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;
1115 * TX_WRITE2 relies on the data previously
1116 * written by the TX_WRITE that caused
1117 * EALREADY. We zero out the BP because
1118 * it is the old, currently-on-disk BP,
1119 * so there's no need to zio_flush() its
1120 * vdevs (flushing would needlesly hurt
1121 * performance, and doesn't work on
1131 zfs_get_done(zgd, error);
1138 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
1140 znode_t *zp = ITOZ(ip);
1141 zfsvfs_t *zfsvfs = ITOZSB(ip);
1147 if (flag & V_ACE_MASK)
1148 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1150 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1157 * Lookup an entry in a directory, or an extended attribute directory.
1158 * If it exists, return a held inode reference for it.
1160 * IN: dip - inode of directory to search.
1161 * nm - name of entry to lookup.
1162 * flags - LOOKUP_XATTR set if looking for an attribute.
1163 * cr - credentials of caller.
1164 * direntflags - directory lookup flags
1165 * realpnp - returned pathname.
1167 * OUT: ipp - inode of located entry, NULL if not found.
1169 * RETURN: 0 on success, error code on failure.
1176 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
1177 cred_t *cr, int *direntflags, pathname_t *realpnp)
1179 znode_t *zdp = ITOZ(dip);
1180 zfsvfs_t *zfsvfs = ITOZSB(dip);
1184 * Fast path lookup, however we must skip DNLC lookup
1185 * for case folding or normalizing lookups because the
1186 * DNLC code only stores the passed in name. This means
1187 * creating 'a' and removing 'A' on a case insensitive
1188 * file system would work, but DNLC still thinks 'a'
1189 * exists and won't let you create it again on the next
1190 * pass through fast path.
1192 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1194 if (!S_ISDIR(dip->i_mode)) {
1195 return (SET_ERROR(ENOTDIR));
1196 } else if (zdp->z_sa_hdl == NULL) {
1197 return (SET_ERROR(EIO));
1200 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1201 error = zfs_fastaccesschk_execute(zdp, cr);
1209 } else if (!zdp->z_zfsvfs->z_norm &&
1210 (zdp->z_zfsvfs->z_case == ZFS_CASE_SENSITIVE)) {
1212 vnode_t *tvp = dnlc_lookup(dvp, nm);
1215 error = zfs_fastaccesschk_execute(zdp, cr);
1220 if (tvp == DNLC_NO_VNODE) {
1222 return (SET_ERROR(ENOENT));
1225 return (specvp_check(vpp, cr));
1228 #endif /* HAVE_DNLC */
1237 if (flags & LOOKUP_XATTR) {
1239 * We don't allow recursive attributes..
1240 * Maybe someday we will.
1242 if (zdp->z_pflags & ZFS_XATTR) {
1244 return (SET_ERROR(EINVAL));
1247 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1253 * Do we have permission to get into attribute directory?
1256 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1266 if (!S_ISDIR(dip->i_mode)) {
1268 return (SET_ERROR(ENOTDIR));
1272 * Check accessibility of directory.
1275 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1280 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1281 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1283 return (SET_ERROR(EILSEQ));
1286 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1287 if ((error == 0) && (*ipp))
1288 zfs_inode_update(ITOZ(*ipp));
1295 * Attempt to create a new entry in a directory. If the entry
1296 * already exists, truncate the file if permissible, else return
1297 * an error. Return the ip of the created or trunc'd file.
1299 * IN: dip - inode of directory to put new file entry in.
1300 * name - name of new file entry.
1301 * vap - attributes of new file.
1302 * excl - flag indicating exclusive or non-exclusive mode.
1303 * mode - mode to open file with.
1304 * cr - credentials of caller.
1305 * flag - large file flag [UNUSED].
1306 * vsecp - ACL to be set
1308 * OUT: ipp - inode of created or trunc'd entry.
1310 * RETURN: 0 on success, error code on failure.
1313 * dip - ctime|mtime updated if new entry created
1314 * ip - ctime|mtime always, atime if new
1319 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1320 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1322 znode_t *zp, *dzp = ITOZ(dip);
1323 zfsvfs_t *zfsvfs = ITOZSB(dip);
1331 zfs_acl_ids_t acl_ids;
1332 boolean_t fuid_dirtied;
1333 boolean_t have_acl = B_FALSE;
1334 boolean_t waited = B_FALSE;
1337 * If we have an ephemeral id, ACL, or XVATTR then
1338 * make sure file system is at proper version
1344 if (zfsvfs->z_use_fuids == B_FALSE &&
1345 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1346 return (SET_ERROR(EINVAL));
1349 return (SET_ERROR(EINVAL));
1354 zilog = zfsvfs->z_log;
1356 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1357 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1359 return (SET_ERROR(EILSEQ));
1362 if (vap->va_mask & ATTR_XVATTR) {
1363 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1364 crgetuid(cr), cr, vap->va_mode)) != 0) {
1372 if (*name == '\0') {
1374 * Null component name refers to the directory itself.
1381 /* possible igrab(zp) */
1384 if (flag & FIGNORECASE)
1387 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1391 zfs_acl_ids_free(&acl_ids);
1392 if (strcmp(name, "..") == 0)
1393 error = SET_ERROR(EISDIR);
1401 uint64_t projid = ZFS_DEFAULT_PROJID;
1404 * Create a new file object and update the directory
1407 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1409 zfs_acl_ids_free(&acl_ids);
1414 * We only support the creation of regular files in
1415 * extended attribute directories.
1418 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1420 zfs_acl_ids_free(&acl_ids);
1421 error = SET_ERROR(EINVAL);
1425 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1426 cr, vsecp, &acl_ids)) != 0)
1430 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode))
1431 projid = zfs_inherit_projid(dzp);
1432 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) {
1433 zfs_acl_ids_free(&acl_ids);
1434 error = SET_ERROR(EDQUOT);
1438 tx = dmu_tx_create(os);
1440 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1441 ZFS_SA_BASE_ATTR_SIZE);
1443 fuid_dirtied = zfsvfs->z_fuid_dirty;
1445 zfs_fuid_txhold(zfsvfs, tx);
1446 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1447 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1448 if (!zfsvfs->z_use_sa &&
1449 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1450 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1451 0, acl_ids.z_aclp->z_acl_bytes);
1454 error = dmu_tx_assign(tx,
1455 (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1457 zfs_dirent_unlock(dl);
1458 if (error == ERESTART) {
1464 zfs_acl_ids_free(&acl_ids);
1469 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1471 error = zfs_link_create(dl, zp, tx, ZNEW);
1474 * Since, we failed to add the directory entry for it,
1475 * delete the newly created dnode.
1477 zfs_znode_delete(zp, tx);
1478 remove_inode_hash(ZTOI(zp));
1479 zfs_acl_ids_free(&acl_ids);
1485 zfs_fuid_sync(zfsvfs, tx);
1487 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1488 if (flag & FIGNORECASE)
1490 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1491 vsecp, acl_ids.z_fuidp, vap);
1492 zfs_acl_ids_free(&acl_ids);
1495 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1498 zfs_acl_ids_free(&acl_ids);
1502 * A directory entry already exists for this name.
1505 * Can't truncate an existing file if in exclusive mode.
1508 error = SET_ERROR(EEXIST);
1512 * Can't open a directory for writing.
1514 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1515 error = SET_ERROR(EISDIR);
1519 * Verify requested access to file.
1521 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1525 mutex_enter(&dzp->z_lock);
1527 mutex_exit(&dzp->z_lock);
1530 * Truncate regular files if requested.
1532 if (S_ISREG(ZTOI(zp)->i_mode) &&
1533 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1534 /* we can't hold any locks when calling zfs_freesp() */
1536 zfs_dirent_unlock(dl);
1539 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1545 zfs_dirent_unlock(dl);
1551 zfs_inode_update(dzp);
1552 zfs_inode_update(zp);
1556 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1557 zil_commit(zilog, 0);
1565 zfs_tmpfile(struct inode *dip, vattr_t *vap, int excl,
1566 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1568 znode_t *zp = NULL, *dzp = ITOZ(dip);
1569 zfsvfs_t *zfsvfs = ITOZSB(dip);
1575 zfs_acl_ids_t acl_ids;
1576 uint64_t projid = ZFS_DEFAULT_PROJID;
1577 boolean_t fuid_dirtied;
1578 boolean_t have_acl = B_FALSE;
1579 boolean_t waited = B_FALSE;
1582 * If we have an ephemeral id, ACL, or XVATTR then
1583 * make sure file system is at proper version
1589 if (zfsvfs->z_use_fuids == B_FALSE &&
1590 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1591 return (SET_ERROR(EINVAL));
1597 if (vap->va_mask & ATTR_XVATTR) {
1598 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1599 crgetuid(cr), cr, vap->va_mode)) != 0) {
1609 * Create a new file object and update the directory
1612 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1614 zfs_acl_ids_free(&acl_ids);
1618 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1619 cr, vsecp, &acl_ids)) != 0)
1623 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode))
1624 projid = zfs_inherit_projid(dzp);
1625 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) {
1626 zfs_acl_ids_free(&acl_ids);
1627 error = SET_ERROR(EDQUOT);
1631 tx = dmu_tx_create(os);
1633 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1634 ZFS_SA_BASE_ATTR_SIZE);
1635 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1637 fuid_dirtied = zfsvfs->z_fuid_dirty;
1639 zfs_fuid_txhold(zfsvfs, tx);
1640 if (!zfsvfs->z_use_sa &&
1641 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1642 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1643 0, acl_ids.z_aclp->z_acl_bytes);
1645 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1647 if (error == ERESTART) {
1653 zfs_acl_ids_free(&acl_ids);
1658 zfs_mknode(dzp, vap, tx, cr, IS_TMPFILE, &zp, &acl_ids);
1661 zfs_fuid_sync(zfsvfs, tx);
1663 /* Add to unlinked set */
1665 zfs_unlinked_add(zp, tx);
1666 zfs_acl_ids_free(&acl_ids);
1674 zfs_inode_update(dzp);
1675 zfs_inode_update(zp);
1684 * Remove an entry from a directory.
1686 * IN: dip - inode of directory to remove entry from.
1687 * name - name of entry to remove.
1688 * cr - credentials of caller.
