4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
25 * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
26 * Copyright 2017 Nexenta Systems, Inc.
29 /* Portions Copyright 2007 Jeremy Teo */
30 /* Portions Copyright 2010 Robert Milkowski */
33 #include <sys/types.h>
34 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/sysmacros.h>
38 #include <sys/resource.h>
40 #include <sys/vfs_opreg.h>
44 #include <sys/taskq.h>
46 #include <sys/vmsystm.h>
47 #include <sys/atomic.h>
49 #include <sys/pathname.h>
50 #include <sys/cmn_err.h>
51 #include <sys/errno.h>
52 #include <sys/unistd.h>
53 #include <sys/zfs_dir.h>
54 #include <sys/zfs_acl.h>
55 #include <sys/zfs_ioctl.h>
56 #include <sys/fs/zfs.h>
58 #include <sys/dmu_objset.h>
64 #include <sys/dirent.h>
65 #include <sys/policy.h>
66 #include <sys/sunddi.h>
69 #include "fs/fs_subr.h"
70 #include <sys/zfs_ctldir.h>
71 #include <sys/zfs_fuid.h>
72 #include <sys/zfs_sa.h>
73 #include <sys/zfs_vnops.h>
75 #include <sys/zfs_rlock.h>
76 #include <sys/extdirent.h>
77 #include <sys/kidmap.h>
82 #include <sys/sa_impl.h>
87 * Each vnode op performs some logical unit of work. To do this, the ZPL must
88 * properly lock its in-core state, create a DMU transaction, do the work,
89 * record this work in the intent log (ZIL), commit the DMU transaction,
90 * and wait for the intent log to commit if it is a synchronous operation.
91 * Moreover, the vnode ops must work in both normal and log replay context.
92 * The ordering of events is important to avoid deadlocks and references
93 * to freed memory. The example below illustrates the following Big Rules:
95 * (1) A check must be made in each zfs thread for a mounted file system.
96 * This is done avoiding races using ZFS_ENTER(zfsvfs).
97 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
98 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
99 * can return EIO from the calling function.
101 * (2) iput() should always be the last thing except for zil_commit()
102 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
103 * First, if it's the last reference, the vnode/znode
104 * can be freed, so the zp may point to freed memory. Second, the last
105 * reference will call zfs_zinactive(), which may induce a lot of work --
106 * pushing cached pages (which acquires range locks) and syncing out
107 * cached atime changes. Third, zfs_zinactive() may require a new tx,
108 * which could deadlock the system if you were already holding one.
109 * If you must call iput() within a tx then use zfs_iput_async().
111 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
112 * as they can span dmu_tx_assign() calls.
114 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
115 * dmu_tx_assign(). This is critical because we don't want to block
116 * while holding locks.
118 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
119 * reduces lock contention and CPU usage when we must wait (note that if
120 * throughput is constrained by the storage, nearly every transaction
123 * Note, in particular, that if a lock is sometimes acquired before
124 * the tx assigns, and sometimes after (e.g. z_lock), then failing
125 * to use a non-blocking assign can deadlock the system. The scenario:
127 * Thread A has grabbed a lock before calling dmu_tx_assign().
128 * Thread B is in an already-assigned tx, and blocks for this lock.
129 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
130 * forever, because the previous txg can't quiesce until B's tx commits.
132 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
133 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
134 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT,
135 * to indicate that this operation has already called dmu_tx_wait().
136 * This will ensure that we don't retry forever, waiting a short bit
139 * (5) If the operation succeeded, generate the intent log entry for it
140 * before dropping locks. This ensures that the ordering of events
141 * in the intent log matches the order in which they actually occurred.
142 * During ZIL replay the zfs_log_* functions will update the sequence
143 * number to indicate the zil transaction has replayed.
145 * (6) At the end of each vnode op, the DMU tx must always commit,
146 * regardless of whether there were any errors.
148 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
149 * to ensure that synchronous semantics are provided when necessary.
151 * In general, this is how things should be ordered in each vnode op:
153 * ZFS_ENTER(zfsvfs); // exit if unmounted
155 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
156 * rw_enter(...); // grab any other locks you need
157 * tx = dmu_tx_create(...); // get DMU tx
158 * dmu_tx_hold_*(); // hold each object you might modify
159 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
161 * rw_exit(...); // drop locks
162 * zfs_dirent_unlock(dl); // unlock directory entry
163 * iput(...); // release held vnodes
164 * if (error == ERESTART) {
170 * dmu_tx_abort(tx); // abort DMU tx
171 * ZFS_EXIT(zfsvfs); // finished in zfs
172 * return (error); // really out of space
174 * error = do_real_work(); // do whatever this VOP does
176 * zfs_log_*(...); // on success, make ZIL entry
177 * dmu_tx_commit(tx); // commit DMU tx -- error or not
178 * rw_exit(...); // drop locks
179 * zfs_dirent_unlock(dl); // unlock directory entry
180 * iput(...); // release held vnodes
181 * zil_commit(zilog, foid); // synchronous when necessary
182 * ZFS_EXIT(zfsvfs); // finished in zfs
183 * return (error); // done, report error
187 * Virus scanning is unsupported. It would be possible to add a hook
188 * here to performance the required virus scan. This could be done
189 * entirely in the kernel or potentially as an update to invoke a
193 zfs_vscan(struct inode *ip, cred_t *cr, int async)
200 zfs_open(struct inode *ip, int mode, int flag, cred_t *cr)
202 znode_t *zp = ITOZ(ip);
203 zfsvfs_t *zfsvfs = ITOZSB(ip);
208 /* Honor ZFS_APPENDONLY file attribute */
209 if ((mode & FMODE_WRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
210 ((flag & O_APPEND) == 0)) {
212 return (SET_ERROR(EPERM));
215 /* Virus scan eligible files on open */
216 if (!zfs_has_ctldir(zp) && zfsvfs->z_vscan && S_ISREG(ip->i_mode) &&
217 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
218 if (zfs_vscan(ip, cr, 0) != 0) {
220 return (SET_ERROR(EACCES));
224 /* Keep a count of the synchronous opens in the znode */
226 atomic_inc_32(&zp->z_sync_cnt);
234 zfs_close(struct inode *ip, int flag, cred_t *cr)
236 znode_t *zp = ITOZ(ip);
237 zfsvfs_t *zfsvfs = ITOZSB(ip);
242 /* Decrement the synchronous opens in the znode */
244 atomic_dec_32(&zp->z_sync_cnt);
246 if (!zfs_has_ctldir(zp) && zfsvfs->z_vscan && S_ISREG(ip->i_mode) &&
247 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
248 VERIFY(zfs_vscan(ip, cr, 1) == 0);
254 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
256 * Lseek support for finding holes (cmd == SEEK_HOLE) and
257 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
260 zfs_holey_common(struct inode *ip, int cmd, loff_t *off)
262 znode_t *zp = ITOZ(ip);
263 uint64_t noff = (uint64_t)*off; /* new offset */
268 file_sz = zp->z_size;
269 if (noff >= file_sz) {
270 return (SET_ERROR(ENXIO));
273 if (cmd == SEEK_HOLE)
278 error = dmu_offset_next(ZTOZSB(zp)->z_os, zp->z_id, hole, &noff);
281 return (SET_ERROR(ENXIO));
283 /* file was dirty, so fall back to using generic logic */
284 if (error == EBUSY) {
292 * We could find a hole that begins after the logical end-of-file,
293 * because dmu_offset_next() only works on whole blocks. If the
294 * EOF falls mid-block, then indicate that the "virtual hole"
295 * at the end of the file begins at the logical EOF, rather than
296 * at the end of the last block.
298 if (noff > file_sz) {
310 zfs_holey(struct inode *ip, int cmd, loff_t *off)
312 znode_t *zp = ITOZ(ip);
313 zfsvfs_t *zfsvfs = ITOZSB(ip);
319 error = zfs_holey_common(ip, cmd, off);
324 #endif /* SEEK_HOLE && SEEK_DATA */
328 * When a file is memory mapped, we must keep the IO data synchronized
329 * between the DMU cache and the memory mapped pages. What this means:
331 * On Write: If we find a memory mapped page, we write to *both*
332 * the page and the dmu buffer.
335 update_pages(struct inode *ip, int64_t start, int len,
336 objset_t *os, uint64_t oid)
338 struct address_space *mp = ip->i_mapping;
344 off = start & (PAGE_SIZE-1);
345 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
346 nbytes = MIN(PAGE_SIZE - off, len);
348 pp = find_lock_page(mp, start >> PAGE_SHIFT);
350 if (mapping_writably_mapped(mp))
351 flush_dcache_page(pp);
354 (void) dmu_read(os, oid, start+off, nbytes, pb+off,
358 if (mapping_writably_mapped(mp))
359 flush_dcache_page(pp);
361 mark_page_accessed(pp);
374 * When a file is memory mapped, we must keep the IO data synchronized
375 * between the DMU cache and the memory mapped pages. What this means:
377 * On Read: We "read" preferentially from memory mapped pages,
378 * else we default from the dmu buffer.
380 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
381 * the file is memory mapped.
384 mappedread(struct inode *ip, int nbytes, uio_t *uio)
386 struct address_space *mp = ip->i_mapping;
388 znode_t *zp = ITOZ(ip);
395 start = uio->uio_loffset;
396 off = start & (PAGE_SIZE-1);
397 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
398 bytes = MIN(PAGE_SIZE - off, len);
400 pp = find_lock_page(mp, start >> PAGE_SHIFT);
402 ASSERT(PageUptodate(pp));
406 error = uiomove(pb + off, bytes, UIO_READ, uio);
409 if (mapping_writably_mapped(mp))
410 flush_dcache_page(pp);
412 mark_page_accessed(pp);
415 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
428 unsigned long zfs_read_chunk_size = 1024 * 1024; /* Tunable */
429 unsigned long zfs_delete_blocks = DMU_MAX_DELETEBLKCNT;
432 * Read bytes from specified file into supplied buffer.
434 * IN: ip - inode of file to be read from.
435 * uio - structure supplying read location, range info,
437 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
438 * O_DIRECT flag; used to bypass page cache.
439 * cr - credentials of caller.
441 * OUT: uio - updated offset and range, buffer filled.
443 * RETURN: 0 on success, error code on failure.
446 * inode - atime updated if byte count > 0
450 zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
452 znode_t *zp = ITOZ(ip);
453 zfsvfs_t *zfsvfs = ITOZSB(ip);
457 #ifdef HAVE_UIO_ZEROCOPY
459 #endif /* HAVE_UIO_ZEROCOPY */
464 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
466 return (SET_ERROR(EACCES));
470 * Validate file offset
472 if (uio->uio_loffset < (offset_t)0) {
474 return (SET_ERROR(EINVAL));
478 * Fasttrack empty reads
480 if (uio->uio_resid == 0) {
486 * If we're in FRSYNC mode, sync out this znode before reading it.
487 * Only do this for non-snapshots.
490 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
491 zil_commit(zfsvfs->z_log, zp->z_id);
494 * Lock the range against changes.
496 rl = zfs_range_lock(&zp->z_range_lock, uio->uio_loffset, uio->uio_resid,
500 * If we are reading past end-of-file we can skip
501 * to the end; but we might still need to set atime.
503 if (uio->uio_loffset >= zp->z_size) {
508 ASSERT(uio->uio_loffset < zp->z_size);
509 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
511 #ifdef HAVE_UIO_ZEROCOPY
512 if ((uio->uio_extflg == UIO_XUIO) &&
513 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
515 int blksz = zp->z_blksz;
516 uint64_t offset = uio->uio_loffset;
518 xuio = (xuio_t *)uio;
520 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
523 ASSERT(offset + n <= blksz);
526 (void) dmu_xuio_init(xuio, nblk);
528 if (vn_has_cached_data(ip)) {
530 * For simplicity, we always allocate a full buffer
531 * even if we only expect to read a portion of a block.
533 while (--nblk >= 0) {
534 (void) dmu_xuio_add(xuio,
535 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
540 #endif /* HAVE_UIO_ZEROCOPY */
543 nbytes = MIN(n, zfs_read_chunk_size -
544 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
546 if (zp->z_is_mapped && !(ioflag & O_DIRECT)) {
547 error = mappedread(ip, nbytes, uio);
549 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
554 /* convert checksum errors into IO errors */
556 error = SET_ERROR(EIO);
563 zfs_range_unlock(rl);
570 * Write the bytes to a file.
572 * IN: ip - inode of file to be written to.
573 * uio - structure supplying write location, range info,
575 * ioflag - FAPPEND flag set if in append mode.
576 * O_DIRECT flag; used to bypass page cache.
577 * cr - credentials of caller.
579 * OUT: uio - updated offset and range.
581 * RETURN: 0 if success
582 * error code if failure
585 * ip - ctime|mtime updated if byte count > 0
590 zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
592 znode_t *zp = ITOZ(ip);
593 rlim64_t limit = uio->uio_limit;
594 ssize_t start_resid = uio->uio_resid;
598 zfsvfs_t *zfsvfs = ZTOZSB(zp);
603 int max_blksz = zfsvfs->z_max_blksz;
606 const iovec_t *aiov = NULL;
610 sa_bulk_attr_t bulk[4];
611 uint64_t mtime[2], ctime[2];
613 #ifdef HAVE_UIO_ZEROCOPY
615 const iovec_t *iovp = uio->uio_iov;
616 ASSERTV(int iovcnt = uio->uio_iovcnt);
620 * Fasttrack empty write
626 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
632 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
633 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
634 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
636 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
640 * Callers might not be able to detect properly that we are read-only,
641 * so check it explicitly here.
