj
->flags
& OBJ_DEAD
) != 0) {
469 VM_OBJECT_WUNLOCK(obj
);
475 * Temporarily increase the backing VM object's reference
476 * count so that a forced reclamation of its vnode does not
477 * immediately destroy it.
479 vm_object_reference_locked(obj
);
480 VM_OBJECT_WUNLOCK(obj
);
492 sendfile_getsock(struct thread
*td
, int s
, struct file
**sock_fp
,
502 * The socket must be a stream socket and connected.
504 error
= getsock_cap(td
, s
, cap_rights_init(&rights
, CAP_SEND
),
508 *so
= (*sock_fp
)->f_data
;
509 if ((*so
)->so_type
!= SOCK_STREAM
)
511 if (((*so
)->so_state
& SS_ISCONNECTED
) == 0)
517 vn_sendfile(struct file
*fp
, int sockfd
, struct uio
*hdr_uio
,
518 struct uio
*trl_uio
, off_t offset
, size_t nbytes
, off_t
*sent
, int flags
,
521 struct file
*sock_fp
;
523 struct vm_object
*obj
;
525 struct mbuf
*m
, *mh
, *mhtail
;
528 struct sendfile_sync
*sfs
;
530 off_t off
, sbytes
, rem
, obj_size
;
531 int error
, softerr
, bsize
, hdrlen
;
540 error
= sendfile_getobj(td
, fp
, &obj
, &vp
, &shmfd
, &obj_size
, &bsize
);
544 error
= sendfile_getsock(td
, sockfd
, &sock_fp
, &so
);
549 error
= mac_socket_check_send(td
->td_ucred
, so
);
554 SFSTAT_INC(sf_syscalls
);
555 SFSTAT_ADD(sf_rhpages_requested
, SF_READAHEAD(flags
));
557 if (flags
& SF_SYNC
) {
558 sfs
= malloc(sizeof *sfs
, M_TEMP
, M_WAITOK
| M_ZERO
);
559 mtx_init(&sfs
->mtx
, "sendfile", NULL
, MTX_DEF
);
560 cv_init(&sfs
->cv
, "sendfile");
563 rem
= nbytes
? omin(nbytes
, obj_size
- offset
) : obj_size
- offset
;
566 * Protect against multiple writers to the socket.
568 * XXXRW: Historically this has assumed non-interruptibility, so now
569 * we implement that, but possibly shouldn't.
571 (void)sblock(&so
->so_snd
, SBL_WAIT
| SBL_NOINTR
);
574 * Loop through the pages of the file, starting with the requested
575 * offset. Get a file page (do I/O if necessary), map the file page
576 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
578 * This is done in two loops. The inner loop turns as many pages
579 * as it can, up to available socket buffer space, without blocking
580 * into mbufs to have it bulk delivered into the socket send buffer.
581 * The outer loop checks the state and available space of the socket
582 * and takes care of the overall progress.
584 for (off
= offset
; rem
> 0; ) {
588 int nios
, space
, npages
, rhpages
;
592 * Check the socket state for ongoing connection,
593 * no errors and space in socket buffer.
594 * If space is low allow for the remainder of the
595 * file to be processed if it fits the socket buffer.
596 * Otherwise block in waiting for sufficient space
597 * to proceed, or if the socket is nonblocking, return
598 * to userland with EAGAIN while reporting how far
600 * We wait until the socket buffer has significant free
601 * space to do bulk sends. This makes good use of file
602 * system read ahead and allows packet segmentation
603 * offloading hardware to take over lots of work. If
604 * we were not careful here we would send off only one
607 SOCKBUF_LOCK(&so
->so_snd
);
608 if (so
->so_snd
.sb_lowat
< so
->so_snd
.sb_hiwat
/ 2)
609 so
->so_snd
.sb_lowat
= so
->so_snd
.sb_hiwat
/ 2;
611 if (so
->so_snd
.sb_state
& SBS_CANTSENDMORE
) {
613 SOCKBUF_UNLOCK(&so
->so_snd
);
615 } else if (so
->so_error
) {
616 error
= so
->so_error
;
618 SOCKBUF_UNLOCK(&so
->so_snd
);
621 space
= sbspace(&so
->so_snd
);
624 space
< so
->so_snd
.sb_lowat
)) {
625 if (so
->so_state
& SS_NBIO
) {
626 SOCKBUF_UNLOCK(&so
->so_snd
);
631 * sbwait drops the lock while sleeping.
632 * When we loop back to retry_space the
633 * state may have changed and we retest
636 error
= sbwait(&so
->so_snd
);
638 * An error from sbwait usually indicates that we've
639 * been interrupted by a signal. If we've sent anything
640 * then return bytes sent, otherwise return the error.
643 SOCKBUF_UNLOCK(&so
->so_snd
);
648 SOCKBUF_UNLOCK(&so
->so_snd
);
651 * At the beginning of the first loop check if any headers
652 * are specified and copy them into mbufs. Reduce space in
653 * the socket buffer by the size of the header mbuf chain.
