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 * 3. 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>
65 #define EXT_FLAG_SYNC EXT_FLAG_VENDOR1
66 #define EXT_FLAG_NOCACHE EXT_FLAG_VENDOR2
69 * Structure describing a single sendfile(2) I/O, which may consist of
70 * several underlying pager I/Os.
72 * The syscall context allocates the structure and initializes 'nios'
73 * to 1. As sendfile_swapin() runs through pages and starts asynchronous
74 * paging operations, it increments 'nios'.
76 * Every I/O completion calls sendfile_iodone(), which decrements the 'nios',
77 * and the syscall also calls sendfile_iodone() after allocating all mbufs,
78 * linking them and sending to socket. Whoever reaches zero 'nios' is
79 * responsible to * call pru_ready on the socket, to notify it of readyness
92 * Structure used to track requests with SF_SYNC flag.
94 struct sendfile_sync {
100 counter_u64_t sfstat[sizeof(struct sfstat) / sizeof(uint64_t)];
103 sfstat_init(const void *unused)
106 COUNTER_ARRAY_ALLOC(sfstat, sizeof(struct sfstat) / sizeof(uint64_t),
109 SYSINIT(sfstat, SI_SUB_MBUF, SI_ORDER_FIRST, sfstat_init, NULL);
112 sfstat_sysctl(SYSCTL_HANDLER_ARGS)
116 COUNTER_ARRAY_COPY(sfstat, &s, sizeof(s) / sizeof(uint64_t));
118 COUNTER_ARRAY_ZERO(sfstat, sizeof(s) / sizeof(uint64_t));
119 return (SYSCTL_OUT(req, &s, sizeof(s)));
121 SYSCTL_PROC(_kern_ipc, OID_AUTO, sfstat, CTLTYPE_OPAQUE | CTLFLAG_RW,
122 NULL, 0, sfstat_sysctl, "I", "sendfile statistics");
125 * Detach mapped page and release resources back to the system. Called
126 * by mbuf(9) code when last reference to a page is freed.
129 sendfile_free_page(vm_page_t pg, bool nocache)
135 * In either case check for the object going away on us. This can
136 * happen since we don't hold a reference to it. If so, we're
137 * responsible for freeing the page. In 'noncache' case try to free
138 * the page, but only if it is cheap to.
140 if (vm_page_unwire_noq(pg)) {
143 if ((obj = pg->object) == NULL)
147 if (nocache && !vm_page_xbusied(pg) &&
148 VM_OBJECT_TRYWLOCK(obj)) {
149 /* Only free unmapped pages. */
150 if (obj->ref_count == 0 ||
151 !pmap_page_is_mapped(pg))
153 * The busy test before the object is
154 * locked cannot be relied upon.
156 freed = vm_page_try_to_free(pg);
157 VM_OBJECT_WUNLOCK(obj);
161 * If we were asked to not cache the page, place
162 * it near the head of the inactive queue so
163 * that it is reclaimed sooner. Otherwise,
167 vm_page_deactivate_noreuse(pg);
168 else if (pg->queue == PQ_ACTIVE)
169 vm_page_reference(pg);
170 else if (pg->queue != PQ_INACTIVE)
171 vm_page_deactivate(pg);
181 sendfile_free_mext(struct mbuf *m)
187 KASSERT(m->m_flags & M_EXT && m->m_ext.ext_type == EXT_SFBUF,
188 ("%s: m %p !M_EXT or !EXT_SFBUF", __func__, m));
190 sf = m->m_ext.ext_arg1;
191 pg = sf_buf_page(sf);
192 nocache = m->m_ext.ext_flags & EXT_FLAG_NOCACHE;
195 sendfile_free_page(pg, nocache);
197 if (m->m_ext.ext_flags & EXT_FLAG_SYNC) {
198 struct sendfile_sync *sfs = m->m_ext.ext_arg2;
201 KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0"));
202 if (--sfs->count == 0)
204 mtx_unlock(&sfs->mtx);
209 * Helper function to calculate how much data to put into page i of n.
210 * Only first and last pages are special.
