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(fuse);
104 * arg0: verbosity. Higher numbers give more verbose messages
105 * arg1: Textual message
107 SDT_PROBE_DEFINE2(fuse, , 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,
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(fuse, , 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(fuse, , 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(fuse, , io, trace, 1,
158 "direct read of vnode");
159 err = fuse_read_directbackend(vp, uio, cred, fufh);
161 SDT_PROBE2(fuse, , io, trace, 1,
162 "buffered read of vnode");
163 err = fuse_read_biobackend(vp, uio, cred, fufh, pid);
168 * Kludge: simulate write-through caching via write-around
169 * caching. Same effect, as far as never caching dirty data,
170 * but slightly pessimal in that newly written data is not
173 if (directio || fuse_data_cache_mode == FUSE_CACHE_WT) {
174 SDT_PROBE2(fuse, , io, trace, 1,
175 "direct write of vnode");
176 err = fuse_write_directbackend(vp, uio, cred, fufh,
179 SDT_PROBE2(fuse, , io, trace, 1,
180 "buffered write of vnode");
181 err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag,
186 panic("uninterpreted mode passed to fuse_io_dispatch");
192 SDT_PROBE_DEFINE3(fuse, , io, read_bio_backend_start, "int", "int", "int");
193 SDT_PROBE_DEFINE2(fuse, , io, read_bio_backend_feed, "int", "int");
194 SDT_PROBE_DEFINE3(fuse, , io, read_bio_backend_end, "int", "ssize_t", "int");
196 fuse_read_biobackend(struct vnode *vp, struct uio *uio,
197 struct ucred *cred, struct fuse_filehandle *fufh, pid_t pid)
202 int err = 0, n = 0, on = 0;
205 const int biosize = fuse_iosize(vp);
207 if (uio->uio_resid == 0)
209 if (uio->uio_offset < 0)
213 filesize = VTOFUD(vp)->filesize;
216 if (fuse_isdeadfs(vp)) {
220 lbn = uio->uio_offset / biosize;
221 on = uio->uio_offset & (biosize - 1);
223 SDT_PROBE3(fuse, , io, read_bio_backend_start,
224 biosize, (int)lbn, on);
227 * Obtain the buffer cache block. Figure out the buffer size
228 * when we are at EOF. If we are modifying the size of the
229 * buffer based on an EOF condition we need to hold
230 * nfs_rslock() through obtaining the buffer to prevent
231 * a potential writer-appender from messing with n_size.
232 * Otherwise we may accidentally truncate the buffer and
235 * Note that bcount is *not* DEV_BSIZE aligned.
237 if ((off_t)lbn * biosize >= filesize) {
239 } else if ((off_t)(lbn + 1) * biosize > filesize) {
240 bcount = filesize - (off_t)lbn *biosize;
242 bp = getblk(vp, lbn, bcount, PCATCH, 0, 0);
248 * If B_CACHE is not set, we must issue the read. If this
249 * fails, we return an error.
252 if ((bp->b_flags & B_CACHE) == 0) {
253 bp->b_iocmd = BIO_READ;
254 vfs_busy_pages(bp, 0);
255 err = fuse_io_strategy(vp, bp);
262 * on is the offset into the current bp. Figure out how many
263 * bytes we can copy out of the bp. Note that bcount is
264 * NOT DEV_BSIZE aligned.
266 * Then figure out how many bytes we can copy into the uio.
271 n = MIN((unsigned)(bcount - on), uio->uio_resid);
273 SDT_PROBE2(fuse, , io, read_bio_backend_feed,
274 n, n + (int)bp->b_resid);
275 err = uiomove(bp->b_data + on, n, uio);
278 SDT_PROBE3(fuse, , io, read_bio_backend_end, err,
280 } while (err == 0 && uio->uio_resid > 0 && n > 0);
285 SDT_PROBE_DEFINE1(fuse, , io, read_directbackend_start, "struct fuse_read_in*");
286 SDT_PROBE_DEFINE2(fuse, , io, read_directbackend_complete,
287 "struct fuse_dispatcher*", "struct uio*");
290 fuse_read_directbackend(struct vnode *vp, struct uio *uio,
291 struct ucred *cred, struct fuse_filehandle *fufh)
293 struct fuse_dispatcher fdi;
294 struct fuse_read_in *fri;
297 if (uio->uio_resid == 0)
303 * XXX In "normal" case we use an intermediate kernel buffer for
304 * transmitting data from daemon's context to ours. Eventually, we should
305 * get rid of this. Anyway, if the target uio lives in sysspace (we are
306 * called from pageops), and the input data doesn't need kernel-side
307 * processing (we are not called from readdir) we can already invoke
308 * an optimized, "peer-to-peer" I/O routine.
