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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
26 /* Portions Copyright 2007 Jeremy Teo */
27 /* Portions Copyright 2010 Robert Milkowski */
30 #include <sys/types.h>
31 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
37 #include <sys/vfs_opreg.h>
41 #include <sys/taskq.h>
43 #include <sys/vmsystm.h>
44 #include <sys/atomic.h>
46 #include <sys/pathname.h>
47 #include <sys/cmn_err.h>
48 #include <sys/errno.h>
49 #include <sys/unistd.h>
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_acl.h>
52 #include <sys/zfs_ioctl.h>
53 #include <sys/fs/zfs.h>
55 #include <sys/dmu_objset.h>
61 #include <sys/dirent.h>
62 #include <sys/policy.h>
63 #include <sys/sunddi.h>
66 #include "fs/fs_subr.h"
67 #include <sys/zfs_ctldir.h>
68 #include <sys/zfs_fuid.h>
69 #include <sys/zfs_sa.h>
70 #include <sys/zfs_vnops.h>
72 #include <sys/zfs_rlock.h>
73 #include <sys/extdirent.h>
74 #include <sys/kidmap.h>
82 * Each vnode op performs some logical unit of work. To do this, the ZPL must
83 * properly lock its in-core state, create a DMU transaction, do the work,
84 * record this work in the intent log (ZIL), commit the DMU transaction,
85 * and wait for the intent log to commit if it is a synchronous operation.
86 * Moreover, the vnode ops must work in both normal and log replay context.
87 * The ordering of events is important to avoid deadlocks and references
88 * to freed memory. The example below illustrates the following Big Rules:
90 * (1) A check must be made in each zfs thread for a mounted file system.
91 * This is done avoiding races using ZFS_ENTER(zsb).
92 * A ZFS_EXIT(zsb) is needed before all returns. Any znodes
93 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
94 * can return EIO from the calling function.
96 * (2) iput() should always be the last thing except for zil_commit()
97 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
98 * First, if it's the last reference, the vnode/znode
99 * can be freed, so the zp may point to freed memory. Second, the last
100 * reference will call zfs_zinactive(), which may induce a lot of work --
101 * pushing cached pages (which acquires range locks) and syncing out
102 * cached atime changes. Third, zfs_zinactive() may require a new tx,
103 * which could deadlock the system if you were already holding one.
104 * If you must call iput() within a tx then use iput_ASYNC().
106 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
107 * as they can span dmu_tx_assign() calls.
109 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
110 * This is critical because we don't want to block while holding locks.
111 * Note, in particular, that if a lock is sometimes acquired before
112 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
113 * use a non-blocking assign can deadlock the system. The scenario:
115 * Thread A has grabbed a lock before calling dmu_tx_assign().
116 * Thread B is in an already-assigned tx, and blocks for this lock.
117 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
118 * forever, because the previous txg can't quiesce until B's tx commits.
120 * If dmu_tx_assign() returns ERESTART and zsb->z_assign is TXG_NOWAIT,
121 * then drop all locks, call dmu_tx_wait(), and try again.
123 * (5) If the operation succeeded, generate the intent log entry for it
124 * before dropping locks. This ensures that the ordering of events
125 * in the intent log matches the order in which they actually occurred.
126 * During ZIL replay the zfs_log_* functions will update the sequence
127 * number to indicate the zil transaction has replayed.
129 * (6) At the end of each vnode op, the DMU tx must always commit,
130 * regardless of whether there were any errors.
132 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
133 * to ensure that synchronous semantics are provided when necessary.
135 * In general, this is how things should be ordered in each vnode op:
137 * ZFS_ENTER(zsb); // exit if unmounted
139 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
140 * rw_enter(...); // grab any other locks you need
141 * tx = dmu_tx_create(...); // get DMU tx
142 * dmu_tx_hold_*(); // hold each object you might modify
143 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign
145 * rw_exit(...); // drop locks
146 * zfs_dirent_unlock(dl); // unlock directory entry
147 * iput(...); // release held vnodes
148 * if (error == ERESTART) {
153 * dmu_tx_abort(tx); // abort DMU tx
154 * ZFS_EXIT(zsb); // finished in zfs
155 * return (error); // really out of space
157 * error = do_real_work(); // do whatever this VOP does
159 * zfs_log_*(...); // on success, make ZIL entry
160 * dmu_tx_commit(tx); // commit DMU tx -- error or not
161 * rw_exit(...); // drop locks
162 * zfs_dirent_unlock(dl); // unlock directory entry
163 * iput(...); // release held vnodes
164 * zil_commit(zilog, foid); // synchronous when necessary
165 * ZFS_EXIT(zsb); // finished in zfs
166 * return (error); // done, report error
170 * Virus scanning is unsupported. It would be possible to add a hook
171 * here to performance the required virus scan. This could be done
172 * entirely in the kernel or potentially as an update to invoke a
176 zfs_vscan(struct inode *ip, cred_t *cr, int async)
183 zfs_open(struct inode *ip, int mode, int flag, cred_t *cr)
185 znode_t *zp = ITOZ(ip);
186 zfs_sb_t *zsb = ITOZSB(ip);
191 /* Honor ZFS_APPENDONLY file attribute */
192 if ((mode & FMODE_WRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
193 ((flag & O_APPEND) == 0)) {
198 /* Virus scan eligible files on open */
199 if (!zfs_has_ctldir(zp) && zsb->z_vscan && S_ISREG(ip->i_mode) &&
200 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
201 if (zfs_vscan(ip, cr, 0) != 0) {
207 /* Keep a count of the synchronous opens in the znode */
209 atomic_inc_32(&zp->z_sync_cnt);
214 EXPORT_SYMBOL(zfs_open);
218 zfs_close(struct inode *ip, int flag, cred_t *cr)
220 znode_t *zp = ITOZ(ip);
221 zfs_sb_t *zsb = ITOZSB(ip);
227 * Zero the synchronous opens in the znode. Under Linux the
228 * zfs_close() hook is not symmetric with zfs_open(), it is
229 * only called once when the last reference is dropped.
234 if (!zfs_has_ctldir(zp) && zsb->z_vscan && S_ISREG(ip->i_mode) &&
235 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
236 VERIFY(zfs_vscan(ip, cr, 1) == 0);
241 EXPORT_SYMBOL(zfs_close);
243 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
245 * Lseek support for finding holes (cmd == SEEK_HOLE) and
246 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
249 zfs_holey_common(struct inode *ip, int cmd, loff_t *off)
251 znode_t *zp = ITOZ(ip);
252 uint64_t noff = (uint64_t)*off; /* new offset */
257 file_sz = zp->z_size;
258 if (noff >= file_sz) {
262 if (cmd == SEEK_HOLE)
267 error = dmu_offset_next(ZTOZSB(zp)->z_os, zp->z_id, hole, &noff);
270 if ((error == ESRCH) || (noff > file_sz)) {
272 * Handle the virtual hole at the end of file.
288 zfs_holey(struct inode *ip, int cmd, loff_t *off)
290 znode_t *zp = ITOZ(ip);
291 zfs_sb_t *zsb = ITOZSB(ip);
297 error = zfs_holey_common(ip, cmd, off);
302 EXPORT_SYMBOL(zfs_holey);
303 #endif /* SEEK_HOLE && SEEK_DATA */
307 * When a file is memory mapped, we must keep the IO data synchronized
308 * between the DMU cache and the memory mapped pages. What this means:
310 * On Write: If we find a memory mapped page, we write to *both*
311 * the page and the dmu buffer.
314 update_pages(struct inode *ip, int64_t start, int len,
315 objset_t *os, uint64_t oid)
317 struct address_space *mp = ip->i_mapping;
323 off = start & (PAGE_CACHE_SIZE-1);
324 for (start &= PAGE_CACHE_MASK; len > 0; start += PAGE_CACHE_SIZE) {
325 nbytes = MIN(PAGE_CACHE_SIZE - off, len);
327 pp = find_lock_page(mp, start >> PAGE_CACHE_SHIFT);
329 if (mapping_writably_mapped(mp))
330 flush_dcache_page(pp);
333 (void) dmu_read(os, oid, start+off, nbytes, pb+off,
337 if (mapping_writably_mapped(mp))
338 flush_dcache_page(pp);
340 mark_page_accessed(pp);
344 page_cache_release(pp);
353 * When a file is memory mapped, we must keep the IO data synchronized
354 * between the DMU cache and the memory mapped pages. What this means:
356 * On Read: We "read" preferentially from memory mapped pages,
357 * else we default from the dmu buffer.
359 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
360 * the file is memory mapped.
363 mappedread(struct inode *ip, int nbytes, uio_t *uio)
365 struct address_space *mp = ip->i_mapping;
367 znode_t *zp = ITOZ(ip);
368 objset_t *os = ITOZSB(ip)->z_os;
375 start = uio->uio_loffset;
376 off = start & (PAGE_CACHE_SIZE-1);
377 for (start &= PAGE_CACHE_MASK; len > 0; start += PAGE_CACHE_SIZE) {
378 bytes = MIN(PAGE_CACHE_SIZE - off, len);
380 pp = find_lock_page(mp, start >> PAGE_CACHE_SHIFT);
382 ASSERT(PageUptodate(pp));
385 error = uiomove(pb + off, bytes, UIO_READ, uio);
388 if (mapping_writably_mapped(mp))
389 flush_dcache_page(pp);
391 mark_page_accessed(pp);
393 page_cache_release(pp);
395 error = dmu_read_uio(os, zp->z_id, uio, bytes);
407 unsigned long zfs_read_chunk_size = 1024 * 1024; /* Tunable */
410 * Read bytes from specified file into supplied buffer.
412 * IN: ip - inode of file to be read from.
413 * uio - structure supplying read location, range info,
415 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
416 * O_DIRECT flag; used to bypass page cache.
417 * cr - credentials of caller.
419 * OUT: uio - updated offset and range, buffer filled.
421 * RETURN: 0 if success
422 * error code if failure
425 * inode - atime updated if byte count > 0
429 zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
431 znode_t *zp = ITOZ(ip);
432 zfs_sb_t *zsb = ITOZSB(ip);
437 #ifdef HAVE_UIO_ZEROCOPY
439 #endif /* HAVE_UIO_ZEROCOPY */
445 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
451 * Validate file offset
453 if (uio->uio_loffset < (offset_t)0) {
459 * Fasttrack empty reads
461 if (uio->uio_resid == 0) {
467 * Check for mandatory locks
469 if (mandatory_lock(ip) &&
470 !lock_may_read(ip, uio->uio_loffset, uio->uio_resid)) {
476 * If we're in FRSYNC mode, sync out this znode before reading it.
478 if (ioflag & FRSYNC || zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
479 zil_commit(zsb->z_log, zp->z_id);
482 * Lock the range against changes.
484 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
487 * If we are reading past end-of-file we can skip
488 * to the end; but we might still need to set atime.
