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.
25 /* Portions Copyright 2007 Jeremy Teo */
26 /* Portions Copyright 2010 Robert Milkowski */
29 #include <sys/types.h>
30 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
36 #include <sys/vfs_opreg.h>
40 #include <sys/taskq.h>
42 #include <sys/vmsystm.h>
43 #include <sys/atomic.h>
45 #include <sys/pathname.h>
46 #include <sys/cmn_err.h>
47 #include <sys/errno.h>
48 #include <sys/unistd.h>
49 #include <sys/zfs_dir.h>
50 #include <sys/zfs_acl.h>
51 #include <sys/zfs_ioctl.h>
52 #include <sys/fs/zfs.h>
54 #include <sys/dmu_objset.h>
60 #include <sys/dirent.h>
61 #include <sys/policy.h>
62 #include <sys/sunddi.h>
65 #include "fs/fs_subr.h"
66 #include <sys/zfs_fuid.h>
67 #include <sys/zfs_sa.h>
68 #include <sys/zfs_vnops.h>
70 #include <sys/zfs_rlock.h>
71 #include <sys/extdirent.h>
72 #include <sys/kidmap.h>
80 * Each vnode op performs some logical unit of work. To do this, the ZPL must
81 * properly lock its in-core state, create a DMU transaction, do the work,
82 * record this work in the intent log (ZIL), commit the DMU transaction,
83 * and wait for the intent log to commit if it is a synchronous operation.
84 * Moreover, the vnode ops must work in both normal and log replay context.
85 * The ordering of events is important to avoid deadlocks and references
86 * to freed memory. The example below illustrates the following Big Rules:
88 * (1) A check must be made in each zfs thread for a mounted file system.
89 * This is done avoiding races using ZFS_ENTER(zsb).
90 * A ZFS_EXIT(zsb) is needed before all returns. Any znodes
91 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
92 * can return EIO from the calling function.
94 * (2) iput() should always be the last thing except for zil_commit()
95 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
96 * First, if it's the last reference, the vnode/znode
97 * can be freed, so the zp may point to freed memory. Second, the last
98 * reference will call zfs_zinactive(), which may induce a lot of work --
99 * pushing cached pages (which acquires range locks) and syncing out
100 * cached atime changes. Third, zfs_zinactive() may require a new tx,
101 * which could deadlock the system if you were already holding one.
102 * If you must call iput() within a tx then use iput_ASYNC().
104 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
105 * as they can span dmu_tx_assign() calls.
107 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
108 * This is critical because we don't want to block while holding locks.
109 * Note, in particular, that if a lock is sometimes acquired before
110 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
111 * use a non-blocking assign can deadlock the system. The scenario:
113 * Thread A has grabbed a lock before calling dmu_tx_assign().
114 * Thread B is in an already-assigned tx, and blocks for this lock.
115 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
116 * forever, because the previous txg can't quiesce until B's tx commits.
118 * If dmu_tx_assign() returns ERESTART and zsb->z_assign is TXG_NOWAIT,
119 * then drop all locks, call dmu_tx_wait(), and try again.
121 * (5) If the operation succeeded, generate the intent log entry for it
122 * before dropping locks. This ensures that the ordering of events
123 * in the intent log matches the order in which they actually occurred.
124 * During ZIL replay the zfs_log_* functions will update the sequence
125 * number to indicate the zil transaction has replayed.
127 * (6) At the end of each vnode op, the DMU tx must always commit,
128 * regardless of whether there were any errors.
130 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
131 * to ensure that synchronous semantics are provided when necessary.