643 if (zfs_is_readonly(zfsvfs)) {
645 return (SET_ERROR(EROFS));
649 * If immutable or not appending then return EPERM
651 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
652 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
653 (uio->uio_loffset < zp->z_size))) {
655 return (SET_ERROR(EPERM));
658 zilog = zfsvfs->z_log;
661 * Validate file offset
663 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
666 return (SET_ERROR(EINVAL));
670 * Pre-fault the pages to ensure slow (eg NFS) pages
672 * Skip this if uio contains loaned arc_buf.
674 #ifdef HAVE_UIO_ZEROCOPY
675 if ((uio->uio_extflg == UIO_XUIO) &&
676 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
677 xuio = (xuio_t *)uio;
680 uio_prefaultpages(MIN(n, max_blksz), uio);
683 * If in append mode, set the io offset pointer to eof.
685 if (ioflag & FAPPEND) {
687 * Obtain an appending range lock to guarantee file append
688 * semantics. We reset the write offset once we have the lock.
690 rl = zfs_range_lock(&zp->z_range_lock, 0, n, RL_APPEND);
692 if (rl->r_len == UINT64_MAX) {
694 * We overlocked the file because this write will cause
695 * the file block size to increase.
696 * Note that zp_size cannot change with this lock held.
700 uio->uio_loffset = woff;
703 * Note that if the file block size will change as a result of
704 * this write, then this range lock will lock the entire file
705 * so that we can re-write the block safely.
707 rl = zfs_range_lock(&zp->z_range_lock, woff, n, RL_WRITER);
711 zfs_range_unlock(rl);
713 return (SET_ERROR(EFBIG));
716 if ((woff + n) > limit || woff > (limit - n))
719 /* Will this write extend the file length? */
720 write_eof = (woff + n > zp->z_size);
722 end_size = MAX(zp->z_size, woff + n);
725 * Write the file in reasonable size chunks. Each chunk is written
726 * in a separate transaction; this keeps the intent log records small
727 * and allows us to do more fine-grained space accounting.
731 woff = uio->uio_loffset;
732 if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT,
733 KUID_TO_SUID(ip->i_uid)) ||
734 zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT,
735 KGID_TO_SGID(ip->i_gid)) ||
736 (zp->z_projid != ZFS_DEFAULT_PROJID &&
737 zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT,
740 dmu_return_arcbuf(abuf);
741 error = SET_ERROR(EDQUOT);
745 if (xuio && abuf == NULL) {
746 #ifdef HAVE_UIO_ZEROCOPY
747 ASSERT(i_iov < iovcnt);
748 ASSERT3U(uio->uio_segflg, !=, UIO_BVEC);
750 abuf = dmu_xuio_arcbuf(xuio, i_iov);
751 dmu_xuio_clear(xuio, i_iov);
752 ASSERT((aiov->iov_base == abuf->b_data) ||
753 ((char *)aiov->iov_base - (char *)abuf->b_data +
754 aiov->iov_len == arc_buf_size(abuf)));
757 } else if (abuf == NULL && n >= max_blksz &&
758 woff >= zp->z_size &&
759 P2PHASE(woff, max_blksz) == 0 &&
760 zp->z_blksz == max_blksz) {
762 * This write covers a full block. "Borrow" a buffer
763 * from the dmu so that we can fill it before we enter
764 * a transaction. This avoids the possibility of
765 * holding up the transaction if the data copy hangs
766 * up on a pagefault (e.g., from an NFS server mapping).
770 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
772 ASSERT(abuf != NULL);
773 ASSERT(arc_buf_size(abuf) == max_blksz);
774 if ((error = uiocopy(abuf->b_data, max_blksz,
775 UIO_WRITE, uio, &cbytes))) {
776 dmu_return_arcbuf(abuf);
779 ASSERT(cbytes == max_blksz);
783 * Start a transaction.
785 tx = dmu_tx_create(zfsvfs->z_os);
786 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
787 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
788 zfs_sa_upgrade_txholds(tx, zp);
789 error = dmu_tx_assign(tx, TXG_WAIT);
793 dmu_return_arcbuf(abuf);
798 * If zfs_range_lock() over-locked we grow the blocksize
799 * and then reduce the lock range. This will only happen
800 * on the first iteration since zfs_range_reduce() will
801 * shrink down r_len to the appropriate size.
803 if (rl->r_len == UINT64_MAX) {
806 if (zp->z_blksz > max_blksz) {
808 * File's blocksize is already larger than the
809 * "recordsize" property. Only let it grow to
810 * the next power of 2.
812 ASSERT(!ISP2(zp->z_blksz));
813 new_blksz = MIN(end_size,
814 1 << highbit64(zp->z_blksz));
816 new_blksz = MIN(end_size, max_blksz);
818 zfs_grow_blocksize(zp, new_blksz, tx);
819 zfs_range_reduce(rl, woff, n);
823 * XXX - should we really limit each write to z_max_blksz?
824 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
826 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
829 tx_bytes = uio->uio_resid;
830 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
832 tx_bytes -= uio->uio_resid;
835 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
837 * If this is not a full block write, but we are
838 * extending the file past EOF and this data starts
839 * block-aligned, use assign_arcbuf(). Otherwise,
840 * write via dmu_write().
842 if (tx_bytes < max_blksz && (!write_eof ||
843 aiov->iov_base != abuf->b_data)) {
845 dmu_write(zfsvfs->z_os, zp->z_id, woff,
846 /* cppcheck-suppress nullPointer */
847 aiov->iov_len, aiov->iov_base, tx);
848 dmu_return_arcbuf(abuf);
849 xuio_stat_wbuf_copied();
851 ASSERT(xuio || tx_bytes == max_blksz);
852 dmu_assign_arcbuf_by_dbuf(
853 sa_get_db(zp->z_sa_hdl), woff, abuf, tx);
855 ASSERT(tx_bytes <= uio->uio_resid);
856 uioskip(uio, tx_bytes);
858 if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT)) {
859 update_pages(ip, woff,
860 tx_bytes, zfsvfs->z_os, zp->z_id);
864 * If we made no progress, we're done. If we made even
865 * partial progress, update the znode and ZIL accordingly.
868 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
869 (void *)&zp->z_size, sizeof (uint64_t), tx);
876 * Clear Set-UID/Set-GID bits on successful write if not
877 * privileged and at least one of the execute bits is set.
879 * It would be nice to to this after all writes have
880 * been done, but that would still expose the ISUID/ISGID
881 * to another app after the partial write is committed.
883 * Note: we don't call zfs_fuid_map_id() here because
884 * user 0 is not an ephemeral uid.
886 mutex_enter(&zp->z_acl_lock);
887 uid = KUID_TO_SUID(ip->i_uid);
888 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
889 (S_IXUSR >> 6))) != 0 &&
890 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
891 secpolicy_vnode_setid_retain(cr,
892 ((zp->z_mode & S_ISUID) != 0 && uid == 0)) != 0) {
894 zp->z_mode &= ~(S_ISUID | S_ISGID);
895 ip->i_mode = newmode = zp->z_mode;
896 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
897 (void *)&newmode, sizeof (uint64_t), tx);
899 mutex_exit(&zp->z_acl_lock);
901 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
904 * Update the file size (zp_size) if it has changed;
905 * account for possible concurrent updates.
907 while ((end_size = zp->z_size) < uio->uio_loffset) {
908 (void) atomic_cas_64(&zp->z_size, end_size,
913 * If we are replaying and eof is non zero then force
914 * the file size to the specified eof. Note, there's no
915 * concurrency during replay.
917 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
918 zp->z_size = zfsvfs->z_replay_eof;
920 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
922 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag,
928 ASSERT(tx_bytes == nbytes);
932 uio_prefaultpages(MIN(n, max_blksz), uio);
935 zfs_inode_update(zp);
936 zfs_range_unlock(rl);
939 * If we're in replay mode, or we made no progress, return error.
940 * Otherwise, it's at least a partial write, so it's successful.
942 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
947 if (ioflag & (FSYNC | FDSYNC) ||
948 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
949 zil_commit(zilog, zp->z_id);
956 * Drop a reference on the passed inode asynchronously. This ensures
957 * that the caller will never drop the last reference on an inode in
958 * the current context. Doing so while holding open a tx could result
959 * in a deadlock if iput_final() re-enters the filesystem code.
962 zfs_iput_async(struct inode *ip)
964 objset_t *os = ITOZSB(ip)->z_os;
966 ASSERT(atomic_read(&ip->i_count) > 0);
969 if (atomic_read(&ip->i_count) == 1)
970 VERIFY(taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os)),
971 (task_func_t *)iput, ip, TQ_SLEEP) != TASKQID_INVALID);
977 zfs_get_done(zgd_t *zgd, int error)
979 znode_t *zp = zgd->zgd_private;
982 dmu_buf_rele(zgd->zgd_db, zgd);
984 zfs_range_unlock(zgd->zgd_rl);
987 * Release the vnode asynchronously as we currently have the
988 * txg stopped from syncing.
990 zfs_iput_async(ZTOI(zp));
992 if (error == 0 && zgd->zgd_bp)
993 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
995 kmem_free(zgd, sizeof (zgd_t));
999 static int zil_fault_io = 0;
1003 * Get data to generate a TX_WRITE intent log record.
1006 zfs_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
1008 zfsvfs_t *zfsvfs = arg;
1009 objset_t *os = zfsvfs->z_os;
1011 uint64_t object = lr->lr_foid;
1012 uint64_t offset = lr->lr_offset;
1013 uint64_t size = lr->lr_length;
1018 ASSERT3P(lwb, !=, NULL);
1019 ASSERT3P(zio, !=, NULL);
1020 ASSERT3U(size, !=, 0);
1023 * Nothing to do if the file has been removed
1025 if (zfs_zget(zfsvfs, object, &zp) != 0)
1026 return (SET_ERROR(ENOENT));
1027 if (zp->z_unlinked) {
1029 * Release the vnode asynchronously as we currently have the
1030 * txg stopped from syncing.
1032 zfs_iput_async(ZTOI(zp));
1033 return (SET_ERROR(ENOENT));
1036 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1038 zgd->zgd_private = zp;
1041 * Write records come in two flavors: immediate and indirect.
1042 * For small writes it's cheaper to store the data with the
1043 * log record (immediate); for large writes it's cheaper to
1044 * sync the data and get a pointer to it (indirect) so that
1045 * we don't have to write the data twice.
1047 if (buf != NULL) { /* immediate write */
1048 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset, size,
1050 /* test for truncation needs to be done while range locked */
1051 if (offset >= zp->z_size) {
1052 error = SET_ERROR(ENOENT);
1054 error = dmu_read(os, object, offset, size, buf,
1055 DMU_READ_NO_PREFETCH);
1057 ASSERT(error == 0 || error == ENOENT);
1058 } else { /* indirect write */
1060 * Have to lock the whole block to ensure when it's
1061 * written out and its checksum is being calculated
1062 * that no one can change the data. We need to re-check
1063 * blocksize after we get the lock in case it's changed!
1068 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1070 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset,
1072 if (zp->z_blksz == size)
1075 zfs_range_unlock(zgd->zgd_rl);
1077 /* test for truncation needs to be done while range locked */
1078 if (lr->lr_offset >= zp->z_size)
1079 error = SET_ERROR(ENOENT);
1082 error = SET_ERROR(EIO);
1087 error = dmu_buf_hold(os, object, offset, zgd, &db,
1088 DMU_READ_NO_PREFETCH);
1091 blkptr_t *bp = &lr->lr_blkptr;
1096 ASSERT(db->db_offset == offset);
1097 ASSERT(db->db_size == size);
1099 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1101 ASSERT(error || lr->lr_length <= size);
1104 * On success, we need to wait for the write I/O
1105 * initiated by dmu_sync() to complete before we can
1106 * release this dbuf. We will finish everything up
1107 * in the zfs_get_done() callback.
1112 if (error == EALREADY) {
1113 lr->lr_common.lrc_txtype = TX_WRITE2;
1119 zfs_get_done(zgd, error);
1126 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
1128 znode_t *zp = ITOZ(ip);
1129 zfsvfs_t *zfsvfs = ITOZSB(ip);
1135 if (flag & V_ACE_MASK)
1136 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1138 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1145 * Lookup an entry in a directory, or an extended attribute directory.
1146 * If it exists, return a held inode reference for it.
1148 * IN: dip - inode of directory to search.
1149 * nm - name of entry to lookup.
1150 * flags - LOOKUP_XATTR set if looking for an attribute.
1151 * cr - credentials of caller.
1152 * direntflags - directory lookup flags
1153 * realpnp - returned pathname.
1155 * OUT: ipp - inode of located entry, NULL if not found.
1157 * RETURN: 0 on success, error code on failure.
1164 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
1165 cred_t *cr, int *direntflags, pathname_t *realpnp)
1167 znode_t *zdp = ITOZ(dip);
1168 zfsvfs_t *zfsvfs = ITOZSB(dip);
1172 * Fast path lookup, however we must skip DNLC lookup
1173 * for case folding or normalizing lookups because the
1174 * DNLC code only stores the passed in name. This means
1175 * creating 'a' and removing 'A' on a case insensitive
1176 * file system would work, but DNLC still thinks 'a'
1177 * exists and won't let you create it again on the next
1178 * pass through fast path.