310 while (uio->uio_resid > 0) {
311 fdi.iosize = sizeof(*fri);
312 fdisp_make_vp(&fdi, FUSE_READ, vp, uio->uio_td, cred);
314 fri->fh = fufh->fh_id;
315 fri->offset = uio->uio_offset;
316 fri->size = MIN(uio->uio_resid,
317 fuse_get_mpdata(vp->v_mount)->max_read);
319 SDT_PROBE1(fuse, , io, read_directbackend_start, fri);
321 if ((err = fdisp_wait_answ(&fdi)))
324 SDT_PROBE2(fuse, , io, read_directbackend_complete,
327 if ((err = uiomove(fdi.answ, MIN(fri->size, fdi.iosize), uio)))
329 if (fdi.iosize < fri->size)
339 fuse_write_directbackend(struct vnode *vp, struct uio *uio,
340 struct ucred *cred, struct fuse_filehandle *fufh, int ioflag)
342 struct fuse_vnode_data *fvdat = VTOFUD(vp);
343 struct fuse_write_in *fwi;
344 struct fuse_write_out *fwo;
345 struct fuse_dispatcher fdi;
348 off_t as_written_offset;
351 bool direct_io = fufh->fuse_open_flags & FOPEN_DIRECT_IO;
353 if (uio->uio_resid == 0)
355 if (ioflag & IO_APPEND)
356 uio_setoffset(uio, fvdat->filesize);
360 while (uio->uio_resid > 0) {
361 chunksize = MIN(uio->uio_resid,
362 fuse_get_mpdata(vp->v_mount)->max_write);
364 fdi.iosize = sizeof(*fwi) + chunksize;
365 fdisp_make_vp(&fdi, FUSE_WRITE, vp, uio->uio_td, cred);
368 fwi->fh = fufh->fh_id;
369 fwi->offset = uio->uio_offset;
370 fwi->size = chunksize;
371 fwi_data = (char *)fdi.indata + sizeof(*fwi);
373 if ((err = uiomove(fwi_data, chunksize, uio)))
377 if ((err = fdisp_wait_answ(&fdi)))
380 fwo = ((struct fuse_write_out *)fdi.answ);
382 /* Adjust the uio in the case of short writes */
383 diff = fwi->size - fwo->size;
384 as_written_offset = uio->uio_offset - diff;
386 if (as_written_offset - diff > fvdat->filesize &&
387 fuse_data_cache_mode != FUSE_CACHE_UC) {
388 fuse_vnode_setsize(vp, cred, as_written_offset);
389 fvdat->flag &= ~FN_SIZECHANGE;
393 printf("WARNING: misbehaving FUSE filesystem "
394 "wrote more data than we provided it\n");
397 } else if (diff > 0) {
400 printf("WARNING: misbehaving FUSE filesystem: "
401 "short writes are only allowed with "
404 if (ioflag & IO_DIRECT) {
406 uio->uio_resid += diff;
407 uio->uio_offset -= diff;
410 /* Resend the unwritten portion of data */
411 fdi.iosize = sizeof(*fwi) + diff;
412 /* Refresh fdi without clearing data buffer */
413 fdisp_refresh_vp(&fdi, FUSE_WRITE, vp,
416 MPASS2(fwi == fdi.indata, "FUSE dispatcher "
417 "reallocated despite no increase in "
419 void *src = (char*)fwi_data + fwo->size;
420 memmove(fwi_data, src, diff);
421 fwi->fh = fufh->fh_id;
422 fwi->offset = as_written_offset;
434 SDT_PROBE_DEFINE6(fuse, , io, write_biobackend_start, "int64_t", "int", "int",
435 "struct uio*", "int", "bool");
436 SDT_PROBE_DEFINE2(fuse, , io, write_biobackend_append_race, "long", "int");
439 fuse_write_biobackend(struct vnode *vp, struct uio *uio,
440 struct ucred *cred, struct fuse_filehandle *fufh, int ioflag, pid_t pid)
442 struct fuse_vnode_data *fvdat = VTOFUD(vp);
448 const int biosize = fuse_iosize(vp);
450 KASSERT(uio->uio_rw == UIO_WRITE, ("ncl_write mode"));
451 if (vp->v_type != VREG)
453 if (uio->uio_offset < 0)
455 if (uio->uio_resid == 0)
457 if (ioflag & IO_APPEND)
458 uio_setoffset(uio, fvdat->filesize);
461 * Find all of this file's B_NEEDCOMMIT buffers. If our writes
462 * would exceed the local maximum per-file write commit size when
463 * combined with those, we must decide whether to flush,
464 * go synchronous, or return err. We don't bother checking
465 * IO_UNIT -- we just make all writes atomic anyway, as there's
466 * no point optimizing for something that really won't ever happen.
