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)) {
653 #ifdef HAVE_UIO_ZEROCOPY
655 * Pre-fault the pages to ensure slow (eg NFS) pages
657 * Skip this if uio contains loaned arc_buf.
659 if ((uio->uio_extflg == UIO_XUIO) &&
660 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
661 xuio = (xuio_t *)uio;
663 uio_prefaultpages(MIN(n, max_blksz), uio);
664 #endif /* HAVE_UIO_ZEROCOPY */
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 loff_t *pos = &(ctx->pos);
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)
2029 prefetch = zp->z_zn_prefetch;
2032 * Initialize the iterator cursor.
2036 * Start iteration from the beginning of the directory.
2038 zap_cursor_init(&zc, os, zp->z_id);
2041 * The offset is a serialized cursor.
2043 zap_cursor_init_serialized(&zc, os, zp->z_id, *pos);
2047 * Transform to file-system independent format
2054 * Special case `.', `..', and `.zfs'.
2057 (void) strcpy(zap.za_name, ".");
2058 zap.za_normalization_conflict = 0;
2060 } else if (*pos == 1) {
2061 (void) strcpy(zap.za_name, "..");
2062 zap.za_normalization_conflict = 0;
2064 } else if (*pos == 2 && zfs_show_ctldir(zp)) {
2065 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2066 zap.za_normalization_conflict = 0;
2067 objnum = ZFSCTL_INO_ROOT;
2072 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2073 if (error == ENOENT)
2080 * Allow multiple entries provided the first entry is
2081 * the object id. Non-zpl consumers may safely make
2082 * use of the additional space.
2084 * XXX: This should be a feature flag for compatibility
2086 if (zap.za_integer_length != 8 ||
2087 zap.za_num_integers == 0) {
2088 cmn_err(CE_WARN, "zap_readdir: bad directory "
2089 "entry, obj = %lld, offset = %lld, "
2090 "length = %d, num = %lld\n",
2091 (u_longlong_t)zp->z_id,
2093 zap.za_integer_length,
2094 (u_longlong_t)zap.za_num_integers);
2099 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2102 done = !dir_emit(ctx, zap.za_name, strlen(zap.za_name),
2103 objnum, ZFS_DIRENT_TYPE(zap.za_first_integer));
2107 /* Prefetch znode */
2109 dmu_prefetch(os, objnum, 0, 0);
2112 if (*pos > 2 || (*pos == 2 && !zfs_show_ctldir(zp))) {
2113 zap_cursor_advance(&zc);
2114 *pos = zap_cursor_serialize(&zc);
2119 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2122 zap_cursor_fini(&zc);
2123 if (error == ENOENT)
2126 ZFS_ACCESSTIME_STAMP(zsb, zp);
2127 zfs_inode_update(zp);
2134 EXPORT_SYMBOL(zfs_readdir);
2136 ulong_t zfs_fsync_sync_cnt = 4;
2139 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2141 znode_t *zp = ITOZ(ip);
2142 zfs_sb_t *zsb = ITOZSB(ip);
2144 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2146 if (zsb->z_os->os_sync != ZFS_SYNC_DISABLED) {
2149 zil_commit(zsb->z_log, zp->z_id);
2154 EXPORT_SYMBOL(zfs_fsync);
2158 * Get the requested file attributes and place them in the provided
2161 * IN: ip - inode of file.
2162 * vap - va_mask identifies requested attributes.
2163 * If ATTR_XVATTR set, then optional attrs are requested
2164 * flags - ATTR_NOACLCHECK (CIFS server context)
2165 * cr - credentials of caller.
2167 * OUT: vap - attribute values.
2169 * RETURN: 0 (always succeeds)
2173 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2175 znode_t *zp = ITOZ(ip);
2176 zfs_sb_t *zsb = ITOZSB(ip);
2179 uint64_t mtime[2], ctime[2];
2180 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2181 xoptattr_t *xoap = NULL;
2182 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2183 sa_bulk_attr_t bulk[2];
2189 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2191 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
2192 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
2194 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2200 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2201 * Also, if we are the owner don't bother, since owner should
2202 * always be allowed to read basic attributes of file.
2204 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2205 (vap->va_uid != crgetuid(cr))) {
2206 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2214 * Return all attributes. It's cheaper to provide the answer
2215 * than to determine whether we were asked the question.
2218 mutex_enter(&zp->z_lock);
2219 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2220 vap->va_mode = zp->z_mode;
2221 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2222 vap->va_nodeid = zp->z_id;
2223 if ((zp->z_id == zsb->z_root) && zfs_show_ctldir(zp))
2224 links = zp->z_links + 1;
2226 links = zp->z_links;
2227 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2228 vap->va_size = i_size_read(ip);
2229 vap->va_rdev = ip->i_rdev;
2230 vap->va_seq = ip->i_generation;
2233 * Add in any requested optional attributes and the create time.
2234 * Also set the corresponding bits in the returned attribute bitmap.
2236 if ((xoap = xva_getxoptattr(xvap)) != NULL && zsb->z_use_fuids) {
2237 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2239 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2240 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2243 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2244 xoap->xoa_readonly =
2245 ((zp->z_pflags & ZFS_READONLY) != 0);
2246 XVA_SET_RTN(xvap, XAT_READONLY);
2249 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2251 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2252 XVA_SET_RTN(xvap, XAT_SYSTEM);
2255 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2257 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2258 XVA_SET_RTN(xvap, XAT_HIDDEN);
2261 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2262 xoap->xoa_nounlink =
2263 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2264 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2267 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2268 xoap->xoa_immutable =
2269 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2270 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2273 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2274 xoap->xoa_appendonly =
2275 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2276 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2279 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2281 ((zp->z_pflags & ZFS_NODUMP) != 0);
2282 XVA_SET_RTN(xvap, XAT_NODUMP);
2285 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2287 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2288 XVA_SET_RTN(xvap, XAT_OPAQUE);
2291 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2292 xoap->xoa_av_quarantined =
2293 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2294 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2297 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2298 xoap->xoa_av_modified =
2299 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2300 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2303 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2304 S_ISREG(ip->i_mode)) {
2305 zfs_sa_get_scanstamp(zp, xvap);
2308 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2311 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zsb),
2312 times, sizeof (times));
2313 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2314 XVA_SET_RTN(xvap, XAT_CREATETIME);
2317 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2318 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2319 XVA_SET_RTN(xvap, XAT_REPARSE);
2321 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2322 xoap->xoa_generation = zp->z_gen;
2323 XVA_SET_RTN(xvap, XAT_GEN);
2326 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2328 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2329 XVA_SET_RTN(xvap, XAT_OFFLINE);
2332 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2334 ((zp->z_pflags & ZFS_SPARSE) != 0);
2335 XVA_SET_RTN(xvap, XAT_SPARSE);
2339 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2340 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2341 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2343 mutex_exit(&zp->z_lock);
2345 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2347 if (zp->z_blksz == 0) {
2349 * Block size hasn't been set; suggest maximal I/O transfers.