1180 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1182 if (!S_ISDIR(dip->i_mode)) {
1183 return (SET_ERROR(ENOTDIR));
1184 } else if (zdp->z_sa_hdl == NULL) {
1185 return (SET_ERROR(EIO));
1188 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1189 error = zfs_fastaccesschk_execute(zdp, cr);
1197 } else if (!zdp->z_zfsvfs->z_norm &&
1198 (zdp->z_zfsvfs->z_case == ZFS_CASE_SENSITIVE)) {
1200 vnode_t *tvp = dnlc_lookup(dvp, nm);
1203 error = zfs_fastaccesschk_execute(zdp, cr);
1208 if (tvp == DNLC_NO_VNODE) {
1210 return (SET_ERROR(ENOENT));
1213 return (specvp_check(vpp, cr));
1216 #endif /* HAVE_DNLC */
1225 if (flags & LOOKUP_XATTR) {
1227 * We don't allow recursive attributes..
1228 * Maybe someday we will.
1230 if (zdp->z_pflags & ZFS_XATTR) {
1232 return (SET_ERROR(EINVAL));
1235 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1241 * Do we have permission to get into attribute directory?
1244 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1254 if (!S_ISDIR(dip->i_mode)) {
1256 return (SET_ERROR(ENOTDIR));
1260 * Check accessibility of directory.
1263 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1268 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1269 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1271 return (SET_ERROR(EILSEQ));
1274 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1275 if ((error == 0) && (*ipp))
1276 zfs_inode_update(ITOZ(*ipp));
1283 * Attempt to create a new entry in a directory. If the entry
1284 * already exists, truncate the file if permissible, else return
1285 * an error. Return the ip of the created or trunc'd file.
1287 * IN: dip - inode of directory to put new file entry in.
1288 * name - name of new file entry.
1289 * vap - attributes of new file.
1290 * excl - flag indicating exclusive or non-exclusive mode.
1291 * mode - mode to open file with.
1292 * cr - credentials of caller.
1293 * flag - large file flag [UNUSED].
1294 * vsecp - ACL to be set
1296 * OUT: ipp - inode of created or trunc'd entry.
1298 * RETURN: 0 on success, error code on failure.
1301 * dip - ctime|mtime updated if new entry created
1302 * ip - ctime|mtime always, atime if new
1307 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1308 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1310 znode_t *zp, *dzp = ITOZ(dip);
1311 zfsvfs_t *zfsvfs = ITOZSB(dip);
1319 zfs_acl_ids_t acl_ids;
1320 boolean_t fuid_dirtied;
1321 boolean_t have_acl = B_FALSE;
1322 boolean_t waited = B_FALSE;
1325 * If we have an ephemeral id, ACL, or XVATTR then
1326 * make sure file system is at proper version
1332 if (zfsvfs->z_use_fuids == B_FALSE &&
1333 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1334 return (SET_ERROR(EINVAL));
1337 return (SET_ERROR(EINVAL));
1342 zilog = zfsvfs->z_log;
1344 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1345 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1347 return (SET_ERROR(EILSEQ));
1350 if (vap->va_mask & ATTR_XVATTR) {
1351 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1352 crgetuid(cr), cr, vap->va_mode)) != 0) {
1360 if (*name == '\0') {
1362 * Null component name refers to the directory itself.
1369 /* possible igrab(zp) */
1372 if (flag & FIGNORECASE)
1375 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1379 zfs_acl_ids_free(&acl_ids);
1380 if (strcmp(name, "..") == 0)
1381 error = SET_ERROR(EISDIR);
1389 uint64_t projid = ZFS_DEFAULT_PROJID;
1392 * Create a new file object and update the directory
1395 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1397 zfs_acl_ids_free(&acl_ids);
1402 * We only support the creation of regular files in
1403 * extended attribute directories.
1406 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1408 zfs_acl_ids_free(&acl_ids);
1409 error = SET_ERROR(EINVAL);
1413 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1414 cr, vsecp, &acl_ids)) != 0)
1418 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode))
1419 projid = zfs_inherit_projid(dzp);
1420 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) {
1421 zfs_acl_ids_free(&acl_ids);
1422 error = SET_ERROR(EDQUOT);
1426 tx = dmu_tx_create(os);
1428 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1429 ZFS_SA_BASE_ATTR_SIZE);
1431 fuid_dirtied = zfsvfs->z_fuid_dirty;
1433 zfs_fuid_txhold(zfsvfs, tx);
1434 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1435 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1436 if (!zfsvfs->z_use_sa &&
1437 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1438 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1439 0, acl_ids.z_aclp->z_acl_bytes);
1441 error = dmu_tx_assign(tx,
1442 (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1444 zfs_dirent_unlock(dl);
1445 if (error == ERESTART) {
1451 zfs_acl_ids_free(&acl_ids);
1456 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1459 zfs_fuid_sync(zfsvfs, tx);
1461 (void) zfs_link_create(dl, zp, tx, ZNEW);
1462 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1463 if (flag & FIGNORECASE)
1465 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1466 vsecp, acl_ids.z_fuidp, vap);
1467 zfs_acl_ids_free(&acl_ids);
1470 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1473 zfs_acl_ids_free(&acl_ids);
1477 * A directory entry already exists for this name.
1480 * Can't truncate an existing file if in exclusive mode.
1483 error = SET_ERROR(EEXIST);
1487 * Can't open a directory for writing.
1489 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1490 error = SET_ERROR(EISDIR);
1494 * Verify requested access to file.
1496 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1500 mutex_enter(&dzp->z_lock);
1502 mutex_exit(&dzp->z_lock);
1505 * Truncate regular files if requested.
1507 if (S_ISREG(ZTOI(zp)->i_mode) &&
1508 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1509 /* we can't hold any locks when calling zfs_freesp() */
1511 zfs_dirent_unlock(dl);
1514 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1520 zfs_dirent_unlock(dl);
1526 zfs_inode_update(dzp);
1527 zfs_inode_update(zp);
1531 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1532 zil_commit(zilog, 0);
1540 zfs_tmpfile(struct inode *dip, vattr_t *vap, int excl,
1541 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1543 znode_t *zp = NULL, *dzp = ITOZ(dip);
1544 zfsvfs_t *zfsvfs = ITOZSB(dip);
1550 zfs_acl_ids_t acl_ids;
1551 uint64_t projid = ZFS_DEFAULT_PROJID;
1552 boolean_t fuid_dirtied;
1553 boolean_t have_acl = B_FALSE;
1554 boolean_t waited = B_FALSE;
1557 * If we have an ephemeral id, ACL, or XVATTR then
1558 * make sure file system is at proper version
1564 if (zfsvfs->z_use_fuids == B_FALSE &&
1565 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1566 return (SET_ERROR(EINVAL));
1572 if (vap->va_mask & ATTR_XVATTR) {
1573 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1574 crgetuid(cr), cr, vap->va_mode)) != 0) {
1584 * Create a new file object and update the directory
1587 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1589 zfs_acl_ids_free(&acl_ids);
1593 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1594 cr, vsecp, &acl_ids)) != 0)
1598 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode))
1599 projid = zfs_inherit_projid(dzp);
1600 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) {
1601 zfs_acl_ids_free(&acl_ids);
1602 error = SET_ERROR(EDQUOT);
1606 tx = dmu_tx_create(os);
1608 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1609 ZFS_SA_BASE_ATTR_SIZE);
1610 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1612 fuid_dirtied = zfsvfs->z_fuid_dirty;
1614 zfs_fuid_txhold(zfsvfs, tx);
1615 if (!zfsvfs->z_use_sa &&
1616 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1617 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1618 0, acl_ids.z_aclp->z_acl_bytes);
1620 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1622 if (error == ERESTART) {
1628 zfs_acl_ids_free(&acl_ids);
1633 zfs_mknode(dzp, vap, tx, cr, IS_TMPFILE, &zp, &acl_ids);
1636 zfs_fuid_sync(zfsvfs, tx);
1638 /* Add to unlinked set */
1640 zfs_unlinked_add(zp, tx);
1641 zfs_acl_ids_free(&acl_ids);
1649 zfs_inode_update(dzp);
1650 zfs_inode_update(zp);
1659 * Remove an entry from a directory.
1661 * IN: dip - inode of directory to remove entry from.
1662 * name - name of entry to remove.
1663 * cr - credentials of caller.
1665 * RETURN: 0 if success
1666 * error code if failure
1670 * ip - ctime (if nlink > 0)
1673 uint64_t null_xattr = 0;
1677 zfs_remove(struct inode *dip, char *name, cred_t *cr, int flags)
1679 znode_t *zp, *dzp = ITOZ(dip);
1682 zfsvfs_t *zfsvfs = ITOZSB(dip);
1684 uint64_t acl_obj, xattr_obj;
1685 uint64_t xattr_obj_unlinked = 0;
1690 boolean_t may_delete_now, delete_now = FALSE;
1691 boolean_t unlinked, toobig = FALSE;
1693 pathname_t *realnmp = NULL;
1697 boolean_t waited = B_FALSE;
1700 return (SET_ERROR(EINVAL));
1704 zilog = zfsvfs->z_log;
1706 if (flags & FIGNORECASE) {
1716 * Attempt to lock directory; fail if entry doesn't exist.
1718 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1728 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1733 * Need to use rmdir for removing directories.
1735 if (S_ISDIR(ip->i_mode)) {
1736 error = SET_ERROR(EPERM);
1742 dnlc_remove(dvp, realnmp->pn_buf);
1744 dnlc_remove(dvp, name);
1745 #endif /* HAVE_DNLC */
1747 mutex_enter(&zp->z_lock);
1748 may_delete_now = atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped);
1749 mutex_exit(&zp->z_lock);
1752 * We may delete the znode now, or we may put it in the unlinked set;
1753 * it depends on whether we're the last link, and on whether there are
1754 * other holds on the inode. So we dmu_tx_hold() the right things to
1755 * allow for either case.
1758 tx = dmu_tx_create(zfsvfs->z_os);
1759 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1760 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1761 zfs_sa_upgrade_txholds(tx, zp);
1762 zfs_sa_upgrade_txholds(tx, dzp);
1763 if (may_delete_now) {
1764 toobig = zp->z_size > zp->z_blksz * zfs_delete_blocks;
1765 /* if the file is too big, only hold_free a token amount */
1766 dmu_tx_hold_free(tx, zp->z_id, 0,
1767 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1770 /* are there any extended attributes? */
1771 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1772 &xattr_obj, sizeof (xattr_obj));
1773 if (error == 0 && xattr_obj) {
1774 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1776 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1777 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1780 mutex_enter(&zp->z_lock);
1781 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1782 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1783 mutex_exit(&zp->z_lock);
1785 /* charge as an update -- would be nice not to charge at all */
1786 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1789 * Mark this transaction as typically resulting in a net free of space
1791 dmu_tx_mark_netfree(tx);
1793 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1795 zfs_dirent_unlock(dl);
1796 if (error == ERESTART) {
1816 * Remove the directory entry.
1818 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1827 * Hold z_lock so that we can make sure that the ACL obj
1828 * hasn't changed. Could have been deleted due to
1831 mutex_enter(&zp->z_lock);
1832 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1833 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1834 delete_now = may_delete_now && !toobig &&
1835 atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped) &&
1836 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1841 if (xattr_obj_unlinked) {
1842 ASSERT3U(ZTOI(xzp)->i_nlink, ==, 2);
1843 mutex_enter(&xzp->z_lock);
1844 xzp->z_unlinked = 1;
1845 clear_nlink(ZTOI(xzp));
1847 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1848 &links, sizeof (links), tx);
1849 ASSERT3U(error, ==, 0);
1850 mutex_exit(&xzp->z_lock);
1851 zfs_unlinked_add(xzp, tx);
1854 error = sa_remove(zp->z_sa_hdl,
1855 SA_ZPL_XATTR(zfsvfs), tx);
1857 error = sa_update(zp->z_sa_hdl,
1858 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1859 sizeof (uint64_t), tx);
1863 * Add to the unlinked set because a new reference could be
1864 * taken concurrently resulting in a deferred destruction.
1866 zfs_unlinked_add(zp, tx);
1867 mutex_exit(&zp->z_lock);
1868 } else if (unlinked) {
1869 mutex_exit(&zp->z_lock);
1870 zfs_unlinked_add(zp, tx);
1874 if (flags & FIGNORECASE)
1876 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1883 zfs_dirent_unlock(dl);
1884 zfs_inode_update(dzp);
1885 zfs_inode_update(zp);
1893 zfs_inode_update(xzp);
1894 zfs_iput_async(ZTOI(xzp));
1897 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1898 zil_commit(zilog, 0);
1905 * Create a new directory and insert it into dip using the name
1906 * provided. Return a pointer to the inserted directory.
1908 * IN: dip - inode of directory to add subdir to.
1909 * dirname - name of new directory.
1910 * vap - attributes of new directory.
1911 * cr - credentials of caller.
1912 * vsecp - ACL to be set
1914 * OUT: ipp - inode of created directory.
