2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2007-2009 Google Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above
14 * copyright notice, this list of conditions and the following disclaimer
15 * in the documentation and/or other materials provided with the
17 * * Neither the name of Google Inc. nor the names of its
18 * contributors may be used to endorse or promote products derived from
19 * this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33 * Copyright (C) 2005 Csaba Henk.
34 * All rights reserved.
36 * Redistribution and use in source and binary forms, with or without
37 * modification, are permitted provided that the following conditions
39 * 1. Redistributions of source code must retain the above copyright
40 * notice, this list of conditions and the following disclaimer.
41 * 2. Redistributions in binary form must reproduce the above copyright
42 * notice, this list of conditions and the following disclaimer in the
43 * documentation and/or other materials provided with the distribution.
45 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD$");
61 #include <sys/types.h>
62 #include <sys/module.h>
63 #include <sys/systm.h>
64 #include <sys/errno.h>
65 #include <sys/param.h>
66 #include <sys/kernel.h>
69 #include <sys/malloc.h>
70 #include <sys/queue.h>
73 #include <sys/mutex.h>
74 #include <sys/rwlock.h>
76 #include <sys/mount.h>
77 #include <sys/vnode.h>
79 #include <sys/unistd.h>
80 #include <sys/filedesc.h>
82 #include <sys/fcntl.h>
85 #include <sys/sysctl.h>
88 #include <vm/vm_extern.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_object.h>
95 #include "fuse_file.h"
96 #include "fuse_node.h"
97 #include "fuse_internal.h"
101 SDT_PROVIDER_DECLARE(fusefs);
104 * arg0: verbosity. Higher numbers give more verbose messages
105 * arg1: Textual message
107 SDT_PROBE_DEFINE2(fusefs, , io, trace, "int", "char*");
110 fuse_read_directbackend(struct vnode *vp, struct uio *uio,
111 struct ucred *cred, struct fuse_filehandle *fufh);
113 fuse_read_biobackend(struct vnode *vp, struct uio *uio, int ioflag,
114 struct ucred *cred, struct fuse_filehandle *fufh, pid_t pid);
116 fuse_write_directbackend(struct vnode *vp, struct uio *uio,
117 struct ucred *cred, struct fuse_filehandle *fufh, int ioflag);
119 fuse_write_biobackend(struct vnode *vp, struct uio *uio,
120 struct ucred *cred, struct fuse_filehandle *fufh, int ioflag, pid_t pid);
122 SDT_PROBE_DEFINE5(fusefs, , io, io_dispatch, "struct vnode*", "struct uio*",
123 "int", "struct ucred*", "struct fuse_filehandle*");
125 fuse_io_dispatch(struct vnode *vp, struct uio *uio, int ioflag,
126 struct ucred *cred, pid_t pid)
128 struct fuse_filehandle *fufh;
132 MPASS(vp->v_type == VREG || vp->v_type == VDIR);
134 fflag = (uio->uio_rw == UIO_READ) ? FREAD : FWRITE;
135 err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
137 printf("FUSE: io dispatch: filehandles are closed\n");
140 SDT_PROBE5(fusefs, , io, io_dispatch, vp, uio, ioflag, cred, fufh);
143 * Ideally, when the daemon asks for direct io at open time, the
144 * standard file flag should be set according to this, so that would
145 * just change the default mode, which later on could be changed via
147 * But this doesn't work, the O_DIRECT flag gets cleared at some point
148 * (don't know where). So to make any use of the Fuse direct_io option,
149 * we hardwire it into the file's private data (similarly to Linux,
