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 <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/capsicum.h>
36 #include <sys/kernel.h>
38 #include <sys/mutex.h>
39 #include <sys/sysproto.h>
40 #include <sys/malloc.h>
43 #include <sys/mount.h>
45 #include <sys/protosw.h>
46 #include <sys/rwlock.h>
47 #include <sys/sf_buf.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/syscallsubr.h>
51 #include <sys/sysctl.h>
52 #include <sys/vnode.h>
56 #include <security/audit/audit.h>
57 #include <security/mac/mac_framework.h>
60 #include <vm/vm_object.h>
61 #include <vm/vm_pager.h>
63 #define EXT_FLAG_SYNC EXT_FLAG_VENDOR1
64 #define EXT_FLAG_NOCACHE EXT_FLAG_VENDOR2
67 * Structure describing a single sendfile(2) I/O, which may consist of
68 * several underlying pager I/Os.
70 * The syscall context allocates the structure and initializes 'nios'
71 * to 1. As sendfile_swapin() runs through pages and starts asynchronous
72 * paging operations, it increments 'nios'.
74 * Every I/O completion calls sendfile_iodone(), which decrements the 'nios',
75 * and the syscall also calls sendfile_iodone() after allocating all mbufs,
76 * linking them and sending to socket. Whoever reaches zero 'nios' is
77 * responsible to * call pru_ready on the socket, to notify it of readyness
90 * Structure used to track requests with SF_SYNC flag.
92 struct sendfile_sync {
98 counter_u64_t sfstat[sizeof(struct sfstat) / sizeof(uint64_t)];
101 sfstat_init(const void *unused)
104 COUNTER_ARRAY_ALLOC(sfstat, sizeof(struct sfstat) / sizeof(uint64_t),
107 SYSINIT(sfstat, SI_SUB_MBUF, SI_ORDER_FIRST, sfstat_init, NULL);
110 sfstat_sysctl(SYSCTL_HANDLER_ARGS)
114 COUNTER_ARRAY_COPY(sfstat, &s, sizeof(s) / sizeof(uint64_t));
116 COUNTER_ARRAY_ZERO(sfstat, sizeof(s) / sizeof(uint64_t));
117 return (SYSCTL_OUT(req, &s, sizeof(s)));
119 SYSCTL_PROC(_kern_ipc, OID_AUTO, sfstat, CTLTYPE_OPAQUE | CTLFLAG_RW,
120 NULL, 0, sfstat_sysctl, "I", "sendfile statistics");
123 * Detach mapped page and release resources back to the system. Called
124 * by mbuf(9) code when last reference to a page is freed.
127 sendfile_free_page(vm_page_t pg, bool nocache)
133 * In either case check for the object going away on us. This can
134 * happen since we don't hold a reference to it. If so, we're
135 * responsible for freeing the page. In 'noncache' case try to free
136 * the page, but only if it is cheap to.
138 if (vm_page_unwire_noq(pg)) {
141 if ((obj = pg->object) == NULL)
145 if (nocache && !vm_page_xbusied(pg) &&
146 VM_OBJECT_TRYWLOCK(obj)) {
147 /* Only free unmapped pages. */
148 if (obj->ref_count == 0 ||
149 !pmap_page_is_mapped(pg))
151 * The busy test before the object is
152 * locked cannot be relied upon.
154 freed = vm_page_try_to_free(pg);
155 VM_OBJECT_WUNLOCK(obj);
159 * If we were asked to not cache the page, place
160 * it near the head of the inactive queue so
161 * that it is reclaimed sooner. Otherwise,
165 vm_page_deactivate_noreuse(pg);
166 else if (vm_page_active(pg))
167 vm_page_reference(pg);
169 vm_page_deactivate(pg);
177 sendfile_free_mext(struct mbuf *m)
183 KASSERT(m->m_flags & M_EXT && m->m_ext.ext_type == EXT_SFBUF,
184 ("%s: m %p !M_EXT or !EXT_SFBUF", __func__, m));
186 sf = m->m_ext.ext_arg1;
187 pg = sf_buf_page(sf);
188 nocache = m->m_ext.ext_flags & EXT_FLAG_NOCACHE;
191 sendfile_free_page(pg, nocache);
193 if (m->m_ext.ext_flags & EXT_FLAG_SYNC) {
194 struct sendfile_sync *sfs = m->m_ext.ext_arg2;
197 KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0"));
198 if (--sfs->count == 0)
200 mtx_unlock(&sfs->mtx);
205 * Helper function to calculate how much data to put into page i of n.