213 xfsize(int i, int n, off_t off, off_t len)
217 return (omin(PAGE_SIZE - (off & PAGE_MASK), len));
219 if (i == n - 1 && ((off + len) & PAGE_MASK) > 0)
220 return ((off + len) & PAGE_MASK);
226 * Helper function to get offset within object for i page.
228 static inline vm_ooffset_t
229 vmoff(int i, off_t off)
233 return ((vm_ooffset_t)off);
235 return (trunc_page(off + i * PAGE_SIZE));
239 * Helper function used when allocation of a page or sf_buf failed.
240 * Pretend as if we don't have enough space, subtract xfsize() of
241 * all pages that failed.
244 fixspace(int old, int new, off_t off, int *space)
247 KASSERT(old > new, ("%s: old %d new %d", __func__, old, new));
249 /* Subtract last one. */
250 *space -= xfsize(old - 1, old, off, *space);
254 /* There was only one page. */
257 /* Subtract first one. */
259 *space -= xfsize(0, old, off, *space);
263 /* Rest of pages are full sized. */
264 *space -= (old - new) * PAGE_SIZE;
266 KASSERT(*space >= 0, ("%s: space went backwards", __func__));
270 * I/O completion callback.
273 sendfile_iodone(void *arg, vm_page_t *pg, int count, int error)
275 struct sf_io *sfio = arg;
276 struct socket *so = sfio->so;
278 for (int i = 0; i < count; i++)
279 if (pg[i] != bogus_page)
280 vm_page_xunbusy(pg[i]);
285 if (!refcount_release(&sfio->nios))
288 CURVNET_SET(so->so_vnet);
293 * I/O operation failed. The state of data in the socket
294 * is now inconsistent, and all what we can do is to tear
295 * it down. Protocol abort method would tear down protocol
296 * state, free all ready mbufs and detach not ready ones.
297 * We will free the mbufs corresponding to this I/O manually.
299 * The socket would be marked with EIO and made available
300 * for read, so that application receives EIO on next
301 * syscall and eventually closes the socket.
303 so->so_proto->pr_usrreqs->pru_abort(so);
307 for (int i = 0; i < sfio->npages; i++)
310 (void )(so->so_proto->pr_usrreqs->pru_ready)(so, sfio->m,
320 * Iterate through pages vector and request paging for non-valid pages.
323 sendfile_swapin(vm_object_t obj, struct sf_io *sfio, off_t off, off_t len,
324 int npages, int rhpages, int flags)
326 vm_page_t *pa = sfio->pa;
330 flags = (flags & SF_NODISKIO) ? VM_ALLOC_NOWAIT : 0;
333 * First grab all the pages and wire them. Note that we grab
334 * only required pages. Readahead pages are dealt with later.
336 VM_OBJECT_WLOCK(obj);
338 grabbed = vm_page_grab_pages(obj, OFF_TO_IDX(off),
339 VM_ALLOC_NORMAL | VM_ALLOC_WIRED | flags, pa, npages);
340 if (grabbed < npages) {
341 for (int i = grabbed; i < npages; i++)
347 for (int i = 0; i < npages;) {
350 /* Skip valid pages. */
351 if (vm_page_is_valid(pa[i], vmoff(i, off) & PAGE_MASK,
352 xfsize(i, npages, off, len))) {
353 vm_page_xunbusy(pa[i]);
354 SFSTAT_INC(sf_pages_valid);
360 * Next page is invalid. Check if it belongs to pager. It
361 * may not be there, which is a regular situation for shmem
362 * pager. For vnode pager this happens only in case of
365 * Important feature of vm_pager_has_page() is the hint
366 * stored in 'a', about how many pages we can pagein after
367 * this page in a single I/O.
369 if (!vm_pager_has_page(obj, OFF_TO_IDX(vmoff(i, off)), NULL,
371 pmap_zero_page(pa[i]);
372 pa[i]->valid = VM_PAGE_BITS_ALL;
373 MPASS(pa[i]->dirty == 0);
374 vm_page_xunbusy(pa[i]);
380 * We want to pagein as many pages as possible, limited only
381 * by the 'a' hint and actual request.