1690 * RETURN: 0 if success
1691 * error code if failure
1695 * ip - ctime (if nlink > 0)
1698 uint64_t null_xattr = 0;
1702 zfs_remove(struct inode *dip, char *name, cred_t *cr, int flags)
1704 znode_t *zp, *dzp = ITOZ(dip);
1707 zfsvfs_t *zfsvfs = ITOZSB(dip);
1709 uint64_t acl_obj, xattr_obj;
1710 uint64_t xattr_obj_unlinked = 0;
1715 boolean_t may_delete_now, delete_now = FALSE;
1716 boolean_t unlinked, toobig = FALSE;
1718 pathname_t *realnmp = NULL;
1722 boolean_t waited = B_FALSE;
1725 return (SET_ERROR(EINVAL));
1729 zilog = zfsvfs->z_log;
1731 if (flags & FIGNORECASE) {
1741 * Attempt to lock directory; fail if entry doesn't exist.
1743 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1753 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1758 * Need to use rmdir for removing directories.
1760 if (S_ISDIR(ip->i_mode)) {
1761 error = SET_ERROR(EPERM);
1767 dnlc_remove(dvp, realnmp->pn_buf);
1769 dnlc_remove(dvp, name);
1770 #endif /* HAVE_DNLC */
1772 mutex_enter(&zp->z_lock);
1773 may_delete_now = atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped);
1774 mutex_exit(&zp->z_lock);
1777 * We may delete the znode now, or we may put it in the unlinked set;
1778 * it depends on whether we're the last link, and on whether there are
1779 * other holds on the inode. So we dmu_tx_hold() the right things to
1780 * allow for either case.
1783 tx = dmu_tx_create(zfsvfs->z_os);
1784 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1785 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1786 zfs_sa_upgrade_txholds(tx, zp);
1787 zfs_sa_upgrade_txholds(tx, dzp);
1788 if (may_delete_now) {
1789 toobig = zp->z_size > zp->z_blksz * zfs_delete_blocks;
1790 /* if the file is too big, only hold_free a token amount */
1791 dmu_tx_hold_free(tx, zp->z_id, 0,
1792 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1795 /* are there any extended attributes? */
1796 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1797 &xattr_obj, sizeof (xattr_obj));
1798 if (error == 0 && xattr_obj) {
1799 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1801 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1802 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1805 mutex_enter(&zp->z_lock);
1806 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1807 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1808 mutex_exit(&zp->z_lock);
1810 /* charge as an update -- would be nice not to charge at all */
1811 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1814 * Mark this transaction as typically resulting in a net free of space
1816 dmu_tx_mark_netfree(tx);
1818 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1820 zfs_dirent_unlock(dl);
1821 if (error == ERESTART) {
1841 * Remove the directory entry.
1843 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1852 * Hold z_lock so that we can make sure that the ACL obj
1853 * hasn't changed. Could have been deleted due to
1856 mutex_enter(&zp->z_lock);
1857 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1858 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1859 delete_now = may_delete_now && !toobig &&
1860 atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped) &&
1861 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1866 if (xattr_obj_unlinked) {
1867 ASSERT3U(ZTOI(xzp)->i_nlink, ==, 2);
1868 mutex_enter(&xzp->z_lock);
1869 xzp->z_unlinked = 1;
1870 clear_nlink(ZTOI(xzp));
1872 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1873 &links, sizeof (links), tx);
1874 ASSERT3U(error, ==, 0);
1875 mutex_exit(&xzp->z_lock);
1876 zfs_unlinked_add(xzp, tx);
1879 error = sa_remove(zp->z_sa_hdl,
1880 SA_ZPL_XATTR(zfsvfs), tx);
1882 error = sa_update(zp->z_sa_hdl,
1883 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1884 sizeof (uint64_t), tx);
1888 * Add to the unlinked set because a new reference could be
1889 * taken concurrently resulting in a deferred destruction.
1891 zfs_unlinked_add(zp, tx);
1892 mutex_exit(&zp->z_lock);
1893 } else if (unlinked) {
1894 mutex_exit(&zp->z_lock);
1895 zfs_unlinked_add(zp, tx);
1899 if (flags & FIGNORECASE)
1901 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1908 zfs_dirent_unlock(dl);
1909 zfs_inode_update(dzp);
1910 zfs_inode_update(zp);
1918 zfs_inode_update(xzp);
1919 zfs_iput_async(ZTOI(xzp));
1922 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1923 zil_commit(zilog, 0);
1930 * Create a new directory and insert it into dip using the name
1931 * provided. Return a pointer to the inserted directory.
1933 * IN: dip - inode of directory to add subdir to.
1934 * dirname - name of new directory.
1935 * vap - attributes of new directory.
1936 * cr - credentials of caller.
1937 * vsecp - ACL to be set
1939 * OUT: ipp - inode of created directory.
1941 * RETURN: 0 if success
1942 * error code if failure
1945 * dip - ctime|mtime updated
1946 * ipp - ctime|mtime|atime updated
1950 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1951 cred_t *cr, int flags, vsecattr_t *vsecp)
1953 znode_t *zp, *dzp = ITOZ(dip);
1954 zfsvfs_t *zfsvfs = ITOZSB(dip);
1962 gid_t gid = crgetgid(cr);
1963 zfs_acl_ids_t acl_ids;
1964 boolean_t fuid_dirtied;
1965 boolean_t waited = B_FALSE;
1967 ASSERT(S_ISDIR(vap->va_mode));
1970 * If we have an ephemeral id, ACL, or XVATTR then
1971 * make sure file system is at proper version
1975 if (zfsvfs->z_use_fuids == B_FALSE &&
1976 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1977 return (SET_ERROR(EINVAL));
1979 if (dirname == NULL)
1980 return (SET_ERROR(EINVAL));
1984 zilog = zfsvfs->z_log;
1986 if (dzp->z_pflags & ZFS_XATTR) {
1988 return (SET_ERROR(EINVAL));
1991 if (zfsvfs->z_utf8 && u8_validate(dirname,
1992 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1994 return (SET_ERROR(EILSEQ));
1996 if (flags & FIGNORECASE)
1999 if (vap->va_mask & ATTR_XVATTR) {
2000 if ((error = secpolicy_xvattr((xvattr_t *)vap,
2001 crgetuid(cr), cr, vap->va_mode)) != 0) {
2007 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2008 vsecp, &acl_ids)) != 0) {
2013 * First make sure the new directory doesn't exist.
2015 * Existence is checked first to make sure we don't return
2016 * EACCES instead of EEXIST which can cause some applications
2022 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2024 zfs_acl_ids_free(&acl_ids);
2029 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
2030 zfs_acl_ids_free(&acl_ids);
2031 zfs_dirent_unlock(dl);
2036 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) {
2037 zfs_acl_ids_free(&acl_ids);
2038 zfs_dirent_unlock(dl);
2040 return (SET_ERROR(EDQUOT));
2044 * Add a new entry to the directory.
2046 tx = dmu_tx_create(zfsvfs->z_os);
2047 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2048 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2049 fuid_dirtied = zfsvfs->z_fuid_dirty;
2051 zfs_fuid_txhold(zfsvfs, tx);
2052 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2053 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2054 acl_ids.z_aclp->z_acl_bytes);
2057 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2058 ZFS_SA_BASE_ATTR_SIZE);
2060 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
2062 zfs_dirent_unlock(dl);
2063 if (error == ERESTART) {
2069 zfs_acl_ids_free(&acl_ids);
2078 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2081 * Now put new name in parent dir.
2083 error = zfs_link_create(dl, zp, tx, ZNEW);
2085 zfs_znode_delete(zp, tx);
2086 remove_inode_hash(ZTOI(zp));
2091 zfs_fuid_sync(zfsvfs, tx);
2095 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2096 if (flags & FIGNORECASE)
2098 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2099 acl_ids.z_fuidp, vap);
2102 zfs_acl_ids_free(&acl_ids);
2106 zfs_dirent_unlock(dl);
2108 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2109 zil_commit(zilog, 0);
2114 zfs_inode_update(dzp);
2115 zfs_inode_update(zp);
2122 * Remove a directory subdir entry. If the current working
2123 * directory is the same as the subdir to be removed, the
2126 * IN: dip - inode of directory to remove from.
2127 * name - name of directory to be removed.
2128 * cwd - inode of current working directory.
2129 * cr - credentials of caller.
2130 * flags - case flags
2132 * RETURN: 0 on success, error code on failure.
2135 * dip - ctime|mtime updated
2139 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
2142 znode_t *dzp = ITOZ(dip);
2145 zfsvfs_t *zfsvfs = ITOZSB(dip);
2151 boolean_t waited = B_FALSE;
2154 return (SET_ERROR(EINVAL));
2158 zilog = zfsvfs->z_log;
2160 if (flags & FIGNORECASE)
2166 * Attempt to lock directory; fail if entry doesn't exist.
2168 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2176 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
2180 if (!S_ISDIR(ip->i_mode)) {
2181 error = SET_ERROR(ENOTDIR);
2186 error = SET_ERROR(EINVAL);
2191 * Grab a lock on the directory to make sure that no one is
2192 * trying to add (or lookup) entries while we are removing it.
2194 rw_enter(&zp->z_name_lock, RW_WRITER);
2197 * Grab a lock on the parent pointer to make sure we play well
2198 * with the treewalk and directory rename code.
2200 rw_enter(&zp->z_parent_lock, RW_WRITER);
2202 tx = dmu_tx_create(zfsvfs->z_os);
2203 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2204 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2205 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2206 zfs_sa_upgrade_txholds(tx, zp);
2207 zfs_sa_upgrade_txholds(tx, dzp);
2208 dmu_tx_mark_netfree(tx);
2209 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
2211 rw_exit(&zp->z_parent_lock);
2212 rw_exit(&zp->z_name_lock);
2213 zfs_dirent_unlock(dl);
2214 if (error == ERESTART) {
2227 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2230 uint64_t txtype = TX_RMDIR;
2231 if (flags & FIGNORECASE)
2233 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2238 rw_exit(&zp->z_parent_lock);
2239 rw_exit(&zp->z_name_lock);
2241 zfs_dirent_unlock(dl);
2243 zfs_inode_update(dzp);
2244 zfs_inode_update(zp);
2247 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2248 zil_commit(zilog, 0);
2255 * Read as many directory entries as will fit into the provided
2256 * dirent buffer from the given directory cursor position.
2258 * IN: ip - inode of directory to read.
2259 * dirent - buffer for directory entries.
2261 * OUT: dirent - filler buffer of directory entries.
2263 * RETURN: 0 if success
2264 * error code if failure
2267 * ip - atime updated
2269 * Note that the low 4 bits of the cookie returned by zap is always zero.