2351 vap->va_blksize = zsb->z_max_blksz;
2357 EXPORT_SYMBOL(zfs_getattr);
2360 * Get the basic file attributes and place them in the provided kstat
2361 * structure. The inode is assumed to be the authoritative source
2362 * for most of the attributes. However, the znode currently has the
2363 * authoritative atime, blksize, and block count.
2365 * IN: ip - inode of file.
2367 * OUT: sp - kstat values.
2369 * RETURN: 0 (always succeeds)
2373 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2375 znode_t *zp = ITOZ(ip);
2376 zfs_sb_t *zsb = ITOZSB(ip);
2381 mutex_enter(&zp->z_lock);
2383 generic_fillattr(ip, sp);
2384 ZFS_TIME_DECODE(&sp->atime, zp->z_atime);
2386 sa_object_size(zp->z_sa_hdl, (uint32_t *)&sp->blksize, &sp->blocks);
2387 if (unlikely(zp->z_blksz == 0)) {
2389 * Block size hasn't been set; suggest maximal I/O transfers.
2391 sp->blksize = zsb->z_max_blksz;
2394 mutex_exit(&zp->z_lock);
2400 EXPORT_SYMBOL(zfs_getattr_fast);
2403 * Set the file attributes to the values contained in the
2406 * IN: ip - inode of file to be modified.
2407 * vap - new attribute values.
2408 * If ATTR_XVATTR set, then optional attrs are being set
2409 * flags - ATTR_UTIME set if non-default time values provided.
2410 * - ATTR_NOACLCHECK (CIFS context only).
2411 * cr - credentials of caller.
2413 * RETURN: 0 if success
2414 * error code if failure
2417 * ip - ctime updated, mtime updated if size changed.
2421 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2423 znode_t *zp = ITOZ(ip);
2424 zfs_sb_t *zsb = ITOZSB(ip);
2428 xvattr_t *tmpxvattr;
2429 uint_t mask = vap->va_mask;
2433 uint64_t new_uid, new_gid;
2435 uint64_t mtime[2], ctime[2];
2437 int need_policy = FALSE;
2439 zfs_fuid_info_t *fuidp = NULL;
2440 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2443 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2444 boolean_t fuid_dirtied = B_FALSE;
2445 sa_bulk_attr_t *bulk, *xattr_bulk;
2446 int count = 0, xattr_count = 0;
2457 * Make sure that if we have ephemeral uid/gid or xvattr specified
2458 * that file system is at proper version level
2461 if (zsb->z_use_fuids == B_FALSE &&
2462 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2463 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2464 (mask & ATTR_XVATTR))) {
2469 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2474 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2480 * If this is an xvattr_t, then get a pointer to the structure of
2481 * optional attributes. If this is NULL, then we have a vattr_t.
2483 xoap = xva_getxoptattr(xvap);
2485 tmpxvattr = kmem_alloc(sizeof(xvattr_t), KM_SLEEP);
2486 xva_init(tmpxvattr);
2488 bulk = kmem_alloc(sizeof(sa_bulk_attr_t) * 7, KM_SLEEP);
2489 xattr_bulk = kmem_alloc(sizeof(sa_bulk_attr_t) * 7, KM_SLEEP);
2492 * Immutable files can only alter immutable bit and atime
2494 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2495 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2496 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2501 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2507 * Verify timestamps doesn't overflow 32 bits.
2508 * ZFS can handle large timestamps, but 32bit syscalls can't
2509 * handle times greater than 2039. This check should be removed
2510 * once large timestamps are fully supported.
2512 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2513 if (((mask & ATTR_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2514 ((mask & ATTR_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2524 /* Can this be moved to before the top label? */
2525 if (zfs_is_readonly(zsb)) {
2531 * First validate permissions
2534 if (mask & ATTR_SIZE) {
2535 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2539 truncate_setsize(ip, vap->va_size);
2542 * XXX - Note, we are not providing any open
2543 * mode flags here (like FNDELAY), so we may
2544 * block if there are locks present... this
2545 * should be addressed in openat().
2547 /* XXX - would it be OK to generate a log record here? */
2548 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2553 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2554 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2555 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2556 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2557 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2558 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2559 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2560 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2561 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2565 if (mask & (ATTR_UID|ATTR_GID)) {
2566 int idmask = (mask & (ATTR_UID|ATTR_GID));
2571 * NOTE: even if a new mode is being set,
2572 * we may clear S_ISUID/S_ISGID bits.
2575 if (!(mask & ATTR_MODE))
2576 vap->va_mode = zp->z_mode;
2579 * Take ownership or chgrp to group we are a member of
2582 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
2583 take_group = (mask & ATTR_GID) &&
2584 zfs_groupmember(zsb, vap->va_gid, cr);
2587 * If both ATTR_UID and ATTR_GID are set then take_owner and
2588 * take_group must both be set in order to allow taking
2591 * Otherwise, send the check through secpolicy_vnode_setattr()
2595 if (((idmask == (ATTR_UID|ATTR_GID)) &&
2596 take_owner && take_group) ||
2597 ((idmask == ATTR_UID) && take_owner) ||
2598 ((idmask == ATTR_GID) && take_group)) {
2599 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2600 skipaclchk, cr) == 0) {
2602 * Remove setuid/setgid for non-privileged users
2604 (void) secpolicy_setid_clear(vap, cr);
2605 trim_mask = (mask & (ATTR_UID|ATTR_GID));
2614 mutex_enter(&zp->z_lock);
2615 oldva.va_mode = zp->z_mode;
2616 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2617 if (mask & ATTR_XVATTR) {
2619 * Update xvattr mask to include only those attributes
2620 * that are actually changing.