383 count = min(a + 1, npages - i);
386 * We should not pagein into a valid page, thus we first trim
387 * any valid pages off the end of request, and substitute
388 * to bogus_page those, that are in the middle.
390 for (j = i + count - 1; j > i; j--) {
391 if (vm_page_is_valid(pa[j], vmoff(j, off) & PAGE_MASK,
392 xfsize(j, npages, off, len))) {
398 for (j = i + 1; j < i + count - 1; j++)
399 if (vm_page_is_valid(pa[j], vmoff(j, off) & PAGE_MASK,
400 xfsize(j, npages, off, len))) {
401 vm_page_xunbusy(pa[j]);
402 SFSTAT_INC(sf_pages_valid);
403 SFSTAT_INC(sf_pages_bogus);
407 refcount_acquire(&sfio->nios);
408 rv = vm_pager_get_pages_async(obj, pa + i, count, NULL,
409 i + count == npages ? &rhpages : NULL,
410 &sendfile_iodone, sfio);
411 KASSERT(rv == VM_PAGER_OK, ("%s: pager fail obj %p page %p",
412 __func__, obj, pa[i]));
414 SFSTAT_INC(sf_iocnt);
415 SFSTAT_ADD(sf_pages_read, count);
416 if (i + count == npages)
417 SFSTAT_ADD(sf_rhpages_read, rhpages);
420 * Restore the valid page pointers. They are already
421 * unbusied, but still wired.
423 for (j = i; j < i + count; j++)
424 if (pa[j] == bogus_page) {
425 pa[j] = vm_page_lookup(obj,
426 OFF_TO_IDX(vmoff(j, off)));
427 KASSERT(pa[j], ("%s: page %p[%d] disappeared",
435 VM_OBJECT_WUNLOCK(obj);
437 if (nios == 0 && npages != 0)
438 SFSTAT_INC(sf_noiocnt);
444 sendfile_getobj(struct thread *td, struct file *fp, vm_object_t *obj_res,
445 struct vnode **vp_res, struct shmfd **shmfd_res, off_t *obj_size,
456 shmfd = *shmfd_res = NULL;
460 * The file descriptor must be a regular file and have a
463 if (fp->f_type == DTYPE_VNODE) {
465 vn_lock(vp, LK_SHARED | LK_RETRY);
466 if (vp->v_type != VREG) {
470 *bsize = vp->v_mount->mnt_stat.f_iosize;
471 error = VOP_GETATTR(vp, &va, td->td_ucred);
474 *obj_size = va.va_size;
480 } else if (fp->f_type == DTYPE_SHM) {
483 obj = shmfd->shm_object;
484 *obj_size = shmfd->shm_size;
490 VM_OBJECT_WLOCK(obj);
491 if ((obj->flags & OBJ_DEAD) != 0) {
492 VM_OBJECT_WUNLOCK(obj);
498 * Temporarily increase the backing VM object's reference
499 * count so that a forced reclamation of its vnode does not
500 * immediately destroy it.
502 vm_object_reference_locked(obj);
503 VM_OBJECT_WUNLOCK(obj);
515 sendfile_getsock(struct thread *td, int s, struct file **sock_fp,
525 * The socket must be a stream socket and connected.
527 error = getsock_cap(td, s, cap_rights_init(&rights, CAP_SEND),
528 sock_fp, NULL, NULL);
531 *so = (*sock_fp)->f_data;
532 if ((*so)->so_type != SOCK_STREAM)
538 vn_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
539 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
542 struct file *sock_fp;
544 struct vm_object *obj;
546 struct mbuf *m, *mh, *mhtail;
549 struct sendfile_sync *sfs;
551 off_t off, sbytes, rem, obj_size;
552 int error, softerr, bsize, hdrlen;
561 error = sendfile_getobj(td, fp, &obj, &vp, &shmfd, &obj_size, &bsize);
565 error = sendfile_getsock(td, sockfd, &sock_fp, &so);
570 error = mac_socket_check_send(td->td_ucred, so);
575 SFSTAT_INC(sf_syscalls);
576 SFSTAT_ADD(sf_rhpages_requested, SF_READAHEAD(flags));
578 if (flags & SF_SYNC) {
579 sfs = malloc(sizeof *sfs, M_TEMP, M_WAITOK | M_ZERO);
580 mtx_init(&sfs->mtx, "sendfile", NULL, MTX_DEF);
581 cv_init(&sfs->cv, "sendfile");
584 rem = nbytes ? omin(nbytes, obj_size - offset) : obj_size - offset;
587 * Protect against multiple writers to the socket.