490 if (uio->uio_loffset >= zp->z_size) {
495 ASSERT(uio->uio_loffset < zp->z_size);
496 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
498 #ifdef HAVE_UIO_ZEROCOPY
499 if ((uio->uio_extflg == UIO_XUIO) &&
500 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
502 int blksz = zp->z_blksz;
503 uint64_t offset = uio->uio_loffset;
505 xuio = (xuio_t *)uio;
507 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
510 ASSERT(offset + n <= blksz);
513 (void) dmu_xuio_init(xuio, nblk);
515 if (vn_has_cached_data(ip)) {
517 * For simplicity, we always allocate a full buffer
518 * even if we only expect to read a portion of a block.
520 while (--nblk >= 0) {
521 (void) dmu_xuio_add(xuio,
522 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
527 #endif /* HAVE_UIO_ZEROCOPY */
530 nbytes = MIN(n, zfs_read_chunk_size -
531 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
533 if (zp->z_is_mapped && !(ioflag & O_DIRECT))
534 error = mappedread(ip, nbytes, uio);
536 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
539 /* convert checksum errors into IO errors */
548 zfs_range_unlock(rl);
550 ZFS_ACCESSTIME_STAMP(zsb, zp);
551 zfs_inode_update(zp);
555 EXPORT_SYMBOL(zfs_read);
558 * Write the bytes to a file.
560 * IN: ip - inode of file to be written to.
561 * uio - structure supplying write location, range info,
563 * ioflag - FAPPEND flag set if in append mode.
564 * O_DIRECT flag; used to bypass page cache.
565 * cr - credentials of caller.
567 * OUT: uio - updated offset and range.
569 * RETURN: 0 if success
570 * error code if failure
573 * ip - ctime|mtime updated if byte count > 0
578 zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
580 znode_t *zp = ITOZ(ip);
581 rlim64_t limit = uio->uio_limit;
582 ssize_t start_resid = uio->uio_resid;
586 zfs_sb_t *zsb = ZTOZSB(zp);
591 int max_blksz = zsb->z_max_blksz;
594 iovec_t *aiov = NULL;
597 iovec_t *iovp = uio->uio_iov;
600 sa_bulk_attr_t bulk[4];
601 uint64_t mtime[2], ctime[2];
602 ASSERTV(int iovcnt = uio->uio_iovcnt);
605 * Fasttrack empty write
611 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
617 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
618 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
619 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb), NULL, &zp->z_size, 8);
620 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
624 * If immutable or not appending then return EPERM
626 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
627 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
628 (uio->uio_loffset < zp->z_size))) {
636 * Validate file offset
638 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
645 * Check for mandatory locks before calling zfs_range_lock()
646 * in order to prevent a deadlock with locks set via fcntl().
648 if (mandatory_lock(ip) && !lock_may_write(ip, woff, n)) {
654 * Pre-fault the pages to ensure slow (eg NFS) pages
656 * Skip this if uio contains loaned arc_buf.
658 #ifdef HAVE_UIO_ZEROCOPY
659 if ((uio->uio_extflg == UIO_XUIO) &&
660 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
661 xuio = (xuio_t *)uio;
664 uio_prefaultpages(MIN(n, max_blksz), uio);
667 * If in append mode, set the io offset pointer to eof.
669 if (ioflag & FAPPEND) {
671 * Obtain an appending range lock to guarantee file append
672 * semantics. We reset the write offset once we have the lock.
674 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
676 if (rl->r_len == UINT64_MAX) {
678 * We overlocked the file because this write will cause
679 * the file block size to increase.
680 * Note that zp_size cannot change with this lock held.
684 uio->uio_loffset = woff;
687 * Note that if the file block size will change as a result of
688 * this write, then this range lock will lock the entire file
689 * so that we can re-write the block safely.
691 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
695 zfs_range_unlock(rl);
700 if ((woff + n) > limit || woff > (limit - n))
703 /* Will this write extend the file length? */
704 write_eof = (woff + n > zp->z_size);
706 end_size = MAX(zp->z_size, woff + n);
709 * Write the file in reasonable size chunks. Each chunk is written
710 * in a separate transaction; this keeps the intent log records small
711 * and allows us to do more fine-grained space accounting.
715 woff = uio->uio_loffset;
717 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
718 zfs_owner_overquota(zsb, zp, B_TRUE)) {
720 dmu_return_arcbuf(abuf);
725 if (xuio && abuf == NULL) {
726 ASSERT(i_iov < iovcnt);
728 abuf = dmu_xuio_arcbuf(xuio, i_iov);
729 dmu_xuio_clear(xuio, i_iov);
730 ASSERT((aiov->iov_base == abuf->b_data) ||
731 ((char *)aiov->iov_base - (char *)abuf->b_data +
732 aiov->iov_len == arc_buf_size(abuf)));
734 } else if (abuf == NULL && n >= max_blksz &&
735 woff >= zp->z_size &&
736 P2PHASE(woff, max_blksz) == 0 &&
737 zp->z_blksz == max_blksz) {
739 * This write covers a full block. "Borrow" a buffer
740 * from the dmu so that we can fill it before we enter
741 * a transaction. This avoids the possibility of
742 * holding up the transaction if the data copy hangs
743 * up on a pagefault (e.g., from an NFS server mapping).
747 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
749 ASSERT(abuf != NULL);
750 ASSERT(arc_buf_size(abuf) == max_blksz);
751 if ((error = uiocopy(abuf->b_data, max_blksz,
752 UIO_WRITE, uio, &cbytes))) {
753 dmu_return_arcbuf(abuf);
756 ASSERT(cbytes == max_blksz);
760 * Start a transaction.
762 tx = dmu_tx_create(zsb->z_os);
763 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
764 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
765 zfs_sa_upgrade_txholds(tx, zp);
766 error = dmu_tx_assign(tx, TXG_NOWAIT);
768 if (error == ERESTART) {
775 dmu_return_arcbuf(abuf);
780 * If zfs_range_lock() over-locked we grow the blocksize
781 * and then reduce the lock range. This will only happen
782 * on the first iteration since zfs_range_reduce() will
783 * shrink down r_len to the appropriate size.
785 if (rl->r_len == UINT64_MAX) {
788 if (zp->z_blksz > max_blksz) {
789 ASSERT(!ISP2(zp->z_blksz));
790 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
792 new_blksz = MIN(end_size, max_blksz);
794 zfs_grow_blocksize(zp, new_blksz, tx);
795 zfs_range_reduce(rl, woff, n);
799 * XXX - should we really limit each write to z_max_blksz?
800 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
802 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
805 tx_bytes = uio->uio_resid;
806 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
808 tx_bytes -= uio->uio_resid;
811 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
813 * If this is not a full block write, but we are
814 * extending the file past EOF and this data starts
815 * block-aligned, use assign_arcbuf(). Otherwise,
816 * write via dmu_write().
818 if (tx_bytes < max_blksz && (!write_eof ||
819 aiov->iov_base != abuf->b_data)) {
821 dmu_write(zsb->z_os, zp->z_id, woff,
822 aiov->iov_len, aiov->iov_base, tx);
823 dmu_return_arcbuf(abuf);
824 xuio_stat_wbuf_copied();
826 ASSERT(xuio || tx_bytes == max_blksz);
827 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
830 ASSERT(tx_bytes <= uio->uio_resid);
831 uioskip(uio, tx_bytes);
834 if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT))
835 update_pages(ip, woff, tx_bytes, zsb->z_os, zp->z_id);
838 * If we made no progress, we're done. If we made even
839 * partial progress, update the znode and ZIL accordingly.
842 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
843 (void *)&zp->z_size, sizeof (uint64_t), tx);
850 * Clear Set-UID/Set-GID bits on successful write if not
851 * privileged and at least one of the excute bits is set.
853 * It would be nice to to this after all writes have
854 * been done, but that would still expose the ISUID/ISGID
855 * to another app after the partial write is committed.
857 * Note: we don't call zfs_fuid_map_id() here because
858 * user 0 is not an ephemeral uid.
860 mutex_enter(&zp->z_acl_lock);
861 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
862 (S_IXUSR >> 6))) != 0 &&
863 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
864 secpolicy_vnode_setid_retain(cr,
865 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
867 zp->z_mode &= ~(S_ISUID | S_ISGID);
868 newmode = zp->z_mode;
869 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zsb),
870 (void *)&newmode, sizeof (uint64_t), tx);
872 mutex_exit(&zp->z_acl_lock);
874 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
878 * Update the file size (zp_size) if it has changed;
879 * account for possible concurrent updates.
881 while ((end_size = zp->z_size) < uio->uio_loffset) {
882 (void) atomic_cas_64(&zp->z_size, end_size,
887 * If we are replaying and eof is non zero then force
888 * the file size to the specified eof. Note, there's no
889 * concurrency during replay.
891 if (zsb->z_replay && zsb->z_replay_eof != 0)
892 zp->z_size = zsb->z_replay_eof;
894 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
896 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
901 ASSERT(tx_bytes == nbytes);
905 uio_prefaultpages(MIN(n, max_blksz), uio);
908 zfs_range_unlock(rl);
911 * If we're in replay mode, or we made no progress, return error.
912 * Otherwise, it's at least a partial write, so it's successful.
914 if (zsb->z_replay || uio->uio_resid == start_resid) {
919 if (ioflag & (FSYNC | FDSYNC) ||
920 zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
921 zil_commit(zilog, zp->z_id);
923 zfs_inode_update(zp);
927 EXPORT_SYMBOL(zfs_write);
930 iput_async(struct inode *ip, taskq_t *taskq)
932 ASSERT(atomic_read(&ip->i_count) > 0);
933 if (atomic_read(&ip->i_count) == 1)
934 taskq_dispatch(taskq, (task_func_t *)iput, ip, TQ_PUSHPAGE);
940 zfs_get_done(zgd_t *zgd, int error)
942 znode_t *zp = zgd->zgd_private;
943 objset_t *os = ZTOZSB(zp)->z_os;
946 dmu_buf_rele(zgd->zgd_db, zgd);
948 zfs_range_unlock(zgd->zgd_rl);
951 * Release the vnode asynchronously as we currently have the
952 * txg stopped from syncing.
954 iput_async(ZTOI(zp), dsl_pool_iput_taskq(dmu_objset_pool(os)));
956 if (error == 0 && zgd->zgd_bp)
957 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
959 kmem_free(zgd, sizeof (zgd_t));
963 static int zil_fault_io = 0;
967 * Get data to generate a TX_WRITE intent log record.
970 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
973 objset_t *os = zsb->z_os;
975 uint64_t object = lr->lr_foid;
976 uint64_t offset = lr->lr_offset;
977 uint64_t size = lr->lr_length;
978 blkptr_t *bp = &lr->lr_blkptr;
987 * Nothing to do if the file has been removed
989 if (zfs_zget(zsb, object, &zp) != 0)
991 if (zp->z_unlinked) {
993 * Release the vnode asynchronously as we currently have the
994 * txg stopped from syncing.
996 iput_async(ZTOI(zp), dsl_pool_iput_taskq(dmu_objset_pool(os)));
1000 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_PUSHPAGE);
1001 zgd->zgd_zilog = zsb->z_log;
1002 zgd->zgd_private = zp;
1005 * Write records come in two flavors: immediate and indirect.