2270 * This allows us to use the low range for "special" directory entries:
2271 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2272 * we use the offset 2 for the '.zfs' directory.
2276 zfs_readdir(struct inode *ip, struct dir_context *ctx, cred_t *cr)
2278 znode_t *zp = ITOZ(ip);
2279 zfsvfs_t *zfsvfs = ITOZSB(ip);
2282 zap_attribute_t zap;
2288 uint64_t offset; /* must be unsigned; checks for < 1 */
2293 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2294 &parent, sizeof (parent))) != 0)
2298 * Quit if directory has been removed (posix)
2306 prefetch = zp->z_zn_prefetch;
2309 * Initialize the iterator cursor.
2313 * Start iteration from the beginning of the directory.
2315 zap_cursor_init(&zc, os, zp->z_id);
2318 * The offset is a serialized cursor.
2320 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2324 * Transform to file-system independent format
2329 * Special case `.', `..', and `.zfs'.
2332 (void) strcpy(zap.za_name, ".");
2333 zap.za_normalization_conflict = 0;
2336 } else if (offset == 1) {
2337 (void) strcpy(zap.za_name, "..");
2338 zap.za_normalization_conflict = 0;
2341 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2342 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2343 zap.za_normalization_conflict = 0;
2344 objnum = ZFSCTL_INO_ROOT;
2350 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2351 if (error == ENOENT)
2358 * Allow multiple entries provided the first entry is
2359 * the object id. Non-zpl consumers may safely make
2360 * use of the additional space.
2362 * XXX: This should be a feature flag for compatibility
2364 if (zap.za_integer_length != 8 ||
2365 zap.za_num_integers == 0) {
2366 cmn_err(CE_WARN, "zap_readdir: bad directory "
2367 "entry, obj = %lld, offset = %lld, "
2368 "length = %d, num = %lld\n",
2369 (u_longlong_t)zp->z_id,
2370 (u_longlong_t)offset,
2371 zap.za_integer_length,
2372 (u_longlong_t)zap.za_num_integers);
2373 error = SET_ERROR(ENXIO);
2377 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2378 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2381 done = !dir_emit(ctx, zap.za_name, strlen(zap.za_name),
2386 /* Prefetch znode */
2388 dmu_prefetch(os, objnum, 0, 0, 0,
2389 ZIO_PRIORITY_SYNC_READ);
2393 * Move to the next entry, fill in the previous offset.
2395 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2396 zap_cursor_advance(&zc);
2397 offset = zap_cursor_serialize(&zc);
2403 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2406 zap_cursor_fini(&zc);
2407 if (error == ENOENT)
2415 ulong_t zfs_fsync_sync_cnt = 4;
2418 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2420 znode_t *zp = ITOZ(ip);
2421 zfsvfs_t *zfsvfs = ITOZSB(ip);
2423 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2425 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2428 zil_commit(zfsvfs->z_log, zp->z_id);
2431 tsd_set(zfs_fsyncer_key, NULL);
2438 * Get the requested file attributes and place them in the provided
2441 * IN: ip - inode of file.
2442 * vap - va_mask identifies requested attributes.
2443 * If ATTR_XVATTR set, then optional attrs are requested
2444 * flags - ATTR_NOACLCHECK (CIFS server context)
2445 * cr - credentials of caller.
2447 * OUT: vap - attribute values.
2449 * RETURN: 0 (always succeeds)
2453 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2455 znode_t *zp = ITOZ(ip);
2456 zfsvfs_t *zfsvfs = ITOZSB(ip);
2459 uint64_t atime[2], mtime[2], ctime[2];
2460 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2461 xoptattr_t *xoap = NULL;
2462 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2463 sa_bulk_attr_t bulk[3];
2469 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2471 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
2472 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2473 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2475 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2481 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2482 * Also, if we are the owner don't bother, since owner should
2483 * always be allowed to read basic attributes of file.
2485 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2486 (vap->va_uid != crgetuid(cr))) {
2487 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2495 * Return all attributes. It's cheaper to provide the answer
2496 * than to determine whether we were asked the question.
2499 mutex_enter(&zp->z_lock);
2500 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2501 vap->va_mode = zp->z_mode;
2502 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2503 vap->va_nodeid = zp->z_id;
2504 if ((zp->z_id == zfsvfs->z_root) && zfs_show_ctldir(zp))
2505 links = ZTOI(zp)->i_nlink + 1;
2507 links = ZTOI(zp)->i_nlink;
2508 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2509 vap->va_size = i_size_read(ip);
2510 vap->va_rdev = ip->i_rdev;
2511 vap->va_seq = ip->i_generation;
2514 * Add in any requested optional attributes and the create time.
2515 * Also set the corresponding bits in the returned attribute bitmap.
2517 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2518 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2520 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2521 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2524 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2525 xoap->xoa_readonly =
2526 ((zp->z_pflags & ZFS_READONLY) != 0);
2527 XVA_SET_RTN(xvap, XAT_READONLY);
2530 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2532 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2533 XVA_SET_RTN(xvap, XAT_SYSTEM);
2536 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2538 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2539 XVA_SET_RTN(xvap, XAT_HIDDEN);
2542 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2543 xoap->xoa_nounlink =
2544 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2545 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2548 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2549 xoap->xoa_immutable =
2550 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2551 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2554 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2555 xoap->xoa_appendonly =
2556 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2557 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2560 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2562 ((zp->z_pflags & ZFS_NODUMP) != 0);
2563 XVA_SET_RTN(xvap, XAT_NODUMP);
2566 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2568 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2569 XVA_SET_RTN(xvap, XAT_OPAQUE);
2572 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2573 xoap->xoa_av_quarantined =
2574 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2575 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2578 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2579 xoap->xoa_av_modified =
2580 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2581 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2584 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2585 S_ISREG(ip->i_mode)) {
2586 zfs_sa_get_scanstamp(zp, xvap);
2589 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2592 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2593 times, sizeof (times));
2594 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2595 XVA_SET_RTN(xvap, XAT_CREATETIME);
2598 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2599 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2600 XVA_SET_RTN(xvap, XAT_REPARSE);
2602 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2603 xoap->xoa_generation = ip->i_generation;
2604 XVA_SET_RTN(xvap, XAT_GEN);
2607 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2609 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2610 XVA_SET_RTN(xvap, XAT_OFFLINE);
2613 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2615 ((zp->z_pflags & ZFS_SPARSE) != 0);
2616 XVA_SET_RTN(xvap, XAT_SPARSE);
2619 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
2620 xoap->xoa_projinherit =
2621 ((zp->z_pflags & ZFS_PROJINHERIT) != 0);
2622 XVA_SET_RTN(xvap, XAT_PROJINHERIT);
2625 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
2626 xoap->xoa_projid = zp->z_projid;
2627 XVA_SET_RTN(xvap, XAT_PROJID);
2631 ZFS_TIME_DECODE(&vap->va_atime, atime);
2632 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2633 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2635 mutex_exit(&zp->z_lock);
2637 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2639 if (zp->z_blksz == 0) {
2641 * Block size hasn't been set; suggest maximal I/O transfers.
2643 vap->va_blksize = zfsvfs->z_max_blksz;
2651 * Get the basic file attributes and place them in the provided kstat
2652 * structure. The inode is assumed to be the authoritative source
2653 * for most of the attributes. However, the znode currently has the
2654 * authoritative atime, blksize, and block count.
2656 * IN: ip - inode of file.
2658 * OUT: sp - kstat values.
2660 * RETURN: 0 (always succeeds)
2664 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2666 znode_t *zp = ITOZ(ip);
2667 zfsvfs_t *zfsvfs = ITOZSB(ip);
2669 u_longlong_t nblocks;
2674 mutex_enter(&zp->z_lock);
2676 generic_fillattr(ip, sp);
2678 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2679 sp->blksize = blksize;
2680 sp->blocks = nblocks;
2682 if (unlikely(zp->z_blksz == 0)) {
2684 * Block size hasn't been set; suggest maximal I/O transfers.
2686 sp->blksize = zfsvfs->z_max_blksz;
2689 mutex_exit(&zp->z_lock);
2692 * Required to prevent NFS client from detecting different inode
2693 * numbers of snapshot root dentry before and after snapshot mount.
2695 if (zfsvfs->z_issnap) {
2696 if (ip->i_sb->s_root->d_inode == ip)
2697 sp->ino = ZFSCTL_INO_SNAPDIRS -
2698 dmu_objset_id(zfsvfs->z_os);
2707 * For the operation of changing file's user/group/project, we need to
2708 * handle not only the main object that is assigned to the file directly,
2709 * but also the ones that are used by the file via hidden xattr directory.
2711 * Because the xattr directory may contains many EA entries, as to it may
2712 * be impossible to change all of them via the transaction of changing the
2713 * main object's user/group/project attributes. Then we have to change them
2714 * via other multiple independent transactions one by one. It may be not good
2715 * solution, but we have no better idea yet.
2718 zfs_setattr_dir(znode_t *dzp)
2720 struct inode *dxip = ZTOI(dzp);
2721 struct inode *xip = NULL;
2722 zfsvfs_t *zfsvfs = ITOZSB(dxip);
2723 objset_t *os = zfsvfs->z_os;
2725 zap_attribute_t zap;
2728 dmu_tx_t *tx = NULL;
2730 sa_bulk_attr_t bulk[4];
2734 zap_cursor_init(&zc, os, dzp->z_id);
2735 while ((err = zap_cursor_retrieve(&zc, &zap)) == 0) {
2736 if (zap.za_integer_length != 8 || zap.za_num_integers != 1) {
2741 err = zfs_dirent_lock(&dl, dzp, (char *)zap.za_name, &zp,
2742 ZEXISTS, NULL, NULL);
2749 if (KUID_TO_SUID(xip->i_uid) == KUID_TO_SUID(dxip->i_uid) &&
2750 KGID_TO_SGID(xip->i_gid) == KGID_TO_SGID(dxip->i_gid) &&
2751 zp->z_projid == dzp->z_projid)
2754 tx = dmu_tx_create(os);
2755 if (!(zp->z_pflags & ZFS_PROJID))
2756 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2758 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2760 err = dmu_tx_assign(tx, TXG_WAIT);
2764 mutex_enter(&dzp->z_lock);
2766 if (KUID_TO_SUID(xip->i_uid) != KUID_TO_SUID(dxip->i_uid)) {
2767 xip->i_uid = dxip->i_uid;
2768 uid = zfs_uid_read(dxip);
2769 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
2770 &uid, sizeof (uid));
2773 if (KGID_TO_SGID(xip->i_gid) != KGID_TO_SGID(dxip->i_gid)) {
2774 xip->i_gid = dxip->i_gid;
2775 gid = zfs_gid_read(dxip);
2776 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
2777 &gid, sizeof (gid));
2780 if (zp->z_projid != dzp->z_projid) {
2781 if (!(zp->z_pflags & ZFS_PROJID)) {
2782 zp->z_pflags |= ZFS_PROJID;
2783 SA_ADD_BULK_ATTR(bulk, count,
2784 SA_ZPL_FLAGS(zfsvfs), NULL, &zp->z_pflags,
2785 sizeof (zp->z_pflags));
2788 zp->z_projid = dzp->z_projid;
2789 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PROJID(zfsvfs),
2790 NULL, &zp->z_projid, sizeof (zp->z_projid));
2793 mutex_exit(&dzp->z_lock);
2795 if (likely(count > 0)) {
2796 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
2802 if (err != 0 && err != ENOENT)
2809 zfs_dirent_unlock(dl);
2811 zap_cursor_advance(&zc);
2818 zfs_dirent_unlock(dl);
2820 zap_cursor_fini(&zc);
2822 return (err == ENOENT ? 0 : err);
2826 * Set the file attributes to the values contained in the
2829 * IN: ip - inode of file to be modified.