589 * XXXRW: Historically this has assumed non-interruptibility, so now
590 * we implement that, but possibly shouldn't.
592 (void)sblock(&so->so_snd, SBL_WAIT | SBL_NOINTR);
595 * Loop through the pages of the file, starting with the requested
596 * offset. Get a file page (do I/O if necessary), map the file page
597 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
599 * This is done in two loops. The inner loop turns as many pages
600 * as it can, up to available socket buffer space, without blocking
601 * into mbufs to have it bulk delivered into the socket send buffer.
602 * The outer loop checks the state and available space of the socket
603 * and takes care of the overall progress.
605 for (off = offset; rem > 0; ) {
609 int nios, space, npages, rhpages;
613 * Check the socket state for ongoing connection,
614 * no errors and space in socket buffer.
615 * If space is low allow for the remainder of the
616 * file to be processed if it fits the socket buffer.
617 * Otherwise block in waiting for sufficient space
618 * to proceed, or if the socket is nonblocking, return
619 * to userland with EAGAIN while reporting how far
621 * We wait until the socket buffer has significant free
622 * space to do bulk sends. This makes good use of file
623 * system read ahead and allows packet segmentation
624 * offloading hardware to take over lots of work. If
625 * we were not careful here we would send off only one
628 SOCKBUF_LOCK(&so->so_snd);
629 if (so->so_snd.sb_lowat < so->so_snd.sb_hiwat / 2)
630 so->so_snd.sb_lowat = so->so_snd.sb_hiwat / 2;
632 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
634 SOCKBUF_UNLOCK(&so->so_snd);
636 } else if (so->so_error) {
637 error = so->so_error;
639 SOCKBUF_UNLOCK(&so->so_snd);
642 if ((so->so_state & SS_ISCONNECTED) == 0) {
643 SOCKBUF_UNLOCK(&so->so_snd);
648 space = sbspace(&so->so_snd);
651 space < so->so_snd.sb_lowat)) {
652 if (so->so_state & SS_NBIO) {
653 SOCKBUF_UNLOCK(&so->so_snd);
658 * sbwait drops the lock while sleeping.
659 * When we loop back to retry_space the
660 * state may have changed and we retest
663 error = sbwait(&so->so_snd);
665 * An error from sbwait usually indicates that we've
666 * been interrupted by a signal. If we've sent anything
667 * then return bytes sent, otherwise return the error.
670 SOCKBUF_UNLOCK(&so->so_snd);
675 SOCKBUF_UNLOCK(&so->so_snd);
678 * At the beginning of the first loop check if any headers
679 * are specified and copy them into mbufs. Reduce space in
680 * the socket buffer by the size of the header mbuf chain.
681 * Clear hdr_uio here and hdrlen at the end of the first loop.
683 if (hdr_uio != NULL && hdr_uio->uio_resid > 0) {
684 hdr_uio->uio_td = td;
685 hdr_uio->uio_rw = UIO_WRITE;
686 mh = m_uiotombuf(hdr_uio, M_WAITOK, space, 0, 0);
687 hdrlen = m_length(mh, &mhtail);
690 * If header consumed all the socket buffer space,
691 * don't waste CPU cycles and jump to the end.