2622 * the bits will be restored prior to actually setting
2623 * the attributes so the caller thinks they were set.
2625 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2626 if (xoap->xoa_appendonly !=
2627 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2630 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2631 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
2635 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2636 if (xoap->xoa_nounlink !=
2637 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2640 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2641 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
2645 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2646 if (xoap->xoa_immutable !=
2647 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2650 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2651 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
2655 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2656 if (xoap->xoa_nodump !=
2657 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2660 XVA_CLR_REQ(xvap, XAT_NODUMP);
2661 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
2665 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2666 if (xoap->xoa_av_modified !=
2667 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2670 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2671 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
2675 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2676 if ((!S_ISREG(ip->i_mode) &&
2677 xoap->xoa_av_quarantined) ||
2678 xoap->xoa_av_quarantined !=
2679 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2682 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2683 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
2687 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2688 mutex_exit(&zp->z_lock);
2693 if (need_policy == FALSE &&
2694 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2695 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2700 mutex_exit(&zp->z_lock);
2702 if (mask & ATTR_MODE) {
2703 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2704 err = secpolicy_setid_setsticky_clear(ip, vap,
2709 trim_mask |= ATTR_MODE;
2717 * If trim_mask is set then take ownership
2718 * has been granted or write_acl is present and user
2719 * has the ability to modify mode. In that case remove
2720 * UID|GID and or MODE from mask so that
2721 * secpolicy_vnode_setattr() doesn't revoke it.
2725 saved_mask = vap->va_mask;
2726 vap->va_mask &= ~trim_mask;
2728 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
2729 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2734 vap->va_mask |= saved_mask;
2738 * secpolicy_vnode_setattr, or take ownership may have
2741 mask = vap->va_mask;
2743 if ((mask & (ATTR_UID | ATTR_GID))) {
2744 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
2745 &xattr_obj, sizeof (xattr_obj));
2747 if (err == 0 && xattr_obj) {
2748 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
2752 if (mask & ATTR_UID) {
2753 new_uid = zfs_fuid_create(zsb,
2754 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2755 if (new_uid != zp->z_uid &&
2756 zfs_fuid_overquota(zsb, B_FALSE, new_uid)) {
2764 if (mask & ATTR_GID) {
2765 new_gid = zfs_fuid_create(zsb, (uint64_t)vap->va_gid,
2766 cr, ZFS_GROUP, &fuidp);
2767 if (new_gid != zp->z_gid &&
2768 zfs_fuid_overquota(zsb, B_TRUE, new_gid)) {
2776 tx = dmu_tx_create(zsb->z_os);
2778 if (mask & ATTR_MODE) {
2779 uint64_t pmode = zp->z_mode;
2781 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2783 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
2785 mutex_enter(&zp->z_lock);
2786 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2788 * Are we upgrading ACL from old V0 format
2791 if (zsb->z_version >= ZPL_VERSION_FUID &&
2792 zfs_znode_acl_version(zp) ==
2793 ZFS_ACL_VERSION_INITIAL) {
2794 dmu_tx_hold_free(tx, acl_obj, 0,
2796 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2797 0, aclp->z_acl_bytes);
2799 dmu_tx_hold_write(tx, acl_obj, 0,
2802 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2803 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2804 0, aclp->z_acl_bytes);
2806 mutex_exit(&zp->z_lock);
2807 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2809 if ((mask & ATTR_XVATTR) &&
2810 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2811 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2813 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2817 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
2820 fuid_dirtied = zsb->z_fuid_dirty;
2822 zfs_fuid_txhold(zsb, tx);
2824 zfs_sa_upgrade_txholds(tx, zp);
2826 err = dmu_tx_assign(tx, TXG_NOWAIT);
2828 if (err == ERESTART)
2835 * Set each attribute requested.
2836 * We group settings according to the locks they need to acquire.
2838 * Note: you cannot set ctime directly, although it will be
2839 * updated as a side-effect of calling this function.
2843 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2844 mutex_enter(&zp->z_acl_lock);
2845 mutex_enter(&zp->z_lock);
2847 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
2848 &zp->z_pflags, sizeof (zp->z_pflags));
2851 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2852 mutex_enter(&attrzp->z_acl_lock);
2853 mutex_enter(&attrzp->z_lock);
2854 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2855 SA_ZPL_FLAGS(zsb), NULL, &attrzp->z_pflags,
2856 sizeof (attrzp->z_pflags));
2859 if (mask & (ATTR_UID|ATTR_GID)) {
2861 if (mask & ATTR_UID) {
2862 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zsb), NULL,
2863 &new_uid, sizeof (new_uid));
2864 zp->z_uid = new_uid;
2866 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2867 SA_ZPL_UID(zsb), NULL, &new_uid,
2869 attrzp->z_uid = new_uid;
2873 if (mask & ATTR_GID) {
2874 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zsb),
2875 NULL, &new_gid, sizeof (new_gid));
2876 zp->z_gid = new_gid;
2878 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2879 SA_ZPL_GID(zsb), NULL, &new_gid,
2881 attrzp->z_gid = new_gid;
2884 if (!(mask & ATTR_MODE)) {
2885 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb),
2886 NULL, &new_mode, sizeof (new_mode));
2887 new_mode = zp->z_mode;
2889 err = zfs_acl_chown_setattr(zp);
2892 err = zfs_acl_chown_setattr(attrzp);
2897 if (mask & ATTR_MODE) {
2898 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb), NULL,
2899 &new_mode, sizeof (new_mode));
2900 zp->z_mode = new_mode;
2901 ASSERT3P(aclp, !=, NULL);
2902 err = zfs_aclset_common(zp, aclp, cr, tx);
2904 if (zp->z_acl_cached)
2905 zfs_acl_free(zp->z_acl_cached);
2906 zp->z_acl_cached = aclp;
2911 if (mask & ATTR_ATIME) {
2912 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
2913 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zsb), NULL,
2914 &zp->z_atime, sizeof (zp->z_atime));
2917 if (mask & ATTR_MTIME) {
2918 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
2919 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL,
2920 mtime, sizeof (mtime));
2923 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2924 if (mask & ATTR_SIZE && !(mask & ATTR_MTIME)) {
2925 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb),
2926 NULL, mtime, sizeof (mtime));
2927 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
2928 &ctime, sizeof (ctime));
2929 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
2931 } else if (mask != 0) {
2932 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
2933 &ctime, sizeof (ctime));
2934 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
2937 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2938 SA_ZPL_CTIME(zsb), NULL,
2939 &ctime, sizeof (ctime));
2940 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
2941 mtime, ctime, B_TRUE);
2945 * Do this after setting timestamps to prevent timestamp
2946 * update from toggling bit
2949 if (xoap && (mask & ATTR_XVATTR)) {
2952 * restore trimmed off masks
2953 * so that return masks can be set for caller.