2830 * vap - new attribute values.
2831 * If ATTR_XVATTR set, then optional attrs are being set
2832 * flags - ATTR_UTIME set if non-default time values provided.
2833 * - ATTR_NOACLCHECK (CIFS context only).
2834 * cr - credentials of caller.
2836 * RETURN: 0 if success
2837 * error code if failure
2840 * ip - ctime updated, mtime updated if size changed.
2844 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2846 znode_t *zp = ITOZ(ip);
2847 zfsvfs_t *zfsvfs = ITOZSB(ip);
2848 objset_t *os = zfsvfs->z_os;
2852 xvattr_t *tmpxvattr;
2853 uint_t mask = vap->va_mask;
2854 uint_t saved_mask = 0;
2857 uint64_t new_kuid = 0, new_kgid = 0, new_uid, new_gid;
2859 uint64_t mtime[2], ctime[2], atime[2];
2860 uint64_t projid = ZFS_INVALID_PROJID;
2862 int need_policy = FALSE;
2864 zfs_fuid_info_t *fuidp = NULL;
2865 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2868 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2869 boolean_t fuid_dirtied = B_FALSE;
2870 boolean_t handle_eadir = B_FALSE;
2871 sa_bulk_attr_t *bulk, *xattr_bulk;
2872 int count = 0, xattr_count = 0, bulks = 8;
2881 * If this is a xvattr_t, then get a pointer to the structure of
2882 * optional attributes. If this is NULL, then we have a vattr_t.
2884 xoap = xva_getxoptattr(xvap);
2885 if (xoap != NULL && (mask & ATTR_XVATTR)) {
2886 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
2887 if (!dmu_objset_projectquota_enabled(os) ||
2888 (!S_ISREG(ip->i_mode) && !S_ISDIR(ip->i_mode))) {
2890 return (SET_ERROR(ENOTSUP));
2893 projid = xoap->xoa_projid;
2894 if (unlikely(projid == ZFS_INVALID_PROJID)) {
2896 return (SET_ERROR(EINVAL));
2899 if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID)
2900 projid = ZFS_INVALID_PROJID;
2905 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) &&
2906 (xoap->xoa_projinherit !=
2907 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) &&
2908 (!dmu_objset_projectquota_enabled(os) ||
2909 (!S_ISREG(ip->i_mode) && !S_ISDIR(ip->i_mode)))) {
2911 return (SET_ERROR(ENOTSUP));
2915 zilog = zfsvfs->z_log;
2918 * Make sure that if we have ephemeral uid/gid or xvattr specified
2919 * that file system is at proper version level
2922 if (zfsvfs->z_use_fuids == B_FALSE &&
2923 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2924 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2925 (mask & ATTR_XVATTR))) {
2927 return (SET_ERROR(EINVAL));
2930 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2932 return (SET_ERROR(EISDIR));
2935 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2937 return (SET_ERROR(EINVAL));
2940 tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP);
2941 xva_init(tmpxvattr);
2943 bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * bulks, KM_SLEEP);
2944 xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * bulks, KM_SLEEP);
2947 * Immutable files can only alter immutable bit and atime
2949 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2950 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2951 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2952 err = SET_ERROR(EPERM);
2956 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2957 err = SET_ERROR(EPERM);
2962 * Verify timestamps doesn't overflow 32 bits.
2963 * ZFS can handle large timestamps, but 32bit syscalls can't
2964 * handle times greater than 2039. This check should be removed
2965 * once large timestamps are fully supported.
2967 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2968 if (((mask & ATTR_ATIME) &&
2969 TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2970 ((mask & ATTR_MTIME) &&
2971 TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2972 err = SET_ERROR(EOVERFLOW);
2981 /* Can this be moved to before the top label? */
2982 if (zfs_is_readonly(zfsvfs)) {
2983 err = SET_ERROR(EROFS);
2988 * First validate permissions
2991 if (mask & ATTR_SIZE) {
2992 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2997 * XXX - Note, we are not providing any open
2998 * mode flags here (like FNDELAY), so we may
2999 * block if there are locks present... this
3000 * should be addressed in openat().
3002 /* XXX - would it be OK to generate a log record here? */
3003 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3008 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
3009 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3010 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3011 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3012 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3013 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3014 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3015 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3016 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3020 if (mask & (ATTR_UID|ATTR_GID)) {
3021 int idmask = (mask & (ATTR_UID|ATTR_GID));
3026 * NOTE: even if a new mode is being set,
3027 * we may clear S_ISUID/S_ISGID bits.
3030 if (!(mask & ATTR_MODE))
3031 vap->va_mode = zp->z_mode;
3034 * Take ownership or chgrp to group we are a member of
3037 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
3038 take_group = (mask & ATTR_GID) &&
3039 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3042 * If both ATTR_UID and ATTR_GID are set then take_owner and
3043 * take_group must both be set in order to allow taking
3046 * Otherwise, send the check through secpolicy_vnode_setattr()
3050 if (((idmask == (ATTR_UID|ATTR_GID)) &&
3051 take_owner && take_group) ||
3052 ((idmask == ATTR_UID) && take_owner) ||
3053 ((idmask == ATTR_GID) && take_group)) {
3054 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3055 skipaclchk, cr) == 0) {
3057 * Remove setuid/setgid for non-privileged users
3059 (void) secpolicy_setid_clear(vap, cr);
3060 trim_mask = (mask & (ATTR_UID|ATTR_GID));
3069 mutex_enter(&zp->z_lock);
3070 oldva.va_mode = zp->z_mode;
3071 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3072 if (mask & ATTR_XVATTR) {
3074 * Update xvattr mask to include only those attributes
3075 * that are actually changing.
3077 * the bits will be restored prior to actually setting
3078 * the attributes so the caller thinks they were set.
3080 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3081 if (xoap->xoa_appendonly !=
3082 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3085 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3086 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
3090 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
3091 if (xoap->xoa_projinherit !=
3092 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) {
3095 XVA_CLR_REQ(xvap, XAT_PROJINHERIT);
3096 XVA_SET_REQ(tmpxvattr, XAT_PROJINHERIT);
3100 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3101 if (xoap->xoa_nounlink !=
3102 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3105 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3106 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
3110 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3111 if (xoap->xoa_immutable !=
3112 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3115 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3116 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
3120 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3121 if (xoap->xoa_nodump !=
3122 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3125 XVA_CLR_REQ(xvap, XAT_NODUMP);
3126 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
3130 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3131 if (xoap->xoa_av_modified !=
3132 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3135 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3136 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
3140 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3141 if ((!S_ISREG(ip->i_mode) &&
3142 xoap->xoa_av_quarantined) ||
3143 xoap->xoa_av_quarantined !=
3144 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3147 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3148 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
3152 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3153 mutex_exit(&zp->z_lock);
3154 err = SET_ERROR(EPERM);
3158 if (need_policy == FALSE &&
3159 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3160 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3165 mutex_exit(&zp->z_lock);
3167 if (mask & ATTR_MODE) {
3168 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3169 err = secpolicy_setid_setsticky_clear(ip, vap,
3174 trim_mask |= ATTR_MODE;
3182 * If trim_mask is set then take ownership
3183 * has been granted or write_acl is present and user
3184 * has the ability to modify mode. In that case remove
3185 * UID|GID and or MODE from mask so that
3186 * secpolicy_vnode_setattr() doesn't revoke it.
3190 saved_mask = vap->va_mask;
3191 vap->va_mask &= ~trim_mask;
3193 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
3194 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3199 vap->va_mask |= saved_mask;
3203 * secpolicy_vnode_setattr, or take ownership may have
3206 mask = vap->va_mask;
3208 if ((mask & (ATTR_UID | ATTR_GID)) || projid != ZFS_INVALID_PROJID) {
3209 handle_eadir = B_TRUE;
3210 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3211 &xattr_obj, sizeof (xattr_obj));
3213 if (err == 0 && xattr_obj) {
3214 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
3218 if (mask & ATTR_UID) {
3219 new_kuid = zfs_fuid_create(zfsvfs,
3220 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3221 if (new_kuid != KUID_TO_SUID(ZTOI(zp)->i_uid) &&
3222 zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT,
3226 err = SET_ERROR(EDQUOT);
3231 if (mask & ATTR_GID) {
3232 new_kgid = zfs_fuid_create(zfsvfs,
3233 (uint64_t)vap->va_gid, cr, ZFS_GROUP, &fuidp);
3234 if (new_kgid != KGID_TO_SGID(ZTOI(zp)->i_gid) &&
3235 zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT,
3239 err = SET_ERROR(EDQUOT);
3244 if (projid != ZFS_INVALID_PROJID &&
3245 zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) {
3252 tx = dmu_tx_create(os);
3254 if (mask & ATTR_MODE) {
3255 uint64_t pmode = zp->z_mode;
3257 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3259 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
3261 mutex_enter(&zp->z_lock);
3262 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3264 * Are we upgrading ACL from old V0 format
3267 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3268 zfs_znode_acl_version(zp) ==
3269 ZFS_ACL_VERSION_INITIAL) {
3270 dmu_tx_hold_free(tx, acl_obj, 0,
3272 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3273 0, aclp->z_acl_bytes);
3275 dmu_tx_hold_write(tx, acl_obj, 0,
3278 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3279 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3280 0, aclp->z_acl_bytes);
3282 mutex_exit(&zp->z_lock);
3283 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3285 if (((mask & ATTR_XVATTR) &&
3286 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ||
3287 (projid != ZFS_INVALID_PROJID &&
3288 !(zp->z_pflags & ZFS_PROJID)))
3289 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3291 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3295 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3298 fuid_dirtied = zfsvfs->z_fuid_dirty;
3300 zfs_fuid_txhold(zfsvfs, tx);
3302 zfs_sa_upgrade_txholds(tx, zp);
3304 err = dmu_tx_assign(tx, TXG_WAIT);
3310 * Set each attribute requested.