152 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
154 switch (uio->uio_rw) {
157 SDT_PROBE2(fusefs, , io, trace, 1,
158 "direct read of vnode");
159 err = fuse_read_directbackend(vp, uio, cred, fufh);
161 SDT_PROBE2(fusefs, , io, trace, 1,
162 "buffered read of vnode");
163 err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh,
169 * Kludge: simulate write-through caching via write-around
170 * caching. Same effect, as far as never caching dirty data,
171 * but slightly pessimal in that newly written data is not
174 if (directio || fuse_data_cache_mode == FUSE_CACHE_WT) {
177 SDT_PROBE2(fusefs, , io, trace, 1,
178 "direct write of vnode");
179 start = uio->uio_offset;
180 end = start + uio->uio_resid;
181 v_inval_buf_range(vp, start, end, fuse_iosize(vp));
182 err = fuse_write_directbackend(vp, uio, cred, fufh,
185 SDT_PROBE2(fusefs, , io, trace, 1,
186 "buffered write of vnode");
187 err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag,
192 panic("uninterpreted mode passed to fuse_io_dispatch");
198 SDT_PROBE_DEFINE3(fusefs, , io, read_bio_backend_start, "int", "int", "int");
199 SDT_PROBE_DEFINE2(fusefs, , io, read_bio_backend_feed, "int", "int");
200 SDT_PROBE_DEFINE3(fusefs, , io, read_bio_backend_end, "int", "ssize_t", "int");
202 fuse_read_biobackend(struct vnode *vp, struct uio *uio, int ioflag,
203 struct ucred *cred, struct fuse_filehandle *fufh, pid_t pid)
208 int err, n = 0, on = 0;
211 const int biosize = fuse_iosize(vp);
213 if (uio->uio_offset < 0)
216 filesize = VTOFUD(vp)->filesize;
218 for (err = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
219 if (fuse_isdeadfs(vp)) {
223 if (filesize - uio->uio_offset <= 0)
225 lbn = uio->uio_offset / biosize;
226 on = uio->uio_offset & (biosize - 1);
228 SDT_PROBE3(fusefs, , io, read_bio_backend_start,
229 biosize, (int)lbn, on);
231 if ((off_t)lbn * biosize >= filesize) {
233 } else if ((off_t)(lbn + 1) * biosize > filesize) {
234 bcount = filesize - (off_t)lbn *biosize;
239 /* TODO: readahead. See ext2_read for an example */
240 err = bread(vp, lbn, bcount, NOCRED, &bp);
248 * on is the offset into the current bp. Figure out how many
249 * bytes we can copy out of the bp. Note that bcount is
250 * NOT DEV_BSIZE aligned.
252 * Then figure out how many bytes we can copy into the uio.
257 n = MIN((unsigned)(bcount - on), uio->uio_resid);
259 SDT_PROBE2(fusefs, , io, read_bio_backend_feed,
260 n, n + (int)bp->b_resid);
261 err = uiomove(bp->b_data + on, n, uio);
263 vfs_bio_brelse(bp, ioflag);
264 SDT_PROBE3(fusefs, , io, read_bio_backend_end, err,
271 SDT_PROBE_DEFINE1(fusefs, , io, read_directbackend_start,
272 "struct fuse_read_in*");
273 SDT_PROBE_DEFINE2(fusefs, , io, read_directbackend_complete,
274 "struct fuse_dispatcher*", "struct uio*");
277 fuse_read_directbackend(struct vnode *vp, struct uio *uio,
278 struct ucred *cred, struct fuse_filehandle *fufh)
280 struct fuse_dispatcher fdi;
281 struct fuse_read_in *fri;
284 if (uio->uio_resid == 0)
290 * XXX In "normal" case we use an intermediate kernel buffer for
291 * transmitting data from daemon's context to ours. Eventually, we should
292 * get rid of this. Anyway, if the target uio lives in sysspace (we are
293 * called from pageops), and the input data doesn't need kernel-side
294 * processing (we are not called from readdir) we can already invoke
295 * an optimized, "peer-to-peer" I/O routine.
