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 extern vm_page_t bogus_page;
68 * Structure describing a single sendfile(2) I/O, which may consist of
69 * several underlying pager I/Os.
71 * The syscall context allocates the structure and initializes 'nios'
72 * to 1. As sendfile_swapin() runs through pages and starts asynchronous
73 * paging operations, it increments 'nios'.
75 * Every I/O completion calls sendfile_iodone(), which decrements the 'nios',
76 * and the syscall also calls sendfile_iodone() after allocating all mbufs,
77 * linking them and sending to socket. Whoever reaches zero 'nios' is
78 * responsible to * call pru_ready on the socket, to notify it of readyness
91 * Structure used to track requests with SF_SYNC flag.
93 struct sendfile_sync {
99 counter_u64_t sfstat[sizeof(struct sfstat) / sizeof(uint64_t)];
102 sfstat_init(const void *unused)
105 COUNTER_ARRAY_ALLOC(sfstat, sizeof(struct sfstat) / sizeof(uint64_t),
108 SYSINIT(sfstat, SI_SUB_MBUF, SI_ORDER_FIRST, sfstat_init, NULL);
111 sfstat_sysctl(SYSCTL_HANDLER_ARGS)
115 COUNTER_ARRAY_COPY(sfstat, &s, sizeof(s) / sizeof(uint64_t));
117 COUNTER_ARRAY_ZERO(sfstat, sizeof(s) / sizeof(uint64_t));
118 return (SYSCTL_OUT(req, &s, sizeof(s)));
120 SYSCTL_PROC(_kern_ipc, OID_AUTO, sfstat, CTLTYPE_OPAQUE | CTLFLAG_RW,
121 NULL, 0, sfstat_sysctl, "I", "sendfile statistics");
124 * Detach mapped page and release resources back to the system. Called
125 * by mbuf(9) code when last reference to a page is freed.
128 sf_ext_free(void *arg1, void *arg2)
130 struct sf_buf *sf = arg1;
131 struct sendfile_sync *sfs = arg2;
132 vm_page_t pg = sf_buf_page(sf);
138 * Check for the object going away on us. This can
139 * happen since we don't hold a reference to it.
140 * If so, we're responsible for freeing the page.
142 if (vm_page_unwire(pg, PQ_INACTIVE) && pg->object == NULL)
148 KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0"));
149 if (--sfs->count == 0)
151 mtx_unlock(&sfs->mtx);
156 * Same as above, but forces the page to be detached from the object
157 * and go into free pool.
160 sf_ext_free_nocache(void *arg1, void *arg2)
162 struct sf_buf *sf = arg1;
163 struct sendfile_sync *sfs = arg2;
164 vm_page_t pg = sf_buf_page(sf);
169 if (vm_page_unwire(pg, PQ_NONE)) {
172 /* Try to free the page, but only if it is cheap to. */
173 if ((obj = pg->object) == NULL)
175 else if (!vm_page_xbusied(pg) && VM_OBJECT_TRYWLOCK(obj)) {
177 VM_OBJECT_WUNLOCK(obj);
179 vm_page_deactivate(pg);
185 KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0"));
186 if (--sfs->count == 0)
188 mtx_unlock(&sfs->mtx);
193 * Helper function to calculate how much data to put into page i of n.
194 * Only first and last pages are special.
197 xfsize(int i, int n, off_t off, off_t len)
201 return (omin(PAGE_SIZE - (off & PAGE_MASK), len));
203 if (i == n - 1 && ((off + len) & PAGE_MASK) > 0)
204 return ((off + len) & PAGE_MASK);
210 * Helper function to get offset within object for i page.
212 static inline vm_offset_t
213 vmoff(int i, off_t off)
217 return ((vm_offset_t)off);
219 return (trunc_page(off + i * PAGE_SIZE));
223 * Helper function used when allocation of a page or sf_buf failed.
224 * Pretend as if we don't have enough space, subtract xfsize() of
225 * all pages that failed.