1916 * RETURN: 0 if success
1917 * error code if failure
1920 * dip - ctime|mtime updated
1921 * ipp - ctime|mtime|atime updated
1925 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1926 cred_t *cr, int flags, vsecattr_t *vsecp)
1928 znode_t *zp, *dzp = ITOZ(dip);
1929 zfsvfs_t *zfsvfs = ITOZSB(dip);
1937 gid_t gid = crgetgid(cr);
1938 zfs_acl_ids_t acl_ids;
1939 boolean_t fuid_dirtied;
1940 boolean_t waited = B_FALSE;
1942 ASSERT(S_ISDIR(vap->va_mode));
1945 * If we have an ephemeral id, ACL, or XVATTR then
1946 * make sure file system is at proper version
1950 if (zfsvfs->z_use_fuids == B_FALSE &&
1951 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1952 return (SET_ERROR(EINVAL));
1954 if (dirname == NULL)
1955 return (SET_ERROR(EINVAL));
1959 zilog = zfsvfs->z_log;
1961 if (dzp->z_pflags & ZFS_XATTR) {
1963 return (SET_ERROR(EINVAL));
1966 if (zfsvfs->z_utf8 && u8_validate(dirname,
1967 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1969 return (SET_ERROR(EILSEQ));
1971 if (flags & FIGNORECASE)
1974 if (vap->va_mask & ATTR_XVATTR) {
1975 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1976 crgetuid(cr), cr, vap->va_mode)) != 0) {
1982 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1983 vsecp, &acl_ids)) != 0) {
1988 * First make sure the new directory doesn't exist.
1990 * Existence is checked first to make sure we don't return
1991 * EACCES instead of EEXIST which can cause some applications
1997 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1999 zfs_acl_ids_free(&acl_ids);
2004 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
2005 zfs_acl_ids_free(&acl_ids);
2006 zfs_dirent_unlock(dl);
2011 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) {
2012 zfs_acl_ids_free(&acl_ids);
2013 zfs_dirent_unlock(dl);
2015 return (SET_ERROR(EDQUOT));
2019 * Add a new entry to the directory.
2021 tx = dmu_tx_create(zfsvfs->z_os);
2022 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2023 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2024 fuid_dirtied = zfsvfs->z_fuid_dirty;
2026 zfs_fuid_txhold(zfsvfs, tx);
2027 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2028 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2029 acl_ids.z_aclp->z_acl_bytes);
2032 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2033 ZFS_SA_BASE_ATTR_SIZE);
2035 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
2037 zfs_dirent_unlock(dl);
2038 if (error == ERESTART) {
2044 zfs_acl_ids_free(&acl_ids);
2053 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2056 zfs_fuid_sync(zfsvfs, tx);
2059 * Now put new name in parent dir.
2061 (void) zfs_link_create(dl, zp, tx, ZNEW);
2065 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2066 if (flags & FIGNORECASE)
2068 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2069 acl_ids.z_fuidp, vap);
2071 zfs_acl_ids_free(&acl_ids);
2075 zfs_dirent_unlock(dl);
2077 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2078 zil_commit(zilog, 0);
2080 zfs_inode_update(dzp);
2081 zfs_inode_update(zp);
2087 * Remove a directory subdir entry. If the current working
2088 * directory is the same as the subdir to be removed, the
2091 * IN: dip - inode of directory to remove from.
2092 * name - name of directory to be removed.
2093 * cwd - inode of current working directory.
2094 * cr - credentials of caller.
2095 * flags - case flags
2097 * RETURN: 0 on success, error code on failure.
2100 * dip - ctime|mtime updated
2104 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
2107 znode_t *dzp = ITOZ(dip);
2110 zfsvfs_t *zfsvfs = ITOZSB(dip);
2116 boolean_t waited = B_FALSE;
2119 return (SET_ERROR(EINVAL));
2123 zilog = zfsvfs->z_log;
2125 if (flags & FIGNORECASE)
2131 * Attempt to lock directory; fail if entry doesn't exist.
2133 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2141 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
2145 if (!S_ISDIR(ip->i_mode)) {
2146 error = SET_ERROR(ENOTDIR);
2151 error = SET_ERROR(EINVAL);
2156 * Grab a lock on the directory to make sure that no one is
2157 * trying to add (or lookup) entries while we are removing it.
2159 rw_enter(&zp->z_name_lock, RW_WRITER);
2162 * Grab a lock on the parent pointer to make sure we play well
2163 * with the treewalk and directory rename code.
2165 rw_enter(&zp->z_parent_lock, RW_WRITER);
2167 tx = dmu_tx_create(zfsvfs->z_os);
2168 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2169 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2170 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2171 zfs_sa_upgrade_txholds(tx, zp);
2172 zfs_sa_upgrade_txholds(tx, dzp);
2173 dmu_tx_mark_netfree(tx);
2174 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
2176 rw_exit(&zp->z_parent_lock);
2177 rw_exit(&zp->z_name_lock);
2178 zfs_dirent_unlock(dl);
2179 if (error == ERESTART) {
2192 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2195 uint64_t txtype = TX_RMDIR;
2196 if (flags & FIGNORECASE)
2198 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2203 rw_exit(&zp->z_parent_lock);
2204 rw_exit(&zp->z_name_lock);
2206 zfs_dirent_unlock(dl);
2208 zfs_inode_update(dzp);
2209 zfs_inode_update(zp);
2212 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2213 zil_commit(zilog, 0);
2220 * Read as many directory entries as will fit into the provided
2221 * dirent buffer from the given directory cursor position.
2223 * IN: ip - inode of directory to read.
2224 * dirent - buffer for directory entries.
2226 * OUT: dirent - filler buffer of directory entries.
2228 * RETURN: 0 if success
2229 * error code if failure
2232 * ip - atime updated
2234 * Note that the low 4 bits of the cookie returned by zap is always zero.
2235 * This allows us to use the low range for "special" directory entries:
2236 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2237 * we use the offset 2 for the '.zfs' directory.
2241 zfs_readdir(struct inode *ip, struct dir_context *ctx, cred_t *cr)
2243 znode_t *zp = ITOZ(ip);
2244 zfsvfs_t *zfsvfs = ITOZSB(ip);
2247 zap_attribute_t zap;
2253 uint64_t offset; /* must be unsigned; checks for < 1 */
2258 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2259 &parent, sizeof (parent))) != 0)
2263 * Quit if directory has been removed (posix)
2271 prefetch = zp->z_zn_prefetch;
2274 * Initialize the iterator cursor.
2278 * Start iteration from the beginning of the directory.
2280 zap_cursor_init(&zc, os, zp->z_id);
2283 * The offset is a serialized cursor.
2285 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2289 * Transform to file-system independent format
2294 * Special case `.', `..', and `.zfs'.
2297 (void) strcpy(zap.za_name, ".");
2298 zap.za_normalization_conflict = 0;
2301 } else if (offset == 1) {
2302 (void) strcpy(zap.za_name, "..");
2303 zap.za_normalization_conflict = 0;
2306 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2307 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2308 zap.za_normalization_conflict = 0;
2309 objnum = ZFSCTL_INO_ROOT;
2315 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2316 if (error == ENOENT)
2323 * Allow multiple entries provided the first entry is
2324 * the object id. Non-zpl consumers may safely make
2325 * use of the additional space.
2327 * XXX: This should be a feature flag for compatibility
2329 if (zap.za_integer_length != 8 ||
2330 zap.za_num_integers == 0) {
2331 cmn_err(CE_WARN, "zap_readdir: bad directory "
2332 "entry, obj = %lld, offset = %lld, "
2333 "length = %d, num = %lld\n",
2334 (u_longlong_t)zp->z_id,
2335 (u_longlong_t)offset,
2336 zap.za_integer_length,
2337 (u_longlong_t)zap.za_num_integers);
2338 error = SET_ERROR(ENXIO);
2342 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2343 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2346 done = !dir_emit(ctx, zap.za_name, strlen(zap.za_name),
2351 /* Prefetch znode */
2353 dmu_prefetch(os, objnum, 0, 0, 0,
2354 ZIO_PRIORITY_SYNC_READ);
2358 * Move to the next entry, fill in the previous offset.
2360 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2361 zap_cursor_advance(&zc);
2362 offset = zap_cursor_serialize(&zc);
2368 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2371 zap_cursor_fini(&zc);
2372 if (error == ENOENT)
2380 ulong_t zfs_fsync_sync_cnt = 4;
2383 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2385 znode_t *zp = ITOZ(ip);
2386 zfsvfs_t *zfsvfs = ITOZSB(ip);
2388 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2390 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2393 zil_commit(zfsvfs->z_log, zp->z_id);
2396 tsd_set(zfs_fsyncer_key, NULL);
2403 * Get the requested file attributes and place them in the provided
2406 * IN: ip - inode of file.
2407 * vap - va_mask identifies requested attributes.
2408 * If ATTR_XVATTR set, then optional attrs are requested
2409 * flags - ATTR_NOACLCHECK (CIFS server context)
2410 * cr - credentials of caller.
2412 * OUT: vap - attribute values.
2414 * RETURN: 0 (always succeeds)
2418 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2420 znode_t *zp = ITOZ(ip);
2421 zfsvfs_t *zfsvfs = ITOZSB(ip);
2424 uint64_t atime[2], mtime[2], ctime[2];
2425 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2426 xoptattr_t *xoap = NULL;
2427 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2428 sa_bulk_attr_t bulk[3];
2434 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2436 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
2437 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2438 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2440 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2446 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2447 * Also, if we are the owner don't bother, since owner should
2448 * always be allowed to read basic attributes of file.
2450 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2451 (vap->va_uid != crgetuid(cr))) {
2452 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2460 * Return all attributes. It's cheaper to provide the answer
2461 * than to determine whether we were asked the question.
2464 mutex_enter(&zp->z_lock);
2465 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2466 vap->va_mode = zp->z_mode;
2467 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2468 vap->va_nodeid = zp->z_id;
2469 if ((zp->z_id == zfsvfs->z_root) && zfs_show_ctldir(zp))
2470 links = ZTOI(zp)->i_nlink + 1;
2472 links = ZTOI(zp)->i_nlink;
2473 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2474 vap->va_size = i_size_read(ip);
2475 vap->va_rdev = ip->i_rdev;
2476 vap->va_seq = ip->i_generation;
2479 * Add in any requested optional attributes and the create time.
2480 * Also set the corresponding bits in the returned attribute bitmap.
2482 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2483 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2485 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2486 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2489 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2490 xoap->xoa_readonly =
2491 ((zp->z_pflags & ZFS_READONLY) != 0);
2492 XVA_SET_RTN(xvap, XAT_READONLY);
2495 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2497 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2498 XVA_SET_RTN(xvap, XAT_SYSTEM);
2501 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2503 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2504 XVA_SET_RTN(xvap, XAT_HIDDEN);
2507 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2508 xoap->xoa_nounlink =
2509 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2510 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2513 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2514 xoap->xoa_immutable =
2515 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2516 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2519 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2520 xoap->xoa_appendonly =
2521 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2522 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2525 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2527 ((zp->z_pflags & ZFS_NODUMP) != 0);
2528 XVA_SET_RTN(xvap, XAT_NODUMP);
2531 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2533 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2534 XVA_SET_RTN(xvap, XAT_OPAQUE);
2537 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2538 xoap->xoa_av_quarantined =
2539 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2540 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2543 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2544 xoap->xoa_av_modified =
2545 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2546 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2549 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2550 S_ISREG(ip->i_mode)) {
2551 zfs_sa_get_scanstamp(zp, xvap);
2554 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2557 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2558 times, sizeof (times));
2559 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2560 XVA_SET_RTN(xvap, XAT_CREATETIME);
2563 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2564 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2565 XVA_SET_RTN(xvap, XAT_REPARSE);
2567 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2568 xoap->xoa_generation = ip->i_generation;
2569 XVA_SET_RTN(xvap, XAT_GEN);
2572 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2574 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2575 XVA_SET_RTN(xvap, XAT_OFFLINE);
2578 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2580 ((zp->z_pflags & ZFS_SPARSE) != 0);
2581 XVA_SET_RTN(xvap, XAT_SPARSE);
2584 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
2585 xoap->xoa_projinherit =
2586 ((zp->z_pflags & ZFS_PROJINHERIT) != 0);
2587 XVA_SET_RTN(xvap, XAT_PROJINHERIT);
2590 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
2591 xoap->xoa_projid = zp->z_projid;
2592 XVA_SET_RTN(xvap, XAT_PROJID);
2596 ZFS_TIME_DECODE(&vap->va_atime, atime);
2597 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2598 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2600 mutex_exit(&zp->z_lock);
2602 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2604 if (zp->z_blksz == 0) {
2606 * Block size hasn't been set; suggest maximal I/O transfers.
2608 vap->va_blksize = zfsvfs->z_max_blksz;
2616 * Get the basic file attributes and place them in the provided kstat
2617 * structure. The inode is assumed to be the authoritative source
2618 * for most of the attributes. However, the znode currently has the
2619 * authoritative atime, blksize, and block count.
2621 * IN: ip - inode of file.
2623 * OUT: sp - kstat values.
2625 * RETURN: 0 (always succeeds)
2629 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2631 znode_t *zp = ITOZ(ip);
2632 zfsvfs_t *zfsvfs = ITOZSB(ip);
2634 u_longlong_t nblocks;
2639 mutex_enter(&zp->z_lock);
2641 generic_fillattr(ip, sp);
2643 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2644 sp->blksize = blksize;
2645 sp->blocks = nblocks;
2647 if (unlikely(zp->z_blksz == 0)) {
2649 * Block size hasn't been set; suggest maximal I/O transfers.
2651 sp->blksize = zfsvfs->z_max_blksz;
2654 mutex_exit(&zp->z_lock);
2657 * Required to prevent NFS client from detecting different inode
2658 * numbers of snapshot root dentry before and after snapshot mount.