297 while (uio->uio_resid > 0) {
298 fdi.iosize = sizeof(*fri);
299 fdisp_make_vp(&fdi, FUSE_READ, vp, uio->uio_td, cred);
301 fri->fh = fufh->fh_id;
302 fri->offset = uio->uio_offset;
303 fri->size = MIN(uio->uio_resid,
304 fuse_get_mpdata(vp->v_mount)->max_read);
306 SDT_PROBE1(fusefs, , io, read_directbackend_start, fri);
308 if ((err = fdisp_wait_answ(&fdi)))
311 SDT_PROBE2(fusefs, , io, read_directbackend_complete,
314 if ((err = uiomove(fdi.answ, MIN(fri->size, fdi.iosize), uio)))
316 if (fdi.iosize < fri->size)
326 fuse_write_directbackend(struct vnode *vp, struct uio *uio,
327 struct ucred *cred, struct fuse_filehandle *fufh, int ioflag)
329 struct fuse_vnode_data *fvdat = VTOFUD(vp);
330 struct fuse_write_in *fwi;
331 struct fuse_write_out *fwo;
332 struct fuse_dispatcher fdi;
335 off_t as_written_offset;
338 bool direct_io = fufh->fuse_open_flags & FOPEN_DIRECT_IO;
340 if (uio->uio_resid == 0)
342 if (ioflag & IO_APPEND)
343 uio_setoffset(uio, fvdat->filesize);
347 while (uio->uio_resid > 0) {
348 chunksize = MIN(uio->uio_resid,
349 fuse_get_mpdata(vp->v_mount)->max_write);
351 fdi.iosize = sizeof(*fwi) + chunksize;
352 fdisp_make_vp(&fdi, FUSE_WRITE, vp, uio->uio_td, cred);
355 fwi->fh = fufh->fh_id;
356 fwi->offset = uio->uio_offset;
357 fwi->size = chunksize;
358 fwi_data = (char *)fdi.indata + sizeof(*fwi);
360 if ((err = uiomove(fwi_data, chunksize, uio)))
364 err = fdisp_wait_answ(&fdi);
365 if (err == ERESTART || err == EINTR || err == EWOULDBLOCK) {
367 * Rewind the uio so dofilewrite will know it's
370 uio->uio_resid += fwi->size;
371 uio->uio_offset -= fwi->size;
373 * Change ERESTART into EINTR because we can't rewind
374 * uio->uio_iov. Basically, once uiomove(9) has been
375 * called, it's impossible to restart a syscall.
384 fwo = ((struct fuse_write_out *)fdi.answ);
386 /* Adjust the uio in the case of short writes */
387 diff = fwi->size - fwo->size;
388 as_written_offset = uio->uio_offset - diff;
390 if (as_written_offset - diff > fvdat->filesize &&
391 fuse_data_cache_mode != FUSE_CACHE_UC) {
392 fuse_vnode_setsize(vp, cred, as_written_offset);
393 fvdat->flag &= ~FN_SIZECHANGE;
397 printf("WARNING: misbehaving FUSE filesystem "
398 "wrote more data than we provided it\n");
401 } else if (diff > 0) {
404 printf("WARNING: misbehaving FUSE filesystem: "
405 "short writes are only allowed with "
408 if (ioflag & IO_DIRECT) {
410 uio->uio_resid += diff;
411 uio->uio_offset -= diff;
414 /* Resend the unwritten portion of data */
415 fdi.iosize = sizeof(*fwi) + diff;
416 /* Refresh fdi without clearing data buffer */
417 fdisp_refresh_vp(&fdi, FUSE_WRITE, vp,
420 MPASS2(fwi == fdi.indata, "FUSE dispatcher "
421 "reallocated despite no increase in "
423 void *src = (char*)fwi_data + fwo->size;
424 memmove(fwi_data, src, diff);
425 fwi->fh = fufh->fh_id;
426 fwi->offset = as_written_offset;
438 SDT_PROBE_DEFINE6(fusefs, , io, write_biobackend_start, "int64_t", "int", "int",
439 "struct uio*", "int", "bool");
440 SDT_PROBE_DEFINE2(fusefs, , io, write_biobackend_append_race, "long", "int");
443 fuse_write_biobackend(struct vnode *vp, struct uio *uio,
444 struct ucred *cred, struct fuse_filehandle *fufh, int ioflag, pid_t pid)
446 struct fuse_vnode_data *fvdat = VTOFUD(vp);
452 const int biosize = fuse_iosize(vp);
454 KASSERT(uio->uio_rw == UIO_WRITE, ("ncl_write mode"));
455 if (vp->v_type != VREG)
457 if (uio->uio_offset < 0)
459 if (uio->uio_resid == 0)
461 if (ioflag & IO_APPEND)
462 uio_setoffset(uio, fvdat->filesize);
465 * Find all of this file's B_NEEDCOMMIT buffers. If our writes
466 * would exceed the local maximum per-file write commit size when
467 * combined with those, we must decide whether to flush,
468 * go synchronous, or return err. We don't bother checking
469 * IO_UNIT -- we just make all writes atomic anyway, as there's
470 * no point optimizing for something that really won't ever happen.