3311 * We group settings according to the locks they need to acquire.
3313 * Note: you cannot set ctime directly, although it will be
3314 * updated as a side-effect of calling this function.
3317 if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) {
3319 * For the existed object that is upgraded from old system,
3320 * its on-disk layout has no slot for the project ID attribute.
3321 * But quota accounting logic needs to access related slots by
3322 * offset directly. So we need to adjust old objects' layout
3323 * to make the project ID to some unified and fixed offset.
3326 err = sa_add_projid(attrzp->z_sa_hdl, tx, projid);
3328 err = sa_add_projid(zp->z_sa_hdl, tx, projid);
3330 if (unlikely(err == EEXIST))
3335 projid = ZFS_INVALID_PROJID;
3338 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3339 mutex_enter(&zp->z_acl_lock);
3340 mutex_enter(&zp->z_lock);
3342 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3343 &zp->z_pflags, sizeof (zp->z_pflags));
3346 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3347 mutex_enter(&attrzp->z_acl_lock);
3348 mutex_enter(&attrzp->z_lock);
3349 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3350 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3351 sizeof (attrzp->z_pflags));
3352 if (projid != ZFS_INVALID_PROJID) {
3353 attrzp->z_projid = projid;
3354 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3355 SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid,
3356 sizeof (attrzp->z_projid));
3360 if (mask & (ATTR_UID|ATTR_GID)) {
3362 if (mask & ATTR_UID) {
3363 ZTOI(zp)->i_uid = SUID_TO_KUID(new_kuid);
3364 new_uid = zfs_uid_read(ZTOI(zp));
3365 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3366 &new_uid, sizeof (new_uid));
3368 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3369 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3371 ZTOI(attrzp)->i_uid = SUID_TO_KUID(new_uid);
3375 if (mask & ATTR_GID) {
3376 ZTOI(zp)->i_gid = SGID_TO_KGID(new_kgid);
3377 new_gid = zfs_gid_read(ZTOI(zp));
3378 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3379 NULL, &new_gid, sizeof (new_gid));
3381 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3382 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3384 ZTOI(attrzp)->i_gid = SGID_TO_KGID(new_kgid);
3387 if (!(mask & ATTR_MODE)) {
3388 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3389 NULL, &new_mode, sizeof (new_mode));
3390 new_mode = zp->z_mode;
3392 err = zfs_acl_chown_setattr(zp);
3395 err = zfs_acl_chown_setattr(attrzp);
3400 if (mask & ATTR_MODE) {
3401 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3402 &new_mode, sizeof (new_mode));
3403 zp->z_mode = ZTOI(zp)->i_mode = new_mode;
3404 ASSERT3P(aclp, !=, NULL);
3405 err = zfs_aclset_common(zp, aclp, cr, tx);
3407 if (zp->z_acl_cached)
3408 zfs_acl_free(zp->z_acl_cached);
3409 zp->z_acl_cached = aclp;
3413 if ((mask & ATTR_ATIME) || zp->z_atime_dirty) {
3414 zp->z_atime_dirty = 0;
3415 ZFS_TIME_ENCODE(&ip->i_atime, atime);
3416 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3417 &atime, sizeof (atime));
3420 if (mask & (ATTR_MTIME | ATTR_SIZE)) {
3421 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3422 ZTOI(zp)->i_mtime = timespec_trunc(vap->va_mtime,
3423 ZTOI(zp)->i_sb->s_time_gran);
3425 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3426 mtime, sizeof (mtime));
3429 if (mask & (ATTR_CTIME | ATTR_SIZE)) {
3430 ZFS_TIME_ENCODE(&vap->va_ctime, ctime);
3431 ZTOI(zp)->i_ctime = timespec_trunc(vap->va_ctime,
3432 ZTOI(zp)->i_sb->s_time_gran);
3433 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3434 ctime, sizeof (ctime));
3437 if (projid != ZFS_INVALID_PROJID) {
3438 zp->z_projid = projid;
3439 SA_ADD_BULK_ATTR(bulk, count,
3440 SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid,
3441 sizeof (zp->z_projid));
3444 if (attrzp && mask) {
3445 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3446 SA_ZPL_CTIME(zfsvfs), NULL, &ctime,
3451 * Do this after setting timestamps to prevent timestamp
3452 * update from toggling bit
3455 if (xoap && (mask & ATTR_XVATTR)) {
3458 * restore trimmed off masks
3459 * so that return masks can be set for caller.
3462 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
3463 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3465 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
3466 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3468 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
3469 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3471 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
3472 XVA_SET_REQ(xvap, XAT_NODUMP);
3474 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
3475 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3477 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
3478 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3480 if (XVA_ISSET_REQ(tmpxvattr, XAT_PROJINHERIT)) {
3481 XVA_SET_REQ(xvap, XAT_PROJINHERIT);
3484 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3485 ASSERT(S_ISREG(ip->i_mode));
3487 zfs_xvattr_set(zp, xvap, tx);
3491 zfs_fuid_sync(zfsvfs, tx);
3494 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3496 mutex_exit(&zp->z_lock);
3497 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3498 mutex_exit(&zp->z_acl_lock);
3501 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3502 mutex_exit(&attrzp->z_acl_lock);
3503 mutex_exit(&attrzp->z_lock);
3506 if (err == 0 && xattr_count > 0) {
3507 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3516 zfs_fuid_info_free(fuidp);
3524 if (err == ERESTART)
3528 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3531 if (err2 == 0 && handle_eadir)
3532 err2 = zfs_setattr_dir(attrzp);
3535 zfs_inode_update(zp);
3539 if (os->os_sync == ZFS_SYNC_ALWAYS)
3540 zil_commit(zilog, 0);
3543 kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * bulks);
3544 kmem_free(bulk, sizeof (sa_bulk_attr_t) * bulks);
3545 kmem_free(tmpxvattr, sizeof (xvattr_t));
3550 typedef struct zfs_zlock {
3551 krwlock_t *zl_rwlock; /* lock we acquired */
3552 znode_t *zl_znode; /* znode we held */
3553 struct zfs_zlock *zl_next; /* next in list */
3557 * Drop locks and release vnodes that were held by zfs_rename_lock().
3560 zfs_rename_unlock(zfs_zlock_t **zlpp)
3564 while ((zl = *zlpp) != NULL) {
3565 if (zl->zl_znode != NULL)
3566 zfs_iput_async(ZTOI(zl->zl_znode));
3567 rw_exit(zl->zl_rwlock);
3568 *zlpp = zl->zl_next;
3569 kmem_free(zl, sizeof (*zl));
3574 * Search back through the directory tree, using the ".." entries.
3575 * Lock each directory in the chain to prevent concurrent renames.
3576 * Fail any attempt to move a directory into one of its own descendants.
3577 * XXX - z_parent_lock can overlap with map or grow locks
3580 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3584 uint64_t rootid = ZTOZSB(zp)->z_root;
3585 uint64_t oidp = zp->z_id;
3586 krwlock_t *rwlp = &szp->z_parent_lock;
3587 krw_t rw = RW_WRITER;
3590 * First pass write-locks szp and compares to zp->z_id.
3591 * Later passes read-lock zp and compare to zp->z_parent.
3594 if (!rw_tryenter(rwlp, rw)) {
3596 * Another thread is renaming in this path.
3597 * Note that if we are a WRITER, we don't have any
3598 * parent_locks held yet.
3600 if (rw == RW_READER && zp->z_id > szp->z_id) {
3602 * Drop our locks and restart
3604 zfs_rename_unlock(&zl);
3608 rwlp = &szp->z_parent_lock;
3613 * Wait for other thread to drop its locks
3619 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3620 zl->zl_rwlock = rwlp;
3621 zl->zl_znode = NULL;
3622 zl->zl_next = *zlpp;
3625 if (oidp == szp->z_id) /* We're a descendant of szp */
3626 return (SET_ERROR(EINVAL));
3628 if (oidp == rootid) /* We've hit the top */
3631 if (rw == RW_READER) { /* i.e. not the first pass */
3632 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3637 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3638 &oidp, sizeof (oidp));
3639 rwlp = &zp->z_parent_lock;
3642 } while (zp->z_id != sdzp->z_id);
3648 * Move an entry from the provided source directory to the target
3649 * directory. Change the entry name as indicated.
3651 * IN: sdip - Source directory containing the "old entry".
3652 * snm - Old entry name.
3653 * tdip - Target directory to contain the "new entry".
3654 * tnm - New entry name.
3655 * cr - credentials of caller.
3656 * flags - case flags
3658 * RETURN: 0 on success, error code on failure.
3661 * sdip,tdip - ctime|mtime updated
3665 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3666 cred_t *cr, int flags)
3668 znode_t *tdzp, *szp, *tzp;
3669 znode_t *sdzp = ITOZ(sdip);
3670 zfsvfs_t *zfsvfs = ITOZSB(sdip);
3672 zfs_dirlock_t *sdl, *tdl;
3675 int cmp, serr, terr;
3678 boolean_t waited = B_FALSE;
3680 if (snm == NULL || tnm == NULL)
3681 return (SET_ERROR(EINVAL));
3684 ZFS_VERIFY_ZP(sdzp);
3685 zilog = zfsvfs->z_log;
3688 ZFS_VERIFY_ZP(tdzp);
3691 * We check i_sb because snapshots and the ctldir must have different
3694 if (tdip->i_sb != sdip->i_sb || zfsctl_is_node(tdip)) {
3696 return (SET_ERROR(EXDEV));
3699 if (zfsvfs->z_utf8 && u8_validate(tnm,
3700 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3702 return (SET_ERROR(EILSEQ));
3705 if (flags & FIGNORECASE)
3714 * This is to prevent the creation of links into attribute space
3715 * by renaming a linked file into/outof an attribute directory.