2956 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
2957 XVA_SET_REQ(xvap, XAT_APPENDONLY);
2959 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
2960 XVA_SET_REQ(xvap, XAT_NOUNLINK);
2962 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
2963 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
2965 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
2966 XVA_SET_REQ(xvap, XAT_NODUMP);
2968 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
2969 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
2971 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
2972 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
2975 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2976 ASSERT(S_ISREG(ip->i_mode));
2978 zfs_xvattr_set(zp, xvap, tx);
2982 zfs_fuid_sync(zsb, tx);
2985 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2987 mutex_exit(&zp->z_lock);
2988 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2989 mutex_exit(&zp->z_acl_lock);
2992 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2993 mutex_exit(&attrzp->z_acl_lock);
2994 mutex_exit(&attrzp->z_lock);
2997 if (err == 0 && attrzp) {
2998 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3009 zfs_fuid_info_free(fuidp);
3015 if (err == ERESTART)
3018 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3020 zfs_inode_update(zp);
3024 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3025 zil_commit(zilog, 0);
3028 kmem_free(xattr_bulk, sizeof(sa_bulk_attr_t) * 7);
3029 kmem_free(bulk, sizeof(sa_bulk_attr_t) * 7);
3030 kmem_free(tmpxvattr, sizeof(xvattr_t));
3034 EXPORT_SYMBOL(zfs_setattr);
3036 typedef struct zfs_zlock {
3037 krwlock_t *zl_rwlock; /* lock we acquired */
3038 znode_t *zl_znode; /* znode we held */
3039 struct zfs_zlock *zl_next; /* next in list */
3043 * Drop locks and release vnodes that were held by zfs_rename_lock().
3046 zfs_rename_unlock(zfs_zlock_t **zlpp)
3050 while ((zl = *zlpp) != NULL) {
3051 if (zl->zl_znode != NULL)
3052 iput(ZTOI(zl->zl_znode));
3053 rw_exit(zl->zl_rwlock);
3054 *zlpp = zl->zl_next;
3055 kmem_free(zl, sizeof (*zl));
3060 * Search back through the directory tree, using the ".." entries.
3061 * Lock each directory in the chain to prevent concurrent renames.
3062 * Fail any attempt to move a directory into one of its own descendants.
3063 * XXX - z_parent_lock can overlap with map or grow locks
3066 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3070 uint64_t rootid = ZTOZSB(zp)->z_root;
3071 uint64_t oidp = zp->z_id;
3072 krwlock_t *rwlp = &szp->z_parent_lock;
3073 krw_t rw = RW_WRITER;
3076 * First pass write-locks szp and compares to zp->z_id.
3077 * Later passes read-lock zp and compare to zp->z_parent.
3080 if (!rw_tryenter(rwlp, rw)) {
3082 * Another thread is renaming in this path.
3083 * Note that if we are a WRITER, we don't have any
3084 * parent_locks held yet.
3086 if (rw == RW_READER && zp->z_id > szp->z_id) {
3088 * Drop our locks and restart
3090 zfs_rename_unlock(&zl);
3094 rwlp = &szp->z_parent_lock;
3099 * Wait for other thread to drop its locks
3105 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3106 zl->zl_rwlock = rwlp;
3107 zl->zl_znode = NULL;
3108 zl->zl_next = *zlpp;
3111 if (oidp == szp->z_id) /* We're a descendant of szp */
3114 if (oidp == rootid) /* We've hit the top */
3117 if (rw == RW_READER) { /* i.e. not the first pass */
3118 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3123 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3124 &oidp, sizeof (oidp));
3125 rwlp = &zp->z_parent_lock;
3128 } while (zp->z_id != sdzp->z_id);
3134 * Move an entry from the provided source directory to the target
3135 * directory. Change the entry name as indicated.
3137 * IN: sdip - Source directory containing the "old entry".
3138 * snm - Old entry name.
3139 * tdip - Target directory to contain the "new entry".
3140 * tnm - New entry name.
3141 * cr - credentials of caller.
3142 * flags - case flags
3144 * RETURN: 0 if success
3145 * error code if failure
3148 * sdip,tdip - ctime|mtime updated
3152 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3153 cred_t *cr, int flags)
3155 znode_t *tdzp, *szp, *tzp;
3156 znode_t *sdzp = ITOZ(sdip);
3157 zfs_sb_t *zsb = ITOZSB(sdip);
3159 zfs_dirlock_t *sdl, *tdl;
3162 int cmp, serr, terr;
3167 ZFS_VERIFY_ZP(sdzp);
3170 if (tdip->i_sb != sdip->i_sb) {
3176 ZFS_VERIFY_ZP(tdzp);
3177 if (zsb->z_utf8 && u8_validate(tnm,
3178 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3183 if (flags & FIGNORECASE)
3192 * This is to prevent the creation of links into attribute space
3193 * by renaming a linked file into/outof an attribute directory.
3194 * See the comment in zfs_link() for why this is considered bad.
3196 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3202 * Lock source and target directory entries. To prevent deadlock,
3203 * a lock ordering must be defined. We lock the directory with
3204 * the smallest object id first, or if it's a tie, the one with
3205 * the lexically first name.
3207 if (sdzp->z_id < tdzp->z_id) {
3209 } else if (sdzp->z_id > tdzp->z_id) {
3213 * First compare the two name arguments without
3214 * considering any case folding.
3216 int nofold = (zsb->z_norm & ~U8_TEXTPREP_TOUPPER);
3218 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3219 ASSERT(error == 0 || !zsb->z_utf8);
3222 * POSIX: "If the old argument and the new argument
3223 * both refer to links to the same existing file,
3224 * the rename() function shall return successfully
3225 * and perform no other action."