206 * Only first and last pages are special.
209 xfsize(int i, int n, off_t off, off_t len)
213 return (omin(PAGE_SIZE - (off & PAGE_MASK), len));
215 if (i == n - 1 && ((off + len) & PAGE_MASK) > 0)
216 return ((off + len) & PAGE_MASK);
222 * Helper function to get offset within object for i page.
224 static inline vm_ooffset_t
225 vmoff(int i, off_t off)
229 return ((vm_ooffset_t)off);
231 return (trunc_page(off + i * PAGE_SIZE));
235 * Helper function used when allocation of a page or sf_buf failed.
236 * Pretend as if we don't have enough space, subtract xfsize() of
237 * all pages that failed.
240 fixspace(int old, int new, off_t off, int *space)
243 KASSERT(old > new, ("%s: old %d new %d", __func__, old, new));
245 /* Subtract last one. */
246 *space -= xfsize(old - 1, old, off, *space);
250 /* There was only one page. */
253 /* Subtract first one. */
255 *space -= xfsize(0, old, off, *space);
259 /* Rest of pages are full sized. */
260 *space -= (old - new) * PAGE_SIZE;
262 KASSERT(*space >= 0, ("%s: space went backwards", __func__));
266 * I/O completion callback.
269 sendfile_iodone(void *arg, vm_page_t *pg, int count, int error)
271 struct sf_io *sfio = arg;
272 struct socket *so = sfio->so;
274 for (int i = 0; i < count; i++)
275 if (pg[i] != bogus_page)
276 vm_page_xunbusy(pg[i]);
281 if (!refcount_release(&sfio->nios))
284 CURVNET_SET(so->so_vnet);
289 * I/O operation failed. The state of data in the socket
290 * is now inconsistent, and all what we can do is to tear
291 * it down. Protocol abort method would tear down protocol
292 * state, free all ready mbufs and detach not ready ones.
293 * We will free the mbufs corresponding to this I/O manually.
295 * The socket would be marked with EIO and made available
296 * for read, so that application receives EIO on next
297 * syscall and eventually closes the socket.
299 so->so_proto->pr_usrreqs->pru_abort(so);
303 for (int i = 0; i < sfio->npages; i++)
306 (void )(so->so_proto->pr_usrreqs->pru_ready)(so, sfio->m,
316 * Iterate through pages vector and request paging for non-valid pages.
319 sendfile_swapin(vm_object_t obj, struct sf_io *sfio, off_t off, off_t len,
320 int npages, int rhpages, int flags)
322 vm_page_t *pa = sfio->pa;
326 flags = (flags & SF_NODISKIO) ? VM_ALLOC_NOWAIT : 0;
329 * First grab all the pages and wire them. Note that we grab
330 * only required pages. Readahead pages are dealt with later.
332 VM_OBJECT_WLOCK(obj);
334 grabbed = vm_page_grab_pages(obj, OFF_TO_IDX(off),
335 VM_ALLOC_NORMAL | VM_ALLOC_WIRED | flags, pa, npages);
336 if (grabbed < npages) {
337 for (int i = grabbed; i < npages; i++)
343 for (int i = 0; i < npages;) {
344 int j, a, count, rv __unused;
346 /* Skip valid pages. */
347 if (vm_page_is_valid(pa[i], vmoff(i, off) & PAGE_MASK,
348 xfsize(i, npages, off, len))) {
349 vm_page_xunbusy(pa[i]);
350 SFSTAT_INC(sf_pages_valid);
356 * Next page is invalid. Check if it belongs to pager. It
357 * may not be there, which is a regular situation for shmem
358 * pager. For vnode pager this happens only in case of
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 if (!vm_pager_has_page(obj, OFF_TO_IDX(vmoff(i, off)), NULL,
367 pmap_zero_page(pa[i]);
368 pa[i]->valid = VM_PAGE_BITS_ALL;
369 MPASS(pa[i]->dirty == 0);
370 vm_page_xunbusy(pa[i]);
376 * We want to pagein as many pages as possible, limited only
377 * by the 'a' hint and actual request.