2660 if (zfsvfs->z_issnap) {
2661 if (ip->i_sb->s_root->d_inode == ip)
2662 sp->ino = ZFSCTL_INO_SNAPDIRS -
2663 dmu_objset_id(zfsvfs->z_os);
2672 * For the operation of changing file's user/group/project, we need to
2673 * handle not only the main object that is assigned to the file directly,
2674 * but also the ones that are used by the file via hidden xattr directory.
2676 * Because the xattr directory may contains many EA entries, as to it may
2677 * be impossible to change all of them via the transaction of changing the
2678 * main object's user/group/project attributes. Then we have to change them
2679 * via other multiple independent transactions one by one. It may be not good
2680 * solution, but we have no better idea yet.
2683 zfs_setattr_dir(znode_t *dzp)
2685 struct inode *dxip = ZTOI(dzp);
2686 struct inode *xip = NULL;
2687 zfsvfs_t *zfsvfs = ITOZSB(dxip);
2688 objset_t *os = zfsvfs->z_os;
2690 zap_attribute_t zap;
2693 dmu_tx_t *tx = NULL;
2695 sa_bulk_attr_t bulk[4];
2699 zap_cursor_init(&zc, os, dzp->z_id);
2700 while ((err = zap_cursor_retrieve(&zc, &zap)) == 0) {
2701 if (zap.za_integer_length != 8 || zap.za_num_integers != 1) {
2706 err = zfs_dirent_lock(&dl, dzp, (char *)zap.za_name, &zp,
2707 ZEXISTS, NULL, NULL);
2714 if (KUID_TO_SUID(xip->i_uid) == KUID_TO_SUID(dxip->i_uid) &&
2715 KGID_TO_SGID(xip->i_gid) == KGID_TO_SGID(dxip->i_gid) &&
2716 zp->z_projid == dzp->z_projid)
2719 tx = dmu_tx_create(os);
2720 if (!(zp->z_pflags & ZFS_PROJID))
2721 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2723 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2725 err = dmu_tx_assign(tx, TXG_WAIT);
2729 mutex_enter(&dzp->z_lock);
2731 if (KUID_TO_SUID(xip->i_uid) != KUID_TO_SUID(dxip->i_uid)) {
2732 xip->i_uid = dxip->i_uid;
2733 uid = zfs_uid_read(dxip);
2734 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
2735 &uid, sizeof (uid));
2738 if (KGID_TO_SGID(xip->i_gid) != KGID_TO_SGID(dxip->i_gid)) {
2739 xip->i_gid = dxip->i_gid;
2740 gid = zfs_gid_read(dxip);
2741 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
2742 &gid, sizeof (gid));
2745 if (zp->z_projid != dzp->z_projid) {
2746 if (!(zp->z_pflags & ZFS_PROJID)) {
2747 zp->z_pflags |= ZFS_PROJID;
2748 SA_ADD_BULK_ATTR(bulk, count,
2749 SA_ZPL_FLAGS(zfsvfs), NULL, &zp->z_pflags,
2750 sizeof (zp->z_pflags));
2753 zp->z_projid = dzp->z_projid;
2754 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PROJID(zfsvfs),
2755 NULL, &zp->z_projid, sizeof (zp->z_projid));
2758 mutex_exit(&dzp->z_lock);
2760 if (likely(count > 0)) {
2761 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
2767 if (err != 0 && err != ENOENT)
2774 zfs_dirent_unlock(dl);
2776 zap_cursor_advance(&zc);
2783 zfs_dirent_unlock(dl);
2785 zap_cursor_fini(&zc);
2787 return (err == ENOENT ? 0 : err);
2791 * Set the file attributes to the values contained in the
2794 * IN: ip - inode of file to be modified.
2795 * vap - new attribute values.
2796 * If ATTR_XVATTR set, then optional attrs are being set
2797 * flags - ATTR_UTIME set if non-default time values provided.
2798 * - ATTR_NOACLCHECK (CIFS context only).
2799 * cr - credentials of caller.
2801 * RETURN: 0 if success
2802 * error code if failure
2805 * ip - ctime updated, mtime updated if size changed.
2809 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2811 znode_t *zp = ITOZ(ip);
2812 zfsvfs_t *zfsvfs = ITOZSB(ip);
2813 objset_t *os = zfsvfs->z_os;
2817 xvattr_t *tmpxvattr;
2818 uint_t mask = vap->va_mask;
2819 uint_t saved_mask = 0;
2822 uint64_t new_kuid = 0, new_kgid = 0, new_uid, new_gid;
2824 uint64_t mtime[2], ctime[2], atime[2];
2825 uint64_t projid = ZFS_INVALID_PROJID;
2827 int need_policy = FALSE;
2829 zfs_fuid_info_t *fuidp = NULL;
2830 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2833 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2834 boolean_t fuid_dirtied = B_FALSE;
2835 boolean_t handle_eadir = B_FALSE;
2836 sa_bulk_attr_t *bulk, *xattr_bulk;
2837 int count = 0, xattr_count = 0, bulks = 8;
2846 * If this is a xvattr_t, then get a pointer to the structure of
2847 * optional attributes. If this is NULL, then we have a vattr_t.
2849 xoap = xva_getxoptattr(xvap);
2850 if (xoap != NULL && (mask & ATTR_XVATTR)) {
2851 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
2852 if (!dmu_objset_projectquota_enabled(os) ||
2853 (!S_ISREG(ip->i_mode) && !S_ISDIR(ip->i_mode))) {
2855 return (SET_ERROR(ENOTSUP));
2858 projid = xoap->xoa_projid;
2859 if (unlikely(projid == ZFS_INVALID_PROJID)) {
2861 return (SET_ERROR(EINVAL));
2864 if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID)
2865 projid = ZFS_INVALID_PROJID;
2870 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) &&
2871 (xoap->xoa_projinherit !=
2872 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) &&
2873 (!dmu_objset_projectquota_enabled(os) ||
2874 (!S_ISREG(ip->i_mode) && !S_ISDIR(ip->i_mode)))) {
2876 return (SET_ERROR(ENOTSUP));
2880 zilog = zfsvfs->z_log;
2883 * Make sure that if we have ephemeral uid/gid or xvattr specified
2884 * that file system is at proper version level
2887 if (zfsvfs->z_use_fuids == B_FALSE &&
2888 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2889 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2890 (mask & ATTR_XVATTR))) {
2892 return (SET_ERROR(EINVAL));
2895 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2897 return (SET_ERROR(EISDIR));
2900 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2902 return (SET_ERROR(EINVAL));
2905 tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP);
2906 xva_init(tmpxvattr);
2908 bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * bulks, KM_SLEEP);
2909 xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * bulks, KM_SLEEP);
2912 * Immutable files can only alter immutable bit and atime
2914 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2915 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2916 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2917 err = SET_ERROR(EPERM);
2921 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2922 err = SET_ERROR(EPERM);
2927 * Verify timestamps doesn't overflow 32 bits.
2928 * ZFS can handle large timestamps, but 32bit syscalls can't
2929 * handle times greater than 2039. This check should be removed
2930 * once large timestamps are fully supported.
2932 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2933 if (((mask & ATTR_ATIME) &&
2934 TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2935 ((mask & ATTR_MTIME) &&
2936 TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2937 err = SET_ERROR(EOVERFLOW);
2946 /* Can this be moved to before the top label? */
2947 if (zfs_is_readonly(zfsvfs)) {
2948 err = SET_ERROR(EROFS);
2953 * First validate permissions
2956 if (mask & ATTR_SIZE) {
2957 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2962 * XXX - Note, we are not providing any open
2963 * mode flags here (like FNDELAY), so we may
2964 * block if there are locks present... this
2965 * should be addressed in openat().
2967 /* XXX - would it be OK to generate a log record here? */
2968 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2973 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2974 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2975 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2976 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2977 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2978 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2979 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2980 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2981 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2985 if (mask & (ATTR_UID|ATTR_GID)) {
2986 int idmask = (mask & (ATTR_UID|ATTR_GID));
2991 * NOTE: even if a new mode is being set,
2992 * we may clear S_ISUID/S_ISGID bits.
2995 if (!(mask & ATTR_MODE))
2996 vap->va_mode = zp->z_mode;
2999 * Take ownership or chgrp to group we are a member of
3002 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
3003 take_group = (mask & ATTR_GID) &&
3004 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3007 * If both ATTR_UID and ATTR_GID are set then take_owner and
3008 * take_group must both be set in order to allow taking
3011 * Otherwise, send the check through secpolicy_vnode_setattr()
3015 if (((idmask == (ATTR_UID|ATTR_GID)) &&
3016 take_owner && take_group) ||
3017 ((idmask == ATTR_UID) && take_owner) ||
3018 ((idmask == ATTR_GID) && take_group)) {
3019 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3020 skipaclchk, cr) == 0) {
3022 * Remove setuid/setgid for non-privileged users
3024 (void) secpolicy_setid_clear(vap, cr);
3025 trim_mask = (mask & (ATTR_UID|ATTR_GID));
3034 mutex_enter(&zp->z_lock);
3035 oldva.va_mode = zp->z_mode;
3036 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3037 if (mask & ATTR_XVATTR) {
3039 * Update xvattr mask to include only those attributes
3040 * that are actually changing.
3042 * the bits will be restored prior to actually setting
3043 * the attributes so the caller thinks they were set.
3045 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3046 if (xoap->xoa_appendonly !=
3047 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3050 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3051 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
3055 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
3056 if (xoap->xoa_projinherit !=
3057 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) {
3060 XVA_CLR_REQ(xvap, XAT_PROJINHERIT);
3061 XVA_SET_REQ(tmpxvattr, XAT_PROJINHERIT);
3065 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3066 if (xoap->xoa_nounlink !=
3067 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3070 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3071 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
3075 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3076 if (xoap->xoa_immutable !=
3077 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3080 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3081 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
3085 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3086 if (xoap->xoa_nodump !=
3087 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3090 XVA_CLR_REQ(xvap, XAT_NODUMP);
3091 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
3095 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3096 if (xoap->xoa_av_modified !=
3097 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3100 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3101 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
3105 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3106 if ((!S_ISREG(ip->i_mode) &&
3107 xoap->xoa_av_quarantined) ||
3108 xoap->xoa_av_quarantined !=
3109 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3112 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3113 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
3117 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3118 mutex_exit(&zp->z_lock);
3119 err = SET_ERROR(EPERM);
3123 if (need_policy == FALSE &&
3124 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3125 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3130 mutex_exit(&zp->z_lock);
3132 if (mask & ATTR_MODE) {
3133 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3134 err = secpolicy_setid_setsticky_clear(ip, vap,
3139 trim_mask |= ATTR_MODE;
3147 * If trim_mask is set then take ownership
3148 * has been granted or write_acl is present and user
3149 * has the ability to modify mode. In that case remove
3150 * UID|GID and or MODE from mask so that
3151 * secpolicy_vnode_setattr() doesn't revoke it.
3155 saved_mask = vap->va_mask;
3156 vap->va_mask &= ~trim_mask;
3158 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
3159 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3164 vap->va_mask |= saved_mask;
3168 * secpolicy_vnode_setattr, or take ownership may have
3171 mask = vap->va_mask;
3173 if ((mask & (ATTR_UID | ATTR_GID)) || projid != ZFS_INVALID_PROJID) {
3174 handle_eadir = B_TRUE;
3175 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3176 &xattr_obj, sizeof (xattr_obj));
3178 if (err == 0 && xattr_obj) {
3179 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
3183 if (mask & ATTR_UID) {
3184 new_kuid = zfs_fuid_create(zfsvfs,
3185 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3186 if (new_kuid != KUID_TO_SUID(ZTOI(zp)->i_uid) &&
3187 zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT,
3191 err = SET_ERROR(EDQUOT);
3196 if (mask & ATTR_GID) {
3197 new_kgid = zfs_fuid_create(zfsvfs,
3198 (uint64_t)vap->va_gid, cr, ZFS_GROUP, &fuidp);
3199 if (new_kgid != KGID_TO_SGID(ZTOI(zp)->i_gid) &&
3200 zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT,
3204 err = SET_ERROR(EDQUOT);
3209 if (projid != ZFS_INVALID_PROJID &&
3210 zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) {
3217 tx = dmu_tx_create(os);
3219 if (mask & ATTR_MODE) {
3220 uint64_t pmode = zp->z_mode;
3222 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3224 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
3226 mutex_enter(&zp->z_lock);
3227 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3229 * Are we upgrading ACL from old V0 format
3232 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3233 zfs_znode_acl_version(zp) ==
3234 ZFS_ACL_VERSION_INITIAL) {
3235 dmu_tx_hold_free(tx, acl_obj, 0,
3237 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3238 0, aclp->z_acl_bytes);
3240 dmu_tx_hold_write(tx, acl_obj, 0,
3243 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3244 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3245 0, aclp->z_acl_bytes);
3247 mutex_exit(&zp->z_lock);
3248 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3250 if (((mask & ATTR_XVATTR) &&
3251 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ||
3252 (projid != ZFS_INVALID_PROJID &&
3253 !(zp->z_pflags & ZFS_PROJID)))
3254 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3256 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3260 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3263 fuid_dirtied = zfsvfs->z_fuid_dirty;
3265 zfs_fuid_txhold(zfsvfs, tx);
3267 zfs_sa_upgrade_txholds(tx, zp);
3269 err = dmu_tx_assign(tx, TXG_WAIT);
3275 * Set each attribute requested.
3276 * We group settings according to the locks they need to acquire.
3278 * Note: you cannot set ctime directly, although it will be
3279 * updated as a side-effect of calling this function.