469 if (fuse_isdeadfs(vp)) {
473 lbn = uio->uio_offset / biosize;
474 on = uio->uio_offset & (biosize - 1);
475 n = MIN((unsigned)(biosize - on), uio->uio_resid);
479 * Handle direct append and file extension cases, calculate
480 * unaligned buffer size.
482 if (uio->uio_offset == fvdat->filesize && n) {
484 * Get the buffer (in its pre-append state to maintain
485 * B_CACHE if it was previously set). Resize the
486 * nfsnode after we have locked the buffer to prevent
487 * readers from reading garbage.
490 SDT_PROBE6(fuse, , io, write_biobackend_start,
491 lbn, on, n, uio, bcount, true);
492 bp = getblk(vp, lbn, bcount, PCATCH, 0, 0);
497 err = fuse_vnode_setsize(vp, cred,
498 uio->uio_offset + n);
503 save = bp->b_flags & B_CACHE;
505 allocbuf(bp, bcount);
510 * Obtain the locked cache block first, and then
511 * adjust the file's size as appropriate.
514 if ((off_t)lbn * biosize + bcount < fvdat->filesize) {
515 if ((off_t)(lbn + 1) * biosize < fvdat->filesize)
518 bcount = fvdat->filesize -
521 SDT_PROBE6(fuse, , io, write_biobackend_start,
522 lbn, on, n, uio, bcount, false);
523 bp = getblk(vp, lbn, bcount, PCATCH, 0, 0);
524 if (bp && uio->uio_offset + n > fvdat->filesize) {
525 err = fuse_vnode_setsize(vp, cred,
526 uio->uio_offset + n);
539 * Issue a READ if B_CACHE is not set. In special-append
540 * mode, B_CACHE is based on the buffer prior to the write
541 * op and is typically set, avoiding the read. If a read
542 * is required in special append mode, the server will
543 * probably send us a short-read since we extended the file
544 * on our end, resulting in b_resid == 0 and, thusly,
545 * B_CACHE getting set.
547 * We can also avoid issuing the read if the write covers
548 * the entire buffer. We have to make sure the buffer state
549 * is reasonable in this case since we will not be initiating
550 * I/O. See the comments in kern/vfs_bio.c's getblk() for
553 * B_CACHE may also be set due to the buffer being cached
557 if (on == 0 && n == bcount) {
558 bp->b_flags |= B_CACHE;
559 bp->b_flags &= ~B_INVAL;
560 bp->b_ioflags &= ~BIO_ERROR;
562 if ((bp->b_flags & B_CACHE) == 0) {
563 bp->b_iocmd = BIO_READ;
564 vfs_busy_pages(bp, 0);
565 fuse_io_strategy(vp, bp);
566 if ((err = bp->b_error)) {
571 if (bp->b_wcred == NOCRED)
572 bp->b_wcred = crhold(cred);
575 * If dirtyend exceeds file size, chop it down. This should
576 * not normally occur but there is an append race where it
577 * might occur XXX, so we log it.
579 * If the chopping creates a reverse-indexed or degenerate
580 * situation with dirtyoff/end, we 0 both of them.
583 if (bp->b_dirtyend > bcount) {
584 SDT_PROBE2(fuse, , io, write_biobackend_append_race,
585 (long)bp->b_blkno * biosize,
586 bp->b_dirtyend - bcount);
587 bp->b_dirtyend = bcount;
589 if (bp->b_dirtyoff >= bp->b_dirtyend)
590 bp->b_dirtyoff = bp->b_dirtyend = 0;
593 * If the new write will leave a contiguous dirty
594 * area, just update the b_dirtyoff and b_dirtyend,
595 * otherwise force a write rpc of the old dirty area.