379 count = min(a + 1, npages - i);
382 * We should not pagein into a valid page, thus we first trim
383 * any valid pages off the end of request, and substitute
384 * to bogus_page those, that are in the middle.
386 for (j = i + count - 1; j > i; j--) {
387 if (vm_page_is_valid(pa[j], vmoff(j, off) & PAGE_MASK,
388 xfsize(j, npages, off, len))) {
394 for (j = i + 1; j < i + count - 1; j++)
395 if (vm_page_is_valid(pa[j], vmoff(j, off) & PAGE_MASK,
396 xfsize(j, npages, off, len))) {
397 vm_page_xunbusy(pa[j]);
398 SFSTAT_INC(sf_pages_valid);
399 SFSTAT_INC(sf_pages_bogus);
403 refcount_acquire(&sfio->nios);
404 rv = vm_pager_get_pages_async(obj, pa + i, count, NULL,
405 i + count == npages ? &rhpages : NULL,
406 &sendfile_iodone, sfio);
407 KASSERT(rv == VM_PAGER_OK, ("%s: pager fail obj %p page %p",
408 __func__, obj, pa[i]));
410 SFSTAT_INC(sf_iocnt);
411 SFSTAT_ADD(sf_pages_read, count);
412 if (i + count == npages)
413 SFSTAT_ADD(sf_rhpages_read, rhpages);
416 * Restore the valid page pointers. They are already
417 * unbusied, but still wired.
419 for (j = i; j < i + count; j++)
420 if (pa[j] == bogus_page) {
421 pa[j] = vm_page_lookup(obj,
422 OFF_TO_IDX(vmoff(j, off)));
423 KASSERT(pa[j], ("%s: page %p[%d] disappeared",
431 VM_OBJECT_WUNLOCK(obj);
433 if (nios == 0 && npages != 0)
434 SFSTAT_INC(sf_noiocnt);
440 sendfile_getobj(struct thread *td, struct file *fp, vm_object_t *obj_res,
441 struct vnode **vp_res, struct shmfd **shmfd_res, off_t *obj_size,
452 shmfd = *shmfd_res = NULL;
456 * The file descriptor must be a regular file and have a
459 if (fp->f_type == DTYPE_VNODE) {
461 vn_lock(vp, LK_SHARED | LK_RETRY);
462 if (vp->v_type != VREG) {
466 *bsize = vp->v_mount->mnt_stat.f_iosize;
467 error = VOP_GETATTR(vp, &va, td->td_ucred);
470 *obj_size = va.va_size;
476 } else if (fp->f_type == DTYPE_SHM) {
479 obj = shmfd->shm_object;
480 *obj_size = shmfd->shm_size;
486 VM_OBJECT_WLOCK(obj);
487 if ((obj->flags & OBJ_DEAD) != 0) {
488 VM_OBJECT_WUNLOCK(obj);
494 * Temporarily increase the backing VM object's reference
495 * count so that a forced reclamation of its vnode does not
496 * immediately destroy it.
498 vm_object_reference_locked(obj);
499 VM_OBJECT_WUNLOCK(obj);
511 sendfile_getsock(struct thread *td, int s, struct file **sock_fp,
520 * The socket must be a stream socket and connected.