3282 if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) {
3284 * For the existed object that is upgraded from old system,
3285 * its on-disk layout has no slot for the project ID attribute.
3286 * But quota accounting logic needs to access related slots by
3287 * offset directly. So we need to adjust old objects' layout
3288 * to make the project ID to some unified and fixed offset.
3291 err = sa_add_projid(attrzp->z_sa_hdl, tx, projid);
3293 err = sa_add_projid(zp->z_sa_hdl, tx, projid);
3295 if (unlikely(err == EEXIST))
3300 projid = ZFS_INVALID_PROJID;
3303 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3304 mutex_enter(&zp->z_acl_lock);
3305 mutex_enter(&zp->z_lock);
3307 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3308 &zp->z_pflags, sizeof (zp->z_pflags));
3311 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3312 mutex_enter(&attrzp->z_acl_lock);
3313 mutex_enter(&attrzp->z_lock);
3314 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3315 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3316 sizeof (attrzp->z_pflags));
3317 if (projid != ZFS_INVALID_PROJID) {
3318 attrzp->z_projid = projid;
3319 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3320 SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid,
3321 sizeof (attrzp->z_projid));
3325 if (mask & (ATTR_UID|ATTR_GID)) {
3327 if (mask & ATTR_UID) {
3328 ZTOI(zp)->i_uid = SUID_TO_KUID(new_kuid);
3329 new_uid = zfs_uid_read(ZTOI(zp));
3330 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3331 &new_uid, sizeof (new_uid));
3333 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3334 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3336 ZTOI(attrzp)->i_uid = SUID_TO_KUID(new_uid);
3340 if (mask & ATTR_GID) {
3341 ZTOI(zp)->i_gid = SGID_TO_KGID(new_kgid);
3342 new_gid = zfs_gid_read(ZTOI(zp));
3343 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3344 NULL, &new_gid, sizeof (new_gid));
3346 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3347 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3349 ZTOI(attrzp)->i_gid = SGID_TO_KGID(new_kgid);
3352 if (!(mask & ATTR_MODE)) {
3353 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3354 NULL, &new_mode, sizeof (new_mode));
3355 new_mode = zp->z_mode;
3357 err = zfs_acl_chown_setattr(zp);
3360 err = zfs_acl_chown_setattr(attrzp);
3365 if (mask & ATTR_MODE) {
3366 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3367 &new_mode, sizeof (new_mode));
3368 zp->z_mode = ZTOI(zp)->i_mode = new_mode;
3369 ASSERT3P(aclp, !=, NULL);
3370 err = zfs_aclset_common(zp, aclp, cr, tx);
3372 if (zp->z_acl_cached)
3373 zfs_acl_free(zp->z_acl_cached);
3374 zp->z_acl_cached = aclp;
3378 if ((mask & ATTR_ATIME) || zp->z_atime_dirty) {
3379 zp->z_atime_dirty = 0;
3380 ZFS_TIME_ENCODE(&ip->i_atime, atime);
3381 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3382 &atime, sizeof (atime));
3385 if (mask & (ATTR_MTIME | ATTR_SIZE)) {
3386 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3387 ZTOI(zp)->i_mtime = timespec_trunc(vap->va_mtime,
3388 ZTOI(zp)->i_sb->s_time_gran);
3390 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3391 mtime, sizeof (mtime));
3394 if (mask & (ATTR_CTIME | ATTR_SIZE)) {
3395 ZFS_TIME_ENCODE(&vap->va_ctime, ctime);
3396 ZTOI(zp)->i_ctime = timespec_trunc(vap->va_ctime,
3397 ZTOI(zp)->i_sb->s_time_gran);
3398 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3399 ctime, sizeof (ctime));
3402 if (projid != ZFS_INVALID_PROJID) {
3403 zp->z_projid = projid;
3404 SA_ADD_BULK_ATTR(bulk, count,
3405 SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid,
3406 sizeof (zp->z_projid));
3409 if (attrzp && mask) {
3410 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3411 SA_ZPL_CTIME(zfsvfs), NULL, &ctime,
3416 * Do this after setting timestamps to prevent timestamp
3417 * update from toggling bit
3420 if (xoap && (mask & ATTR_XVATTR)) {
3423 * restore trimmed off masks
3424 * so that return masks can be set for caller.
3427 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
3428 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3430 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
3431 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3433 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
3434 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3436 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
3437 XVA_SET_REQ(xvap, XAT_NODUMP);
3439 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
3440 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3442 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
3443 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3445 if (XVA_ISSET_REQ(tmpxvattr, XAT_PROJINHERIT)) {
3446 XVA_SET_REQ(xvap, XAT_PROJINHERIT);
3449 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3450 ASSERT(S_ISREG(ip->i_mode));
3452 zfs_xvattr_set(zp, xvap, tx);
3456 zfs_fuid_sync(zfsvfs, tx);
3459 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3461 mutex_exit(&zp->z_lock);
3462 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3463 mutex_exit(&zp->z_acl_lock);
3466 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3467 mutex_exit(&attrzp->z_acl_lock);
3468 mutex_exit(&attrzp->z_lock);
3471 if (err == 0 && xattr_count > 0) {
3472 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3481 zfs_fuid_info_free(fuidp);
3489 if (err == ERESTART)
3493 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3496 if (err2 == 0 && handle_eadir)
3497 err2 = zfs_setattr_dir(attrzp);
3500 zfs_inode_update(zp);
3504 if (os->os_sync == ZFS_SYNC_ALWAYS)
3505 zil_commit(zilog, 0);
3508 kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * bulks);
3509 kmem_free(bulk, sizeof (sa_bulk_attr_t) * bulks);
3510 kmem_free(tmpxvattr, sizeof (xvattr_t));
3515 typedef struct zfs_zlock {
3516 krwlock_t *zl_rwlock; /* lock we acquired */
3517 znode_t *zl_znode; /* znode we held */
3518 struct zfs_zlock *zl_next; /* next in list */
3522 * Drop locks and release vnodes that were held by zfs_rename_lock().
3525 zfs_rename_unlock(zfs_zlock_t **zlpp)
3529 while ((zl = *zlpp) != NULL) {
3530 if (zl->zl_znode != NULL)
3531 zfs_iput_async(ZTOI(zl->zl_znode));
3532 rw_exit(zl->zl_rwlock);
3533 *zlpp = zl->zl_next;
3534 kmem_free(zl, sizeof (*zl));
3539 * Search back through the directory tree, using the ".." entries.
3540 * Lock each directory in the chain to prevent concurrent renames.
3541 * Fail any attempt to move a directory into one of its own descendants.
3542 * XXX - z_parent_lock can overlap with map or grow locks
3545 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3549 uint64_t rootid = ZTOZSB(zp)->z_root;
3550 uint64_t oidp = zp->z_id;
3551 krwlock_t *rwlp = &szp->z_parent_lock;
3552 krw_t rw = RW_WRITER;
3555 * First pass write-locks szp and compares to zp->z_id.
3556 * Later passes read-lock zp and compare to zp->z_parent.
3559 if (!rw_tryenter(rwlp, rw)) {
3561 * Another thread is renaming in this path.
3562 * Note that if we are a WRITER, we don't have any
3563 * parent_locks held yet.
3565 if (rw == RW_READER && zp->z_id > szp->z_id) {
3567 * Drop our locks and restart
3569 zfs_rename_unlock(&zl);
3573 rwlp = &szp->z_parent_lock;
3578 * Wait for other thread to drop its locks
3584 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3585 zl->zl_rwlock = rwlp;
3586 zl->zl_znode = NULL;
3587 zl->zl_next = *zlpp;
3590 if (oidp == szp->z_id) /* We're a descendant of szp */
3591 return (SET_ERROR(EINVAL));
3593 if (oidp == rootid) /* We've hit the top */
3596 if (rw == RW_READER) { /* i.e. not the first pass */
3597 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3602 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3603 &oidp, sizeof (oidp));
3604 rwlp = &zp->z_parent_lock;
3607 } while (zp->z_id != sdzp->z_id);
3613 * Move an entry from the provided source directory to the target
3614 * directory. Change the entry name as indicated.
3616 * IN: sdip - Source directory containing the "old entry".
3617 * snm - Old entry name.
3618 * tdip - Target directory to contain the "new entry".
3619 * tnm - New entry name.
3620 * cr - credentials of caller.
3621 * flags - case flags
3623 * RETURN: 0 on success, error code on failure.
3626 * sdip,tdip - ctime|mtime updated
3630 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3631 cred_t *cr, int flags)
3633 znode_t *tdzp, *szp, *tzp;
3634 znode_t *sdzp = ITOZ(sdip);
3635 zfsvfs_t *zfsvfs = ITOZSB(sdip);
3637 zfs_dirlock_t *sdl, *tdl;
3640 int cmp, serr, terr;
3643 boolean_t waited = B_FALSE;
3645 if (snm == NULL || tnm == NULL)
3646 return (SET_ERROR(EINVAL));
3649 ZFS_VERIFY_ZP(sdzp);
3650 zilog = zfsvfs->z_log;
3653 ZFS_VERIFY_ZP(tdzp);
3656 * We check i_sb because snapshots and the ctldir must have different
3659 if (tdip->i_sb != sdip->i_sb || zfsctl_is_node(tdip)) {
3661 return (SET_ERROR(EXDEV));
3664 if (zfsvfs->z_utf8 && u8_validate(tnm,
3665 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3667 return (SET_ERROR(EILSEQ));
3670 if (flags & FIGNORECASE)
3679 * This is to prevent the creation of links into attribute space
3680 * by renaming a linked file into/outof an attribute directory.
3681 * See the comment in zfs_link() for why this is considered bad.
3683 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3685 return (SET_ERROR(EINVAL));
3689 * Lock source and target directory entries. To prevent deadlock,
3690 * a lock ordering must be defined. We lock the directory with
3691 * the smallest object id first, or if it's a tie, the one with
3692 * the lexically first name.
3694 if (sdzp->z_id < tdzp->z_id) {
3696 } else if (sdzp->z_id > tdzp->z_id) {
3700 * First compare the two name arguments without
3701 * considering any case folding.
3703 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3705 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3706 ASSERT(error == 0 || !zfsvfs->z_utf8);
3709 * POSIX: "If the old argument and the new argument
3710 * both refer to links to the same existing file,
3711 * the rename() function shall return successfully
3712 * and perform no other action."
3718 * If the file system is case-folding, then we may
3719 * have some more checking to do. A case-folding file
3720 * system is either supporting mixed case sensitivity
3721 * access or is completely case-insensitive. Note
3722 * that the file system is always case preserving.
3724 * In mixed sensitivity mode case sensitive behavior
3725 * is the default. FIGNORECASE must be used to
3726 * explicitly request case insensitive behavior.
3728 * If the source and target names provided differ only
3729 * by case (e.g., a request to rename 'tim' to 'Tim'),
3730 * we will treat this as a special case in the
3731 * case-insensitive mode: as long as the source name
3732 * is an exact match, we will allow this to proceed as
3733 * a name-change request.
3735 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3736 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3737 flags & FIGNORECASE)) &&
3738 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3741 * case preserving rename request, require exact
3750 * If the source and destination directories are the same, we should
3751 * grab the z_name_lock of that directory only once.
3755 rw_enter(&sdzp->z_name_lock, RW_READER);
3759 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3760 ZEXISTS | zflg, NULL, NULL);
3761 terr = zfs_dirent_lock(&tdl,
3762 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3764 terr = zfs_dirent_lock(&tdl,
3765 tdzp, tnm, &tzp, zflg, NULL, NULL);
3766 serr = zfs_dirent_lock(&sdl,
3767 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3773 * Source entry invalid or not there.
3776 zfs_dirent_unlock(tdl);
3782 rw_exit(&sdzp->z_name_lock);
3784 if (strcmp(snm, "..") == 0)
3790 zfs_dirent_unlock(sdl);
3794 rw_exit(&sdzp->z_name_lock);
3796 if (strcmp(tnm, "..") == 0)
3803 * If we are using project inheritance, means if the directory has
3804 * ZFS_PROJINHERIT set, then its descendant directories will inherit
3805 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
3806 * such case, we only allow renames into our tree when the project
3809 if (tdzp->z_pflags & ZFS_PROJINHERIT &&
3810 tdzp->z_projid != szp->z_projid) {
3811 error = SET_ERROR(EXDEV);
3816 * Must have write access at the source to remove the old entry
3817 * and write access at the target to create the new entry.
3818 * Note that if target and source are the same, this can be
3819 * done in a single check.
3822 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3825 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3827 * Check to make sure rename is valid.
3828 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3830 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3835 * Does target exist?
3839 * Source and target must be the same type.
3841 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3842 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3843 error = SET_ERROR(ENOTDIR);
3847 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3848 error = SET_ERROR(EISDIR);
3853 * POSIX dictates that when the source and target
3854 * entries refer to the same file object, rename
3855 * must do nothing and exit without error.