654 * Clear hdr_uio here and hdrlen at the end of the first loop.
656 if (hdr_uio
!= NULL
&& hdr_uio
->uio_resid
> 0) {
657 hdr_uio
->uio_td
= td
;
658 hdr_uio
->uio_rw
= UIO_WRITE
;
659 hdr_uio
->uio_resid
= min(hdr_uio
->uio_resid
, space
);
660 mh
= m_uiotombuf(hdr_uio
, M_WAITOK
, 0, 0, 0);
661 hdrlen
= m_length(mh
, &mhtail
);
667 error
= vn_lock(vp
, LK_SHARED
);
670 error
= VOP_GETATTR(vp
, &va
, td
->td_ucred
);
671 if (error
!= 0 || off
>= va
.va_size
) {
675 if (va
.va_size
!= obj_size
) {
677 rem
+= va
.va_size
- obj_size
;
678 else if (offset
+ nbytes
> va
.va_size
)
679 rem
-= (offset
+ nbytes
- va
.va_size
);
680 obj_size
= va
.va_size
;
687 npages
= howmany(space
+ (off
& PAGE_MASK
), PAGE_SIZE
);
690 * Calculate maximum allowed number of pages for readahead
691 * at this iteration. First, we allow readahead up to "rem".
692 * If application wants more, let it be, but there is no
693 * reason to go above MAXPHYS. Also check against "obj_size",
694 * since vm_pager_has_page() can hint beyond EOF.
696 rhpages
= howmany(rem
+ (off
& PAGE_MASK
), PAGE_SIZE
) - npages
;
697 rhpages
+= SF_READAHEAD(flags
);
698 rhpages
= min(howmany(MAXPHYS
, PAGE_SIZE
), rhpages
);
699 rhpages
= min(howmany(obj_size
- trunc_page(off
), PAGE_SIZE
) -
702 sfio
= malloc(sizeof(struct sf_io
) +
703 npages
* sizeof(vm_page_t
), M_TEMP
, M_WAITOK
);
704 refcount_init(&sfio
->nios
, 1);
707 nios
= sendfile_swapin(obj
, sfio
, off
, space
, npages
, rhpages
,
711 * Loop and construct maximum sized mbuf chain to be bulk
712 * dumped into socket buffer.
715 for (int i
= 0; i
< npages
; i
++) {
719 * If a page wasn't grabbed successfully, then
720 * trim the array. Can happen only with SF_NODISKIO.
724 fixspace(npages
, i
, off
, &space
);
731 * Get a sendfile buf. When allocating the
732 * first buffer for mbuf chain, we usually
733 * wait as long as necessary, but this wait
734 * can be interrupted. For consequent
735 * buffers, do not sleep, since several
736 * threads might exhaust the buffers and then
739 sf
= sf_buf_alloc(pa
[i
],
740 m
!= NULL
? SFB_NOWAIT
: SFB_CATCH
);
742 SFSTAT_INC(sf_allocfail
);
743 for (int j
= i
; j
< npages
; j
++) {
745 vm_page_unwire(pa
[j
], PQ_INACTIVE
);
746 vm_page_unlock(pa
[j
]);
750 fixspace(npages
, i
, off
, &space
);
755 m0
= m_get(M_WAITOK
, MT_DATA
);
756 m0
->m_ext
.ext_buf
= (char *)sf_buf_kva(sf
);
757 m0
->m_ext
.ext_size
= PAGE_SIZE
;
758 m0
->m_ext
.ext_arg1
= sf
;
759 m0
->m_ext
.ext_arg2
= sfs
;
761 * SF_NOCACHE sets the page as being freed upon send.
762 * However, we ignore it for the last page in 'space',
763 * if the page is truncated, and we got more data to
764 * send (rem > space), or if we have readahead
765 * configured (rhpages > 0).
767 if ((flags
& SF_NOCACHE
) == 0 ||
769 ((off
+ space
) & PAGE_MASK
) &&
770 (rem
> space
|| rhpages
> 0)))
771 m0
->m_ext
.ext_type
= EXT_SFBUF
;
773 m0
->m_ext
.ext_type
= EXT_SFBUF_NOCACHE
;
774 m0
->m_ext
.ext_flags
= EXT_FLAG_EMBREF
;
775 m0
->m_ext
.ext_count
= 1;
776 m0
->m_flags
|= (M_EXT
| M_RDONLY
);
778 m0
->m_flags
|= M_NOTREADY
;
779 m0
->m_data
= (char *)sf_buf_kva(sf
) +
780 (vmoff(i
, off
) & PAGE_MASK
);
781 m0
->m_len
= xfsize(i
, npages
, off
, space
);
786 /* Append to mbuf chain. */
796 mtx_unlock(&sfs
->mtx
);
803 /* Keep track of bytes processed. */
807 /* Prepend header, if any. */
815 KASSERT(softerr
, ("%s: m NULL, no error", __func__
));
821 /* Add the buffer chain to the socket buffer. */
822 KASSERT(m_length(m
, NULL
) == space
+ hdrlen
,
823 ("%s: mlen %u space %d hdrlen %d",
824 __func__
, m_length(m
, NULL
), space
, hdrlen
));
826 CURVNET_SET(so
->so_vnet
);
829 * If sendfile_swapin() didn't initiate any I/Os,
830 * which happens if all data is cached in VM, then
831 * we can send data right now without the
832 * PRUS_NOTREADY flag.