702 error = vn_lock(vp, LK_SHARED);
705 error = VOP_GETATTR(vp, &va, td->td_ucred);
706 if (error != 0 || off >= va.va_size) {
710 if (va.va_size != obj_size) {
711 obj_size = va.va_size;
713 omin(nbytes + offset, obj_size) : obj_size;
721 npages = howmany(space + (off & PAGE_MASK), PAGE_SIZE);
724 * Calculate maximum allowed number of pages for readahead
725 * at this iteration. If SF_USER_READAHEAD was set, we don't
726 * do any heuristics and use exactly the value supplied by
727 * application. Otherwise, we allow readahead up to "rem".
728 * If application wants more, let it be, but there is no
729 * reason to go above MAXPHYS. Also check against "obj_size",
730 * since vm_pager_has_page() can hint beyond EOF.
732 if (flags & SF_USER_READAHEAD) {
733 rhpages = SF_READAHEAD(flags);
735 rhpages = howmany(rem + (off & PAGE_MASK), PAGE_SIZE) -
737 rhpages += SF_READAHEAD(flags);
739 rhpages = min(howmany(MAXPHYS, PAGE_SIZE), rhpages);
740 rhpages = min(howmany(obj_size - trunc_page(off), PAGE_SIZE) -
743 sfio = malloc(sizeof(struct sf_io) +
744 npages * sizeof(vm_page_t), M_TEMP, M_WAITOK);
745 refcount_init(&sfio->nios, 1);
749 nios = sendfile_swapin(obj, sfio, off, space, npages, rhpages,
753 * Loop and construct maximum sized mbuf chain to be bulk
754 * dumped into socket buffer.
757 for (int i = 0; i < npages; i++) {
761 * If a page wasn't grabbed successfully, then
762 * trim the array. Can happen only with SF_NODISKIO.
766 fixspace(npages, i, off, &space);
773 * Get a sendfile buf. When allocating the
774 * first buffer for mbuf chain, we usually
775 * wait as long as necessary, but this wait
776 * can be interrupted. For consequent
777 * buffers, do not sleep, since several
778 * threads might exhaust the buffers and then
781 sf = sf_buf_alloc(pa[i],
782 m != NULL ? SFB_NOWAIT : SFB_CATCH);
784 SFSTAT_INC(sf_allocfail);
785 for (int j = i; j < npages; j++) {
787 vm_page_unwire(pa[j], PQ_INACTIVE);
788 vm_page_unlock(pa[j]);
792 fixspace(npages, i, off, &space);
797 m0 = m_get(M_WAITOK, MT_DATA);
798 m0->m_ext.ext_buf = (char *)sf_buf_kva(sf);
799 m0->m_ext.ext_size = PAGE_SIZE;
800 m0->m_ext.ext_arg1 = sf;
801 m0->m_ext.ext_type = EXT_SFBUF;
802 m0->m_ext.ext_flags = EXT_FLAG_EMBREF;
803 m0->m_ext.ext_free = sendfile_free_mext;
805 * SF_NOCACHE sets the page as being freed upon send.
806 * However, we ignore it for the last page in 'space',
807 * if the page is truncated, and we got more data to
808 * send (rem > space), or if we have readahead
809 * configured (rhpages > 0).
811 if ((flags & SF_NOCACHE) &&
813 !((off + space) & PAGE_MASK) ||
814 !(rem > space || rhpages > 0)))
815 m0->m_ext.ext_flags |= EXT_FLAG_NOCACHE;
817 m0->m_ext.ext_flags |= EXT_FLAG_SYNC;
818 m0->m_ext.ext_arg2 = sfs;
821 mtx_unlock(&sfs->mtx);
823 m0->m_ext.ext_count = 1;
824 m0->m_flags |= (M_EXT | M_RDONLY);
826 m0->m_flags |= M_NOTREADY;
827 m0->m_data = (char *)sf_buf_kva(sf) +
828 (vmoff(i, off) & PAGE_MASK);
829 m0->m_len = xfsize(i, npages, off, space);
834 /* Append to mbuf chain. */
845 /* Keep track of bytes processed. */
849 /* Prepend header, if any. */
858 KASSERT(softerr, ("%s: m NULL, no error", __func__));
864 /* Add the buffer chain to the socket buffer. */
865 KASSERT(m_length(m, NULL) == space + hdrlen,
866 ("%s: mlen %u space %d hdrlen %d",
867 __func__, m_length(m, NULL), space, hdrlen));
869 CURVNET_SET(so->so_vnet);
872 * If sendfile_swapin() didn't initiate any I/Os,
873 * which happens if all data is cached in VM, then
874 * we can send data right now without the
875 * PRUS_NOTREADY flag.