3716 * See the comment in zfs_link() for why this is considered bad.
3718 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3720 return (SET_ERROR(EINVAL));
3724 * Lock source and target directory entries. To prevent deadlock,
3725 * a lock ordering must be defined. We lock the directory with
3726 * the smallest object id first, or if it's a tie, the one with
3727 * the lexically first name.
3729 if (sdzp->z_id < tdzp->z_id) {
3731 } else if (sdzp->z_id > tdzp->z_id) {
3735 * First compare the two name arguments without
3736 * considering any case folding.
3738 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3740 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3741 ASSERT(error == 0 || !zfsvfs->z_utf8);
3744 * POSIX: "If the old argument and the new argument
3745 * both refer to links to the same existing file,
3746 * the rename() function shall return successfully
3747 * and perform no other action."
3753 * If the file system is case-folding, then we may
3754 * have some more checking to do. A case-folding file
3755 * system is either supporting mixed case sensitivity
3756 * access or is completely case-insensitive. Note
3757 * that the file system is always case preserving.
3759 * In mixed sensitivity mode case sensitive behavior
3760 * is the default. FIGNORECASE must be used to
3761 * explicitly request case insensitive behavior.
3763 * If the source and target names provided differ only
3764 * by case (e.g., a request to rename 'tim' to 'Tim'),
3765 * we will treat this as a special case in the
3766 * case-insensitive mode: as long as the source name
3767 * is an exact match, we will allow this to proceed as
3768 * a name-change request.
3770 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3771 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3772 flags & FIGNORECASE)) &&
3773 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3776 * case preserving rename request, require exact
3785 * If the source and destination directories are the same, we should
3786 * grab the z_name_lock of that directory only once.
3790 rw_enter(&sdzp->z_name_lock, RW_READER);
3794 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3795 ZEXISTS | zflg, NULL, NULL);
3796 terr = zfs_dirent_lock(&tdl,
3797 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3799 terr = zfs_dirent_lock(&tdl,
3800 tdzp, tnm, &tzp, zflg, NULL, NULL);
3801 serr = zfs_dirent_lock(&sdl,
3802 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3808 * Source entry invalid or not there.
3811 zfs_dirent_unlock(tdl);
3817 rw_exit(&sdzp->z_name_lock);
3819 if (strcmp(snm, "..") == 0)
3825 zfs_dirent_unlock(sdl);
3829 rw_exit(&sdzp->z_name_lock);
3831 if (strcmp(tnm, "..") == 0)
3838 * If we are using project inheritance, means if the directory has
3839 * ZFS_PROJINHERIT set, then its descendant directories will inherit
3840 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
3841 * such case, we only allow renames into our tree when the project
3844 if (tdzp->z_pflags & ZFS_PROJINHERIT &&
3845 tdzp->z_projid != szp->z_projid) {
3846 error = SET_ERROR(EXDEV);
3851 * Must have write access at the source to remove the old entry
3852 * and write access at the target to create the new entry.
3853 * Note that if target and source are the same, this can be
3854 * done in a single check.
3857 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3860 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3862 * Check to make sure rename is valid.
3863 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3865 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3870 * Does target exist?
3874 * Source and target must be the same type.
3876 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3877 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3878 error = SET_ERROR(ENOTDIR);
3882 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3883 error = SET_ERROR(EISDIR);
3888 * POSIX dictates that when the source and target
3889 * entries refer to the same file object, rename
3890 * must do nothing and exit without error.
3892 if (szp->z_id == tzp->z_id) {
3898 tx = dmu_tx_create(zfsvfs->z_os);
3899 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3900 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3901 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3902 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3904 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3905 zfs_sa_upgrade_txholds(tx, tdzp);
3908 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3909 zfs_sa_upgrade_txholds(tx, tzp);
3912 zfs_sa_upgrade_txholds(tx, szp);
3913 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3914 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
3917 zfs_rename_unlock(&zl);
3918 zfs_dirent_unlock(sdl);
3919 zfs_dirent_unlock(tdl);
3922 rw_exit(&sdzp->z_name_lock);
3924 if (error == ERESTART) {
3941 if (tzp) /* Attempt to remove the existing target */
3942 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3945 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3947 szp->z_pflags |= ZFS_AV_MODIFIED;
3948 if (tdzp->z_pflags & ZFS_PROJINHERIT)
3949 szp->z_pflags |= ZFS_PROJINHERIT;
3951 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3952 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3955 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3957 zfs_log_rename(zilog, tx, TX_RENAME |
3958 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3959 sdl->dl_name, tdzp, tdl->dl_name, szp);
3962 * At this point, we have successfully created
3963 * the target name, but have failed to remove
3964 * the source name. Since the create was done
3965 * with the ZRENAMING flag, there are
3966 * complications; for one, the link count is
3967 * wrong. The easiest way to deal with this
3968 * is to remove the newly created target, and
3969 * return the original error. This must
3970 * succeed; fortunately, it is very unlikely to
3971 * fail, since we just created it.
3973 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3974 ZRENAMING, NULL), ==, 0);
3978 * If we had removed the existing target, subsequent
3979 * call to zfs_link_create() to add back the same entry
3980 * but, the new dnode (szp) should not fail.
3982 ASSERT(tzp == NULL);
3989 zfs_rename_unlock(&zl);
3991 zfs_dirent_unlock(sdl);
3992 zfs_dirent_unlock(tdl);
3994 zfs_inode_update(sdzp);
3996 rw_exit(&sdzp->z_name_lock);
3999 zfs_inode_update(tdzp);
4001 zfs_inode_update(szp);
4004 zfs_inode_update(tzp);
4008 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4009 zil_commit(zilog, 0);
4016 * Insert the indicated symbolic reference entry into the directory.
4018 * IN: dip - Directory to contain new symbolic link.
4019 * link - Name for new symlink entry.
4020 * vap - Attributes of new entry.
4021 * target - Target path of new symlink.
4023 * cr - credentials of caller.
4024 * flags - case flags
4026 * RETURN: 0 on success, error code on failure.
4029 * dip - ctime|mtime updated
4033 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
4034 struct inode **ipp, cred_t *cr, int flags)
4036 znode_t *zp, *dzp = ITOZ(dip);
4039 zfsvfs_t *zfsvfs = ITOZSB(dip);
4041 uint64_t len = strlen(link);
4044 zfs_acl_ids_t acl_ids;
4045 boolean_t fuid_dirtied;
4046 uint64_t txtype = TX_SYMLINK;
4047 boolean_t waited = B_FALSE;
4049 ASSERT(S_ISLNK(vap->va_mode));
4052 return (SET_ERROR(EINVAL));
4056 zilog = zfsvfs->z_log;
4058 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4059 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4061 return (SET_ERROR(EILSEQ));
4063 if (flags & FIGNORECASE)
4066 if (len > MAXPATHLEN) {
4068 return (SET_ERROR(ENAMETOOLONG));
4071 if ((error = zfs_acl_ids_create(dzp, 0,
4072 vap, cr, NULL, &acl_ids)) != 0) {
4080 * Attempt to lock directory; fail if entry already exists.
4082 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4084 zfs_acl_ids_free(&acl_ids);
4089 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
4090 zfs_acl_ids_free(&acl_ids);
4091 zfs_dirent_unlock(dl);
4096 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, ZFS_DEFAULT_PROJID)) {
4097 zfs_acl_ids_free(&acl_ids);
4098 zfs_dirent_unlock(dl);
4100 return (SET_ERROR(EDQUOT));
4102 tx = dmu_tx_create(zfsvfs->z_os);
4103 fuid_dirtied = zfsvfs->z_fuid_dirty;
4104 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4105 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4106 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4107 ZFS_SA_BASE_ATTR_SIZE + len);
4108 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4109 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4110 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4111 acl_ids.z_aclp->z_acl_bytes);
4114 zfs_fuid_txhold(zfsvfs, tx);
4115 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
4117 zfs_dirent_unlock(dl);
4118 if (error == ERESTART) {
4124 zfs_acl_ids_free(&acl_ids);
4131 * Create a new object for the symlink.
4132 * for version 4 ZPL datsets the symlink will be an SA attribute
4134 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4137 zfs_fuid_sync(zfsvfs, tx);
4139 mutex_enter(&zp->z_lock);
4141 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4144 zfs_sa_symlink(zp, link, len, tx);
4145 mutex_exit(&zp->z_lock);
4148 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4149 &zp->z_size, sizeof (zp->z_size), tx);
4151 * Insert the new object into the directory.
4153 error = zfs_link_create(dl, zp, tx, ZNEW);
4155 zfs_znode_delete(zp, tx);
4156 remove_inode_hash(ZTOI(zp));
4158 if (flags & FIGNORECASE)
4160 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4162 zfs_inode_update(dzp);
4163 zfs_inode_update(zp);
4166 zfs_acl_ids_free(&acl_ids);
4170 zfs_dirent_unlock(dl);
4175 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4176 zil_commit(zilog, 0);
4186 * Return, in the buffer contained in the provided uio structure,
4187 * the symbolic path referred to by ip.
4189 * IN: ip - inode of symbolic link
4190 * uio - structure to contain the link path.
4191 * cr - credentials of caller.
4193 * RETURN: 0 if success
4194 * error code if failure
4197 * ip - atime updated
4201 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
4203 znode_t *zp = ITOZ(ip);
4204 zfsvfs_t *zfsvfs = ITOZSB(ip);
4210 mutex_enter(&zp->z_lock);
4212 error = sa_lookup_uio(zp->z_sa_hdl,
4213 SA_ZPL_SYMLINK(zfsvfs), uio);
4215 error = zfs_sa_readlink(zp, uio);
4216 mutex_exit(&zp->z_lock);
4223 * Insert a new entry into directory tdip referencing sip.
4225 * IN: tdip - Directory to contain new entry.
4226 * sip - inode of new entry.
4227 * name - name of new entry.
4228 * cr - credentials of caller.
4230 * RETURN: 0 if success
4231 * error code if failure
4234 * tdip - ctime|mtime updated
4235 * sip - ctime updated
4239 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr,
4242 znode_t *dzp = ITOZ(tdip);
4244 zfsvfs_t *zfsvfs = ITOZSB(tdip);
4252 boolean_t waited = B_FALSE;
4253 boolean_t is_tmpfile = 0;
4256 is_tmpfile = (sip->i_nlink == 0 && (sip->i_state & I_LINKABLE));
4258 ASSERT(S_ISDIR(tdip->i_mode));
4261 return (SET_ERROR(EINVAL));
4265 zilog = zfsvfs->z_log;
4268 * POSIX dictates that we return EPERM here.