3231 * If the file system is case-folding, then we may
3232 * have some more checking to do. A case-folding file
3233 * system is either supporting mixed case sensitivity
3234 * access or is completely case-insensitive. Note
3235 * that the file system is always case preserving.
3237 * In mixed sensitivity mode case sensitive behavior
3238 * is the default. FIGNORECASE must be used to
3239 * explicitly request case insensitive behavior.
3241 * If the source and target names provided differ only
3242 * by case (e.g., a request to rename 'tim' to 'Tim'),
3243 * we will treat this as a special case in the
3244 * case-insensitive mode: as long as the source name
3245 * is an exact match, we will allow this to proceed as
3246 * a name-change request.
3248 if ((zsb->z_case == ZFS_CASE_INSENSITIVE ||
3249 (zsb->z_case == ZFS_CASE_MIXED &&
3250 flags & FIGNORECASE)) &&
3251 u8_strcmp(snm, tnm, 0, zsb->z_norm, U8_UNICODE_LATEST,
3254 * case preserving rename request, require exact
3263 * If the source and destination directories are the same, we should
3264 * grab the z_name_lock of that directory only once.
3268 rw_enter(&sdzp->z_name_lock, RW_READER);
3272 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3273 ZEXISTS | zflg, NULL, NULL);
3274 terr = zfs_dirent_lock(&tdl,
3275 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3277 terr = zfs_dirent_lock(&tdl,
3278 tdzp, tnm, &tzp, zflg, NULL, NULL);
3279 serr = zfs_dirent_lock(&sdl,
3280 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3286 * Source entry invalid or not there.
3289 zfs_dirent_unlock(tdl);
3295 rw_exit(&sdzp->z_name_lock);
3297 if (strcmp(snm, "..") == 0)
3303 zfs_dirent_unlock(sdl);
3307 rw_exit(&sdzp->z_name_lock);
3309 if (strcmp(tnm, "..") == 0)
3316 * Must have write access at the source to remove the old entry
3317 * and write access at the target to create the new entry.
3318 * Note that if target and source are the same, this can be
3319 * done in a single check.
3322 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3325 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3327 * Check to make sure rename is valid.
3328 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3330 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3335 * Does target exist?
3339 * Source and target must be the same type.
3341 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3342 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3347 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3353 * POSIX dictates that when the source and target
3354 * entries refer to the same file object, rename
3355 * must do nothing and exit without error.
3357 if (szp->z_id == tzp->z_id) {
3363 tx = dmu_tx_create(zsb->z_os);
3364 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3365 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3366 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3367 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3369 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3370 zfs_sa_upgrade_txholds(tx, tdzp);
3373 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3374 zfs_sa_upgrade_txholds(tx, tzp);
3377 zfs_sa_upgrade_txholds(tx, szp);
3378 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
3379 error = dmu_tx_assign(tx, TXG_NOWAIT);
3382 zfs_rename_unlock(&zl);
3383 zfs_dirent_unlock(sdl);
3384 zfs_dirent_unlock(tdl);
3387 rw_exit(&sdzp->z_name_lock);
3392 if (error == ERESTART) {
3402 if (tzp) /* Attempt to remove the existing target */
3403 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3406 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3408 szp->z_pflags |= ZFS_AV_MODIFIED;
3410 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zsb),
3411 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3414 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3416 zfs_log_rename(zilog, tx, TX_RENAME |
3417 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3418 sdl->dl_name, tdzp, tdl->dl_name, szp);
3421 * At this point, we have successfully created
3422 * the target name, but have failed to remove
3423 * the source name. Since the create was done
3424 * with the ZRENAMING flag, there are
3425 * complications; for one, the link count is
3426 * wrong. The easiest way to deal with this
3427 * is to remove the newly created target, and
3428 * return the original error. This must
3429 * succeed; fortunately, it is very unlikely to
3430 * fail, since we just created it.
3432 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3433 ZRENAMING, NULL), ==, 0);
3441 zfs_rename_unlock(&zl);
3443 zfs_dirent_unlock(sdl);
3444 zfs_dirent_unlock(tdl);
3446 zfs_inode_update(sdzp);
3448 rw_exit(&sdzp->z_name_lock);
3451 zfs_inode_update(tdzp);
3453 zfs_inode_update(szp);
3456 zfs_inode_update(tzp);
3460 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3461 zil_commit(zilog, 0);
3466 EXPORT_SYMBOL(zfs_rename);
3469 * Insert the indicated symbolic reference entry into the directory.
3471 * IN: dip - Directory to contain new symbolic link.
3472 * link - Name for new symlink entry.
3473 * vap - Attributes of new entry.
3474 * target - Target path of new symlink.
3476 * cr - credentials of caller.
3477 * flags - case flags
3479 * RETURN: 0 if success
3480 * error code if failure
3483 * dip - ctime|mtime updated
3487 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
3488 struct inode **ipp, cred_t *cr, int flags)
3490 znode_t *zp, *dzp = ITOZ(dip);
3493 zfs_sb_t *zsb = ITOZSB(dip);
3495 uint64_t len = strlen(link);
3498 zfs_acl_ids_t acl_ids;
3499 boolean_t fuid_dirtied;
3500 uint64_t txtype = TX_SYMLINK;
3502 ASSERT(S_ISLNK(vap->va_mode));
3508 if (zsb->z_utf8 && u8_validate(name, strlen(name),
3509 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3513 if (flags & FIGNORECASE)
3516 if (len > MAXPATHLEN) {
3518 return (ENAMETOOLONG);
3521 if ((error = zfs_acl_ids_create(dzp, 0,
3522 vap, cr, NULL, &acl_ids)) != 0) {
3530 * Attempt to lock directory; fail if entry already exists.
3532 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3534 zfs_acl_ids_free(&acl_ids);
3539 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3540 zfs_acl_ids_free(&acl_ids);
3541 zfs_dirent_unlock(dl);
3546 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
3547 zfs_acl_ids_free(&acl_ids);
3548 zfs_dirent_unlock(dl);
3552 tx = dmu_tx_create(zsb->z_os);
3553 fuid_dirtied = zsb->z_fuid_dirty;
3554 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3555 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3556 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3557 ZFS_SA_BASE_ATTR_SIZE + len);
3558 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3559 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3560 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3561 acl_ids.z_aclp->z_acl_bytes);
3564 zfs_fuid_txhold(zsb, tx);
3565 error = dmu_tx_assign(tx, TXG_NOWAIT);
3567 zfs_dirent_unlock(dl);
3568 if (error == ERESTART) {
3573 zfs_acl_ids_free(&acl_ids);
3580 * Create a new object for the symlink.