1006 * For small writes it's cheaper to store the data with the
1007 * log record (immediate); for large writes it's cheaper to
1008 * sync the data and get a pointer to it (indirect) so that
1009 * we don't have to write the data twice.
1011 if (buf != NULL) { /* immediate write */
1012 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1013 /* test for truncation needs to be done while range locked */
1014 if (offset >= zp->z_size) {
1017 error = dmu_read(os, object, offset, size, buf,
1018 DMU_READ_NO_PREFETCH);
1020 ASSERT(error == 0 || error == ENOENT);
1021 } else { /* indirect write */
1023 * Have to lock the whole block to ensure when it's
1024 * written out and it's checksum is being calculated
1025 * that no one can change the data. We need to re-check
1026 * blocksize after we get the lock in case it's changed!
1031 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1033 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1035 if (zp->z_blksz == size)
1038 zfs_range_unlock(zgd->zgd_rl);
1040 /* test for truncation needs to be done while range locked */
1041 if (lr->lr_offset >= zp->z_size)
1050 error = dmu_buf_hold(os, object, offset, zgd, &db,
1051 DMU_READ_NO_PREFETCH);
1057 ASSERT(db->db_offset == offset);
1058 ASSERT(db->db_size == size);
1060 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1062 ASSERT(error || lr->lr_length <= zp->z_blksz);
1065 * On success, we need to wait for the write I/O
1066 * initiated by dmu_sync() to complete before we can
1067 * release this dbuf. We will finish everything up
1068 * in the zfs_get_done() callback.
1073 if (error == EALREADY) {
1074 lr->lr_common.lrc_txtype = TX_WRITE2;
1080 zfs_get_done(zgd, error);
1087 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
1089 znode_t *zp = ITOZ(ip);
1090 zfs_sb_t *zsb = ITOZSB(ip);
1096 if (flag & V_ACE_MASK)
1097 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1099 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1104 EXPORT_SYMBOL(zfs_access);
1107 * Lookup an entry in a directory, or an extended attribute directory.
1108 * If it exists, return a held inode reference for it.
1110 * IN: dip - inode of directory to search.
1111 * nm - name of entry to lookup.
1112 * flags - LOOKUP_XATTR set if looking for an attribute.
1113 * cr - credentials of caller.
1114 * direntflags - directory lookup flags
1115 * realpnp - returned pathname.
1117 * OUT: ipp - inode of located entry, NULL if not found.
1119 * RETURN: 0 if success
1120 * error code if failure
1127 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
1128 cred_t *cr, int *direntflags, pathname_t *realpnp)
1130 znode_t *zdp = ITOZ(dip);
1131 zfs_sb_t *zsb = ITOZSB(dip);
1135 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1137 if (!S_ISDIR(dip->i_mode)) {
1139 } else if (zdp->z_sa_hdl == NULL) {
1143 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1144 error = zfs_fastaccesschk_execute(zdp, cr);
1153 vnode_t *tvp = dnlc_lookup(dvp, nm);
1156 error = zfs_fastaccesschk_execute(zdp, cr);
1161 if (tvp == DNLC_NO_VNODE) {
1166 return (specvp_check(vpp, cr));
1169 #endif /* HAVE_DNLC */
1178 if (flags & LOOKUP_XATTR) {
1180 * We don't allow recursive attributes..
1181 * Maybe someday we will.
1183 if (zdp->z_pflags & ZFS_XATTR) {
1188 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1194 * Do we have permission to get into attribute directory?
1197 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1207 if (!S_ISDIR(dip->i_mode)) {
1213 * Check accessibility of directory.
1216 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1221 if (zsb->z_utf8 && u8_validate(nm, strlen(nm),
1222 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1227 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1228 if ((error == 0) && (*ipp))
1229 zfs_inode_update(ITOZ(*ipp));
1234 EXPORT_SYMBOL(zfs_lookup);
1237 * Attempt to create a new entry in a directory. If the entry
1238 * already exists, truncate the file if permissible, else return
1239 * an error. Return the ip of the created or trunc'd file.
1241 * IN: dip - inode of directory to put new file entry in.
1242 * name - name of new file entry.
1243 * vap - attributes of new file.
1244 * excl - flag indicating exclusive or non-exclusive mode.
1245 * mode - mode to open file with.
1246 * cr - credentials of caller.
1247 * flag - large file flag [UNUSED].
1248 * vsecp - ACL to be set
1250 * OUT: ipp - inode of created or trunc'd entry.
1252 * RETURN: 0 if success
1253 * error code if failure
1256 * dip - ctime|mtime updated if new entry created
1257 * ip - ctime|mtime always, atime if new
1262 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1263 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1265 znode_t *zp, *dzp = ITOZ(dip);
1266 zfs_sb_t *zsb = ITOZSB(dip);
1274 zfs_acl_ids_t acl_ids;
1275 boolean_t fuid_dirtied;
1276 boolean_t have_acl = B_FALSE;
1279 * If we have an ephemeral id, ACL, or XVATTR then
1280 * make sure file system is at proper version
1286 if (zsb->z_use_fuids == B_FALSE &&
1287 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1295 if (zsb->z_utf8 && u8_validate(name, strlen(name),
1296 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1301 if (vap->va_mask & ATTR_XVATTR) {
1302 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1303 crgetuid(cr), cr, vap->va_mode)) != 0) {
1311 if (*name == '\0') {
1313 * Null component name refers to the directory itself.
1320 /* possible igrab(zp) */
1323 if (flag & FIGNORECASE)
1326 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1330 zfs_acl_ids_free(&acl_ids);
1331 if (strcmp(name, "..") == 0)
1342 * Create a new file object and update the directory
1345 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1347 zfs_acl_ids_free(&acl_ids);
1352 * We only support the creation of regular files in
1353 * extended attribute directories.
1356 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1358 zfs_acl_ids_free(&acl_ids);
1363 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1364 cr, vsecp, &acl_ids)) != 0)
1368 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1369 zfs_acl_ids_free(&acl_ids);
1374 tx = dmu_tx_create(os);
1376 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1377 ZFS_SA_BASE_ATTR_SIZE);
1379 fuid_dirtied = zsb->z_fuid_dirty;
1381 zfs_fuid_txhold(zsb, tx);
1382 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1383 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1384 if (!zsb->z_use_sa &&
1385 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1386 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1387 0, acl_ids.z_aclp->z_acl_bytes);
1389 error = dmu_tx_assign(tx, TXG_NOWAIT);
1391 zfs_dirent_unlock(dl);
1392 if (error == ERESTART) {
1397 zfs_acl_ids_free(&acl_ids);
1402 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1405 zfs_fuid_sync(zsb, tx);
1407 (void) zfs_link_create(dl, zp, tx, ZNEW);
1408 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1409 if (flag & FIGNORECASE)
1411 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1412 vsecp, acl_ids.z_fuidp, vap);
1413 zfs_acl_ids_free(&acl_ids);
1416 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1419 zfs_acl_ids_free(&acl_ids);
1423 * A directory entry already exists for this name.
1426 * Can't truncate an existing file if in exclusive mode.
1433 * Can't open a directory for writing.
1435 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1440 * Verify requested access to file.
1442 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1446 mutex_enter(&dzp->z_lock);
1448 mutex_exit(&dzp->z_lock);
1451 * Truncate regular files if requested.
1453 if (S_ISREG(ZTOI(zp)->i_mode) &&
1454 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1455 /* we can't hold any locks when calling zfs_freesp() */
1456 zfs_dirent_unlock(dl);
1458 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1464 zfs_dirent_unlock(dl);
1470 zfs_inode_update(dzp);
1471 zfs_inode_update(zp);
1475 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1476 zil_commit(zilog, 0);
1481 EXPORT_SYMBOL(zfs_create);
1484 * Remove an entry from a directory.
1486 * IN: dip - inode of directory to remove entry from.
1487 * name - name of entry to remove.
1488 * cr - credentials of caller.
1490 * RETURN: 0 if success
1491 * error code if failure
1495 * ip - ctime (if nlink > 0)
1498 uint64_t null_xattr = 0;
1502 zfs_remove(struct inode *dip, char *name, cred_t *cr)
1504 znode_t *zp, *dzp = ITOZ(dip);
1507 zfs_sb_t *zsb = ITOZSB(dip);
1510 uint64_t xattr_obj_unlinked = 0;
1516 pathname_t *realnmp = NULL;
1517 #ifdef HAVE_PN_UTILS
1519 #endif /* HAVE_PN_UTILS */
1527 #ifdef HAVE_PN_UTILS
1528 if (flags & FIGNORECASE) {
1533 #endif /* HAVE_PN_UTILS */
1539 * Attempt to lock directory; fail if entry doesn't exist.
1541 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1543 #ifdef HAVE_PN_UTILS
1546 #endif /* HAVE_PN_UTILS */
1553 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1558 * Need to use rmdir for removing directories.
1560 if (S_ISDIR(ip->i_mode)) {
1567 dnlc_remove(dvp, realnmp->pn_buf);
1569 dnlc_remove(dvp, name);
1570 #endif /* HAVE_DNLC */
1573 * We never delete the znode and always place it in the unlinked
1574 * set. The dentry cache will always hold the last reference and
1575 * is responsible for safely freeing the znode.
1578 tx = dmu_tx_create(zsb->z_os);
1579 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1580 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1581 zfs_sa_upgrade_txholds(tx, zp);
1582 zfs_sa_upgrade_txholds(tx, dzp);
1584 /* are there any extended attributes? */
1585 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1586 &xattr_obj, sizeof (xattr_obj));
1587 if (error == 0 && xattr_obj) {
1588 error = zfs_zget(zsb, xattr_obj, &xzp);
1590 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1591 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1594 /* charge as an update -- would be nice not to charge at all */
1595 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
1597 error = dmu_tx_assign(tx, TXG_NOWAIT);
1599 zfs_dirent_unlock(dl);
1603 if (error == ERESTART) {
1608 #ifdef HAVE_PN_UTILS
1611 #endif /* HAVE_PN_UTILS */
1618 * Remove the directory entry.
1620 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1629 * Hold z_lock so that we can make sure that the ACL obj
1630 * hasn't changed. Could have been deleted due to
1633 mutex_enter(&zp->z_lock);
1634 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1635 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1636 mutex_exit(&zp->z_lock);
1637 zfs_unlinked_add(zp, tx);
1641 #ifdef HAVE_PN_UTILS
1642 if (flags & FIGNORECASE)
1644 #endif /* HAVE_PN_UTILS */
1645 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1649 #ifdef HAVE_PN_UTILS
1652 #endif /* HAVE_PN_UTILS */
1654 zfs_dirent_unlock(dl);
1655 zfs_inode_update(dzp);
1656 zfs_inode_update(zp);
1658 zfs_inode_update(xzp);
1664 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1665 zil_commit(zilog, 0);
1670 EXPORT_SYMBOL(zfs_remove);
1673 * Create a new directory and insert it into dip using the name
1674 * provided. Return a pointer to the inserted directory.
1676 * IN: dip - inode of directory to add subdir to.