473 if (fuse_isdeadfs(vp)) {
477 lbn = uio->uio_offset / biosize;
478 on = uio->uio_offset & (biosize - 1);
479 n = MIN((unsigned)(biosize - on), uio->uio_resid);
483 * Handle direct append and file extension cases, calculate
484 * unaligned buffer size.
486 if (uio->uio_offset == fvdat->filesize && n) {
488 * Get the buffer (in its pre-append state to maintain
489 * B_CACHE if it was previously set). Resize the
490 * nfsnode after we have locked the buffer to prevent
491 * readers from reading garbage.
494 SDT_PROBE6(fusefs, , io, write_biobackend_start,
495 lbn, on, n, uio, bcount, true);
496 bp = getblk(vp, lbn, bcount, PCATCH, 0, 0);
501 err = fuse_vnode_setsize(vp, cred,
502 uio->uio_offset + n);
507 save = bp->b_flags & B_CACHE;
509 allocbuf(bp, bcount);
514 * Obtain the locked cache block first, and then
515 * adjust the file's size as appropriate.
518 if ((off_t)lbn * biosize + bcount < fvdat->filesize) {
519 if ((off_t)(lbn + 1) * biosize < fvdat->filesize)
522 bcount = fvdat->filesize -
525 SDT_PROBE6(fusefs, , io, write_biobackend_start,
526 lbn, on, n, uio, bcount, false);
527 bp = getblk(vp, lbn, bcount, PCATCH, 0, 0);
528 if (bp && uio->uio_offset + n > fvdat->filesize) {
529 err = fuse_vnode_setsize(vp, cred,
530 uio->uio_offset + n);
543 * Issue a READ if B_CACHE is not set. In special-append
544 * mode, B_CACHE is based on the buffer prior to the write
545 * op and is typically set, avoiding the read. If a read
546 * is required in special append mode, the server will
547 * probably send us a short-read since we extended the file
548 * on our end, resulting in b_resid == 0 and, thusly,
549 * B_CACHE getting set.
551 * We can also avoid issuing the read if the write covers
552 * the entire buffer. We have to make sure the buffer state
553 * is reasonable in this case since we will not be initiating
554 * I/O. See the comments in kern/vfs_bio.c's getblk() for
557 * B_CACHE may also be set due to the buffer being cached
561 if (on == 0 && n == bcount) {
562 bp->b_flags |= B_CACHE;
563 bp->b_flags &= ~B_INVAL;
564 bp->b_ioflags &= ~BIO_ERROR;
566 if ((bp->b_flags & B_CACHE) == 0) {
567 bp->b_iocmd = BIO_READ;
568 vfs_busy_pages(bp, 0);
569 fuse_io_strategy(vp, bp);
570 if ((err = bp->b_error)) {
575 if (bp->b_wcred == NOCRED)
576 bp->b_wcred = crhold(cred);
579 * If dirtyend exceeds file size, chop it down. This should
580 * not normally occur but there is an append race where it
581 * might occur XXX, so we log it.
583 * If the chopping creates a reverse-indexed or degenerate
584 * situation with dirtyoff/end, we 0 both of them.
587 if (bp->b_dirtyend > bcount) {
588 SDT_PROBE2(fusefs, , io, write_biobackend_append_race,
589 (long)bp->b_blkno * biosize,
590 bp->b_dirtyend - bcount);
591 bp->b_dirtyend = bcount;
593 if (bp->b_dirtyoff >= bp->b_dirtyend)
594 bp->b_dirtyoff = bp->b_dirtyend = 0;
597 * If the new write will leave a contiguous dirty
598 * area, just update the b_dirtyoff and b_dirtyend,
599 * otherwise force a write rpc of the old dirty area.
601 * While it is possible to merge discontiguous writes due to
602 * our having a B_CACHE buffer ( and thus valid read data
603 * for the hole), we don't because it could lead to
604 * significant cache coherency problems with multiple clients,
605 * especially if locking is implemented later on.