3581 * for version 4 ZPL datsets the symlink will be an SA attribute
3583 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3586 zfs_fuid_sync(zsb, tx);
3588 mutex_enter(&zp->z_lock);
3590 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zsb),
3593 zfs_sa_symlink(zp, link, len, tx);
3594 mutex_exit(&zp->z_lock);
3597 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
3598 &zp->z_size, sizeof (zp->z_size), tx);
3600 * Insert the new object into the directory.
3602 (void) zfs_link_create(dl, zp, tx, ZNEW);
3604 if (flags & FIGNORECASE)
3606 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3608 zfs_inode_update(dzp);
3609 zfs_inode_update(zp);
3611 zfs_acl_ids_free(&acl_ids);
3615 zfs_dirent_unlock(dl);
3619 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3620 zil_commit(zilog, 0);
3625 EXPORT_SYMBOL(zfs_symlink);
3628 * Return, in the buffer contained in the provided uio structure,
3629 * the symbolic path referred to by ip.
3631 * IN: ip - inode of symbolic link
3632 * uio - structure to contain the link path.
3633 * cr - credentials of caller.
3635 * RETURN: 0 if success
3636 * error code if failure
3639 * ip - atime updated
3643 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
3645 znode_t *zp = ITOZ(ip);
3646 zfs_sb_t *zsb = ITOZSB(ip);
3652 mutex_enter(&zp->z_lock);
3654 error = sa_lookup_uio(zp->z_sa_hdl,
3655 SA_ZPL_SYMLINK(zsb), uio);
3657 error = zfs_sa_readlink(zp, uio);
3658 mutex_exit(&zp->z_lock);
3660 ZFS_ACCESSTIME_STAMP(zsb, zp);
3661 zfs_inode_update(zp);
3665 EXPORT_SYMBOL(zfs_readlink);
3668 * Insert a new entry into directory tdip referencing sip.
3670 * IN: tdip - Directory to contain new entry.
3671 * sip - inode of new entry.
3672 * name - name of new entry.
3673 * cr - credentials of caller.
3675 * RETURN: 0 if success
3676 * error code if failure
3679 * tdip - ctime|mtime updated
3680 * sip - ctime updated
3684 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr)
3686 znode_t *dzp = ITOZ(tdip);
3688 zfs_sb_t *zsb = ITOZSB(tdip);
3697 ASSERT(S_ISDIR(tdip->i_mode));
3704 * POSIX dictates that we return EPERM here.
3705 * Better choices include ENOTSUP or EISDIR.
3707 if (S_ISDIR(sip->i_mode)) {
3712 if (sip->i_sb != tdip->i_sb) {
3720 /* Prevent links to .zfs/shares files */
3722 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zsb),
3723 &parent, sizeof (uint64_t))) != 0) {
3727 if (parent == zsb->z_shares_dir) {
3732 if (zsb->z_utf8 && u8_validate(name,
3733 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3737 #ifdef HAVE_PN_UTILS
3738 if (flags & FIGNORECASE)
3740 #endif /* HAVE_PN_UTILS */
3743 * We do not support links between attributes and non-attributes
3744 * because of the potential security risk of creating links
3745 * into "normal" file space in order to circumvent restrictions
3746 * imposed in attribute space.
3748 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3753 owner = zfs_fuid_map_id(zsb, szp->z_uid, cr, ZFS_OWNER);
3754 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3759 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3766 * Attempt to lock directory; fail if entry already exists.
3768 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3774 tx = dmu_tx_create(zsb->z_os);
3775 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3776 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3777 zfs_sa_upgrade_txholds(tx, szp);
3778 zfs_sa_upgrade_txholds(tx, dzp);
3779 error = dmu_tx_assign(tx, TXG_NOWAIT);
3781 zfs_dirent_unlock(dl);
3782 if (error == ERESTART) {
3792 error = zfs_link_create(dl, szp, tx, 0);
3795 uint64_t txtype = TX_LINK;
3796 #ifdef HAVE_PN_UTILS
3797 if (flags & FIGNORECASE)
3799 #endif /* HAVE_PN_UTILS */
3800 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3805 zfs_dirent_unlock(dl);
3807 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3808 zil_commit(zilog, 0);
3810 zfs_inode_update(dzp);
3811 zfs_inode_update(szp);
3815 EXPORT_SYMBOL(zfs_link);
3818 zfs_putpage_commit_cb(void *arg, int error)
3820 struct page *pp = arg;
3823 __set_page_dirty_nobuffers(pp);
3825 if (error != ECANCELED)
3831 end_page_writeback(pp);
3835 * Push a page out to disk, once the page is on stable storage the
3836 * registered commit callback will be run as notification of completion.
3838 * IN: ip - page mapped for inode.
3839 * pp - page to push (page is locked)
3840 * wbc - writeback control data
3842 * RETURN: 0 if success
3843 * error code if failure
3846 * ip - ctime|mtime updated
3850 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
3852 znode_t *zp = ITOZ(ip);
3853 zfs_sb_t *zsb = ITOZSB(ip);
3861 uint64_t mtime[2], ctime[2];
3862 sa_bulk_attr_t bulk[3];
3869 ASSERT(PageLocked(pp));
3871 pgoff = page_offset(pp); /* Page byte-offset in file */
3872 offset = i_size_read(ip); /* File length in bytes */
3873 pglen = MIN(PAGE_CACHE_SIZE, /* Page length in bytes */
3874 P2ROUNDUP(offset, PAGE_CACHE_SIZE)-pgoff);
3876 /* Page is beyond end of file */
3877 if (pgoff >= offset) {
3883 /* Truncate page length to end of file */
3884 if (pgoff + pglen > offset)
3885 pglen = offset - pgoff;
3889 * FIXME: Allow mmap writes past its quota. The correct fix
3890 * is to register a page_mkwrite() handler to count the page
3891 * against its quota when it is about to be dirtied.