228 fixspace(int old, int new, off_t off, int *space)
231 KASSERT(old > new, ("%s: old %d new %d", __func__, old, new));
233 /* Subtract last one. */
234 *space -= xfsize(old - 1, old, off, *space);
238 /* There was only one page. */
241 /* Subtract first one. */
243 *space -= xfsize(0, old, off, *space);
247 /* Rest of pages are full sized. */
248 *space -= (old - new) * PAGE_SIZE;
250 KASSERT(*space >= 0, ("%s: space went backwards", __func__));
254 * I/O completion callback.
257 sendfile_iodone(void *arg, vm_page_t *pg, int count, int error)
259 struct sf_io *sfio = arg;
262 for (int i = 0; i < count; i++)
263 if (pg[i] != bogus_page)
264 vm_page_xunbusy(pg[i]);
269 if (!refcount_release(&sfio->nios))
272 so = sfio->sock_fp->f_data;
278 * I/O operation failed. The state of data in the socket
279 * is now inconsistent, and all what we can do is to tear
280 * it down. Protocol abort method would tear down protocol
281 * state, free all ready mbufs and detach not ready ones.
282 * We will free the mbufs corresponding to this I/O manually.
284 * The socket would be marked with EIO and made available
285 * for read, so that application receives EIO on next
286 * syscall and eventually closes the socket.
288 so->so_proto->pr_usrreqs->pru_abort(so);
292 for (int i = 0; i < sfio->npages; i++)
295 CURVNET_SET(so->so_vnet);
296 (void )(so->so_proto->pr_usrreqs->pru_ready)(so, sfio->m,
301 /* XXXGL: curthread */
302 fdrop(sfio->sock_fp, curthread);
307 * Iterate through pages vector and request paging for non-valid pages.
310 sendfile_swapin(vm_object_t obj, struct sf_io *sfio, off_t off, off_t len,
311 int npages, int rhpages, int flags)
313 vm_page_t *pa = sfio->pa;
317 flags = (flags & SF_NODISKIO) ? VM_ALLOC_NOWAIT : 0;
320 * First grab all the pages and wire them. Note that we grab
321 * only required pages. Readahead pages are dealt with later.
323 VM_OBJECT_WLOCK(obj);
324 for (int i = 0; i < npages; i++) {
325 pa[i] = vm_page_grab(obj, OFF_TO_IDX(vmoff(i, off)),
326 VM_ALLOC_WIRED | VM_ALLOC_NORMAL | flags);
334 for (int i = 0; i < npages;) {
337 /* Skip valid pages. */
338 if (vm_page_is_valid(pa[i], vmoff(i, off) & PAGE_MASK,
339 xfsize(i, npages, off, len))) {
340 vm_page_xunbusy(pa[i]);
341 SFSTAT_INC(sf_pages_valid);
347 * Next page is invalid. Check if it belongs to pager. It
348 * may not be there, which is a regular situation for shmem
349 * pager. For vnode pager this happens only in case of
352 * Important feature of vm_pager_has_page() is the hint
353 * stored in 'a', about how many pages we can pagein after
354 * this page in a single I/O.
356 if (!vm_pager_has_page(obj, OFF_TO_IDX(vmoff(i, off)), NULL,
358 pmap_zero_page(pa[i]);
359 pa[i]->valid = VM_PAGE_BITS_ALL;
361 vm_page_xunbusy(pa[i]);
367 * We want to pagein as many pages as possible, limited only
368 * by the 'a' hint and actual request.
370 count = min(a + 1, npages - i);
373 * We should not pagein into a valid page, thus we first trim
374 * any valid pages off the end of request, and substitute
375 * to bogus_page those, that are in the middle.