607 * as an optimization we could theoretically maintain
608 * a linked list of discontinuous areas, but we would still
609 * have to commit them separately so there isn't much
610 * advantage to it except perhaps a bit of asynchronization.
613 if (bp->b_dirtyend > 0 &&
614 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) {
616 * Yes, we mean it. Write out everything to "storage"
617 * immediately, without hesitation. (Apart from other
618 * reasons: the only way to know if a write is valid
619 * if its actually written out.)
622 if (bp->b_error == EINTR) {
628 err = uiomove((char *)bp->b_data + on, n, uio);
631 * Since this block is being modified, it must be written
632 * again and not just committed. Since write clustering does
633 * not work for the stage 1 data write, only the stage 2
634 * commit rpc, we have to clear B_CLUSTEROK as well.
636 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
639 bp->b_ioflags |= BIO_ERROR;
645 * Only update dirtyoff/dirtyend if not a degenerate
649 if (bp->b_dirtyend > 0) {
650 bp->b_dirtyoff = MIN(on, bp->b_dirtyoff);
651 bp->b_dirtyend = MAX((on + n), bp->b_dirtyend);
654 bp->b_dirtyend = on + n;
656 vfs_bio_set_valid(bp, on, n);
661 } while (uio->uio_resid > 0 && n > 0);
663 if (fuse_sync_resize && (fvdat->flag & FN_SIZECHANGE) != 0)
664 fuse_vnode_savesize(vp, cred, pid);
670 fuse_io_strategy(struct vnode *vp, struct buf *bp)
672 struct fuse_filehandle *fufh;
673 struct fuse_vnode_data *fvdat = VTOFUD(vp);
680 /* We don't know the true pid when we're dealing with the cache */
683 const int biosize = fuse_iosize(vp);
685 MPASS(vp->v_type == VREG || vp->v_type == VDIR);
686 MPASS(bp->b_iocmd == BIO_READ || bp->b_iocmd == BIO_WRITE);
688 fflag = bp->b_iocmd == BIO_READ ? FREAD : FWRITE;
689 cred = bp->b_iocmd == BIO_READ ? bp->b_rcred : bp->b_wcred;
690 error = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
691 if (bp->b_iocmd == BIO_READ && error == EBADF) {
693 * This may be a read-modify-write operation on a cached file
694 * opened O_WRONLY. The FUSE protocol allows this.
696 * TODO: eliminate this hacky check once the FUFH table is gone
698 error = fuse_filehandle_get(vp, FWRITE, &fufh, cred, pid);
701 printf("FUSE: strategy: filehandles are closed\n");
702 bp->b_ioflags |= BIO_ERROR;
710 uiop->uio_iovcnt = 1;
711 uiop->uio_segflg = UIO_SYSSPACE;
712 uiop->uio_td = curthread;
715 * clear BIO_ERROR and B_INVAL state prior to initiating the I/O. We
716 * do this here so we do not have to do it in all the code that
719 bp->b_flags &= ~B_INVAL;
720 bp->b_ioflags &= ~BIO_ERROR;
722 KASSERT(!(bp->b_flags & B_DONE),
723 ("fuse_io_strategy: bp %p already marked done", bp));
724 if (bp->b_iocmd == BIO_READ) {
725 io.iov_len = uiop->uio_resid = bp->b_bcount;
726 io.iov_base = bp->b_data;
727 uiop->uio_rw = UIO_READ;
729 uiop->uio_offset = ((off_t)bp->b_blkno) * biosize;
730 error = fuse_read_directbackend(vp, uiop, cred, fufh);
732 /* XXXCEM: Potentially invalid access to cached_attrs here */
733 if ((!error && uiop->uio_resid) ||
734 (fsess_opt_brokenio(vnode_mount(vp)) && error == EIO &&
735 uiop->uio_offset < fvdat->filesize && fvdat->filesize > 0 &&
736 uiop->uio_offset >= fvdat->cached_attrs.va_size)) {
738 * If we had a short read with no error, we must have
739 * hit a file hole. We should zero-fill the remainder.
740 * This can also occur if the server hits the file EOF.
742 * Holes used to be able to occur due to pending
743 * writes, but that is not possible any longer.