1677 * dirname - name of new directory.
1678 * vap - attributes of new directory.
1679 * cr - credentials of caller.
1680 * vsecp - ACL to be set
1682 * OUT: ipp - inode of created directory.
1684 * RETURN: 0 if success
1685 * error code if failure
1688 * dip - ctime|mtime updated
1689 * ipp - ctime|mtime|atime updated
1693 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1694 cred_t *cr, int flags, vsecattr_t *vsecp)
1696 znode_t *zp, *dzp = ITOZ(dip);
1697 zfs_sb_t *zsb = ITOZSB(dip);
1705 gid_t gid = crgetgid(cr);
1706 zfs_acl_ids_t acl_ids;
1707 boolean_t fuid_dirtied;
1709 ASSERT(S_ISDIR(vap->va_mode));
1712 * If we have an ephemeral id, ACL, or XVATTR then
1713 * make sure file system is at proper version
1717 if (zsb->z_use_fuids == B_FALSE &&
1718 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1725 if (dzp->z_pflags & ZFS_XATTR) {
1730 if (zsb->z_utf8 && u8_validate(dirname,
1731 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1735 if (flags & FIGNORECASE)
1738 if (vap->va_mask & ATTR_XVATTR) {
1739 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1740 crgetuid(cr), cr, vap->va_mode)) != 0) {
1746 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1747 vsecp, &acl_ids)) != 0) {
1752 * First make sure the new directory doesn't exist.
1754 * Existence is checked first to make sure we don't return
1755 * EACCES instead of EEXIST which can cause some applications
1761 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1763 zfs_acl_ids_free(&acl_ids);
1768 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
1769 zfs_acl_ids_free(&acl_ids);
1770 zfs_dirent_unlock(dl);
1775 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1776 zfs_acl_ids_free(&acl_ids);
1777 zfs_dirent_unlock(dl);
1783 * Add a new entry to the directory.
1785 tx = dmu_tx_create(zsb->z_os);
1786 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1787 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1788 fuid_dirtied = zsb->z_fuid_dirty;
1790 zfs_fuid_txhold(zsb, tx);
1791 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1792 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1793 acl_ids.z_aclp->z_acl_bytes);
1796 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1797 ZFS_SA_BASE_ATTR_SIZE);
1799 error = dmu_tx_assign(tx, TXG_NOWAIT);
1801 zfs_dirent_unlock(dl);
1802 if (error == ERESTART) {
1807 zfs_acl_ids_free(&acl_ids);
1816 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1819 zfs_fuid_sync(zsb, tx);
1822 * Now put new name in parent dir.
1824 (void) zfs_link_create(dl, zp, tx, ZNEW);
1828 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1829 if (flags & FIGNORECASE)
1831 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1832 acl_ids.z_fuidp, vap);
1834 zfs_acl_ids_free(&acl_ids);
1838 zfs_dirent_unlock(dl);
1840 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1841 zil_commit(zilog, 0);
1843 zfs_inode_update(dzp);
1844 zfs_inode_update(zp);
1848 EXPORT_SYMBOL(zfs_mkdir);
1851 * Remove a directory subdir entry. If the current working
1852 * directory is the same as the subdir to be removed, the
1855 * IN: dip - inode of directory to remove from.
1856 * name - name of directory to be removed.
1857 * cwd - inode of current working directory.
1858 * cr - credentials of caller.
1859 * flags - case flags
1861 * RETURN: 0 if success
1862 * error code if failure
1865 * dip - ctime|mtime updated
1869 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
1872 znode_t *dzp = ITOZ(dip);
1875 zfs_sb_t *zsb = ITOZSB(dip);
1886 if (flags & FIGNORECASE)
1892 * Attempt to lock directory; fail if entry doesn't exist.
1894 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1902 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1906 if (!S_ISDIR(ip->i_mode)) {
1917 * Grab a lock on the directory to make sure that noone is
1918 * trying to add (or lookup) entries while we are removing it.
1920 rw_enter(&zp->z_name_lock, RW_WRITER);
1923 * Grab a lock on the parent pointer to make sure we play well
1924 * with the treewalk and directory rename code.
1926 rw_enter(&zp->z_parent_lock, RW_WRITER);
1928 tx = dmu_tx_create(zsb->z_os);
1929 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1930 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1931 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
1932 zfs_sa_upgrade_txholds(tx, zp);
1933 zfs_sa_upgrade_txholds(tx, dzp);
1934 error = dmu_tx_assign(tx, TXG_NOWAIT);
1936 rw_exit(&zp->z_parent_lock);
1937 rw_exit(&zp->z_name_lock);
1938 zfs_dirent_unlock(dl);
1940 if (error == ERESTART) {
1950 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
1953 uint64_t txtype = TX_RMDIR;
1954 if (flags & FIGNORECASE)
1956 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
1961 rw_exit(&zp->z_parent_lock);
1962 rw_exit(&zp->z_name_lock);
1964 zfs_dirent_unlock(dl);
1966 zfs_inode_update(dzp);
1967 zfs_inode_update(zp);
1970 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1971 zil_commit(zilog, 0);
1976 EXPORT_SYMBOL(zfs_rmdir);
1979 * Read as many directory entries as will fit into the provided
1980 * dirent buffer from the given directory cursor position.
1982 * IN: ip - inode of directory to read.
1983 * dirent - buffer for directory entries.
1985 * OUT: dirent - filler buffer of directory entries.
1987 * RETURN: 0 if success
1988 * error code if failure
1991 * ip - atime updated
1993 * Note that the low 4 bits of the cookie returned by zap is always zero.
1994 * This allows us to use the low range for "special" directory entries:
1995 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
1996 * we use the offset 2 for the '.zfs' directory.
2000 zfs_readdir(struct inode *ip, struct dir_context *ctx, cred_t *cr)
2002 znode_t *zp = ITOZ(ip);
2003 zfs_sb_t *zsb = ITOZSB(ip);
2006 zap_attribute_t zap;
2012 uint64_t offset; /* must be unsigned; checks for < 1 */
2017 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zsb),
2018 &parent, sizeof (parent))) != 0)
2022 * Quit if directory has been removed (posix)
2030 prefetch = zp->z_zn_prefetch;
2033 * Initialize the iterator cursor.
2037 * Start iteration from the beginning of the directory.
2039 zap_cursor_init(&zc, os, zp->z_id);
2042 * The offset is a serialized cursor.
2044 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2048 * Transform to file-system independent format
2053 * Special case `.', `..', and `.zfs'.
2056 (void) strcpy(zap.za_name, ".");
2057 zap.za_normalization_conflict = 0;
2060 } else if (offset == 1) {
2061 (void) strcpy(zap.za_name, "..");
2062 zap.za_normalization_conflict = 0;
2065 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2066 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2067 zap.za_normalization_conflict = 0;
2068 objnum = ZFSCTL_INO_ROOT;
2074 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2075 if (error == ENOENT)
2082 * Allow multiple entries provided the first entry is
2083 * the object id. Non-zpl consumers may safely make
2084 * use of the additional space.
2086 * XXX: This should be a feature flag for compatibility
2088 if (zap.za_integer_length != 8 ||
2089 zap.za_num_integers == 0) {
2090 cmn_err(CE_WARN, "zap_readdir: bad directory "
2091 "entry, obj = %lld, offset = %lld, "
2092 "length = %d, num = %lld\n",
2093 (u_longlong_t)zp->z_id,
2094 (u_longlong_t)offset,
2095 zap.za_integer_length,
2096 (u_longlong_t)zap.za_num_integers);
2101 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2102 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2105 done = !dir_emit(ctx, zap.za_name, strlen(zap.za_name),
2110 /* Prefetch znode */
2112 dmu_prefetch(os, objnum, 0, 0);
2116 * Move to the next entry, fill in the previous offset.
2118 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2119 zap_cursor_advance(&zc);
2120 offset = zap_cursor_serialize(&zc);
2126 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2129 zap_cursor_fini(&zc);
2130 if (error == ENOENT)
2133 ZFS_ACCESSTIME_STAMP(zsb, zp);
2134 zfs_inode_update(zp);
2141 EXPORT_SYMBOL(zfs_readdir);
2143 ulong_t zfs_fsync_sync_cnt = 4;
2146 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2148 znode_t *zp = ITOZ(ip);
2149 zfs_sb_t *zsb = ITOZSB(ip);
2151 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2153 if (zsb->z_os->os_sync != ZFS_SYNC_DISABLED) {
2156 zil_commit(zsb->z_log, zp->z_id);
2161 EXPORT_SYMBOL(zfs_fsync);
2165 * Get the requested file attributes and place them in the provided
2168 * IN: ip - inode of file.
2169 * vap - va_mask identifies requested attributes.
2170 * If ATTR_XVATTR set, then optional attrs are requested
2171 * flags - ATTR_NOACLCHECK (CIFS server context)
2172 * cr - credentials of caller.
2174 * OUT: vap - attribute values.
2176 * RETURN: 0 (always succeeds)
2180 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2182 znode_t *zp = ITOZ(ip);
2183 zfs_sb_t *zsb = ITOZSB(ip);
2186 uint64_t mtime[2], ctime[2];
2187 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2188 xoptattr_t *xoap = NULL;
2189 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2190 sa_bulk_attr_t bulk[2];
2196 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2198 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
2199 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
2201 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2207 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2208 * Also, if we are the owner don't bother, since owner should
2209 * always be allowed to read basic attributes of file.
2211 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2212 (vap->va_uid != crgetuid(cr))) {
2213 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2221 * Return all attributes. It's cheaper to provide the answer
2222 * than to determine whether we were asked the question.
2225 mutex_enter(&zp->z_lock);
2226 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2227 vap->va_mode = zp->z_mode;
2228 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2229 vap->va_nodeid = zp->z_id;
2230 if ((zp->z_id == zsb->z_root) && zfs_show_ctldir(zp))
2231 links = zp->z_links + 1;
2233 links = zp->z_links;
2234 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2235 vap->va_size = i_size_read(ip);
2236 vap->va_rdev = ip->i_rdev;
2237 vap->va_seq = ip->i_generation;
2240 * Add in any requested optional attributes and the create time.
2241 * Also set the corresponding bits in the returned attribute bitmap.