377 for (j = i + count - 1; j > i; j--) {
378 if (vm_page_is_valid(pa[j], vmoff(j, off) & PAGE_MASK,
379 xfsize(j, npages, off, len))) {
385 for (j = i + 1; j < i + count - 1; j++)
386 if (vm_page_is_valid(pa[j], vmoff(j, off) & PAGE_MASK,
387 xfsize(j, npages, off, len))) {
388 vm_page_xunbusy(pa[j]);
389 SFSTAT_INC(sf_pages_valid);
390 SFSTAT_INC(sf_pages_bogus);
394 refcount_acquire(&sfio->nios);
395 rv = vm_pager_get_pages_async(obj, pa + i, count, NULL,
396 i + count == npages ? &rhpages : NULL,
397 &sendfile_iodone, sfio);
398 KASSERT(rv == VM_PAGER_OK, ("%s: pager fail obj %p page %p",
399 __func__, obj, pa[i]));
401 SFSTAT_INC(sf_iocnt);
402 SFSTAT_ADD(sf_pages_read, count);
403 if (i + count == npages)
404 SFSTAT_ADD(sf_rhpages_read, rhpages);
407 * Restore the valid page pointers. They are already
408 * unbusied, but still wired.
410 for (j = i; j < i + count; j++)
411 if (pa[j] == bogus_page) {
412 pa[j] = vm_page_lookup(obj,
413 OFF_TO_IDX(vmoff(j, off)));
414 KASSERT(pa[j], ("%s: page %p[%d] disappeared",
422 VM_OBJECT_WUNLOCK(obj);
424 if (nios == 0 && npages != 0)
425 SFSTAT_INC(sf_noiocnt);
431 sendfile_getobj(struct thread *td, struct file *fp, vm_object_t *obj_res,
432 struct vnode **vp_res, struct shmfd **shmfd_res, off_t *obj_size,
443 shmfd = *shmfd_res = NULL;
447 * The file descriptor must be a regular file and have a
450 if (fp->f_type == DTYPE_VNODE) {
452 vn_lock(vp, LK_SHARED | LK_RETRY);
453 if (vp->v_type != VREG) {
457 *bsize = vp->v_mount->mnt_stat.f_iosize;
458 error = VOP_GETATTR(vp, &va, td->td_ucred);
461 *obj_size = va.va_size;
467 } else if (fp->f_type == DTYPE_SHM) {
470 obj = shmfd->shm_object;
471 *obj_size = shmfd->shm_size;
477 VM_OBJECT_WLOCK(obj);
478 if ((obj->flags & OBJ_DEAD) != 0) {
479 VM_OBJECT_WUNLOCK(obj);
485 * Temporarily increase the backing VM object's reference
486 * count so that a forced reclamation of its vnode does not
487 * immediately destroy it.
489 vm_object_reference_locked(obj);
490 VM_OBJECT_WUNLOCK(obj);
502 sendfile_getsock(struct thread *td, int s, struct file **sock_fp,
512 * The socket must be a stream socket and connected.
514 error = getsock_cap(td, s, cap_rights_init(&rights, CAP_SEND),
515 sock_fp, NULL, NULL);
518 *so = (*sock_fp)->f_data;
519 if ((*so)->so_type != SOCK_STREAM)
521 if (((*so)->so_state & SS_ISCONNECTED) == 0)
527 vn_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
528 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
531 struct file *sock_fp;
533 struct vm_object *obj;
535 struct mbuf *m, *mh, *mhtail;
538 struct sendfile_sync *sfs;
540 off_t off, sbytes, rem, obj_size;
541 int error, softerr, bsize, hdrlen;
550 error = sendfile_getobj(td, fp, &obj, &vp, &shmfd, &obj_size, &bsize);
554 error = sendfile_getsock(td, sockfd, &sock_fp, &so);
559 error = mac_socket_check_send(td->td_ucred, so);
564 SFSTAT_INC(sf_syscalls);
565 SFSTAT_ADD(sf_rhpages_requested, SF_READAHEAD(flags));
567 if (flags & SF_SYNC) {
568 sfs = malloc(sizeof *sfs, M_TEMP, M_WAITOK | M_ZERO);
569 mtx_init(&sfs->mtx, "sendfile", NULL, MTX_DEF);
570 cv_init(&sfs->cv, "sendfile");
573 rem = nbytes ? omin(nbytes, obj_size - offset) : obj_size - offset;
576 * Protect against multiple writers to the socket.