3857 if (szp->z_id == tzp->z_id) {
3863 tx = dmu_tx_create(zfsvfs->z_os);
3864 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3865 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3866 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3867 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3869 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3870 zfs_sa_upgrade_txholds(tx, tdzp);
3873 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3874 zfs_sa_upgrade_txholds(tx, tzp);
3877 zfs_sa_upgrade_txholds(tx, szp);
3878 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3879 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
3882 zfs_rename_unlock(&zl);
3883 zfs_dirent_unlock(sdl);
3884 zfs_dirent_unlock(tdl);
3887 rw_exit(&sdzp->z_name_lock);
3889 if (error == ERESTART) {
3906 if (tzp) /* Attempt to remove the existing target */
3907 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3910 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3912 szp->z_pflags |= ZFS_AV_MODIFIED;
3913 if (tdzp->z_pflags & ZFS_PROJINHERIT)
3914 szp->z_pflags |= ZFS_PROJINHERIT;
3916 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3917 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3920 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3922 zfs_log_rename(zilog, tx, TX_RENAME |
3923 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3924 sdl->dl_name, tdzp, tdl->dl_name, szp);
3927 * At this point, we have successfully created
3928 * the target name, but have failed to remove
3929 * the source name. Since the create was done
3930 * with the ZRENAMING flag, there are
3931 * complications; for one, the link count is
3932 * wrong. The easiest way to deal with this
3933 * is to remove the newly created target, and
3934 * return the original error. This must
3935 * succeed; fortunately, it is very unlikely to
3936 * fail, since we just created it.
3938 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3939 ZRENAMING, NULL), ==, 0);
3947 zfs_rename_unlock(&zl);
3949 zfs_dirent_unlock(sdl);
3950 zfs_dirent_unlock(tdl);
3952 zfs_inode_update(sdzp);
3954 rw_exit(&sdzp->z_name_lock);
3957 zfs_inode_update(tdzp);
3959 zfs_inode_update(szp);
3962 zfs_inode_update(tzp);
3966 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3967 zil_commit(zilog, 0);
3974 * Insert the indicated symbolic reference entry into the directory.
3976 * IN: dip - Directory to contain new symbolic link.
3977 * link - Name for new symlink entry.
3978 * vap - Attributes of new entry.
3979 * target - Target path of new symlink.
3981 * cr - credentials of caller.
3982 * flags - case flags
3984 * RETURN: 0 on success, error code on failure.
3987 * dip - ctime|mtime updated
3991 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
3992 struct inode **ipp, cred_t *cr, int flags)
3994 znode_t *zp, *dzp = ITOZ(dip);
3997 zfsvfs_t *zfsvfs = ITOZSB(dip);
3999 uint64_t len = strlen(link);
4002 zfs_acl_ids_t acl_ids;
4003 boolean_t fuid_dirtied;
4004 uint64_t txtype = TX_SYMLINK;
4005 boolean_t waited = B_FALSE;
4007 ASSERT(S_ISLNK(vap->va_mode));
4010 return (SET_ERROR(EINVAL));
4014 zilog = zfsvfs->z_log;
4016 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4017 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4019 return (SET_ERROR(EILSEQ));
4021 if (flags & FIGNORECASE)
4024 if (len > MAXPATHLEN) {
4026 return (SET_ERROR(ENAMETOOLONG));
4029 if ((error = zfs_acl_ids_create(dzp, 0,
4030 vap, cr, NULL, &acl_ids)) != 0) {
4038 * Attempt to lock directory; fail if entry already exists.
4040 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4042 zfs_acl_ids_free(&acl_ids);
4047 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
4048 zfs_acl_ids_free(&acl_ids);
4049 zfs_dirent_unlock(dl);
4054 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, ZFS_DEFAULT_PROJID)) {
4055 zfs_acl_ids_free(&acl_ids);
4056 zfs_dirent_unlock(dl);
4058 return (SET_ERROR(EDQUOT));
4060 tx = dmu_tx_create(zfsvfs->z_os);
4061 fuid_dirtied = zfsvfs->z_fuid_dirty;
4062 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4063 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4064 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4065 ZFS_SA_BASE_ATTR_SIZE + len);
4066 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4067 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4068 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4069 acl_ids.z_aclp->z_acl_bytes);
4072 zfs_fuid_txhold(zfsvfs, tx);
4073 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
4075 zfs_dirent_unlock(dl);
4076 if (error == ERESTART) {
4082 zfs_acl_ids_free(&acl_ids);
4089 * Create a new object for the symlink.
4090 * for version 4 ZPL datsets the symlink will be an SA attribute
4092 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4095 zfs_fuid_sync(zfsvfs, tx);
4097 mutex_enter(&zp->z_lock);
4099 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4102 zfs_sa_symlink(zp, link, len, tx);
4103 mutex_exit(&zp->z_lock);
4106 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4107 &zp->z_size, sizeof (zp->z_size), tx);
4109 * Insert the new object into the directory.
4111 (void) zfs_link_create(dl, zp, tx, ZNEW);
4113 if (flags & FIGNORECASE)
4115 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4117 zfs_inode_update(dzp);
4118 zfs_inode_update(zp);
4120 zfs_acl_ids_free(&acl_ids);
4124 zfs_dirent_unlock(dl);
4128 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4129 zil_commit(zilog, 0);
4136 * Return, in the buffer contained in the provided uio structure,
4137 * the symbolic path referred to by ip.
4139 * IN: ip - inode of symbolic link
4140 * uio - structure to contain the link path.
4141 * cr - credentials of caller.
4143 * RETURN: 0 if success
4144 * error code if failure
4147 * ip - atime updated
4151 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
4153 znode_t *zp = ITOZ(ip);
4154 zfsvfs_t *zfsvfs = ITOZSB(ip);
4160 mutex_enter(&zp->z_lock);
4162 error = sa_lookup_uio(zp->z_sa_hdl,
4163 SA_ZPL_SYMLINK(zfsvfs), uio);
4165 error = zfs_sa_readlink(zp, uio);
4166 mutex_exit(&zp->z_lock);
4173 * Insert a new entry into directory tdip referencing sip.
4175 * IN: tdip - Directory to contain new entry.
4176 * sip - inode of new entry.
4177 * name - name of new entry.
4178 * cr - credentials of caller.
4180 * RETURN: 0 if success
4181 * error code if failure
4184 * tdip - ctime|mtime updated
4185 * sip - ctime updated
4189 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr,
4192 znode_t *dzp = ITOZ(tdip);
4194 zfsvfs_t *zfsvfs = ITOZSB(tdip);
4202 boolean_t waited = B_FALSE;
4203 boolean_t is_tmpfile = 0;
4206 is_tmpfile = (sip->i_nlink == 0 && (sip->i_state & I_LINKABLE));
4208 ASSERT(S_ISDIR(tdip->i_mode));
4211 return (SET_ERROR(EINVAL));
4215 zilog = zfsvfs->z_log;
4218 * POSIX dictates that we return EPERM here.
4219 * Better choices include ENOTSUP or EISDIR.
4221 if (S_ISDIR(sip->i_mode)) {
4223 return (SET_ERROR(EPERM));
4230 * If we are using project inheritance, means if the directory has
4231 * ZFS_PROJINHERIT set, then its descendant directories will inherit
4232 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
4233 * such case, we only allow hard link creation in our tree when the
4234 * project IDs are the same.
4236 if (dzp->z_pflags & ZFS_PROJINHERIT && dzp->z_projid != szp->z_projid) {
4238 return (SET_ERROR(EXDEV));
4242 * We check i_sb because snapshots and the ctldir must have different
4245 if (sip->i_sb != tdip->i_sb || zfsctl_is_node(sip)) {
4247 return (SET_ERROR(EXDEV));
4250 /* Prevent links to .zfs/shares files */
4252 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4253 &parent, sizeof (uint64_t))) != 0) {
4257 if (parent == zfsvfs->z_shares_dir) {
4259 return (SET_ERROR(EPERM));
4262 if (zfsvfs->z_utf8 && u8_validate(name,
4263 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4265 return (SET_ERROR(EILSEQ));
4267 if (flags & FIGNORECASE)
4271 * We do not support links between attributes and non-attributes
4272 * because of the potential security risk of creating links
4273 * into "normal" file space in order to circumvent restrictions
4274 * imposed in attribute space.
4276 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4278 return (SET_ERROR(EINVAL));
4281 owner = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(sip->i_uid),
4283 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4285 return (SET_ERROR(EPERM));
4288 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
4295 * Attempt to lock directory; fail if entry already exists.
4297 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4303 tx = dmu_tx_create(zfsvfs->z_os);
4304 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4305 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4307 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
4309 zfs_sa_upgrade_txholds(tx, szp);
4310 zfs_sa_upgrade_txholds(tx, dzp);
4311 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
4313 zfs_dirent_unlock(dl);
4314 if (error == ERESTART) {
4324 /* unmark z_unlinked so zfs_link_create will not reject */
4326 szp->z_unlinked = 0;
4327 error = zfs_link_create(dl, szp, tx, 0);
4330 uint64_t txtype = TX_LINK;
4332 * tmpfile is created to be in z_unlinkedobj, so remove it.
4333 * Also, we don't log in ZIL, be cause all previous file
4334 * operation on the tmpfile are ignored by ZIL. Instead we
4335 * always wait for txg to sync to make sure all previous
4336 * operation are sync safe.
4339 VERIFY(zap_remove_int(zfsvfs->z_os,
4340 zfsvfs->z_unlinkedobj, szp->z_id, tx) == 0);
4342 if (flags & FIGNORECASE)
4344 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4346 } else if (is_tmpfile) {
4347 /* restore z_unlinked since when linking failed */
4348 szp->z_unlinked = 1;
4350 txg = dmu_tx_get_txg(tx);
4353 zfs_dirent_unlock(dl);
4355 if (!is_tmpfile && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4356 zil_commit(zilog, 0);
4359 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), txg);
4361 zfs_inode_update(dzp);
4362 zfs_inode_update(szp);
4368 zfs_putpage_commit_cb(void *arg)
4370 struct page *pp = arg;
4373 end_page_writeback(pp);
4377 * Push a page out to disk, once the page is on stable storage the
4378 * registered commit callback will be run as notification of completion.
4380 * IN: ip - page mapped for inode.
4381 * pp - page to push (page is locked)
4382 * wbc - writeback control data
4384 * RETURN: 0 if success
4385 * error code if failure
4388 * ip - ctime|mtime updated
4392 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
4394 znode_t *zp = ITOZ(ip);
4395 zfsvfs_t *zfsvfs = ITOZSB(ip);
4403 uint64_t mtime[2], ctime[2];
4404 sa_bulk_attr_t bulk[3];
4406 struct address_space *mapping;
4411 ASSERT(PageLocked(pp));
4413 pgoff = page_offset(pp); /* Page byte-offset in file */
4414 offset = i_size_read(ip); /* File length in bytes */
4415 pglen = MIN(PAGE_SIZE, /* Page length in bytes */
4416 P2ROUNDUP(offset, PAGE_SIZE)-pgoff);
4418 /* Page is beyond end of file */
4419 if (pgoff >= offset) {
4425 /* Truncate page length to end of file */
4426 if (pgoff + pglen > offset)
4427 pglen = offset - pgoff;
4431 * FIXME: Allow mmap writes past its quota. The correct fix
4432 * is to register a page_mkwrite() handler to count the page
4433 * against its quota when it is about to be dirtied.
4435 if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT,
4436 KUID_TO_SUID(ip->i_uid)) ||
4437 zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT,
4438 KGID_TO_SGID(ip->i_gid)) ||
4439 (zp->z_projid != ZFS_DEFAULT_PROJID &&
4440 zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT,
4447 * The ordering here is critical and must adhere to the following
4448 * rules in order to avoid deadlocking in either zfs_read() or
4449 * zfs_free_range() due to a lock inversion.
4451 * 1) The page must be unlocked prior to acquiring the range lock.
4452 * This is critical because zfs_read() calls find_lock_page()
4453 * which may block on the page lock while holding the range lock.
4455 * 2) Before setting or clearing write back on a page the range lock
4456 * must be held in order to prevent a lock inversion with the
4457 * zfs_free_range() function.
4459 * This presents a problem because upon entering this function the
4460 * page lock is already held. To safely acquire the range lock the
4461 * page lock must be dropped. This creates a window where another
4462 * process could truncate, invalidate, dirty, or write out the page.
4464 * Therefore, after successfully reacquiring the range and page locks
4465 * the current page state is checked. In the common case everything
4466 * will be as is expected and it can be written out. However, if
4467 * the page state has changed it must be handled accordingly.
4469 mapping = pp->mapping;
4470 redirty_page_for_writepage(wbc, pp);
4473 rl = zfs_range_lock(&zp->z_range_lock, pgoff, pglen, RL_WRITER);
4476 /* Page mapping changed or it was no longer dirty, we're done */
4477 if (unlikely((mapping != pp->mapping) || !PageDirty(pp))) {
4479 zfs_range_unlock(rl);
4484 /* Another process started write block if required */
4485 if (PageWriteback(pp)) {
4487 zfs_range_unlock(rl);
4489 if (wbc->sync_mode != WB_SYNC_NONE)
4490 wait_on_page_writeback(pp);
4496 /* Clear the dirty flag the required locks are held */
4497 if (!clear_page_dirty_for_io(pp)) {
4499 zfs_range_unlock(rl);
4505 * Counterpart for redirty_page_for_writepage() above. This page
4506 * was in fact not skipped and should not be counted as if it were.
4508 wbc->pages_skipped--;
4509 set_page_writeback(pp);
4512 tx = dmu_tx_create(zfsvfs->z_os);
4513 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
4514 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4515 zfs_sa_upgrade_txholds(tx, zp);
4517 err = dmu_tx_assign(tx, TXG_NOWAIT);
4519 if (err == ERESTART)
4523 __set_page_dirty_nobuffers(pp);
4525 end_page_writeback(pp);
4526 zfs_range_unlock(rl);
4532 ASSERT3U(pglen, <=, PAGE_SIZE);
4533 dmu_write(zfsvfs->z_os, zp->z_id, pgoff, pglen, va, tx);
4536 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
4537 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
4538 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zfsvfs), NULL,
4541 /* Preserve the mtime and ctime provided by the inode */
4542 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4543 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4544 zp->z_atime_dirty = 0;
4547 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4549 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0,
4550 zfs_putpage_commit_cb, pp);
4553 zfs_range_unlock(rl);
4555 if (wbc->sync_mode != WB_SYNC_NONE) {
4557 * Note that this is rarely called under writepages(), because
4558 * writepages() normally handles the entire commit for
4559 * performance reasons.