4269 * Better choices include ENOTSUP or EISDIR.
4271 if (S_ISDIR(sip->i_mode)) {
4273 return (SET_ERROR(EPERM));
4280 * If we are using project inheritance, means if the directory has
4281 * ZFS_PROJINHERIT set, then its descendant directories will inherit
4282 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
4283 * such case, we only allow hard link creation in our tree when the
4284 * project IDs are the same.
4286 if (dzp->z_pflags & ZFS_PROJINHERIT && dzp->z_projid != szp->z_projid) {
4288 return (SET_ERROR(EXDEV));
4292 * We check i_sb because snapshots and the ctldir must have different
4295 if (sip->i_sb != tdip->i_sb || zfsctl_is_node(sip)) {
4297 return (SET_ERROR(EXDEV));
4300 /* Prevent links to .zfs/shares files */
4302 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4303 &parent, sizeof (uint64_t))) != 0) {
4307 if (parent == zfsvfs->z_shares_dir) {
4309 return (SET_ERROR(EPERM));
4312 if (zfsvfs->z_utf8 && u8_validate(name,
4313 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4315 return (SET_ERROR(EILSEQ));
4317 if (flags & FIGNORECASE)
4321 * We do not support links between attributes and non-attributes
4322 * because of the potential security risk of creating links
4323 * into "normal" file space in order to circumvent restrictions
4324 * imposed in attribute space.
4326 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4328 return (SET_ERROR(EINVAL));
4331 owner = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(sip->i_uid),
4333 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4335 return (SET_ERROR(EPERM));
4338 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
4345 * Attempt to lock directory; fail if entry already exists.
4347 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4353 tx = dmu_tx_create(zfsvfs->z_os);
4354 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4355 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4357 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
4359 zfs_sa_upgrade_txholds(tx, szp);
4360 zfs_sa_upgrade_txholds(tx, dzp);
4361 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
4363 zfs_dirent_unlock(dl);
4364 if (error == ERESTART) {
4374 /* unmark z_unlinked so zfs_link_create will not reject */
4376 szp->z_unlinked = 0;
4377 error = zfs_link_create(dl, szp, tx, 0);
4380 uint64_t txtype = TX_LINK;
4382 * tmpfile is created to be in z_unlinkedobj, so remove it.
4383 * Also, we don't log in ZIL, be cause all previous file
4384 * operation on the tmpfile are ignored by ZIL. Instead we
4385 * always wait for txg to sync to make sure all previous
4386 * operation are sync safe.
4389 VERIFY(zap_remove_int(zfsvfs->z_os,
4390 zfsvfs->z_unlinkedobj, szp->z_id, tx) == 0);
4392 if (flags & FIGNORECASE)
4394 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4396 } else if (is_tmpfile) {
4397 /* restore z_unlinked since when linking failed */
4398 szp->z_unlinked = 1;
4400 txg = dmu_tx_get_txg(tx);
4403 zfs_dirent_unlock(dl);
4405 if (!is_tmpfile && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4406 zil_commit(zilog, 0);
4409 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), txg);
4411 zfs_inode_update(dzp);
4412 zfs_inode_update(szp);
4418 zfs_putpage_commit_cb(void *arg)
4420 struct page *pp = arg;
4423 end_page_writeback(pp);
4427 * Push a page out to disk, once the page is on stable storage the
4428 * registered commit callback will be run as notification of completion.
4430 * IN: ip - page mapped for inode.
4431 * pp - page to push (page is locked)
4432 * wbc - writeback control data
4434 * RETURN: 0 if success
4435 * error code if failure
4438 * ip - ctime|mtime updated
4442 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
4444 znode_t *zp = ITOZ(ip);
4445 zfsvfs_t *zfsvfs = ITOZSB(ip);
4453 uint64_t mtime[2], ctime[2];
4454 sa_bulk_attr_t bulk[3];
4456 struct address_space *mapping;
4461 ASSERT(PageLocked(pp));
4463 pgoff = page_offset(pp); /* Page byte-offset in file */
4464 offset = i_size_read(ip); /* File length in bytes */
4465 pglen = MIN(PAGE_SIZE, /* Page length in bytes */
4466 P2ROUNDUP(offset, PAGE_SIZE)-pgoff);
4468 /* Page is beyond end of file */
4469 if (pgoff >= offset) {
4475 /* Truncate page length to end of file */
4476 if (pgoff + pglen > offset)
4477 pglen = offset - pgoff;
4481 * FIXME: Allow mmap writes past its quota. The correct fix
4482 * is to register a page_mkwrite() handler to count the page
4483 * against its quota when it is about to be dirtied.
4485 if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT,
4486 KUID_TO_SUID(ip->i_uid)) ||
4487 zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT,
4488 KGID_TO_SGID(ip->i_gid)) ||
4489 (zp->z_projid != ZFS_DEFAULT_PROJID &&
4490 zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT,
4497 * The ordering here is critical and must adhere to the following
4498 * rules in order to avoid deadlocking in either zfs_read() or
4499 * zfs_free_range() due to a lock inversion.
4501 * 1) The page must be unlocked prior to acquiring the range lock.
4502 * This is critical because zfs_read() calls find_lock_page()
4503 * which may block on the page lock while holding the range lock.
4505 * 2) Before setting or clearing write back on a page the range lock
4506 * must be held in order to prevent a lock inversion with the
4507 * zfs_free_range() function.
4509 * This presents a problem because upon entering this function the
4510 * page lock is already held. To safely acquire the range lock the
4511 * page lock must be dropped. This creates a window where another
4512 * process could truncate, invalidate, dirty, or write out the page.
4514 * Therefore, after successfully reacquiring the range and page locks
4515 * the current page state is checked. In the common case everything
4516 * will be as is expected and it can be written out. However, if
4517 * the page state has changed it must be handled accordingly.
4519 mapping = pp->mapping;
4520 redirty_page_for_writepage(wbc, pp);
4523 rl = zfs_range_lock(&zp->z_range_lock, pgoff, pglen, RL_WRITER);
4526 /* Page mapping changed or it was no longer dirty, we're done */
4527 if (unlikely((mapping != pp->mapping) || !PageDirty(pp))) {
4529 zfs_range_unlock(rl);
4534 /* Another process started write block if required */
4535 if (PageWriteback(pp)) {
4537 zfs_range_unlock(rl);
4539 if (wbc->sync_mode != WB_SYNC_NONE)
4540 wait_on_page_writeback(pp);
4546 /* Clear the dirty flag the required locks are held */
4547 if (!clear_page_dirty_for_io(pp)) {
4549 zfs_range_unlock(rl);
4555 * Counterpart for redirty_page_for_writepage() above. This page
4556 * was in fact not skipped and should not be counted as if it were.
4558 wbc->pages_skipped--;
4559 set_page_writeback(pp);
4562 tx = dmu_tx_create(zfsvfs->z_os);
4563 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
4564 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4565 zfs_sa_upgrade_txholds(tx, zp);
4567 err = dmu_tx_assign(tx, TXG_NOWAIT);
4569 if (err == ERESTART)
4573 __set_page_dirty_nobuffers(pp);
4575 end_page_writeback(pp);
4576 zfs_range_unlock(rl);
4582 ASSERT3U(pglen, <=, PAGE_SIZE);
4583 dmu_write(zfsvfs->z_os, zp->z_id, pgoff, pglen, va, tx);
4586 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
4587 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
4588 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zfsvfs), NULL,
4591 /* Preserve the mtime and ctime provided by the inode */
4592 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4593 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4594 zp->z_atime_dirty = 0;
4597 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4599 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0,
4600 zfs_putpage_commit_cb, pp);
4603 zfs_range_unlock(rl);
4605 if (wbc->sync_mode != WB_SYNC_NONE) {
4607 * Note that this is rarely called under writepages(), because
4608 * writepages() normally handles the entire commit for
4609 * performance reasons.
4611 zil_commit(zfsvfs->z_log, zp->z_id);
4619 * Update the system attributes when the inode has been dirtied. For the
4620 * moment we only update the mode, atime, mtime, and ctime.
4623 zfs_dirty_inode(struct inode *ip, int flags)
4625 znode_t *zp = ITOZ(ip);
4626 zfsvfs_t *zfsvfs = ITOZSB(ip);
4628 uint64_t mode, atime[2], mtime[2], ctime[2];
4629 sa_bulk_attr_t bulk[4];
4633 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
4641 * This is the lazytime semantic indroduced in Linux 4.0
4642 * This flag will only be called from update_time when lazytime is set.
4643 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4644 * Fortunately mtime and ctime are managed within ZFS itself, so we
4645 * only need to dirty atime.
4647 if (flags == I_DIRTY_TIME) {
4648 zp->z_atime_dirty = 1;
4653 tx = dmu_tx_create(zfsvfs->z_os);
4655 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4656 zfs_sa_upgrade_txholds(tx, zp);
4658 error = dmu_tx_assign(tx, TXG_WAIT);
4664 mutex_enter(&zp->z_lock);
4665 zp->z_atime_dirty = 0;
4667 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
4668 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
4669 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
4670 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
4672 /* Preserve the mode, mtime and ctime provided by the inode */
4673 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4674 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4675 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4680 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4681 mutex_exit(&zp->z_lock);
4691 zfs_inactive(struct inode *ip)
4693 znode_t *zp = ITOZ(ip);
4694 zfsvfs_t *zfsvfs = ITOZSB(ip);
4697 int need_unlock = 0;
4699 /* Only read lock if we haven't already write locked, e.g. rollback */
4700 if (!RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)) {
4702 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4704 if (zp->z_sa_hdl == NULL) {
4706 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4710 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4711 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4713 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4714 zfs_sa_upgrade_txholds(tx, zp);
4715 error = dmu_tx_assign(tx, TXG_WAIT);
4719 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4720 mutex_enter(&zp->z_lock);
4721 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4722 (void *)&atime, sizeof (atime), tx);
4723 zp->z_atime_dirty = 0;
4724 mutex_exit(&zp->z_lock);
4731 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4735 * Bounds-check the seek operation.
4737 * IN: ip - inode seeking within
4738 * ooff - old file offset
4739 * noffp - pointer to new file offset
4740 * ct - caller context
4742 * RETURN: 0 if success
4743 * EINVAL if new offset invalid
4747 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
4749 if (S_ISDIR(ip->i_mode))
4751 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4755 * Fill pages with data from the disk.