578 * XXXRW: Historically this has assumed non-interruptibility, so now
579 * we implement that, but possibly shouldn't.
581 (void)sblock(&so->so_snd, SBL_WAIT | SBL_NOINTR);
584 * Loop through the pages of the file, starting with the requested
585 * offset. Get a file page (do I/O if necessary), map the file page
586 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
588 * This is done in two loops. The inner loop turns as many pages
589 * as it can, up to available socket buffer space, without blocking
590 * into mbufs to have it bulk delivered into the socket send buffer.
591 * The outer loop checks the state and available space of the socket
592 * and takes care of the overall progress.
594 for (off = offset; rem > 0; ) {
598 int nios, space, npages, rhpages;
602 * Check the socket state for ongoing connection,
603 * no errors and space in socket buffer.
604 * If space is low allow for the remainder of the
605 * file to be processed if it fits the socket buffer.
606 * Otherwise block in waiting for sufficient space
607 * to proceed, or if the socket is nonblocking, return
608 * to userland with EAGAIN while reporting how far
610 * We wait until the socket buffer has significant free
611 * space to do bulk sends. This makes good use of file
612 * system read ahead and allows packet segmentation
613 * offloading hardware to take over lots of work. If
614 * we were not careful here we would send off only one
617 SOCKBUF_LOCK(&so->so_snd);
618 if (so->so_snd.sb_lowat < so->so_snd.sb_hiwat / 2)
619 so->so_snd.sb_lowat = so->so_snd.sb_hiwat / 2;
621 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
623 SOCKBUF_UNLOCK(&so->so_snd);
625 } else if (so->so_error) {
626 error = so->so_error;
628 SOCKBUF_UNLOCK(&so->so_snd);
631 space = sbspace(&so->so_snd);
634 space < so->so_snd.sb_lowat)) {
635 if (so->so_state & SS_NBIO) {
636 SOCKBUF_UNLOCK(&so->so_snd);
641 * sbwait drops the lock while sleeping.
642 * When we loop back to retry_space the
643 * state may have changed and we retest
646 error = sbwait(&so->so_snd);
648 * An error from sbwait usually indicates that we've
649 * been interrupted by a signal. If we've sent anything
650 * then return bytes sent, otherwise return the error.
653 SOCKBUF_UNLOCK(&so->so_snd);
658 SOCKBUF_UNLOCK(&so->so_snd);
661 * At the beginning of the first loop check if any headers
662 * are specified and copy them into mbufs. Reduce space in
663 * the socket buffer by the size of the header mbuf chain.
664 * Clear hdr_uio here and hdrlen at the end of the first loop.
666 if (hdr_uio != NULL && hdr_uio->uio_resid > 0) {
667 hdr_uio->uio_td = td;
668 hdr_uio->uio_rw = UIO_WRITE;
669 mh = m_uiotombuf(hdr_uio, M_WAITOK, space, 0, 0);
670 hdrlen = m_length(mh, &mhtail);
673 * If header consumed all the socket buffer space,
674 * don't waste CPU cycles and jump to the end.
685 error = vn_lock(vp, LK_SHARED);
688 error = VOP_GETATTR(vp, &va, td->td_ucred);
689 if (error != 0 || off >= va.va_size) {
693 if (va.va_size != obj_size) {
695 rem += va.va_size - obj_size;
696 else if (offset + nbytes > va.va_size)
697 rem -= (offset + nbytes - va.va_size);
698 obj_size = va.va_size;
705 npages = howmany(space + (off & PAGE_MASK), PAGE_SIZE);
708 * Calculate maximum allowed number of pages for readahead
709 * at this iteration. If SF_USER_READAHEAD was set, we don't
710 * do any heuristics and use exactly the value supplied by
711 * application. Otherwise, we allow readahead up to "rem".