133 * In general, this is how things should be ordered in each vnode op:
135 * ZFS_ENTER(zsb); // exit if unmounted
137 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
138 * rw_enter(...); // grab any other locks you need
139 * tx = dmu_tx_create(...); // get DMU tx
140 * dmu_tx_hold_*(); // hold each object you might modify
141 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign
143 * rw_exit(...); // drop locks
144 * zfs_dirent_unlock(dl); // unlock directory entry
145 * iput(...); // release held vnodes
146 * if (error == ERESTART) {
151 * dmu_tx_abort(tx); // abort DMU tx
152 * ZFS_EXIT(zsb); // finished in zfs
153 * return (error); // really out of space
155 * error = do_real_work(); // do whatever this VOP does
157 * zfs_log_*(...); // on success, make ZIL entry
158 * dmu_tx_commit(tx); // commit DMU tx -- error or not
159 * rw_exit(...); // drop locks
160 * zfs_dirent_unlock(dl); // unlock directory entry
161 * iput(...); // release held vnodes
162 * zil_commit(zilog, foid); // synchronous when necessary
163 * ZFS_EXIT(zsb); // finished in zfs
164 * return (error); // done, report error
168 * Virus scanning is unsupported. It would be possible to add a hook
169 * here to performance the required virus scan. This could be done
170 * entirely in the kernel or potentially as an update to invoke a
174 zfs_vscan(struct inode *ip, cred_t *cr, int async)
181 zfs_open(struct inode *ip, int mode, int flag, cred_t *cr)
183 znode_t *zp = ITOZ(ip);
184 zfs_sb_t *zsb = ITOZSB(ip);
189 /* Honor ZFS_APPENDONLY file attribute */
190 if ((mode & FMODE_WRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
191 ((flag & O_APPEND) == 0)) {
196 /* Virus scan eligible files on open */
197 if (!zfs_has_ctldir(zp) && zsb->z_vscan && S_ISREG(ip->i_mode) &&
198 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
199 if (zfs_vscan(ip, cr, 0) != 0) {
205 /* Keep a count of the synchronous opens in the znode */
207 atomic_inc_32(&zp->z_sync_cnt);
212 EXPORT_SYMBOL(zfs_open);
216 zfs_close(struct inode *ip, int flag, cred_t *cr)
218 znode_t *zp = ITOZ(ip);
219 zfs_sb_t *zsb = ITOZSB(ip);
225 * Zero the synchronous opens in the znode. Under Linux the
226 * zfs_close() hook is not symmetric with zfs_open(), it is
227 * only called once when the last reference is dropped.
232 if (!zfs_has_ctldir(zp) && zsb->z_vscan && S_ISREG(ip->i_mode) &&
233 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
234 VERIFY(zfs_vscan(ip, cr, 1) == 0);
239 EXPORT_SYMBOL(zfs_close);
243 * When a file is memory mapped, we must keep the IO data synchronized
244 * between the DMU cache and the memory mapped pages. What this means:
246 * On Write: If we find a memory mapped page, we write to *both*
247 * the page and the dmu buffer.
250 update_pages(struct inode *ip, int64_t start, int len,
251 objset_t *os, uint64_t oid)
253 struct address_space *mp = ip->i_mapping;
259 off = start & (PAGE_CACHE_SIZE-1);
260 for (start &= PAGE_CACHE_MASK; len > 0; start += PAGE_CACHE_SIZE) {
261 nbytes = MIN(PAGE_CACHE_SIZE - off, len);
263 pp = find_lock_page(mp, start >> PAGE_CACHE_SHIFT);
265 if (mapping_writably_mapped(mp))
266 flush_dcache_page(pp);
269 (void) dmu_read(os, oid, start+off, nbytes, pb+off,
273 if (mapping_writably_mapped(mp))
274 flush_dcache_page(pp);
276 mark_page_accessed(pp);
280 page_cache_release(pp);
289 * When a file is memory mapped, we must keep the IO data synchronized
290 * between the DMU cache and the memory mapped pages. What this means:
292 * On Read: We "read" preferentially from memory mapped pages,
293 * else we default from the dmu buffer.
295 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
296 * the file is memory mapped.
299 mappedread(struct inode *ip, int nbytes, uio_t *uio)
301 struct address_space *mp = ip->i_mapping;
303 znode_t *zp = ITOZ(ip);
304 objset_t *os = ITOZSB(ip)->z_os;
311 start = uio->uio_loffset;
312 off = start & (PAGE_CACHE_SIZE-1);
313 for (start &= PAGE_CACHE_MASK; len > 0; start += PAGE_CACHE_SIZE) {
314 bytes = MIN(PAGE_CACHE_SIZE - off, len);
316 pp = find_lock_page(mp, start >> PAGE_CACHE_SHIFT);
318 ASSERT(PageUptodate(pp));
321 error = uiomove(pb + off, bytes, UIO_READ, uio);
324 if (mapping_writably_mapped(mp))
325 flush_dcache_page(pp);
327 mark_page_accessed(pp);
329 page_cache_release(pp);
331 error = dmu_read_uio(os, zp->z_id, uio, bytes);
343 unsigned long zfs_read_chunk_size = 1024 * 1024; /* Tunable */
346 * Read bytes from specified file into supplied buffer.
348 * IN: ip - inode of file to be read from.
349 * uio - structure supplying read location, range info,
351 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
352 * O_DIRECT flag; used to bypass page cache.
353 * cr - credentials of caller.
355 * OUT: uio - updated offset and range, buffer filled.
357 * RETURN: 0 if success
358 * error code if failure
361 * inode - atime updated if byte count > 0
365 zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
367 znode_t *zp = ITOZ(ip);
368 zfs_sb_t *zsb = ITOZSB(ip);
373 #ifdef HAVE_UIO_ZEROCOPY
375 #endif /* HAVE_UIO_ZEROCOPY */
381 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
387 * Validate file offset
389 if (uio->uio_loffset < (offset_t)0) {
395 * Fasttrack empty reads
397 if (uio->uio_resid == 0) {
403 * Check for mandatory locks
405 if (mandatory_lock(ip) &&
406 !lock_may_read(ip, uio->uio_loffset, uio->uio_resid)) {
412 * If we're in FRSYNC mode, sync out this znode before reading it.