4758 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
4760 znode_t *zp = ITOZ(ip);
4761 zfsvfs_t *zfsvfs = ITOZSB(ip);
4763 struct page *cur_pp;
4764 u_offset_t io_off, total;
4771 io_len = nr_pages << PAGE_SHIFT;
4772 i_size = i_size_read(ip);
4773 io_off = page_offset(pl[0]);
4775 if (io_off + io_len > i_size)
4776 io_len = i_size - io_off;
4779 * Iterate over list of pages and read each page individually.
4782 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4785 cur_pp = pl[page_idx++];
4787 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4791 /* convert checksum errors into IO errors */
4793 err = SET_ERROR(EIO);
4802 * Uses zfs_fillpage to read data from the file and fill the pages.
4804 * IN: ip - inode of file to get data from.
4805 * pl - list of pages to read
4806 * nr_pages - number of pages to read
4808 * RETURN: 0 on success, error code on failure.
4811 * vp - atime updated
4815 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
4817 znode_t *zp = ITOZ(ip);
4818 zfsvfs_t *zfsvfs = ITOZSB(ip);
4827 err = zfs_fillpage(ip, pl, nr_pages);
4834 * Check ZFS specific permissions to memory map a section of a file.
4836 * IN: ip - inode of the file to mmap
4838 * addrp - start address in memory region
4839 * len - length of memory region
4840 * vm_flags- address flags
4842 * RETURN: 0 if success
4843 * error code if failure
4847 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4848 unsigned long vm_flags)
4850 znode_t *zp = ITOZ(ip);
4851 zfsvfs_t *zfsvfs = ITOZSB(ip);
4856 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4857 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4859 return (SET_ERROR(EPERM));
4862 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4863 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4865 return (SET_ERROR(EACCES));
4868 if (off < 0 || len > MAXOFFSET_T - off) {
4870 return (SET_ERROR(ENXIO));
4878 * convoff - converts the given data (start, whence) to the
4882 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4887 if ((lckdat->l_whence == 2) || (whence == 2)) {
4888 if ((error = zfs_getattr(ip, &vap, 0, CRED())))
4892 switch (lckdat->l_whence) {
4894 lckdat->l_start += offset;
4897 lckdat->l_start += vap.va_size;
4902 return (SET_ERROR(EINVAL));
4905 if (lckdat->l_start < 0)
4906 return (SET_ERROR(EINVAL));
4910 lckdat->l_start -= offset;
4913 lckdat->l_start -= vap.va_size;
4918 return (SET_ERROR(EINVAL));
4921 lckdat->l_whence = (short)whence;
4926 * Free or allocate space in a file. Currently, this function only
4927 * supports the `F_FREESP' command. However, this command is somewhat
4928 * misnamed, as its functionality includes the ability to allocate as
4929 * well as free space.
4931 * IN: ip - inode of file to free data in.
4932 * cmd - action to take (only F_FREESP supported).
4933 * bfp - section of file to free/alloc.
4934 * flag - current file open mode flags.
4935 * offset - current file offset.
4936 * cr - credentials of caller [UNUSED].
4938 * RETURN: 0 on success, error code on failure.
4941 * ip - ctime|mtime updated
4945 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4946 offset_t offset, cred_t *cr)
4948 znode_t *zp = ITOZ(ip);
4949 zfsvfs_t *zfsvfs = ITOZSB(ip);
4956 if (cmd != F_FREESP) {
4958 return (SET_ERROR(EINVAL));
4962 * Callers might not be able to detect properly that we are read-only,
4963 * so check it explicitly here.
4965 if (zfs_is_readonly(zfsvfs)) {
4967 return (SET_ERROR(EROFS));
4970 if ((error = convoff(ip, bfp, 0, offset))) {
4975 if (bfp->l_len < 0) {
4977 return (SET_ERROR(EINVAL));
4981 * Permissions aren't checked on Solaris because on this OS
4982 * zfs_space() can only be called with an opened file handle.
4983 * On Linux we can get here through truncate_range() which
4984 * operates directly on inodes, so we need to check access rights.
4986 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4992 len = bfp->l_len; /* 0 means from off to end of file */
4994 error = zfs_freesp(zp, off, len, flag, TRUE);
5002 zfs_fid(struct inode *ip, fid_t *fidp)
5004 znode_t *zp = ITOZ(ip);
5005 zfsvfs_t *zfsvfs = ITOZSB(ip);
5008 uint64_t object = zp->z_id;
5015 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5016 &gen64, sizeof (uint64_t))) != 0) {
5021 gen = (uint32_t)gen64;
5023 size = SHORT_FID_LEN;
5025 zfid = (zfid_short_t *)fidp;
5027 zfid->zf_len = size;
5029 for (i = 0; i < sizeof (zfid->zf_object); i++)
5030 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5032 /* Must have a non-zero generation number to distinguish from .zfs */
5035 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5036 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5044 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
5046 znode_t *zp = ITOZ(ip);
5047 zfsvfs_t *zfsvfs = ITOZSB(ip);
5049 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5053 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5061 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
5063 znode_t *zp = ITOZ(ip);
5064 zfsvfs_t *zfsvfs = ITOZSB(ip);
5066 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5067 zilog_t *zilog = zfsvfs->z_log;
5072 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5074 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5075 zil_commit(zilog, 0);
5081 #ifdef HAVE_UIO_ZEROCOPY
5083 * Tunable, both must be a power of 2.
5085 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
5086 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
5087 * an arcbuf for a partial block read
5089 int zcr_blksz_min = (1 << 10); /* 1K */
5090 int zcr_blksz_max = (1 << 17); /* 128K */
5094 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
5096 znode_t *zp = ITOZ(ip);
5097 zfsvfs_t *zfsvfs = ITOZSB(ip);
5098 int max_blksz = zfsvfs->z_max_blksz;
5099 uio_t *uio = &xuio->xu_uio;
5100 ssize_t size = uio->uio_resid;
5101 offset_t offset = uio->uio_loffset;
5106 int preamble, postamble;
5108 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5109 return (SET_ERROR(EINVAL));
5116 * Loan out an arc_buf for write if write size is bigger than
5117 * max_blksz, and the file's block size is also max_blksz.
5120 if (size < blksz || zp->z_blksz != blksz) {
5122 return (SET_ERROR(EINVAL));
5125 * Caller requests buffers for write before knowing where the
5126 * write offset might be (e.g. NFS TCP write).
5131 preamble = P2PHASE(offset, blksz);
5133 preamble = blksz - preamble;
5138 postamble = P2PHASE(size, blksz);
5141 fullblk = size / blksz;
5142 (void) dmu_xuio_init(xuio,
5143 (preamble != 0) + fullblk + (postamble != 0));
5146 * Have to fix iov base/len for partial buffers. They
5147 * currently represent full arc_buf's.
5150 /* data begins in the middle of the arc_buf */
5151 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5154 (void) dmu_xuio_add(xuio, abuf,
5155 blksz - preamble, preamble);
5158 for (i = 0; i < fullblk; i++) {
5159 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5162 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5166 /* data ends in the middle of the arc_buf */
5167 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5170 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5175 * Loan out an arc_buf for read if the read size is larger than
5176 * the current file block size. Block alignment is not
5177 * considered. Partial arc_buf will be loaned out for read.
5179 blksz = zp->z_blksz;
5180 if (blksz < zcr_blksz_min)
5181 blksz = zcr_blksz_min;
5182 if (blksz > zcr_blksz_max)
5183 blksz = zcr_blksz_max;
5184 /* avoid potential complexity of dealing with it */
5185 if (blksz > max_blksz) {
5187 return (SET_ERROR(EINVAL));
5190 maxsize = zp->z_size - uio->uio_loffset;
5196 return (SET_ERROR(EINVAL));
5201 return (SET_ERROR(EINVAL));
5204 uio->uio_extflg = UIO_XUIO;
5205 XUIO_XUZC_RW(xuio) = ioflag;
5212 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
5216 int ioflag = XUIO_XUZC_RW(xuio);
5218 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5220 i = dmu_xuio_cnt(xuio);
5222 abuf = dmu_xuio_arcbuf(xuio, i);
5224 * if abuf == NULL, it must be a write buffer
5225 * that has been returned in zfs_write().
5228 dmu_return_arcbuf(abuf);
5229 ASSERT(abuf || ioflag == UIO_WRITE);
5232 dmu_xuio_fini(xuio);
5235 #endif /* HAVE_UIO_ZEROCOPY */
5237 #if defined(_KERNEL) && defined(HAVE_SPL)
5238 EXPORT_SYMBOL(zfs_open);
5239 EXPORT_SYMBOL(zfs_close);
5240 EXPORT_SYMBOL(zfs_read);
5241 EXPORT_SYMBOL(zfs_write);
5242 EXPORT_SYMBOL(zfs_access);
5243 EXPORT_SYMBOL(zfs_lookup);
5244 EXPORT_SYMBOL(zfs_create);
5245 EXPORT_SYMBOL(zfs_tmpfile);
5246 EXPORT_SYMBOL(zfs_remove);
5247 EXPORT_SYMBOL(zfs_mkdir);
5248 EXPORT_SYMBOL(zfs_rmdir);
5249 EXPORT_SYMBOL(zfs_readdir);
5250 EXPORT_SYMBOL(zfs_fsync);
5251 EXPORT_SYMBOL(zfs_getattr);
5252 EXPORT_SYMBOL(zfs_getattr_fast);
5253 EXPORT_SYMBOL(zfs_setattr);
5254 EXPORT_SYMBOL(zfs_rename);
5255 EXPORT_SYMBOL(zfs_symlink);
5256 EXPORT_SYMBOL(zfs_readlink);
5257 EXPORT_SYMBOL(zfs_link);
5258 EXPORT_SYMBOL(zfs_inactive);
5259 EXPORT_SYMBOL(zfs_space);
5260 EXPORT_SYMBOL(zfs_fid);
5261 EXPORT_SYMBOL(zfs_getsecattr);
5262 EXPORT_SYMBOL(zfs_setsecattr);
5263 EXPORT_SYMBOL(zfs_getpage);
5264 EXPORT_SYMBOL(zfs_putpage);
5265 EXPORT_SYMBOL(zfs_dirty_inode);
5266 EXPORT_SYMBOL(zfs_map);
5269 module_param(zfs_delete_blocks, ulong, 0644);
5270 MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async");
5271 module_param(zfs_read_chunk_size, long, 0644);
5272 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");