2243 if ((xoap = xva_getxoptattr(xvap)) != NULL && zsb->z_use_fuids) {
2244 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2246 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2247 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2250 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2251 xoap->xoa_readonly =
2252 ((zp->z_pflags & ZFS_READONLY) != 0);
2253 XVA_SET_RTN(xvap, XAT_READONLY);
2256 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2258 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2259 XVA_SET_RTN(xvap, XAT_SYSTEM);
2262 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2264 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2265 XVA_SET_RTN(xvap, XAT_HIDDEN);
2268 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2269 xoap->xoa_nounlink =
2270 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2271 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2274 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2275 xoap->xoa_immutable =
2276 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2277 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2280 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2281 xoap->xoa_appendonly =
2282 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2283 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2286 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2288 ((zp->z_pflags & ZFS_NODUMP) != 0);
2289 XVA_SET_RTN(xvap, XAT_NODUMP);
2292 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2294 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2295 XVA_SET_RTN(xvap, XAT_OPAQUE);
2298 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2299 xoap->xoa_av_quarantined =
2300 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2301 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2304 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2305 xoap->xoa_av_modified =
2306 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2307 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2310 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2311 S_ISREG(ip->i_mode)) {
2312 zfs_sa_get_scanstamp(zp, xvap);
2315 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2318 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zsb),
2319 times, sizeof (times));
2320 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2321 XVA_SET_RTN(xvap, XAT_CREATETIME);
2324 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2325 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2326 XVA_SET_RTN(xvap, XAT_REPARSE);
2328 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2329 xoap->xoa_generation = zp->z_gen;
2330 XVA_SET_RTN(xvap, XAT_GEN);
2333 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2335 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2336 XVA_SET_RTN(xvap, XAT_OFFLINE);
2339 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2341 ((zp->z_pflags & ZFS_SPARSE) != 0);
2342 XVA_SET_RTN(xvap, XAT_SPARSE);
2346 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2347 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2348 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2350 mutex_exit(&zp->z_lock);
2352 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2354 if (zp->z_blksz == 0) {
2356 * Block size hasn't been set; suggest maximal I/O transfers.
2358 vap->va_blksize = zsb->z_max_blksz;
2364 EXPORT_SYMBOL(zfs_getattr);
2367 * Get the basic file attributes and place them in the provided kstat
2368 * structure. The inode is assumed to be the authoritative source
2369 * for most of the attributes. However, the znode currently has the
2370 * authoritative atime, blksize, and block count.
2372 * IN: ip - inode of file.
2374 * OUT: sp - kstat values.
2376 * RETURN: 0 (always succeeds)
2380 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2382 znode_t *zp = ITOZ(ip);
2383 zfs_sb_t *zsb = ITOZSB(ip);
2388 mutex_enter(&zp->z_lock);
2390 generic_fillattr(ip, sp);
2391 ZFS_TIME_DECODE(&sp->atime, zp->z_atime);
2393 sa_object_size(zp->z_sa_hdl, (uint32_t *)&sp->blksize, &sp->blocks);
2394 if (unlikely(zp->z_blksz == 0)) {
2396 * Block size hasn't been set; suggest maximal I/O transfers.
2398 sp->blksize = zsb->z_max_blksz;
2401 mutex_exit(&zp->z_lock);
2407 EXPORT_SYMBOL(zfs_getattr_fast);
2410 * Set the file attributes to the values contained in the
2413 * IN: ip - inode of file to be modified.
2414 * vap - new attribute values.
2415 * If ATTR_XVATTR set, then optional attrs are being set
2416 * flags - ATTR_UTIME set if non-default time values provided.
2417 * - ATTR_NOACLCHECK (CIFS context only).
2418 * cr - credentials of caller.
2420 * RETURN: 0 if success
2421 * error code if failure
2424 * ip - ctime updated, mtime updated if size changed.
2428 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2430 znode_t *zp = ITOZ(ip);
2431 zfs_sb_t *zsb = ITOZSB(ip);
2435 xvattr_t *tmpxvattr;
2436 uint_t mask = vap->va_mask;
2437 uint_t saved_mask = 0;
2440 uint64_t new_uid, new_gid;
2442 uint64_t mtime[2], ctime[2];
2444 int need_policy = FALSE;
2446 zfs_fuid_info_t *fuidp = NULL;
2447 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2450 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2451 boolean_t fuid_dirtied = B_FALSE;
2452 sa_bulk_attr_t *bulk, *xattr_bulk;
2453 int count = 0, xattr_count = 0;
2464 * Make sure that if we have ephemeral uid/gid or xvattr specified
2465 * that file system is at proper version level
2468 if (zsb->z_use_fuids == B_FALSE &&
2469 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2470 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2471 (mask & ATTR_XVATTR))) {
2476 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2481 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2487 * If this is an xvattr_t, then get a pointer to the structure of
2488 * optional attributes. If this is NULL, then we have a vattr_t.
2490 xoap = xva_getxoptattr(xvap);
2492 tmpxvattr = kmem_alloc(sizeof(xvattr_t), KM_SLEEP);
2493 xva_init(tmpxvattr);
2495 bulk = kmem_alloc(sizeof(sa_bulk_attr_t) * 7, KM_SLEEP);
2496 xattr_bulk = kmem_alloc(sizeof(sa_bulk_attr_t) * 7, KM_SLEEP);
2499 * Immutable files can only alter immutable bit and atime
2501 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2502 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2503 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2508 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2514 * Verify timestamps doesn't overflow 32 bits.
2515 * ZFS can handle large timestamps, but 32bit syscalls can't
2516 * handle times greater than 2039. This check should be removed
2517 * once large timestamps are fully supported.
2519 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2520 if (((mask & ATTR_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2521 ((mask & ATTR_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2531 /* Can this be moved to before the top label? */
2532 if (zfs_is_readonly(zsb)) {
2538 * First validate permissions
2541 if (mask & ATTR_SIZE) {
2542 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2546 truncate_setsize(ip, vap->va_size);
2549 * XXX - Note, we are not providing any open
2550 * mode flags here (like FNDELAY), so we may
2551 * block if there are locks present... this
2552 * should be addressed in openat().
2554 /* XXX - would it be OK to generate a log record here? */
2555 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2560 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2561 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2562 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2563 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2564 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2565 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2566 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2567 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2568 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2572 if (mask & (ATTR_UID|ATTR_GID)) {
2573 int idmask = (mask & (ATTR_UID|ATTR_GID));
2578 * NOTE: even if a new mode is being set,
2579 * we may clear S_ISUID/S_ISGID bits.
2582 if (!(mask & ATTR_MODE))
2583 vap->va_mode = zp->z_mode;
2586 * Take ownership or chgrp to group we are a member of
2589 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
2590 take_group = (mask & ATTR_GID) &&
2591 zfs_groupmember(zsb, vap->va_gid, cr);
2594 * If both ATTR_UID and ATTR_GID are set then take_owner and
2595 * take_group must both be set in order to allow taking
2598 * Otherwise, send the check through secpolicy_vnode_setattr()
2602 if (((idmask == (ATTR_UID|ATTR_GID)) &&
2603 take_owner && take_group) ||
2604 ((idmask == ATTR_UID) && take_owner) ||
2605 ((idmask == ATTR_GID) && take_group)) {
2606 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2607 skipaclchk, cr) == 0) {
2609 * Remove setuid/setgid for non-privileged users
2611 (void) secpolicy_setid_clear(vap, cr);
2612 trim_mask = (mask & (ATTR_UID|ATTR_GID));
2621 mutex_enter(&zp->z_lock);
2622 oldva.va_mode = zp->z_mode;
2623 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2624 if (mask & ATTR_XVATTR) {
2626 * Update xvattr mask to include only those attributes
2627 * that are actually changing.
2629 * the bits will be restored prior to actually setting
2630 * the attributes so the caller thinks they were set.
2632 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2633 if (xoap->xoa_appendonly !=
2634 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2637 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2638 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
2642 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2643 if (xoap->xoa_nounlink !=
2644 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2647 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2648 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
2652 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2653 if (xoap->xoa_immutable !=
2654 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2657 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2658 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
2662 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2663 if (xoap->xoa_nodump !=
2664 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2667 XVA_CLR_REQ(xvap, XAT_NODUMP);
2668 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
2672 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2673 if (xoap->xoa_av_modified !=
2674 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2677 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2678 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
2682 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2683 if ((!S_ISREG(ip->i_mode) &&
2684 xoap->xoa_av_quarantined) ||
2685 xoap->xoa_av_quarantined !=
2686 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2689 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2690 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
2694 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2695 mutex_exit(&zp->z_lock);
2700 if (need_policy == FALSE &&
2701 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2702 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2707 mutex_exit(&zp->z_lock);
2709 if (mask & ATTR_MODE) {
2710 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2711 err = secpolicy_setid_setsticky_clear(ip, vap,
2716 trim_mask |= ATTR_MODE;
2724 * If trim_mask is set then take ownership
2725 * has been granted or write_acl is present and user
2726 * has the ability to modify mode. In that case remove
2727 * UID|GID and or MODE from mask so that
2728 * secpolicy_vnode_setattr() doesn't revoke it.
2732 saved_mask = vap->va_mask;
2733 vap->va_mask &= ~trim_mask;
2735 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
2736 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2741 vap->va_mask |= saved_mask;
2745 * secpolicy_vnode_setattr, or take ownership may have
2748 mask = vap->va_mask;
2750 if ((mask & (ATTR_UID | ATTR_GID))) {
2751 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
2752 &xattr_obj, sizeof (xattr_obj));
2754 if (err == 0 && xattr_obj) {
2755 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
2759 if (mask & ATTR_UID) {
2760 new_uid = zfs_fuid_create(zsb,
2761 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2762 if (new_uid != zp->z_uid &&
2763 zfs_fuid_overquota(zsb, B_FALSE, new_uid)) {
2771 if (mask & ATTR_GID) {
2772 new_gid = zfs_fuid_create(zsb, (uint64_t)vap->va_gid,
2773 cr, ZFS_GROUP, &fuidp);
2774 if (new_gid != zp->z_gid &&
2775 zfs_fuid_overquota(zsb, B_TRUE, new_gid)) {
2783 tx = dmu_tx_create(zsb->z_os);
2785 if (mask & ATTR_MODE) {
2786 uint64_t pmode = zp->z_mode;
2788 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2790 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
2792 mutex_enter(&zp->z_lock);
2793 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2795 * Are we upgrading ACL from old V0 format
2798 if (zsb->z_version >= ZPL_VERSION_FUID &&
2799 zfs_znode_acl_version(zp) ==
2800 ZFS_ACL_VERSION_INITIAL) {
2801 dmu_tx_hold_free(tx, acl_obj, 0,
2803 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2804 0, aclp->z_acl_bytes);
2806 dmu_tx_hold_write(tx, acl_obj, 0,
2809 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2810 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2811 0, aclp->z_acl_bytes);
2813 mutex_exit(&zp->z_lock);
2814 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2816 if ((mask & ATTR_XVATTR) &&
2817 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2818 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2820 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2824 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
2827 fuid_dirtied = zsb->z_fuid_dirty;
2829 zfs_fuid_txhold(zsb, tx);
2831 zfs_sa_upgrade_txholds(tx, zp);
2833 err = dmu_tx_assign(tx, TXG_NOWAIT);
2835 if (err == ERESTART)
2842 * Set each attribute requested.
2843 * We group settings according to the locks they need to acquire.
2845 * Note: you cannot set ctime directly, although it will be
2846 * updated as a side-effect of calling this function.