3893 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
3894 zfs_owner_overquota(zsb, zp, B_TRUE)) {
3899 set_page_writeback(pp);
3902 rl = zfs_range_lock(zp, pgoff, pglen, RL_WRITER);
3903 tx = dmu_tx_create(zsb->z_os);
3905 sync = ((zsb->z_os->os_sync == ZFS_SYNC_ALWAYS) ||
3906 (wbc->sync_mode == WB_SYNC_ALL));
3908 dmu_tx_callback_register(tx, zfs_putpage_commit_cb, pp);
3910 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
3912 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3913 zfs_sa_upgrade_txholds(tx, zp);
3914 err = dmu_tx_assign(tx, TXG_NOWAIT);
3916 if (err == ERESTART)
3919 /* Will call all registered commit callbacks */
3923 * For the synchronous case the commit callback must be
3924 * explicitly called because there is no registered callback.
3927 zfs_putpage_commit_cb(pp, ECANCELED);
3929 zfs_range_unlock(rl);
3935 ASSERT3U(pglen, <=, PAGE_CACHE_SIZE);
3936 dmu_write(zsb->z_os, zp->z_id, pgoff, pglen, va, tx);
3939 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
3940 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
3941 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zsb), NULL, &zp->z_pflags, 8);
3943 /* Preserve the mtime and ctime provided by the inode */
3944 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
3945 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
3946 zp->z_atime_dirty = 0;
3949 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
3951 zfs_log_write(zsb->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0);
3954 zfs_range_unlock(rl);
3957 zil_commit(zsb->z_log, zp->z_id);
3958 zfs_putpage_commit_cb(pp, err);
3966 * Update the system attributes when the inode has been dirtied. For the
3967 * moment we're conservative and only update the atime, mtime, and ctime.
3970 zfs_dirty_inode(struct inode *ip, int flags)
3972 znode_t *zp = ITOZ(ip);
3973 zfs_sb_t *zsb = ITOZSB(ip);
3975 uint64_t atime[2], mtime[2], ctime[2];
3976 sa_bulk_attr_t bulk[3];
3983 tx = dmu_tx_create(zsb->z_os);
3985 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3986 zfs_sa_upgrade_txholds(tx, zp);
3988 error = dmu_tx_assign(tx, TXG_WAIT);
3994 mutex_enter(&zp->z_lock);
3995 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zsb), NULL, &atime, 16);
3996 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
3997 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
3999 /* Preserve the mtime and ctime provided by the inode */
4000 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4001 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4002 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4003 zp->z_atime_dirty = 0;
4005 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4006 mutex_exit(&zp->z_lock);
4013 EXPORT_SYMBOL(zfs_dirty_inode);
4017 zfs_inactive(struct inode *ip)
4019 znode_t *zp = ITOZ(ip);
4020 zfs_sb_t *zsb = ITOZSB(ip);
4023 if (zfsctl_is_node(ip)) {
4024 zfsctl_inode_inactive(ip);
4028 rw_enter(&zsb->z_teardown_inactive_lock, RW_READER);
4029 if (zp->z_sa_hdl == NULL) {
4030 rw_exit(&zsb->z_teardown_inactive_lock);
4034 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4035 dmu_tx_t *tx = dmu_tx_create(zsb->z_os);
4037 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4038 zfs_sa_upgrade_txholds(tx, zp);
4039 error = dmu_tx_assign(tx, TXG_WAIT);
4043 mutex_enter(&zp->z_lock);
4044 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zsb),
4045 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4046 zp->z_atime_dirty = 0;
4047 mutex_exit(&zp->z_lock);
4053 rw_exit(&zsb->z_teardown_inactive_lock);
4055 EXPORT_SYMBOL(zfs_inactive);
4058 * Bounds-check the seek operation.
4060 * IN: ip - inode seeking within
4061 * ooff - old file offset
4062 * noffp - pointer to new file offset
4063 * ct - caller context
4065 * RETURN: 0 if success
4066 * EINVAL if new offset invalid
4070 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
4072 if (S_ISDIR(ip->i_mode))
4074 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4076 EXPORT_SYMBOL(zfs_seek);
4079 * Fill pages with data from the disk.
4082 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
4084 znode_t *zp = ITOZ(ip);
4085 zfs_sb_t *zsb = ITOZSB(ip);
4087 struct page *cur_pp;
4088 u_offset_t io_off, total;
4095 io_len = nr_pages << PAGE_CACHE_SHIFT;
4096 i_size = i_size_read(ip);
4097 io_off = page_offset(pl[0]);
4099 if (io_off + io_len > i_size)
4100 io_len = i_size - io_off;
4103 * Iterate over list of pages and read each page individually.
4107 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4111 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4115 /* convert checksum errors into IO errors */
4120 cur_pp = pl[++page_idx];
4127 * Uses zfs_fillpage to read data from the file and fill the pages.
4129 * IN: ip - inode of file to get data from.
4130 * pl - list of pages to read
4131 * nr_pages - number of pages to read
4133 * RETURN: 0 if success
4134 * error code if failure
4137 * vp - atime updated
4141 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
4143 znode_t *zp = ITOZ(ip);
4144 zfs_sb_t *zsb = ITOZSB(ip);
4153 err = zfs_fillpage(ip, pl, nr_pages);
4156 ZFS_ACCESSTIME_STAMP(zsb, zp);
4161 EXPORT_SYMBOL(zfs_getpage);
4164 * Check ZFS specific permissions to memory map a section of a file.
4166 * IN: ip - inode of the file to mmap
4168 * addrp - start address in memory region
4169 * len - length of memory region
4170 * vm_flags- address flags
4172 * RETURN: 0 if success
4173 * error code if failure
4177 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4178 unsigned long vm_flags)
4180 znode_t *zp = ITOZ(ip);
4181 zfs_sb_t *zsb = ITOZSB(ip);
4186 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4187 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4192 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4193 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4198 if (off < 0 || len > MAXOFFSET_T - off) {
4206 EXPORT_SYMBOL(zfs_map);
4209 * convoff - converts the given data (start, whence) to the
4213 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4218 if ((lckdat->l_whence == 2) || (whence == 2)) {
4219 if ((error = zfs_getattr(ip, &vap, 0, CRED()) != 0))
4223 switch (lckdat->l_whence) {
4225 lckdat->l_start += offset;
4228 lckdat->l_start += vap.va_size;
4236 if (lckdat->l_start < 0)
4241 lckdat->l_start -= offset;
4244 lckdat->l_start -= vap.va_size;
4252 lckdat->l_whence = (short)whence;
4257 * Free or allocate space in a file. Currently, this function only
4258 * supports the `F_FREESP' command. However, this command is somewhat
4259 * misnamed, as its functionality includes the ability to allocate as
4260 * well as free space.