4561 zil_commit(zfsvfs->z_log, zp->z_id);
4569 * Update the system attributes when the inode has been dirtied. For the
4570 * moment we only update the mode, atime, mtime, and ctime.
4573 zfs_dirty_inode(struct inode *ip, int flags)
4575 znode_t *zp = ITOZ(ip);
4576 zfsvfs_t *zfsvfs = ITOZSB(ip);
4578 uint64_t mode, atime[2], mtime[2], ctime[2];
4579 sa_bulk_attr_t bulk[4];
4583 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
4591 * This is the lazytime semantic indroduced in Linux 4.0
4592 * This flag will only be called from update_time when lazytime is set.
4593 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4594 * Fortunately mtime and ctime are managed within ZFS itself, so we
4595 * only need to dirty atime.
4597 if (flags == I_DIRTY_TIME) {
4598 zp->z_atime_dirty = 1;
4603 tx = dmu_tx_create(zfsvfs->z_os);
4605 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4606 zfs_sa_upgrade_txholds(tx, zp);
4608 error = dmu_tx_assign(tx, TXG_WAIT);
4614 mutex_enter(&zp->z_lock);
4615 zp->z_atime_dirty = 0;
4617 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
4618 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
4619 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
4620 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
4622 /* Preserve the mode, mtime and ctime provided by the inode */
4623 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4624 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4625 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4630 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4631 mutex_exit(&zp->z_lock);
4641 zfs_inactive(struct inode *ip)
4643 znode_t *zp = ITOZ(ip);
4644 zfsvfs_t *zfsvfs = ITOZSB(ip);
4647 int need_unlock = 0;
4649 /* Only read lock if we haven't already write locked, e.g. rollback */
4650 if (!RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)) {
4652 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4654 if (zp->z_sa_hdl == NULL) {
4656 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4660 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4661 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4663 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4664 zfs_sa_upgrade_txholds(tx, zp);
4665 error = dmu_tx_assign(tx, TXG_WAIT);
4669 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4670 mutex_enter(&zp->z_lock);
4671 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4672 (void *)&atime, sizeof (atime), tx);
4673 zp->z_atime_dirty = 0;
4674 mutex_exit(&zp->z_lock);
4681 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4685 * Bounds-check the seek operation.
4687 * IN: ip - inode seeking within
4688 * ooff - old file offset
4689 * noffp - pointer to new file offset
4690 * ct - caller context
4692 * RETURN: 0 if success
4693 * EINVAL if new offset invalid
4697 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
4699 if (S_ISDIR(ip->i_mode))
4701 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4705 * Fill pages with data from the disk.
4708 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
4710 znode_t *zp = ITOZ(ip);
4711 zfsvfs_t *zfsvfs = ITOZSB(ip);
4713 struct page *cur_pp;
4714 u_offset_t io_off, total;
4721 io_len = nr_pages << PAGE_SHIFT;
4722 i_size = i_size_read(ip);
4723 io_off = page_offset(pl[0]);
4725 if (io_off + io_len > i_size)
4726 io_len = i_size - io_off;
4729 * Iterate over list of pages and read each page individually.
4732 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4735 cur_pp = pl[page_idx++];
4737 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4741 /* convert checksum errors into IO errors */
4743 err = SET_ERROR(EIO);
4752 * Uses zfs_fillpage to read data from the file and fill the pages.
4754 * IN: ip - inode of file to get data from.
4755 * pl - list of pages to read
4756 * nr_pages - number of pages to read
4758 * RETURN: 0 on success, error code on failure.
4761 * vp - atime updated
4765 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
4767 znode_t *zp = ITOZ(ip);
4768 zfsvfs_t *zfsvfs = ITOZSB(ip);
4777 err = zfs_fillpage(ip, pl, nr_pages);
4784 * Check ZFS specific permissions to memory map a section of a file.
4786 * IN: ip - inode of the file to mmap
4788 * addrp - start address in memory region
4789 * len - length of memory region
4790 * vm_flags- address flags
4792 * RETURN: 0 if success
4793 * error code if failure
4797 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4798 unsigned long vm_flags)
4800 znode_t *zp = ITOZ(ip);
4801 zfsvfs_t *zfsvfs = ITOZSB(ip);
4806 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4807 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4809 return (SET_ERROR(EPERM));
4812 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4813 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4815 return (SET_ERROR(EACCES));
4818 if (off < 0 || len > MAXOFFSET_T - off) {
4820 return (SET_ERROR(ENXIO));
4828 * convoff - converts the given data (start, whence) to the
4832 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4837 if ((lckdat->l_whence == 2) || (whence == 2)) {
4838 if ((error = zfs_getattr(ip, &vap, 0, CRED())))
4842 switch (lckdat->l_whence) {
4844 lckdat->l_start += offset;
4847 lckdat->l_start += vap.va_size;
4852 return (SET_ERROR(EINVAL));
4855 if (lckdat->l_start < 0)
4856 return (SET_ERROR(EINVAL));
4860 lckdat->l_start -= offset;
4863 lckdat->l_start -= vap.va_size;
4868 return (SET_ERROR(EINVAL));
4871 lckdat->l_whence = (short)whence;
4876 * Free or allocate space in a file. Currently, this function only
4877 * supports the `F_FREESP' command. However, this command is somewhat
4878 * misnamed, as its functionality includes the ability to allocate as
4879 * well as free space.
4881 * IN: ip - inode of file to free data in.
4882 * cmd - action to take (only F_FREESP supported).
4883 * bfp - section of file to free/alloc.
4884 * flag - current file open mode flags.
4885 * offset - current file offset.
4886 * cr - credentials of caller [UNUSED].
4888 * RETURN: 0 on success, error code on failure.
4891 * ip - ctime|mtime updated
4895 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4896 offset_t offset, cred_t *cr)
4898 znode_t *zp = ITOZ(ip);
4899 zfsvfs_t *zfsvfs = ITOZSB(ip);
4906 if (cmd != F_FREESP) {
4908 return (SET_ERROR(EINVAL));
4912 * Callers might not be able to detect properly that we are read-only,
4913 * so check it explicitly here.
4915 if (zfs_is_readonly(zfsvfs)) {
4917 return (SET_ERROR(EROFS));
4920 if ((error = convoff(ip, bfp, 0, offset))) {
4925 if (bfp->l_len < 0) {
4927 return (SET_ERROR(EINVAL));
4931 * Permissions aren't checked on Solaris because on this OS
4932 * zfs_space() can only be called with an opened file handle.
4933 * On Linux we can get here through truncate_range() which
4934 * operates directly on inodes, so we need to check access rights.
4936 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4942 len = bfp->l_len; /* 0 means from off to end of file */
4944 error = zfs_freesp(zp, off, len, flag, TRUE);
4952 zfs_fid(struct inode *ip, fid_t *fidp)
4954 znode_t *zp = ITOZ(ip);
4955 zfsvfs_t *zfsvfs = ITOZSB(ip);
4958 uint64_t object = zp->z_id;
4965 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4966 &gen64, sizeof (uint64_t))) != 0) {
4971 gen = (uint32_t)gen64;
4973 size = SHORT_FID_LEN;
4975 zfid = (zfid_short_t *)fidp;
4977 zfid->zf_len = size;
4979 for (i = 0; i < sizeof (zfid->zf_object); i++)
4980 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4982 /* Must have a non-zero generation number to distinguish from .zfs */
4985 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4986 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4994 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4996 znode_t *zp = ITOZ(ip);
4997 zfsvfs_t *zfsvfs = ITOZSB(ip);
4999 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5003 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5011 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
5013 znode_t *zp = ITOZ(ip);
5014 zfsvfs_t *zfsvfs = ITOZSB(ip);
5016 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5017 zilog_t *zilog = zfsvfs->z_log;
5022 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5024 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5025 zil_commit(zilog, 0);
5031 #ifdef HAVE_UIO_ZEROCOPY
5033 * Tunable, both must be a power of 2.
5035 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
5036 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
5037 * an arcbuf for a partial block read
5039 int zcr_blksz_min = (1 << 10); /* 1K */
5040 int zcr_blksz_max = (1 << 17); /* 128K */
5044 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
5046 znode_t *zp = ITOZ(ip);
5047 zfsvfs_t *zfsvfs = ITOZSB(ip);
5048 int max_blksz = zfsvfs->z_max_blksz;
5049 uio_t *uio = &xuio->xu_uio;
5050 ssize_t size = uio->uio_resid;
5051 offset_t offset = uio->uio_loffset;
5056 int preamble, postamble;
5058 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5059 return (SET_ERROR(EINVAL));
5066 * Loan out an arc_buf for write if write size is bigger than
5067 * max_blksz, and the file's block size is also max_blksz.
5070 if (size < blksz || zp->z_blksz != blksz) {
5072 return (SET_ERROR(EINVAL));
5075 * Caller requests buffers for write before knowing where the
5076 * write offset might be (e.g. NFS TCP write).
5081 preamble = P2PHASE(offset, blksz);
5083 preamble = blksz - preamble;
5088 postamble = P2PHASE(size, blksz);
5091 fullblk = size / blksz;
5092 (void) dmu_xuio_init(xuio,
5093 (preamble != 0) + fullblk + (postamble != 0));
5096 * Have to fix iov base/len for partial buffers. They
5097 * currently represent full arc_buf's.
5100 /* data begins in the middle of the arc_buf */
5101 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5104 (void) dmu_xuio_add(xuio, abuf,
5105 blksz - preamble, preamble);
5108 for (i = 0; i < fullblk; i++) {
5109 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5112 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5116 /* data ends in the middle of the arc_buf */
5117 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5120 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5125 * Loan out an arc_buf for read if the read size is larger than
5126 * the current file block size. Block alignment is not
5127 * considered. Partial arc_buf will be loaned out for read.
5129 blksz = zp->z_blksz;
5130 if (blksz < zcr_blksz_min)
5131 blksz = zcr_blksz_min;
5132 if (blksz > zcr_blksz_max)
5133 blksz = zcr_blksz_max;
5134 /* avoid potential complexity of dealing with it */
5135 if (blksz > max_blksz) {
5137 return (SET_ERROR(EINVAL));
5140 maxsize = zp->z_size - uio->uio_loffset;
5146 return (SET_ERROR(EINVAL));
5151 return (SET_ERROR(EINVAL));
5154 uio->uio_extflg = UIO_XUIO;
5155 XUIO_XUZC_RW(xuio) = ioflag;
5162 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
5166 int ioflag = XUIO_XUZC_RW(xuio);
5168 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5170 i = dmu_xuio_cnt(xuio);
5172 abuf = dmu_xuio_arcbuf(xuio, i);
5174 * if abuf == NULL, it must be a write buffer
5175 * that has been returned in zfs_write().
5178 dmu_return_arcbuf(abuf);
5179 ASSERT(abuf || ioflag == UIO_WRITE);
5182 dmu_xuio_fini(xuio);
5185 #endif /* HAVE_UIO_ZEROCOPY */
5187 #if defined(_KERNEL) && defined(HAVE_SPL)
5188 EXPORT_SYMBOL(zfs_open);
5189 EXPORT_SYMBOL(zfs_close);
5190 EXPORT_SYMBOL(zfs_read);
5191 EXPORT_SYMBOL(zfs_write);
5192 EXPORT_SYMBOL(zfs_access);
5193 EXPORT_SYMBOL(zfs_lookup);
5194 EXPORT_SYMBOL(zfs_create);
5195 EXPORT_SYMBOL(zfs_tmpfile);
5196 EXPORT_SYMBOL(zfs_remove);
5197 EXPORT_SYMBOL(zfs_mkdir);
5198 EXPORT_SYMBOL(zfs_rmdir);
5199 EXPORT_SYMBOL(zfs_readdir);
5200 EXPORT_SYMBOL(zfs_fsync);
5201 EXPORT_SYMBOL(zfs_getattr);
5202 EXPORT_SYMBOL(zfs_getattr_fast);
5203 EXPORT_SYMBOL(zfs_setattr);
5204 EXPORT_SYMBOL(zfs_rename);
5205 EXPORT_SYMBOL(zfs_symlink);
5206 EXPORT_SYMBOL(zfs_readlink);
5207 EXPORT_SYMBOL(zfs_link);
5208 EXPORT_SYMBOL(zfs_inactive);
5209 EXPORT_SYMBOL(zfs_space);
5210 EXPORT_SYMBOL(zfs_fid);
5211 EXPORT_SYMBOL(zfs_getsecattr);
5212 EXPORT_SYMBOL(zfs_setsecattr);
5213 EXPORT_SYMBOL(zfs_getpage);
5214 EXPORT_SYMBOL(zfs_putpage);
5215 EXPORT_SYMBOL(zfs_dirty_inode);
5216 EXPORT_SYMBOL(zfs_map);
5219 module_param(zfs_delete_blocks, ulong, 0644);
5220 MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async");
5221 module_param(zfs_read_chunk_size, long, 0644);
5222 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");