712 * If application wants more, let it be, but there is no
713 * reason to go above MAXPHYS. Also check against "obj_size",
714 * since vm_pager_has_page() can hint beyond EOF.
716 if (flags & SF_USER_READAHEAD) {
717 rhpages = SF_READAHEAD(flags);
719 rhpages = howmany(rem + (off & PAGE_MASK), PAGE_SIZE) -
721 rhpages += SF_READAHEAD(flags);
723 rhpages = min(howmany(MAXPHYS, PAGE_SIZE), rhpages);
724 rhpages = min(howmany(obj_size - trunc_page(off), PAGE_SIZE) -
727 sfio = malloc(sizeof(struct sf_io) +
728 npages * sizeof(vm_page_t), M_TEMP, M_WAITOK);
729 refcount_init(&sfio->nios, 1);
732 nios = sendfile_swapin(obj, sfio, off, space, npages, rhpages,
736 * Loop and construct maximum sized mbuf chain to be bulk
737 * dumped into socket buffer.
740 for (int i = 0; i < npages; i++) {
744 * If a page wasn't grabbed successfully, then
745 * trim the array. Can happen only with SF_NODISKIO.
749 fixspace(npages, i, off, &space);
756 * Get a sendfile buf. When allocating the
757 * first buffer for mbuf chain, we usually
758 * wait as long as necessary, but this wait
759 * can be interrupted. For consequent
760 * buffers, do not sleep, since several
761 * threads might exhaust the buffers and then
764 sf = sf_buf_alloc(pa[i],
765 m != NULL ? SFB_NOWAIT : SFB_CATCH);
767 SFSTAT_INC(sf_allocfail);
768 for (int j = i; j < npages; j++) {
770 vm_page_unwire(pa[j], PQ_INACTIVE);
771 vm_page_unlock(pa[j]);
775 fixspace(npages, i, off, &space);
780 m0 = m_get(M_WAITOK, MT_DATA);
781 m0->m_ext.ext_buf = (char *)sf_buf_kva(sf);
782 m0->m_ext.ext_size = PAGE_SIZE;
783 m0->m_ext.ext_arg1 = sf;
784 m0->m_ext.ext_arg2 = sfs;
786 * SF_NOCACHE sets the page as being freed upon send.
787 * However, we ignore it for the last page in 'space',
788 * if the page is truncated, and we got more data to
789 * send (rem > space), or if we have readahead
790 * configured (rhpages > 0).
792 if ((flags & SF_NOCACHE) == 0 ||
794 ((off + space) & PAGE_MASK) &&
795 (rem > space || rhpages > 0)))
796 m0->m_ext.ext_type = EXT_SFBUF;
798 m0->m_ext.ext_type = EXT_SFBUF_NOCACHE;
799 m0->m_ext.ext_flags = EXT_FLAG_EMBREF;
800 m0->m_ext.ext_count = 1;
801 m0->m_flags |= (M_EXT | M_RDONLY);
803 m0->m_flags |= M_NOTREADY;
804 m0->m_data = (char *)sf_buf_kva(sf) +
805 (vmoff(i, off) & PAGE_MASK);
806 m0->m_len = xfsize(i, npages, off, space);
811 /* Append to mbuf chain. */
821 mtx_unlock(&sfs->mtx);
828 /* Keep track of bytes processed. */
832 /* Prepend header, if any. */
841 KASSERT(softerr, ("%s: m NULL, no error", __func__));
847 /* Add the buffer chain to the socket buffer. */
848 KASSERT(m_length(m, NULL) == space + hdrlen,
849 ("%s: mlen %u space %d hdrlen %d",
850 __func__, m_length(m, NULL), space, hdrlen));
852 CURVNET_SET(so->so_vnet);
855 * If sendfile_swapin() didn't initiate any I/Os,
856 * which happens if all data is cached in VM, then
857 * we can send data right now without the
858 * PRUS_NOTREADY flag.