597 * While it is possible to merge discontiguous writes due to
598 * our having a B_CACHE buffer ( and thus valid read data
599 * for the hole), we don't because it could lead to
600 * significant cache coherency problems with multiple clients,
601 * especially if locking is implemented later on.
603 * as an optimization we could theoretically maintain
604 * a linked list of discontinuous areas, but we would still
605 * have to commit them separately so there isn't much
606 * advantage to it except perhaps a bit of asynchronization.
609 if (bp->b_dirtyend > 0 &&
610 (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) {
612 * Yes, we mean it. Write out everything to "storage"
613 * immediately, without hesitation. (Apart from other
614 * reasons: the only way to know if a write is valid
615 * if its actually written out.)
618 if (bp->b_error == EINTR) {
624 err = uiomove((char *)bp->b_data + on, n, uio);
627 * Since this block is being modified, it must be written
628 * again and not just committed. Since write clustering does
629 * not work for the stage 1 data write, only the stage 2
630 * commit rpc, we have to clear B_CLUSTEROK as well.
632 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
635 bp->b_ioflags |= BIO_ERROR;
641 * Only update dirtyoff/dirtyend if not a degenerate
645 if (bp->b_dirtyend > 0) {
646 bp->b_dirtyoff = MIN(on, bp->b_dirtyoff);
647 bp->b_dirtyend = MAX((on + n), bp->b_dirtyend);
650 bp->b_dirtyend = on + n;
652 vfs_bio_set_valid(bp, on, n);
657 } while (uio->uio_resid > 0 && n > 0);
659 if (fuse_sync_resize && (fvdat->flag & FN_SIZECHANGE) != 0)
660 fuse_vnode_savesize(vp, cred, pid);
666 fuse_io_strategy(struct vnode *vp, struct buf *bp)
668 struct fuse_filehandle *fufh;
669 struct fuse_vnode_data *fvdat = VTOFUD(vp);
676 /* We don't know the true pid when we're dealing with the cache */
679 const int biosize = fuse_iosize(vp);
681 MPASS(vp->v_type == VREG || vp->v_type == VDIR);
682 MPASS(bp->b_iocmd == BIO_READ || bp->b_iocmd == BIO_WRITE);
684 fflag = bp->b_iocmd == BIO_READ ? FREAD : FWRITE;
685 cred = bp->b_iocmd == BIO_READ ? bp->b_rcred : bp->b_wcred;
686 error = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
687 if (bp->b_iocmd == BIO_READ && error == EBADF) {
689 * This may be a read-modify-write operation on a cached file
690 * opened O_WRONLY. The FUSE protocol allows this.
692 * TODO: eliminate this hacky check once the FUFH table is gone
694 error = fuse_filehandle_get(vp, FWRITE, &fufh, cred, pid);
697 printf("FUSE: strategy: filehandles are closed\n");
698 bp->b_ioflags |= BIO_ERROR;
706 uiop->uio_iovcnt = 1;
707 uiop->uio_segflg = UIO_SYSSPACE;
708 uiop->uio_td = curthread;
711 * clear BIO_ERROR and B_INVAL state prior to initiating the I/O. We
712 * do this here so we do not have to do it in all the code that
715 bp->b_flags &= ~B_INVAL;
716 bp->b_ioflags &= ~BIO_ERROR;
718 KASSERT(!(bp->b_flags & B_DONE),
719 ("fuse_io_strategy: bp %p already marked done", bp));
720 if (bp->b_iocmd == BIO_READ) {
721 io.iov_len = uiop->uio_resid = bp->b_bcount;
722 io.iov_base = bp->b_data;
723 uiop->uio_rw = UIO_READ;
725 uiop->uio_offset = ((off_t)bp->b_blkno) * biosize;
726 error = fuse_read_directbackend(vp, uiop, cred, fufh);
728 /* XXXCEM: Potentially invalid access to cached_attrs here */
729 if ((!error && uiop->uio_resid) ||
730 (fsess_opt_brokenio(vnode_mount(vp)) && error == EIO &&
731 uiop->uio_offset < fvdat->filesize && fvdat->filesize > 0 &&
732 uiop->uio_offset >= fvdat->cached_attrs.va_size)) {
734 * If we had a short read with no error, we must have
735 * hit a file hole. We should zero-fill the remainder.
736 * This can also occur if the server hits the file EOF.