522 error = getsock_cap(td, s, &cap_send_rights,
523 sock_fp, NULL, NULL);
526 *so = (*sock_fp)->f_data;
527 if ((*so)->so_type != SOCK_STREAM)
533 vn_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
534 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
537 struct file *sock_fp;
539 struct vm_object *obj;
541 struct mbuf *m, *mh, *mhtail;
544 struct sendfile_sync *sfs;
546 off_t off, sbytes, rem, obj_size;
547 int error, softerr, bsize, hdrlen;
556 error = sendfile_getobj(td, fp, &obj, &vp, &shmfd, &obj_size, &bsize);
560 error = sendfile_getsock(td, sockfd, &sock_fp, &so);
565 error = mac_socket_check_send(td->td_ucred, so);
570 SFSTAT_INC(sf_syscalls);
571 SFSTAT_ADD(sf_rhpages_requested, SF_READAHEAD(flags));
573 if (flags & SF_SYNC) {
574 sfs = malloc(sizeof *sfs, M_TEMP, M_WAITOK | M_ZERO);
575 mtx_init(&sfs->mtx, "sendfile", NULL, MTX_DEF);
576 cv_init(&sfs->cv, "sendfile");
579 rem = nbytes ? omin(nbytes, obj_size - offset) : obj_size - offset;
582 * Protect against multiple writers to the socket.
584 * XXXRW: Historically this has assumed non-interruptibility, so now
585 * we implement that, but possibly shouldn't.
587 (void)sblock(&so->so_snd, SBL_WAIT | SBL_NOINTR);
590 * Loop through the pages of the file, starting with the requested
591 * offset. Get a file page (do I/O if necessary), map the file page
592 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
594 * This is done in two loops. The inner loop turns as many pages
595 * as it can, up to available socket buffer space, without blocking
596 * into mbufs to have it bulk delivered into the socket send buffer.
597 * The outer loop checks the state and available space of the socket
598 * and takes care of the overall progress.
600 for (off = offset; rem > 0; ) {
604 int nios, space, npages, rhpages;
608 * Check the socket state for ongoing connection,
609 * no errors and space in socket buffer.
610 * If space is low allow for the remainder of the
611 * file to be processed if it fits the socket buffer.
612 * Otherwise block in waiting for sufficient space
613 * to proceed, or if the socket is nonblocking, return
614 * to userland with EAGAIN while reporting how far
616 * We wait until the socket buffer has significant free
617 * space to do bulk sends. This makes good use of file
618 * system read ahead and allows packet segmentation
619 * offloading hardware to take over lots of work. If
620 * we were not careful here we would send off only one
623 SOCKBUF_LOCK(&so->so_snd);
624 if (so->so_snd.sb_lowat < so->so_snd.sb_hiwat / 2)
625 so->so_snd.sb_lowat = so->so_snd.sb_hiwat / 2;
627 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
629 SOCKBUF_UNLOCK(&so->so_snd);
631 } else if (so->so_error) {
632 error = so->so_error;
634 SOCKBUF_UNLOCK(&so->so_snd);
637 if ((so->so_state & SS_ISCONNECTED) == 0) {
638 SOCKBUF_UNLOCK(&so->so_snd);
643 space = sbspace(&so->so_snd);
646 space < so->so_snd.sb_lowat)) {
647 if (so->so_state & SS_NBIO) {
648 SOCKBUF_UNLOCK(&so->so_snd);
653 * sbwait drops the lock while sleeping.
654 * When we loop back to retry_space the
655 * state may have changed and we retest
658 error = sbwait(&so->so_snd);
660 * An error from sbwait usually indicates that we've
661 * been interrupted by a signal. If we've sent anything
662 * then return bytes sent, otherwise return the error.
665 SOCKBUF_UNLOCK(&so->so_snd);
670 SOCKBUF_UNLOCK(&so->so_snd);
673 * At the beginning of the first loop check if any headers
674 * are specified and copy them into mbufs. Reduce space in
675 * the socket buffer by the size of the header mbuf chain.
676 * Clear hdr_uio here and hdrlen at the end of the first loop.
678 if (hdr_uio != NULL && hdr_uio->uio_resid > 0) {
679 hdr_uio->uio_td = td;
680 hdr_uio->uio_rw = UIO_WRITE;
681 mh = m_uiotombuf(hdr_uio, M_WAITOK, space, 0, 0);
682 hdrlen = m_length(mh, &mhtail);
685 * If header consumed all the socket buffer space,
686 * don't waste CPU cycles and jump to the end.