878 error = (*so->so_proto->pr_usrreqs->pru_send)
879 (so, 0, m, NULL, NULL, td);
881 sfio->npages = npages;
883 error = (*so->so_proto->pr_usrreqs->pru_send)
884 (so, PRUS_NOTREADY, m, NULL, NULL, td);
885 sendfile_iodone(sfio, NULL, 0, 0);
889 m = NULL; /* pru_send always consumes */
892 sbytes += space + hdrlen;
902 * Send trailers. Wimp out and use writev(2).
904 if (trl_uio != NULL) {
905 sbunlock(&so->so_snd);
906 error = kern_writev(td, sockfd, trl_uio);
908 sbytes += td->td_retval[0];
913 sbunlock(&so->so_snd);
916 * If there was no error we have to clear td->td_retval[0]
917 * because it may have been set by writev.
920 td->td_retval[0] = 0;
926 vm_object_deallocate(obj);
937 cv_wait(&sfs->cv, &sfs->mtx);
938 KASSERT(sfs->count == 0, ("sendfile sync still busy"));
939 cv_destroy(&sfs->cv);
940 mtx_destroy(&sfs->mtx);
944 if (error == ERESTART)
951 sendfile(struct thread *td, struct sendfile_args *uap, int compat)
954 struct uio *hdr_uio, *trl_uio;
961 * File offset must be positive. If it goes beyond EOF
962 * we send only the header/trailer and no payload data.
968 hdr_uio = trl_uio = NULL;
970 if (uap->hdtr != NULL) {
971 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
974 if (hdtr.headers != NULL) {
975 error = copyinuio(hdtr.headers, hdtr.hdr_cnt,
979 #ifdef COMPAT_FREEBSD4
981 * In FreeBSD < 5.0 the nbytes to send also included
982 * the header. If compat is specified subtract the
983 * header size from nbytes.
986 if (uap->nbytes > hdr_uio->uio_resid)
987 uap->nbytes -= hdr_uio->uio_resid;
993 if (hdtr.trailers != NULL) {
994 error = copyinuio(hdtr.trailers, hdtr.trl_cnt,
1001 AUDIT_ARG_FD(uap->fd);
1004 * sendfile(2) can start at any offset within a file so we require
1005 * CAP_READ+CAP_SEEK = CAP_PREAD.
1007 if ((error = fget_read(td, uap->fd,
1008 cap_rights_init(&rights, CAP_PREAD), &fp)) != 0) {
1012 error = fo_sendfile(fp, uap->s, hdr_uio, trl_uio, uap->offset,
1013 uap->nbytes, &sbytes, uap->flags, td);
1016 if (uap->sbytes != NULL)
1017 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1020 free(hdr_uio, M_IOV);
1021 free(trl_uio, M_IOV);
1028 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1029 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1031 * Send a file specified by 'fd' and starting at 'offset' to a socket
1032 * specified by 's'. Send only 'nbytes' of the file or until EOF if nbytes ==
1033 * 0. Optionally add a header and/or trailer to the socket output. If
1034 * specified, write the total number of bytes sent into *sbytes.
1037 sys_sendfile(struct thread *td, struct sendfile_args *uap)
1040 return (sendfile(td, uap, 0));
1043 #ifdef COMPAT_FREEBSD4
1045 freebsd4_sendfile(struct thread *td, struct freebsd4_sendfile_args *uap)
1047 struct sendfile_args args;
1051 args.offset = uap->offset;
1052 args.nbytes = uap->nbytes;
1053 args.hdtr = uap->hdtr;
1054 args.sbytes = uap->sbytes;
1055 args.flags = uap->flags;
1057 return (sendfile(td, &args, 1));
1059 #endif /* COMPAT_FREEBSD4 */