745 int nread = bp->b_bcount - uiop->uio_resid;
746 int left = uiop->uio_resid;
749 printf("FUSE: Fix broken io: offset %ju, "
750 " resid %zd, file size %ju/%ju\n",
751 (uintmax_t)uiop->uio_offset,
752 uiop->uio_resid, fvdat->filesize,
753 fvdat->cached_attrs.va_size);
757 bzero((char *)bp->b_data + nread, left);
761 bp->b_ioflags |= BIO_ERROR;
766 * If we only need to commit, try to commit
768 if (bp->b_flags & B_NEEDCOMMIT) {
769 SDT_PROBE2(fusefs, , io, trace, 1,
770 "write: B_NEEDCOMMIT flags set");
773 * Setup for actual write
775 if ((off_t)bp->b_blkno * biosize + bp->b_dirtyend >
777 bp->b_dirtyend = fvdat->filesize -
778 (off_t)bp->b_blkno * biosize;
780 if (bp->b_dirtyend > bp->b_dirtyoff) {
781 io.iov_len = uiop->uio_resid = bp->b_dirtyend
783 uiop->uio_offset = (off_t)bp->b_blkno * biosize
785 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff;
786 uiop->uio_rw = UIO_WRITE;
788 error = fuse_write_directbackend(vp, uiop, cred, fufh, 0);
790 if (error == EINTR || error == ETIMEDOUT
791 || (!error && (bp->b_flags & B_NEEDCOMMIT))) {
793 bp->b_flags &= ~(B_INVAL | B_NOCACHE);
794 if ((bp->b_flags & B_PAGING) == 0) {
796 bp->b_flags &= ~B_DONE;
798 if ((error == EINTR || error == ETIMEDOUT) &&
799 (bp->b_flags & B_ASYNC) == 0)
800 bp->b_flags |= B_EINTR;
803 bp->b_ioflags |= BIO_ERROR;
804 bp->b_flags |= B_INVAL;
807 bp->b_dirtyoff = bp->b_dirtyend = 0;
815 bp->b_resid = uiop->uio_resid;
821 fuse_io_flushbuf(struct vnode *vp, int waitfor, struct thread *td)
823 struct vop_fsync_args a = {
825 .a_waitfor = waitfor,
829 return (vop_stdfsync(&a));
833 * Flush and invalidate all dirty buffers. If another process is already
834 * doing the flush, just wait for completion.
837 fuse_io_invalbuf(struct vnode *vp, struct thread *td)
839 struct fuse_vnode_data *fvdat = VTOFUD(vp);
842 if (vp->v_iflag & VI_DOOMED)
845 ASSERT_VOP_ELOCKED(vp, "fuse_io_invalbuf");
847 while (fvdat->flag & FN_FLUSHINPROG) {
848 struct proc *p = td->td_proc;
850 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF)
852 fvdat->flag |= FN_FLUSHWANT;
853 tsleep(&fvdat->flag, PRIBIO + 2, "fusevinv", 2 * hz);
857 if (SIGNOTEMPTY(p->p_siglist) ||
858 SIGNOTEMPTY(td->td_siglist))
865 fvdat->flag |= FN_FLUSHINPROG;
867 if (vp->v_bufobj.bo_object != NULL) {
868 VM_OBJECT_WLOCK(vp->v_bufobj.bo_object);
869 vm_object_page_clean(vp->v_bufobj.bo_object, 0, 0, OBJPC_SYNC);
870 VM_OBJECT_WUNLOCK(vp->v_bufobj.bo_object);
872 error = vinvalbuf(vp, V_SAVE, PCATCH, 0);
874 if (error == ERESTART || error == EINTR) {
875 fvdat->flag &= ~FN_FLUSHINPROG;
876 if (fvdat->flag & FN_FLUSHWANT) {
877 fvdat->flag &= ~FN_FLUSHWANT;
878 wakeup(&fvdat->flag);
882 error = vinvalbuf(vp, V_SAVE, PCATCH, 0);
884 fvdat->flag &= ~FN_FLUSHINPROG;
885 if (fvdat->flag & FN_FLUSHWANT) {
886 fvdat->flag &= ~FN_FLUSHWANT;
887 wakeup(&fvdat->flag);