4262 * IN: ip - inode of file to free data in.
4263 * cmd - action to take (only F_FREESP supported).
4264 * bfp - section of file to free/alloc.
4265 * flag - current file open mode flags.
4266 * offset - current file offset.
4267 * cr - credentials of caller [UNUSED].
4269 * RETURN: 0 if success
4270 * error code if failure
4273 * ip - ctime|mtime updated
4277 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4278 offset_t offset, cred_t *cr)
4280 znode_t *zp = ITOZ(ip);
4281 zfs_sb_t *zsb = ITOZSB(ip);
4288 if (cmd != F_FREESP) {
4293 if ((error = convoff(ip, bfp, 0, offset))) {
4298 if (bfp->l_len < 0) {
4304 * Permissions aren't checked on Solaris because on this OS
4305 * zfs_space() can only be called with an opened file handle.
4306 * On Linux we can get here through truncate_range() which
4307 * operates directly on inodes, so we need to check access rights.
4309 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4315 len = bfp->l_len; /* 0 means from off to end of file */
4317 error = zfs_freesp(zp, off, len, flag, TRUE);
4322 EXPORT_SYMBOL(zfs_space);
4326 zfs_fid(struct inode *ip, fid_t *fidp)
4328 znode_t *zp = ITOZ(ip);
4329 zfs_sb_t *zsb = ITOZSB(ip);
4332 uint64_t object = zp->z_id;
4339 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zsb),
4340 &gen64, sizeof (uint64_t))) != 0) {
4345 gen = (uint32_t)gen64;
4347 size = (zsb->z_parent != zsb) ? LONG_FID_LEN : SHORT_FID_LEN;
4348 if (fidp->fid_len < size) {
4349 fidp->fid_len = size;
4354 zfid = (zfid_short_t *)fidp;
4356 zfid->zf_len = size;
4358 for (i = 0; i < sizeof (zfid->zf_object); i++)
4359 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4361 /* Must have a non-zero generation number to distinguish from .zfs */
4364 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4365 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4367 if (size == LONG_FID_LEN) {
4368 uint64_t objsetid = dmu_objset_id(zsb->z_os);
4371 zlfid = (zfid_long_t *)fidp;
4373 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4374 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4376 /* XXX - this should be the generation number for the objset */
4377 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4378 zlfid->zf_setgen[i] = 0;
4384 EXPORT_SYMBOL(zfs_fid);
4388 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4390 znode_t *zp = ITOZ(ip);
4391 zfs_sb_t *zsb = ITOZSB(ip);
4393 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4397 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4402 EXPORT_SYMBOL(zfs_getsecattr);
4406 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4408 znode_t *zp = ITOZ(ip);
4409 zfs_sb_t *zsb = ITOZSB(ip);
4411 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4412 zilog_t *zilog = zsb->z_log;
4417 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4419 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
4420 zil_commit(zilog, 0);
4425 EXPORT_SYMBOL(zfs_setsecattr);
4427 #ifdef HAVE_UIO_ZEROCOPY
4429 * Tunable, both must be a power of 2.
4431 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4432 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4433 * an arcbuf for a partial block read
4435 int zcr_blksz_min = (1 << 10); /* 1K */
4436 int zcr_blksz_max = (1 << 17); /* 128K */
4440 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
4442 znode_t *zp = ITOZ(ip);
4443 zfs_sb_t *zsb = ITOZSB(ip);
4444 int max_blksz = zsb->z_max_blksz;
4445 uio_t *uio = &xuio->xu_uio;
4446 ssize_t size = uio->uio_resid;
4447 offset_t offset = uio->uio_loffset;
4452 int preamble, postamble;
4454 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4462 * Loan out an arc_buf for write if write size is bigger than
4463 * max_blksz, and the file's block size is also max_blksz.
4466 if (size < blksz || zp->z_blksz != blksz) {
4471 * Caller requests buffers for write before knowing where the
4472 * write offset might be (e.g. NFS TCP write).
4477 preamble = P2PHASE(offset, blksz);
4479 preamble = blksz - preamble;
4484 postamble = P2PHASE(size, blksz);
4487 fullblk = size / blksz;
4488 (void) dmu_xuio_init(xuio,
4489 (preamble != 0) + fullblk + (postamble != 0));
4492 * Have to fix iov base/len for partial buffers. They
4493 * currently represent full arc_buf's.
4496 /* data begins in the middle of the arc_buf */
4497 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4500 (void) dmu_xuio_add(xuio, abuf,
4501 blksz - preamble, preamble);
4504 for (i = 0; i < fullblk; i++) {
4505 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4508 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
4512 /* data ends in the middle of the arc_buf */
4513 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4516 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
4521 * Loan out an arc_buf for read if the read size is larger than
4522 * the current file block size. Block alignment is not
4523 * considered. Partial arc_buf will be loaned out for read.
4525 blksz = zp->z_blksz;
4526 if (blksz < zcr_blksz_min)
4527 blksz = zcr_blksz_min;
4528 if (blksz > zcr_blksz_max)
4529 blksz = zcr_blksz_max;
4530 /* avoid potential complexity of dealing with it */
4531 if (blksz > max_blksz) {
4536 maxsize = zp->z_size - uio->uio_loffset;
4550 uio->uio_extflg = UIO_XUIO;
4551 XUIO_XUZC_RW(xuio) = ioflag;
4558 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
4562 int ioflag = XUIO_XUZC_RW(xuio);
4564 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
4566 i = dmu_xuio_cnt(xuio);
4568 abuf = dmu_xuio_arcbuf(xuio, i);
4570 * if abuf == NULL, it must be a write buffer
4571 * that has been returned in zfs_write().
4574 dmu_return_arcbuf(abuf);
4575 ASSERT(abuf || ioflag == UIO_WRITE);
4578 dmu_xuio_fini(xuio);
4581 #endif /* HAVE_UIO_ZEROCOPY */
4583 #if defined(_KERNEL) && defined(HAVE_SPL)
4584 module_param(zfs_read_chunk_size, long, 0644);
4585 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");