697 error = vn_lock(vp, LK_SHARED);
700 error = VOP_GETATTR(vp, &va, td->td_ucred);
701 if (error != 0 || off >= va.va_size) {
705 if (va.va_size != obj_size) {
706 obj_size = va.va_size;
708 omin(nbytes + offset, obj_size) : obj_size;
716 npages = howmany(space + (off & PAGE_MASK), PAGE_SIZE);
719 * Calculate maximum allowed number of pages for readahead
720 * at this iteration. If SF_USER_READAHEAD was set, we don't
721 * do any heuristics and use exactly the value supplied by
722 * application. Otherwise, we allow readahead up to "rem".
723 * If application wants more, let it be, but there is no
724 * reason to go above MAXPHYS. Also check against "obj_size",
725 * since vm_pager_has_page() can hint beyond EOF.
727 if (flags & SF_USER_READAHEAD) {
728 rhpages = SF_READAHEAD(flags);
730 rhpages = howmany(rem + (off & PAGE_MASK), PAGE_SIZE) -
732 rhpages += SF_READAHEAD(flags);
734 rhpages = min(howmany(MAXPHYS, PAGE_SIZE), rhpages);
735 rhpages = min(howmany(obj_size - trunc_page(off), PAGE_SIZE) -
738 sfio = malloc(sizeof(struct sf_io) +
739 npages * sizeof(vm_page_t), M_TEMP, M_WAITOK);
740 refcount_init(&sfio->nios, 1);
744 nios = sendfile_swapin(obj, sfio, off, space, npages, rhpages,
748 * Loop and construct maximum sized mbuf chain to be bulk
749 * dumped into socket buffer.
752 for (int i = 0; i < npages; i++) {
756 * If a page wasn't grabbed successfully, then
757 * trim the array. Can happen only with SF_NODISKIO.
761 fixspace(npages, i, off, &space);
768 * Get a sendfile buf. When allocating the
769 * first buffer for mbuf chain, we usually
770 * wait as long as necessary, but this wait
771 * can be interrupted. For consequent
772 * buffers, do not sleep, since several
773 * threads might exhaust the buffers and then
776 sf = sf_buf_alloc(pa[i],
777 m != NULL ? SFB_NOWAIT : SFB_CATCH);
779 SFSTAT_INC(sf_allocfail);
780 for (int j = i; j < npages; j++) {
782 vm_page_unwire(pa[j], PQ_INACTIVE);
783 vm_page_unlock(pa[j]);
787 fixspace(npages, i, off, &space);
792 m0 = m_get(M_WAITOK, MT_DATA);
793 m0->m_ext.ext_buf = (char *)sf_buf_kva(sf);
794 m0->m_ext.ext_size = PAGE_SIZE;
795 m0->m_ext.ext_arg1 = sf;
796 m0->m_ext.ext_type = EXT_SFBUF;
797 m0->m_ext.ext_flags = EXT_FLAG_EMBREF;
798 m0->m_ext.ext_free = sendfile_free_mext;
800 * SF_NOCACHE sets the page as being freed upon send.
801 * However, we ignore it for the last page in 'space',
802 * if the page is truncated, and we got more data to
803 * send (rem > space), or if we have readahead
804 * configured (rhpages > 0).
806 if ((flags & SF_NOCACHE) &&
808 !((off + space) & PAGE_MASK) ||
809 !(rem > space || rhpages > 0)))
810 m0->m_ext.ext_flags |= EXT_FLAG_NOCACHE;
812 m0->m_ext.ext_flags |= EXT_FLAG_SYNC;
813 m0->m_ext.ext_arg2 = sfs;
816 mtx_unlock(&sfs->mtx);
818 m0->m_ext.ext_count = 1;
819 m0->m_flags |= (M_EXT | M_RDONLY);
821 m0->m_flags |= M_NOTREADY;
822 m0->m_data = (char *)sf_buf_kva(sf) +
823 (vmoff(i, off) & PAGE_MASK);
824 m0->m_len = xfsize(i, npages, off, space);
829 /* Append to mbuf chain. */
840 /* Keep track of bytes processed. */
844 /* Prepend header, if any. */
853 KASSERT(softerr, ("%s: m NULL, no error", __func__));
859 /* Add the buffer chain to the socket buffer. */
860 KASSERT(m_length(m, NULL) == space + hdrlen,
861 ("%s: mlen %u space %d hdrlen %d",
862 __func__, m_length(m, NULL), space, hdrlen));
864 CURVNET_SET(so->so_vnet);
867 * If sendfile_swapin() didn't initiate any I/Os,
868 * which happens if all data is cached in VM, then
869 * we can send data right now without the
870 * PRUS_NOTREADY flag.