2850 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2851 mutex_enter(&zp->z_acl_lock);
2852 mutex_enter(&zp->z_lock);
2854 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
2855 &zp->z_pflags, sizeof (zp->z_pflags));
2858 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2859 mutex_enter(&attrzp->z_acl_lock);
2860 mutex_enter(&attrzp->z_lock);
2861 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2862 SA_ZPL_FLAGS(zsb), NULL, &attrzp->z_pflags,
2863 sizeof (attrzp->z_pflags));
2866 if (mask & (ATTR_UID|ATTR_GID)) {
2868 if (mask & ATTR_UID) {
2869 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zsb), NULL,
2870 &new_uid, sizeof (new_uid));
2871 zp->z_uid = new_uid;
2873 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2874 SA_ZPL_UID(zsb), NULL, &new_uid,
2876 attrzp->z_uid = new_uid;
2880 if (mask & ATTR_GID) {
2881 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zsb),
2882 NULL, &new_gid, sizeof (new_gid));
2883 zp->z_gid = new_gid;
2885 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2886 SA_ZPL_GID(zsb), NULL, &new_gid,
2888 attrzp->z_gid = new_gid;
2891 if (!(mask & ATTR_MODE)) {
2892 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb),
2893 NULL, &new_mode, sizeof (new_mode));
2894 new_mode = zp->z_mode;
2896 err = zfs_acl_chown_setattr(zp);
2899 err = zfs_acl_chown_setattr(attrzp);
2904 if (mask & ATTR_MODE) {
2905 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb), NULL,
2906 &new_mode, sizeof (new_mode));
2907 zp->z_mode = new_mode;
2908 ASSERT3P(aclp, !=, NULL);
2909 err = zfs_aclset_common(zp, aclp, cr, tx);
2911 if (zp->z_acl_cached)
2912 zfs_acl_free(zp->z_acl_cached);
2913 zp->z_acl_cached = aclp;
2918 if (mask & ATTR_ATIME) {
2919 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
2920 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zsb), NULL,
2921 &zp->z_atime, sizeof (zp->z_atime));
2924 if (mask & ATTR_MTIME) {
2925 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
2926 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL,
2927 mtime, sizeof (mtime));
2930 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2931 if (mask & ATTR_SIZE && !(mask & ATTR_MTIME)) {
2932 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb),
2933 NULL, mtime, sizeof (mtime));
2934 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
2935 &ctime, sizeof (ctime));
2936 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
2938 } else if (mask != 0) {
2939 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
2940 &ctime, sizeof (ctime));
2941 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
2944 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2945 SA_ZPL_CTIME(zsb), NULL,
2946 &ctime, sizeof (ctime));
2947 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
2948 mtime, ctime, B_TRUE);
2952 * Do this after setting timestamps to prevent timestamp
2953 * update from toggling bit
2956 if (xoap && (mask & ATTR_XVATTR)) {
2959 * restore trimmed off masks
2960 * so that return masks can be set for caller.
2963 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
2964 XVA_SET_REQ(xvap, XAT_APPENDONLY);
2966 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
2967 XVA_SET_REQ(xvap, XAT_NOUNLINK);
2969 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
2970 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
2972 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
2973 XVA_SET_REQ(xvap, XAT_NODUMP);
2975 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
2976 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
2978 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
2979 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
2982 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2983 ASSERT(S_ISREG(ip->i_mode));
2985 zfs_xvattr_set(zp, xvap, tx);
2989 zfs_fuid_sync(zsb, tx);
2992 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2994 mutex_exit(&zp->z_lock);
2995 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2996 mutex_exit(&zp->z_acl_lock);
2999 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3000 mutex_exit(&attrzp->z_acl_lock);
3001 mutex_exit(&attrzp->z_lock);
3004 if (err == 0 && attrzp) {
3005 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3016 zfs_fuid_info_free(fuidp);
3022 if (err == ERESTART)
3025 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3027 zfs_inode_update(zp);
3031 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3032 zil_commit(zilog, 0);
3035 kmem_free(xattr_bulk, sizeof(sa_bulk_attr_t) * 7);
3036 kmem_free(bulk, sizeof(sa_bulk_attr_t) * 7);
3037 kmem_free(tmpxvattr, sizeof(xvattr_t));
3041 EXPORT_SYMBOL(zfs_setattr);
3043 typedef struct zfs_zlock {
3044 krwlock_t *zl_rwlock; /* lock we acquired */
3045 znode_t *zl_znode; /* znode we held */
3046 struct zfs_zlock *zl_next; /* next in list */
3050 * Drop locks and release vnodes that were held by zfs_rename_lock().
3053 zfs_rename_unlock(zfs_zlock_t **zlpp)
3057 while ((zl = *zlpp) != NULL) {
3058 if (zl->zl_znode != NULL)
3059 iput(ZTOI(zl->zl_znode));
3060 rw_exit(zl->zl_rwlock);
3061 *zlpp = zl->zl_next;
3062 kmem_free(zl, sizeof (*zl));
3067 * Search back through the directory tree, using the ".." entries.
3068 * Lock each directory in the chain to prevent concurrent renames.
3069 * Fail any attempt to move a directory into one of its own descendants.
3070 * XXX - z_parent_lock can overlap with map or grow locks
3073 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3077 uint64_t rootid = ZTOZSB(zp)->z_root;
3078 uint64_t oidp = zp->z_id;
3079 krwlock_t *rwlp = &szp->z_parent_lock;
3080 krw_t rw = RW_WRITER;
3083 * First pass write-locks szp and compares to zp->z_id.
3084 * Later passes read-lock zp and compare to zp->z_parent.
3087 if (!rw_tryenter(rwlp, rw)) {
3089 * Another thread is renaming in this path.
3090 * Note that if we are a WRITER, we don't have any
3091 * parent_locks held yet.
3093 if (rw == RW_READER && zp->z_id > szp->z_id) {
3095 * Drop our locks and restart
3097 zfs_rename_unlock(&zl);
3101 rwlp = &szp->z_parent_lock;
3106 * Wait for other thread to drop its locks
3112 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3113 zl->zl_rwlock = rwlp;
3114 zl->zl_znode = NULL;
3115 zl->zl_next = *zlpp;
3118 if (oidp == szp->z_id) /* We're a descendant of szp */
3121 if (oidp == rootid) /* We've hit the top */
3124 if (rw == RW_READER) { /* i.e. not the first pass */
3125 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3130 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3131 &oidp, sizeof (oidp));
3132 rwlp = &zp->z_parent_lock;
3135 } while (zp->z_id != sdzp->z_id);
3141 * Move an entry from the provided source directory to the target
3142 * directory. Change the entry name as indicated.
3144 * IN: sdip - Source directory containing the "old entry".
3145 * snm - Old entry name.
3146 * tdip - Target directory to contain the "new entry".
3147 * tnm - New entry name.
3148 * cr - credentials of caller.
3149 * flags - case flags
3151 * RETURN: 0 if success
3152 * error code if failure
3155 * sdip,tdip - ctime|mtime updated
3159 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3160 cred_t *cr, int flags)
3162 znode_t *tdzp, *szp, *tzp;
3163 znode_t *sdzp = ITOZ(sdip);
3164 zfs_sb_t *zsb = ITOZSB(sdip);
3166 zfs_dirlock_t *sdl, *tdl;
3169 int cmp, serr, terr;
3174 ZFS_VERIFY_ZP(sdzp);
3177 if (tdip->i_sb != sdip->i_sb) {
3183 ZFS_VERIFY_ZP(tdzp);
3184 if (zsb->z_utf8 && u8_validate(tnm,
3185 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3190 if (flags & FIGNORECASE)
3199 * This is to prevent the creation of links into attribute space
3200 * by renaming a linked file into/outof an attribute directory.
3201 * See the comment in zfs_link() for why this is considered bad.
3203 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3209 * Lock source and target directory entries. To prevent deadlock,
3210 * a lock ordering must be defined. We lock the directory with
3211 * the smallest object id first, or if it's a tie, the one with
3212 * the lexically first name.
3214 if (sdzp->z_id < tdzp->z_id) {
3216 } else if (sdzp->z_id > tdzp->z_id) {
3220 * First compare the two name arguments without
3221 * considering any case folding.
3223 int nofold = (zsb->z_norm & ~U8_TEXTPREP_TOUPPER);
3225 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3226 ASSERT(error == 0 || !zsb->z_utf8);
3229 * POSIX: "If the old argument and the new argument
3230 * both refer to links to the same existing file,
3231 * the rename() function shall return successfully
3232 * and perform no other action."
3238 * If the file system is case-folding, then we may
3239 * have some more checking to do. A case-folding file
3240 * system is either supporting mixed case sensitivity
3241 * access or is completely case-insensitive. Note
3242 * that the file system is always case preserving.
3244 * In mixed sensitivity mode case sensitive behavior
3245 * is the default. FIGNORECASE must be used to
3246 * explicitly request case insensitive behavior.
3248 * If the source and target names provided differ only
3249 * by case (e.g., a request to rename 'tim' to 'Tim'),
3250 * we will treat this as a special case in the
3251 * case-insensitive mode: as long as the source name
3252 * is an exact match, we will allow this to proceed as
3253 * a name-change request.
3255 if ((zsb->z_case == ZFS_CASE_INSENSITIVE ||
3256 (zsb->z_case == ZFS_CASE_MIXED &&
3257 flags & FIGNORECASE)) &&
3258 u8_strcmp(snm, tnm, 0, zsb->z_norm, U8_UNICODE_LATEST,
3261 * case preserving rename request, require exact
3270 * If the source and destination directories are the same, we should
3271 * grab the z_name_lock of that directory only once.
3275 rw_enter(&sdzp->z_name_lock, RW_READER);
3279 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3280 ZEXISTS | zflg, NULL, NULL);
3281 terr = zfs_dirent_lock(&tdl,
3282 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3284 terr = zfs_dirent_lock(&tdl,
3285 tdzp, tnm, &tzp, zflg, NULL, NULL);
3286 serr = zfs_dirent_lock(&sdl,
3287 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3293 * Source entry invalid or not there.
3296 zfs_dirent_unlock(tdl);
3302 rw_exit(&sdzp->z_name_lock);
3304 if (strcmp(snm, "..") == 0)
3310 zfs_dirent_unlock(sdl);
3314 rw_exit(&sdzp->z_name_lock);
3316 if (strcmp(tnm, "..") == 0)
3323 * Must have write access at the source to remove the old entry
3324 * and write access at the target to create the new entry.
3325 * Note that if target and source are the same, this can be
3326 * done in a single check.
3329 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3332 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3334 * Check to make sure rename is valid.
3335 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3337 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3342 * Does target exist?
3346 * Source and target must be the same type.
3348 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3349 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3354 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3360 * POSIX dictates that when the source and target
3361 * entries refer to the same file object, rename
3362 * must do nothing and exit without error.