861 error = (*so->so_proto->pr_usrreqs->pru_send)
862 (so, 0, m, NULL, NULL, td);
864 sfio->sock_fp = sock_fp;
865 sfio->npages = npages;
867 error = (*so->so_proto->pr_usrreqs->pru_send)
868 (so, PRUS_NOTREADY, m, NULL, NULL, td);
869 sendfile_iodone(sfio, NULL, 0, 0);
873 m = NULL; /* pru_send always consumes */
876 sbytes += space + hdrlen;
886 * Send trailers. Wimp out and use writev(2).
888 if (trl_uio != NULL) {
889 sbunlock(&so->so_snd);
890 error = kern_writev(td, sockfd, trl_uio);
892 sbytes += td->td_retval[0];
897 sbunlock(&so->so_snd);
900 * If there was no error we have to clear td->td_retval[0]
901 * because it may have been set by writev.
904 td->td_retval[0] = 0;
910 vm_object_deallocate(obj);
921 cv_wait(&sfs->cv, &sfs->mtx);
922 KASSERT(sfs->count == 0, ("sendfile sync still busy"));
923 cv_destroy(&sfs->cv);
924 mtx_destroy(&sfs->mtx);
928 if (error == ERESTART)
935 sendfile(struct thread *td, struct sendfile_args *uap, int compat)
938 struct uio *hdr_uio, *trl_uio;
945 * File offset must be positive. If it goes beyond EOF
946 * we send only the header/trailer and no payload data.
951 hdr_uio = trl_uio = NULL;
953 if (uap->hdtr != NULL) {
954 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
957 if (hdtr.headers != NULL) {
958 error = copyinuio(hdtr.headers, hdtr.hdr_cnt,
962 #ifdef COMPAT_FREEBSD4
964 * In FreeBSD < 5.0 the nbytes to send also included
965 * the header. If compat is specified subtract the
966 * header size from nbytes.
969 if (uap->nbytes > hdr_uio->uio_resid)
970 uap->nbytes -= hdr_uio->uio_resid;
976 if (hdtr.trailers != NULL) {
977 error = copyinuio(hdtr.trailers, hdtr.trl_cnt,
984 AUDIT_ARG_FD(uap->fd);
987 * sendfile(2) can start at any offset within a file so we require
988 * CAP_READ+CAP_SEEK = CAP_PREAD.
990 if ((error = fget_read(td, uap->fd,
991 cap_rights_init(&rights, CAP_PREAD), &fp)) != 0) {
995 error = fo_sendfile(fp, uap->s, hdr_uio, trl_uio, uap->offset,
996 uap->nbytes, &sbytes, uap->flags, td);
999 if (uap->sbytes != NULL)
1000 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1003 free(hdr_uio, M_IOV);
1004 free(trl_uio, M_IOV);
1011 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1012 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1014 * Send a file specified by 'fd' and starting at 'offset' to a socket
1015 * specified by 's'. Send only 'nbytes' of the file or until EOF if nbytes ==
1016 * 0. Optionally add a header and/or trailer to the socket output. If
1017 * specified, write the total number of bytes sent into *sbytes.
1020 sys_sendfile(struct thread *td, struct sendfile_args *uap)
1023 return (sendfile(td, uap, 0));
1026 #ifdef COMPAT_FREEBSD4
1028 freebsd4_sendfile(struct thread *td, struct freebsd4_sendfile_args *uap)
1030 struct sendfile_args args;
1034 args.offset = uap->offset;
1035 args.nbytes = uap->nbytes;
1036 args.hdtr = uap->hdtr;
1037 args.sbytes = uap->sbytes;
1038 args.flags = uap->flags;
1040 return (sendfile(td, &args, 1));
1042 #endif /* COMPAT_FREEBSD4 */