738 * Holes used to be able to occur due to pending
739 * writes, but that is not possible any longer.
741 int nread = bp->b_bcount - uiop->uio_resid;
742 int left = uiop->uio_resid;
745 printf("FUSE: Fix broken io: offset %ju, "
746 " resid %zd, file size %ju/%ju\n",
747 (uintmax_t)uiop->uio_offset,
748 uiop->uio_resid, fvdat->filesize,
749 fvdat->cached_attrs.va_size);
753 bzero((char *)bp->b_data + nread, left);
757 bp->b_ioflags |= BIO_ERROR;
762 * If we only need to commit, try to commit
764 if (bp->b_flags & B_NEEDCOMMIT) {
765 SDT_PROBE2(fuse, , io, trace, 1,
766 "write: B_NEEDCOMMIT flags set");
769 * Setup for actual write
771 if ((off_t)bp->b_blkno * biosize + bp->b_dirtyend >
773 bp->b_dirtyend = fvdat->filesize -
774 (off_t)bp->b_blkno * biosize;
776 if (bp->b_dirtyend > bp->b_dirtyoff) {
777 io.iov_len = uiop->uio_resid = bp->b_dirtyend
779 uiop->uio_offset = (off_t)bp->b_blkno * biosize
781 io.iov_base = (char *)bp->b_data + bp->b_dirtyoff;
782 uiop->uio_rw = UIO_WRITE;
784 error = fuse_write_directbackend(vp, uiop, cred, fufh, 0);
786 if (error == EINTR || error == ETIMEDOUT
787 || (!error && (bp->b_flags & B_NEEDCOMMIT))) {
789 bp->b_flags &= ~(B_INVAL | B_NOCACHE);
790 if ((bp->b_flags & B_PAGING) == 0) {
792 bp->b_flags &= ~B_DONE;
794 if ((error == EINTR || error == ETIMEDOUT) &&
795 (bp->b_flags & B_ASYNC) == 0)
796 bp->b_flags |= B_EINTR;
799 bp->b_ioflags |= BIO_ERROR;
800 bp->b_flags |= B_INVAL;
803 bp->b_dirtyoff = bp->b_dirtyend = 0;
811 bp->b_resid = uiop->uio_resid;
817 fuse_io_flushbuf(struct vnode *vp, int waitfor, struct thread *td)
819 struct vop_fsync_args a = {
821 .a_waitfor = waitfor,
825 return (vop_stdfsync(&a));
829 * Flush and invalidate all dirty buffers. If another process is already
830 * doing the flush, just wait for completion.
833 fuse_io_invalbuf(struct vnode *vp, struct thread *td)
835 struct fuse_vnode_data *fvdat = VTOFUD(vp);
838 if (vp->v_iflag & VI_DOOMED)
841 ASSERT_VOP_ELOCKED(vp, "fuse_io_invalbuf");
843 while (fvdat->flag & FN_FLUSHINPROG) {
844 struct proc *p = td->td_proc;
846 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF)
848 fvdat->flag |= FN_FLUSHWANT;
849 tsleep(&fvdat->flag, PRIBIO + 2, "fusevinv", 2 * hz);
853 if (SIGNOTEMPTY(p->p_siglist) ||
854 SIGNOTEMPTY(td->td_siglist))
861 fvdat->flag |= FN_FLUSHINPROG;
863 if (vp->v_bufobj.bo_object != NULL) {
864 VM_OBJECT_WLOCK(vp->v_bufobj.bo_object);
865 vm_object_page_clean(vp->v_bufobj.bo_object, 0, 0, OBJPC_SYNC);
866 VM_OBJECT_WUNLOCK(vp->v_bufobj.bo_object);
868 error = vinvalbuf(vp, V_SAVE, PCATCH, 0);
870 if (error == ERESTART || error == EINTR) {
871 fvdat->flag &= ~FN_FLUSHINPROG;
872 if (fvdat->flag & FN_FLUSHWANT) {
873 fvdat->flag &= ~FN_FLUSHWANT;
874 wakeup(&fvdat->flag);
878 error = vinvalbuf(vp, V_SAVE, PCATCH, 0);
880 fvdat->flag &= ~FN_FLUSHINPROG;
881 if (fvdat->flag & FN_FLUSHWANT) {
882 fvdat->flag &= ~FN_FLUSHWANT;
883 wakeup(&fvdat->flag);