3364 if (szp->z_id == tzp->z_id) {
3370 tx = dmu_tx_create(zsb->z_os);
3371 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3372 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3373 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3374 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3376 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3377 zfs_sa_upgrade_txholds(tx, tdzp);
3380 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3381 zfs_sa_upgrade_txholds(tx, tzp);
3384 zfs_sa_upgrade_txholds(tx, szp);
3385 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
3386 error = dmu_tx_assign(tx, TXG_NOWAIT);
3389 zfs_rename_unlock(&zl);
3390 zfs_dirent_unlock(sdl);
3391 zfs_dirent_unlock(tdl);
3394 rw_exit(&sdzp->z_name_lock);
3399 if (error == ERESTART) {
3409 if (tzp) /* Attempt to remove the existing target */
3410 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3413 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3415 szp->z_pflags |= ZFS_AV_MODIFIED;
3417 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zsb),
3418 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3421 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3423 zfs_log_rename(zilog, tx, TX_RENAME |
3424 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3425 sdl->dl_name, tdzp, tdl->dl_name, szp);
3428 * At this point, we have successfully created
3429 * the target name, but have failed to remove
3430 * the source name. Since the create was done
3431 * with the ZRENAMING flag, there are
3432 * complications; for one, the link count is
3433 * wrong. The easiest way to deal with this
3434 * is to remove the newly created target, and
3435 * return the original error. This must
3436 * succeed; fortunately, it is very unlikely to
3437 * fail, since we just created it.
3439 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3440 ZRENAMING, NULL), ==, 0);
3448 zfs_rename_unlock(&zl);
3450 zfs_dirent_unlock(sdl);
3451 zfs_dirent_unlock(tdl);
3453 zfs_inode_update(sdzp);
3455 rw_exit(&sdzp->z_name_lock);
3458 zfs_inode_update(tdzp);
3460 zfs_inode_update(szp);
3463 zfs_inode_update(tzp);
3467 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3468 zil_commit(zilog, 0);
3473 EXPORT_SYMBOL(zfs_rename);
3476 * Insert the indicated symbolic reference entry into the directory.
3478 * IN: dip - Directory to contain new symbolic link.
3479 * link - Name for new symlink entry.
3480 * vap - Attributes of new entry.
3481 * target - Target path of new symlink.
3483 * cr - credentials of caller.
3484 * flags - case flags
3486 * RETURN: 0 if success
3487 * error code if failure
3490 * dip - ctime|mtime updated
3494 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
3495 struct inode **ipp, cred_t *cr, int flags)
3497 znode_t *zp, *dzp = ITOZ(dip);
3500 zfs_sb_t *zsb = ITOZSB(dip);
3502 uint64_t len = strlen(link);
3505 zfs_acl_ids_t acl_ids;
3506 boolean_t fuid_dirtied;
3507 uint64_t txtype = TX_SYMLINK;
3509 ASSERT(S_ISLNK(vap->va_mode));
3515 if (zsb->z_utf8 && u8_validate(name, strlen(name),
3516 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3520 if (flags & FIGNORECASE)
3523 if (len > MAXPATHLEN) {
3525 return (ENAMETOOLONG);
3528 if ((error = zfs_acl_ids_create(dzp, 0,
3529 vap, cr, NULL, &acl_ids)) != 0) {
3537 * Attempt to lock directory; fail if entry already exists.
3539 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3541 zfs_acl_ids_free(&acl_ids);
3546 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3547 zfs_acl_ids_free(&acl_ids);
3548 zfs_dirent_unlock(dl);
3553 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
3554 zfs_acl_ids_free(&acl_ids);
3555 zfs_dirent_unlock(dl);
3559 tx = dmu_tx_create(zsb->z_os);
3560 fuid_dirtied = zsb->z_fuid_dirty;
3561 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3562 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3563 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3564 ZFS_SA_BASE_ATTR_SIZE + len);
3565 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3566 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3567 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3568 acl_ids.z_aclp->z_acl_bytes);
3571 zfs_fuid_txhold(zsb, tx);
3572 error = dmu_tx_assign(tx, TXG_NOWAIT);
3574 zfs_dirent_unlock(dl);
3575 if (error == ERESTART) {
3580 zfs_acl_ids_free(&acl_ids);
3587 * Create a new object for the symlink.
3588 * for version 4 ZPL datsets the symlink will be an SA attribute
3590 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3593 zfs_fuid_sync(zsb, tx);
3595 mutex_enter(&zp->z_lock);
3597 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zsb),
3600 zfs_sa_symlink(zp, link, len, tx);
3601 mutex_exit(&zp->z_lock);
3604 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
3605 &zp->z_size, sizeof (zp->z_size), tx);
3607 * Insert the new object into the directory.
3609 (void) zfs_link_create(dl, zp, tx, ZNEW);
3611 if (flags & FIGNORECASE)
3613 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3615 zfs_inode_update(dzp);
3616 zfs_inode_update(zp);
3618 zfs_acl_ids_free(&acl_ids);
3622 zfs_dirent_unlock(dl);
3626 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3627 zil_commit(zilog, 0);
3632 EXPORT_SYMBOL(zfs_symlink);
3635 * Return, in the buffer contained in the provided uio structure,
3636 * the symbolic path referred to by ip.
3638 * IN: ip - inode of symbolic link
3639 * uio - structure to contain the link path.
3640 * cr - credentials of caller.
3642 * RETURN: 0 if success
3643 * error code if failure
3646 * ip - atime updated
3650 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
3652 znode_t *zp = ITOZ(ip);
3653 zfs_sb_t *zsb = ITOZSB(ip);
3659 mutex_enter(&zp->z_lock);
3661 error = sa_lookup_uio(zp->z_sa_hdl,
3662 SA_ZPL_SYMLINK(zsb), uio);
3664 error = zfs_sa_readlink(zp, uio);
3665 mutex_exit(&zp->z_lock);
3667 ZFS_ACCESSTIME_STAMP(zsb, zp);
3668 zfs_inode_update(zp);
3672 EXPORT_SYMBOL(zfs_readlink);
3675 * Insert a new entry into directory tdip referencing sip.
3677 * IN: tdip - Directory to contain new entry.
3678 * sip - inode of new entry.
3679 * name - name of new entry.
3680 * cr - credentials of caller.
3682 * RETURN: 0 if success
3683 * error code if failure
3686 * tdip - ctime|mtime updated
3687 * sip - ctime updated
3691 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr)
3693 znode_t *dzp = ITOZ(tdip);
3695 zfs_sb_t *zsb = ITOZSB(tdip);
3704 ASSERT(S_ISDIR(tdip->i_mode));
3711 * POSIX dictates that we return EPERM here.
3712 * Better choices include ENOTSUP or EISDIR.
3714 if (S_ISDIR(sip->i_mode)) {
3719 if (sip->i_sb != tdip->i_sb) {
3727 /* Prevent links to .zfs/shares files */
3729 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zsb),
3730 &parent, sizeof (uint64_t))) != 0) {
3734 if (parent == zsb->z_shares_dir) {
3739 if (zsb->z_utf8 && u8_validate(name,
3740 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3744 #ifdef HAVE_PN_UTILS
3745 if (flags & FIGNORECASE)
3747 #endif /* HAVE_PN_UTILS */
3750 * We do not support links between attributes and non-attributes
3751 * because of the potential security risk of creating links
3752 * into "normal" file space in order to circumvent restrictions
3753 * imposed in attribute space.
3755 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3760 owner = zfs_fuid_map_id(zsb, szp->z_uid, cr, ZFS_OWNER);
3761 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3766 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3773 * Attempt to lock directory; fail if entry already exists.
3775 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3781 tx = dmu_tx_create(zsb->z_os);
3782 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3783 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3784 zfs_sa_upgrade_txholds(tx, szp);
3785 zfs_sa_upgrade_txholds(tx, dzp);
3786 error = dmu_tx_assign(tx, TXG_NOWAIT);
3788 zfs_dirent_unlock(dl);
3789 if (error == ERESTART) {
3799 error = zfs_link_create(dl, szp, tx, 0);
3802 uint64_t txtype = TX_LINK;
3803 #ifdef HAVE_PN_UTILS
3804 if (flags & FIGNORECASE)
3806 #endif /* HAVE_PN_UTILS */
3807 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3812 zfs_dirent_unlock(dl);
3814 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3815 zil_commit(zilog, 0);
3817 zfs_inode_update(dzp);
3818 zfs_inode_update(szp);
3822 EXPORT_SYMBOL(zfs_link);
3825 zfs_putpage_commit_cb(void *arg, int error)
3827 struct page *pp = arg;
3830 __set_page_dirty_nobuffers(pp);
3832 if (error != ECANCELED)
3838 end_page_writeback(pp);
3842 * Push a page out to disk, once the page is on stable storage the
3843 * registered commit callback will be run as notification of completion.
3845 * IN: ip - page mapped for inode.
3846 * pp - page to push (page is locked)
3847 * wbc - writeback control data
3849 * RETURN: 0 if success
3850 * error code if failure
3853 * ip - ctime|mtime updated
3857 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
3859 znode_t *zp = ITOZ(ip);
3860 zfs_sb_t *zsb = ITOZSB(ip);
3868 uint64_t mtime[2], ctime[2];
3869 sa_bulk_attr_t bulk[3];
3876 ASSERT(PageLocked(pp));
3878 pgoff = page_offset(pp); /* Page byte-offset in file */
3879 offset = i_size_read(ip); /* File length in bytes */
3880 pglen = MIN(PAGE_CACHE_SIZE, /* Page length in bytes */
3881 P2ROUNDUP(offset, PAGE_CACHE_SIZE)-pgoff);
3883 /* Page is beyond end of file */
3884 if (pgoff >= offset) {
3890 /* Truncate page length to end of file */
3891 if (pgoff + pglen > offset)
3892 pglen = offset - pgoff;
3896 * FIXME: Allow mmap writes past its quota. The correct fix
3897 * is to register a page_mkwrite() handler to count the page
3898 * against its quota when it is about to be dirtied.
3900 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
3901 zfs_owner_overquota(zsb, zp, B_TRUE)) {
3906 set_page_writeback(pp);
3909 rl = zfs_range_lock(zp, pgoff, pglen, RL_WRITER);
3910 tx = dmu_tx_create(zsb->z_os);
3912 sync = ((zsb->z_os->os_sync == ZFS_SYNC_ALWAYS) ||
3913 (wbc->sync_mode == WB_SYNC_ALL));
3915 dmu_tx_callback_register(tx, zfs_putpage_commit_cb, pp);
3917 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
3919 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3920 zfs_sa_upgrade_txholds(tx, zp);
3921 err = dmu_tx_assign(tx, TXG_NOWAIT);
3923 if (err == ERESTART)
3926 /* Will call all registered commit callbacks */
3930 * For the synchronous case the commit callback must be
3931 * explicitly called because there is no registered callback.
3934 zfs_putpage_commit_cb(pp, ECANCELED);
3936 zfs_range_unlock(rl);
3942 ASSERT3U(pglen, <=, PAGE_CACHE_SIZE);
3943 dmu_write(zsb->z_os, zp->z_id, pgoff, pglen, va, tx);
3946 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
3947 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
3948 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zsb), NULL, &zp->z_pflags, 8);
3950 /* Preserve the mtime and ctime provided by the inode */
3951 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
3952 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
3953 zp->z_atime_dirty = 0;
3956 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
3958 zfs_log_write(zsb->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0);
3961 zfs_range_unlock(rl);
3964 zil_commit(zsb->z_log, zp->z_id);
3965 zfs_putpage_commit_cb(pp, err);
3973 * Update the system attributes when the inode has been dirtied. For the
3974 * moment we only update the mode, atime, mtime, and ctime.