835 error
= (*so
->so_proto
->pr_usrreqs
->pru_send
)
836 (so
, 0, m
, NULL
, NULL
, td
);
838 sfio
->sock_fp
= sock_fp
;
839 sfio
->npages
= npages
;
841 error
= (*so
->so_proto
->pr_usrreqs
->pru_send
)
842 (so
, PRUS_NOTREADY
, m
, NULL
, NULL
, td
);
843 sendfile_iodone(sfio
, NULL
, 0, 0);
847 m
= NULL
; /* pru_send always consumes */
850 sbytes
+= space
+ hdrlen
;
860 * Send trailers. Wimp out and use writev(2).
862 if (trl_uio
!= NULL
) {
863 sbunlock(&so
->so_snd
);
864 error
= kern_writev(td
, sockfd
, trl_uio
);
866 sbytes
+= td
->td_retval
[0];
871 sbunlock(&so
->so_snd
);
874 * If there was no error we have to clear td->td_retval[0]
875 * because it may have been set by writev.
878 td
->td_retval
[0] = 0;
884 vm_object_deallocate(obj
);
895 cv_wait(&sfs
->cv
, &sfs
->mtx
);
896 KASSERT(sfs
->count
== 0, ("sendfile sync still busy"));
897 cv_destroy(&sfs
->cv
);
898 mtx_destroy(&sfs
->mtx
);
902 if (error
== ERESTART
)
909 sendfile(struct thread
*td
, struct sendfile_args
*uap
, int compat
)
912 struct uio
*hdr_uio
, *trl_uio
;
919 * File offset must be positive. If it goes beyond EOF
920 * we send only the header/trailer and no payload data.
925 hdr_uio
= trl_uio
= NULL
;
927 if (uap
->hdtr
!= NULL
) {
928 error
= copyin(uap
->hdtr
, &hdtr
, sizeof(hdtr
));
931 if (hdtr
.headers
!= NULL
) {
932 error
= copyinuio(hdtr
.headers
, hdtr
.hdr_cnt
,
936 #ifdef COMPAT_FREEBSD4
938 * In FreeBSD < 5.0 the nbytes to send also included
939 * the header. If compat is specified subtract the
940 * header size from nbytes.
943 if (uap
->nbytes
> hdr_uio
->uio_resid
)
944 uap
->nbytes
-= hdr_uio
->uio_resid
;
950 if (hdtr
.trailers
!= NULL
) {
951 error
= copyinuio(hdtr
.trailers
, hdtr
.trl_cnt
,
958 AUDIT_ARG_FD(uap
->fd
);
961 * sendfile(2) can start at any offset within a file so we require
962 * CAP_READ+CAP_SEEK = CAP_PREAD.
964 if ((error
= fget_read(td
, uap
->fd
,
965 cap_rights_init(&rights
, CAP_PREAD
), &fp
)) != 0) {
969 error
= fo_sendfile(fp
, uap
->s
, hdr_uio
, trl_uio
, uap
->offset
,
970 uap
->nbytes
, &sbytes
, uap
->flags
, td
);
973 if (uap
->sbytes
!= NULL
)
974 copyout(&sbytes
, uap
->sbytes
, sizeof(off_t
));
977 free(hdr_uio
, M_IOV
);
978 free(trl_uio
, M_IOV
);
985 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
986 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
988 * Send a file specified by 'fd' and starting at 'offset' to a socket
989 * specified by 's'. Send only 'nbytes' of the file or until EOF if nbytes ==
990 * 0. Optionally add a header and/or trailer to the socket output. If
991 * specified, write the total number of bytes sent into *sbytes.
994 sys_sendfile(struct thread
*td
, struct sendfile_args
*uap
)
997 return (sendfile(td
, uap
, 0));
1000 #ifdef COMPAT_FREEBSD4
1002 freebsd4_sendfile(struct thread
*td
, struct freebsd4_sendfile_args
*uap
)
1004 struct sendfile_args args
;
1008 args
.offset
= uap
->offset
;
1009 args
.nbytes
= uap
->nbytes
;
1010 args
.hdtr
= uap
->hdtr
;
1011 args
.sbytes
= uap
->sbytes
;
1012 args
.flags
= uap
->flags
;
1014 return (sendfile(td
, &args
, 1));
1016 #endif /* COMPAT_FREEBSD4 */