873 error = (*so->so_proto->pr_usrreqs->pru_send)
874 (so, 0, m, NULL, NULL, td);
876 sfio->npages = npages;
878 error = (*so->so_proto->pr_usrreqs->pru_send)
879 (so, PRUS_NOTREADY, m, NULL, NULL, td);
880 sendfile_iodone(sfio, NULL, 0, 0);
884 m = NULL; /* pru_send always consumes */
887 sbytes += space + hdrlen;
897 * Send trailers. Wimp out and use writev(2).
899 if (trl_uio != NULL) {
900 sbunlock(&so->so_snd);
901 error = kern_writev(td, sockfd, trl_uio);
903 sbytes += td->td_retval[0];
908 sbunlock(&so->so_snd);
911 * If there was no error we have to clear td->td_retval[0]
912 * because it may have been set by writev.
915 td->td_retval[0] = 0;
921 vm_object_deallocate(obj);
932 cv_wait(&sfs->cv, &sfs->mtx);
933 KASSERT(sfs->count == 0, ("sendfile sync still busy"));
934 cv_destroy(&sfs->cv);
935 mtx_destroy(&sfs->mtx);
939 if (error == ERESTART)
946 sendfile(struct thread *td, struct sendfile_args *uap, int compat)
949 struct uio *hdr_uio, *trl_uio;
955 * File offset must be positive. If it goes beyond EOF
956 * we send only the header/trailer and no payload data.
962 hdr_uio = trl_uio = NULL;
964 if (uap->hdtr != NULL) {
965 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
968 if (hdtr.headers != NULL) {
969 error = copyinuio(hdtr.headers, hdtr.hdr_cnt,
973 #ifdef COMPAT_FREEBSD4
975 * In FreeBSD < 5.0 the nbytes to send also included
976 * the header. If compat is specified subtract the
977 * header size from nbytes.
980 if (uap->nbytes > hdr_uio->uio_resid)
981 uap->nbytes -= hdr_uio->uio_resid;
987 if (hdtr.trailers != NULL) {
988 error = copyinuio(hdtr.trailers, hdtr.trl_cnt,
995 AUDIT_ARG_FD(uap->fd);
998 * sendfile(2) can start at any offset within a file so we require
999 * CAP_READ+CAP_SEEK = CAP_PREAD.
1001 if ((error = fget_read(td, uap->fd, &cap_pread_rights, &fp)) != 0)
1004 error = fo_sendfile(fp, uap->s, hdr_uio, trl_uio, uap->offset,
1005 uap->nbytes, &sbytes, uap->flags, td);
1008 if (uap->sbytes != NULL)
1009 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1012 free(hdr_uio, M_IOV);
1013 free(trl_uio, M_IOV);
1020 * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1021 * struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1023 * Send a file specified by 'fd' and starting at 'offset' to a socket
1024 * specified by 's'. Send only 'nbytes' of the file or until EOF if nbytes ==
1025 * 0. Optionally add a header and/or trailer to the socket output. If
1026 * specified, write the total number of bytes sent into *sbytes.
1029 sys_sendfile(struct thread *td, struct sendfile_args *uap)
1032 return (sendfile(td, uap, 0));
1035 #ifdef COMPAT_FREEBSD4
1037 freebsd4_sendfile(struct thread *td, struct freebsd4_sendfile_args *uap)
1039 struct sendfile_args args;
1043 args.offset = uap->offset;
1044 args.nbytes = uap->nbytes;
1045 args.hdtr = uap->hdtr;
1046 args.sbytes = uap->sbytes;
1047 args.flags = uap->flags;
1049 return (sendfile(td, &args, 1));
1051 #endif /* COMPAT_FREEBSD4 */