3977 zfs_dirty_inode(struct inode *ip, int flags)
3979 znode_t *zp = ITOZ(ip);
3980 zfs_sb_t *zsb = ITOZSB(ip);
3982 uint64_t mode, atime[2], mtime[2], ctime[2];
3983 sa_bulk_attr_t bulk[4];
3990 tx = dmu_tx_create(zsb->z_os);
3992 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3993 zfs_sa_upgrade_txholds(tx, zp);
3995 error = dmu_tx_assign(tx, TXG_WAIT);
4001 mutex_enter(&zp->z_lock);
4002 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zsb), NULL, &mode, 8);
4003 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zsb), NULL, &atime, 16);
4004 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
4005 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
4007 /* Preserve the mode, mtime and ctime provided by the inode */
4008 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4009 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4010 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4014 zp->z_atime_dirty = 0;
4016 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4017 mutex_exit(&zp->z_lock);
4024 EXPORT_SYMBOL(zfs_dirty_inode);
4028 zfs_inactive(struct inode *ip)
4030 znode_t *zp = ITOZ(ip);
4031 zfs_sb_t *zsb = ITOZSB(ip);
4034 if (zfsctl_is_node(ip)) {
4035 zfsctl_inode_inactive(ip);
4039 rw_enter(&zsb->z_teardown_inactive_lock, RW_READER);
4040 if (zp->z_sa_hdl == NULL) {
4041 rw_exit(&zsb->z_teardown_inactive_lock);
4045 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4046 dmu_tx_t *tx = dmu_tx_create(zsb->z_os);
4048 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4049 zfs_sa_upgrade_txholds(tx, zp);
4050 error = dmu_tx_assign(tx, TXG_WAIT);
4054 mutex_enter(&zp->z_lock);
4055 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zsb),
4056 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4057 zp->z_atime_dirty = 0;
4058 mutex_exit(&zp->z_lock);
4064 rw_exit(&zsb->z_teardown_inactive_lock);
4066 EXPORT_SYMBOL(zfs_inactive);
4069 * Bounds-check the seek operation.
4071 * IN: ip - inode seeking within
4072 * ooff - old file offset
4073 * noffp - pointer to new file offset
4074 * ct - caller context
4076 * RETURN: 0 if success
4077 * EINVAL if new offset invalid
4081 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
4083 if (S_ISDIR(ip->i_mode))
4085 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4087 EXPORT_SYMBOL(zfs_seek);
4090 * Fill pages with data from the disk.
4093 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
4095 znode_t *zp = ITOZ(ip);
4096 zfs_sb_t *zsb = ITOZSB(ip);
4098 struct page *cur_pp;
4099 u_offset_t io_off, total;
4106 io_len = nr_pages << PAGE_CACHE_SHIFT;
4107 i_size = i_size_read(ip);
4108 io_off = page_offset(pl[0]);
4110 if (io_off + io_len > i_size)
4111 io_len = i_size - io_off;
4114 * Iterate over list of pages and read each page individually.
4118 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4122 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4126 /* convert checksum errors into IO errors */
4131 cur_pp = pl[++page_idx];
4138 * Uses zfs_fillpage to read data from the file and fill the pages.
4140 * IN: ip - inode of file to get data from.
4141 * pl - list of pages to read
4142 * nr_pages - number of pages to read
4144 * RETURN: 0 if success
4145 * error code if failure
4148 * vp - atime updated
4152 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
4154 znode_t *zp = ITOZ(ip);
4155 zfs_sb_t *zsb = ITOZSB(ip);
4164 err = zfs_fillpage(ip, pl, nr_pages);
4167 ZFS_ACCESSTIME_STAMP(zsb, zp);
4172 EXPORT_SYMBOL(zfs_getpage);
4175 * Check ZFS specific permissions to memory map a section of a file.
4177 * IN: ip - inode of the file to mmap
4179 * addrp - start address in memory region
4180 * len - length of memory region
4181 * vm_flags- address flags
4183 * RETURN: 0 if success
4184 * error code if failure
4188 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4189 unsigned long vm_flags)
4191 znode_t *zp = ITOZ(ip);
4192 zfs_sb_t *zsb = ITOZSB(ip);
4197 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4198 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4203 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4204 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4209 if (off < 0 || len > MAXOFFSET_T - off) {
4217 EXPORT_SYMBOL(zfs_map);
4220 * convoff - converts the given data (start, whence) to the
4224 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4229 if ((lckdat->l_whence == 2) || (whence == 2)) {
4230 if ((error = zfs_getattr(ip, &vap, 0, CRED()) != 0))
4234 switch (lckdat->l_whence) {
4236 lckdat->l_start += offset;
4239 lckdat->l_start += vap.va_size;
4247 if (lckdat->l_start < 0)
4252 lckdat->l_start -= offset;
4255 lckdat->l_start -= vap.va_size;
4263 lckdat->l_whence = (short)whence;
4268 * Free or allocate space in a file. Currently, this function only
4269 * supports the `F_FREESP' command. However, this command is somewhat
4270 * misnamed, as its functionality includes the ability to allocate as
4271 * well as free space.
4273 * IN: ip - inode of file to free data in.
4274 * cmd - action to take (only F_FREESP supported).
4275 * bfp - section of file to free/alloc.
4276 * flag - current file open mode flags.
4277 * offset - current file offset.
4278 * cr - credentials of caller [UNUSED].
4280 * RETURN: 0 if success
4281 * error code if failure
4284 * ip - ctime|mtime updated
4288 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4289 offset_t offset, cred_t *cr)
4291 znode_t *zp = ITOZ(ip);
4292 zfs_sb_t *zsb = ITOZSB(ip);
4299 if (cmd != F_FREESP) {
4304 if ((error = convoff(ip, bfp, 0, offset))) {
4309 if (bfp->l_len < 0) {
4315 * Permissions aren't checked on Solaris because on this OS
4316 * zfs_space() can only be called with an opened file handle.
4317 * On Linux we can get here through truncate_range() which
4318 * operates directly on inodes, so we need to check access rights.
4320 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4326 len = bfp->l_len; /* 0 means from off to end of file */
4328 error = zfs_freesp(zp, off, len, flag, TRUE);
4333 EXPORT_SYMBOL(zfs_space);
4337 zfs_fid(struct inode *ip, fid_t *fidp)
4339 znode_t *zp = ITOZ(ip);
4340 zfs_sb_t *zsb = ITOZSB(ip);
4343 uint64_t object = zp->z_id;
4350 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zsb),
4351 &gen64, sizeof (uint64_t))) != 0) {
4356 gen = (uint32_t)gen64;
4358 size = (zsb->z_parent != zsb) ? LONG_FID_LEN : SHORT_FID_LEN;
4359 if (fidp->fid_len < size) {
4360 fidp->fid_len = size;
4365 zfid = (zfid_short_t *)fidp;
4367 zfid->zf_len = size;
4369 for (i = 0; i < sizeof (zfid->zf_object); i++)
4370 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4372 /* Must have a non-zero generation number to distinguish from .zfs */
4375 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4376 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4378 if (size == LONG_FID_LEN) {
4379 uint64_t objsetid = dmu_objset_id(zsb->z_os);
4382 zlfid = (zfid_long_t *)fidp;
4384 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4385 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4387 /* XXX - this should be the generation number for the objset */
4388 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4389 zlfid->zf_setgen[i] = 0;
4395 EXPORT_SYMBOL(zfs_fid);
4399 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4401 znode_t *zp = ITOZ(ip);
4402 zfs_sb_t *zsb = ITOZSB(ip);
4404 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4408 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4413 EXPORT_SYMBOL(zfs_getsecattr);
4417 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4419 znode_t *zp = ITOZ(ip);
4420 zfs_sb_t *zsb = ITOZSB(ip);
4422 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4423 zilog_t *zilog = zsb->z_log;
4428 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4430 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
4431 zil_commit(zilog, 0);
4436 EXPORT_SYMBOL(zfs_setsecattr);
4438 #ifdef HAVE_UIO_ZEROCOPY
4440 * Tunable, both must be a power of 2.
4442 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4443 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4444 * an arcbuf for a partial block read
4446 int zcr_blksz_min = (1 << 10); /* 1K */
4447 int zcr_blksz_max = (1 << 17); /* 128K */
4451 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
4453 znode_t *zp = ITOZ(ip);
4454 zfs_sb_t *zsb = ITOZSB(ip);
4455 int max_blksz = zsb->z_max_blksz;
4456 uio_t *uio = &xuio->xu_uio;
4457 ssize_t size = uio->uio_resid;
4458 offset_t offset = uio->uio_loffset;
4463 int preamble, postamble;
4465 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4473 * Loan out an arc_buf for write if write size is bigger than
4474 * max_blksz, and the file's block size is also max_blksz.
4477 if (size < blksz || zp->z_blksz != blksz) {
4482 * Caller requests buffers for write before knowing where the
4483 * write offset might be (e.g. NFS TCP write).
4488 preamble = P2PHASE(offset, blksz);
4490 preamble = blksz - preamble;
4495 postamble = P2PHASE(size, blksz);
4498 fullblk = size / blksz;
4499 (void) dmu_xuio_init(xuio,
4500 (preamble != 0) + fullblk + (postamble != 0));
4503 * Have to fix iov base/len for partial buffers. They
4504 * currently represent full arc_buf's.
4507 /* data begins in the middle of the arc_buf */
4508 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4511 (void) dmu_xuio_add(xuio, abuf,
4512 blksz - preamble, preamble);
4515 for (i = 0; i < fullblk; i++) {
4516 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4519 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
4523 /* data ends in the middle of the arc_buf */
4524 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4527 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
4532 * Loan out an arc_buf for read if the read size is larger than
4533 * the current file block size. Block alignment is not
4534 * considered. Partial arc_buf will be loaned out for read.
4536 blksz = zp->z_blksz;
4537 if (blksz < zcr_blksz_min)
4538 blksz = zcr_blksz_min;
4539 if (blksz > zcr_blksz_max)
4540 blksz = zcr_blksz_max;
4541 /* avoid potential complexity of dealing with it */
4542 if (blksz > max_blksz) {
4547 maxsize = zp->z_size - uio->uio_loffset;
4561 uio->uio_extflg = UIO_XUIO;
4562 XUIO_XUZC_RW(xuio) = ioflag;
4569 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
4573 int ioflag = XUIO_XUZC_RW(xuio);
4575 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
4577 i = dmu_xuio_cnt(xuio);
4579 abuf = dmu_xuio_arcbuf(xuio, i);
4581 * if abuf == NULL, it must be a write buffer
4582 * that has been returned in zfs_write().
4585 dmu_return_arcbuf(abuf);
4586 ASSERT(abuf || ioflag == UIO_WRITE);
4589 dmu_xuio_fini(xuio);
4592 #endif /* HAVE_UIO_ZEROCOPY */
4594 #if defined(_KERNEL) && defined(HAVE_SPL)
4595 module_param(zfs_read_chunk_size, long, 0644);
4596 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");