414 if (ioflag & FRSYNC || zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
415 zil_commit(zsb->z_log, zp->z_id);
418 * Lock the range against changes.
420 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
423 * If we are reading past end-of-file we can skip
424 * to the end; but we might still need to set atime.
426 if (uio->uio_loffset >= zp->z_size) {
431 ASSERT(uio->uio_loffset < zp->z_size);
432 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
434 #ifdef HAVE_UIO_ZEROCOPY
435 if ((uio->uio_extflg == UIO_XUIO) &&
436 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
438 int blksz = zp->z_blksz;
439 uint64_t offset = uio->uio_loffset;
441 xuio = (xuio_t *)uio;
443 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
446 ASSERT(offset + n <= blksz);
449 (void) dmu_xuio_init(xuio, nblk);
451 if (vn_has_cached_data(ip)) {
453 * For simplicity, we always allocate a full buffer
454 * even if we only expect to read a portion of a block.
456 while (--nblk >= 0) {
457 (void) dmu_xuio_add(xuio,
458 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
463 #endif /* HAVE_UIO_ZEROCOPY */
466 nbytes = MIN(n, zfs_read_chunk_size -
467 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
469 if (zp->z_is_mapped && !(ioflag & O_DIRECT))
470 error = mappedread(ip, nbytes, uio);
472 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
475 /* convert checksum errors into IO errors */
484 zfs_range_unlock(rl);
486 ZFS_ACCESSTIME_STAMP(zsb, zp);
487 zfs_inode_update(zp);
491 EXPORT_SYMBOL(zfs_read);
494 * Write the bytes to a file.
496 * IN: ip - inode of file to be written to.
497 * uio - structure supplying write location, range info,
499 * ioflag - FAPPEND flag set if in append mode.
500 * O_DIRECT flag; used to bypass page cache.
501 * cr - credentials of caller.
503 * OUT: uio - updated offset and range.
505 * RETURN: 0 if success
506 * error code if failure
509 * ip - ctime|mtime updated if byte count > 0
514 zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
516 znode_t *zp = ITOZ(ip);
517 rlim64_t limit = uio->uio_limit;
518 ssize_t start_resid = uio->uio_resid;
522 zfs_sb_t *zsb = ZTOZSB(zp);
527 int max_blksz = zsb->z_max_blksz;
530 iovec_t *aiov = NULL;
533 iovec_t *iovp = uio->uio_iov;
536 sa_bulk_attr_t bulk[4];
537 uint64_t mtime[2], ctime[2];
538 ASSERTV(int iovcnt = uio->uio_iovcnt);
541 * Fasttrack empty write
547 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
553 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
554 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
555 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb), NULL, &zp->z_size, 8);
556 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
560 * If immutable or not appending then return EPERM
562 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
563 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
564 (uio->uio_loffset < zp->z_size))) {
572 * Validate file offset
574 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
581 * Check for mandatory locks before calling zfs_range_lock()
582 * in order to prevent a deadlock with locks set via fcntl().
584 if (mandatory_lock(ip) && !lock_may_write(ip, woff, n)) {
589 #ifdef HAVE_UIO_ZEROCOPY
591 * Pre-fault the pages to ensure slow (eg NFS) pages
593 * Skip this if uio contains loaned arc_buf.
595 if ((uio->uio_extflg == UIO_XUIO) &&
596 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
597 xuio = (xuio_t *)uio;
599 uio_prefaultpages(MIN(n, max_blksz), uio);
600 #endif /* HAVE_UIO_ZEROCOPY */
603 * If in append mode, set the io offset pointer to eof.
605 if (ioflag & FAPPEND) {
607 * Obtain an appending range lock to guarantee file append
608 * semantics. We reset the write offset once we have the lock.
610 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
612 if (rl->r_len == UINT64_MAX) {
614 * We overlocked the file because this write will cause
615 * the file block size to increase.
616 * Note that zp_size cannot change with this lock held.
620 uio->uio_loffset = woff;
623 * Note that if the file block size will change as a result of
624 * this write, then this range lock will lock the entire file
625 * so that we can re-write the block safely.
627 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
631 zfs_range_unlock(rl);
636 if ((woff + n) > limit || woff > (limit - n))
639 /* Will this write extend the file length? */
640 write_eof = (woff + n > zp->z_size);
642 end_size = MAX(zp->z_size, woff + n);
645 * Write the file in reasonable size chunks. Each chunk is written
646 * in a separate transaction; this keeps the intent log records small
647 * and allows us to do more fine-grained space accounting.
651 woff = uio->uio_loffset;
653 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
654 zfs_owner_overquota(zsb, zp, B_TRUE)) {
656 dmu_return_arcbuf(abuf);
661 if (xuio && abuf == NULL) {
662 ASSERT(i_iov < iovcnt);
664 abuf = dmu_xuio_arcbuf(xuio, i_iov);
665 dmu_xuio_clear(xuio, i_iov);
666 ASSERT((aiov->iov_base == abuf->b_data) ||
667 ((char *)aiov->iov_base - (char *)abuf->b_data +
668 aiov->iov_len == arc_buf_size(abuf)));
670 } else if (abuf == NULL && n >= max_blksz &&
671 woff >= zp->z_size &&
672 P2PHASE(woff, max_blksz) == 0 &&
673 zp->z_blksz == max_blksz) {
675 * This write covers a full block. "Borrow" a buffer
676 * from the dmu so that we can fill it before we enter
677 * a transaction. This avoids the possibility of
678 * holding up the transaction if the data copy hangs
679 * up on a pagefault (e.g., from an NFS server mapping).
683 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
685 ASSERT(abuf != NULL);
686 ASSERT(arc_buf_size(abuf) == max_blksz);
687 if ((error = uiocopy(abuf->b_data, max_blksz,
688 UIO_WRITE, uio, &cbytes))) {
689 dmu_return_arcbuf(abuf);
692 ASSERT(cbytes == max_blksz);
696 * Start a transaction.
698 tx = dmu_tx_create(zsb->z_os);
699 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
700 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
701 zfs_sa_upgrade_txholds(tx, zp);
702 error = dmu_tx_assign(tx, TXG_NOWAIT);
704 if (error == ERESTART) {
711 dmu_return_arcbuf(abuf);
716 * If zfs_range_lock() over-locked we grow the blocksize
717 * and then reduce the lock range. This will only happen
718 * on the first iteration since zfs_range_reduce() will
719 * shrink down r_len to the appropriate size.
721 if (rl->r_len == UINT64_MAX) {
724 if (zp->z_blksz > max_blksz) {
725 ASSERT(!ISP2(zp->z_blksz));
726 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
728 new_blksz = MIN(end_size, max_blksz);
730 zfs_grow_blocksize(zp, new_blksz, tx);
731 zfs_range_reduce(rl, woff, n);
735 * XXX - should we really limit each write to z_max_blksz?
736 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
738 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
741 tx_bytes = uio->uio_resid;
742 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
744 tx_bytes -= uio->uio_resid;
747 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
749 * If this is not a full block write, but we are
750 * extending the file past EOF and this data starts
751 * block-aligned, use assign_arcbuf(). Otherwise,
752 * write via dmu_write().
754 if (tx_bytes < max_blksz && (!write_eof ||
755 aiov->iov_base != abuf->b_data)) {
757 dmu_write(zsb->z_os, zp->z_id, woff,
758 aiov->iov_len, aiov->iov_base, tx);
759 dmu_return_arcbuf(abuf);
760 xuio_stat_wbuf_copied();
762 ASSERT(xuio || tx_bytes == max_blksz);
763 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
766 ASSERT(tx_bytes <= uio->uio_resid);
767 uioskip(uio, tx_bytes);
770 if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT))
771 update_pages(ip, woff, tx_bytes, zsb->z_os, zp->z_id);
774 * If we made no progress, we're done. If we made even
775 * partial progress, update the znode and ZIL accordingly.
778 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
779 (void *)&zp->z_size, sizeof (uint64_t), tx);
786 * Clear Set-UID/Set-GID bits on successful write if not
787 * privileged and at least one of the excute bits is set.
789 * It would be nice to to this after all writes have
790 * been done, but that would still expose the ISUID/ISGID
791 * to another app after the partial write is committed.
793 * Note: we don't call zfs_fuid_map_id() here because
794 * user 0 is not an ephemeral uid.
796 mutex_enter(&zp->z_acl_lock);
797 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
798 (S_IXUSR >> 6))) != 0 &&
799 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
800 secpolicy_vnode_setid_retain(cr,
801 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
803 zp->z_mode &= ~(S_ISUID | S_ISGID);
804 newmode = zp->z_mode;
805 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zsb),
806 (void *)&newmode, sizeof (uint64_t), tx);
808 mutex_exit(&zp->z_acl_lock);
810 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
814 * Update the file size (zp_size) if it has changed;
815 * account for possible concurrent updates.
817 while ((end_size = zp->z_size) < uio->uio_loffset) {
818 (void) atomic_cas_64(&zp->z_size, end_size,
823 * If we are replaying and eof is non zero then force
824 * the file size to the specified eof. Note, there's no
825 * concurrency during replay.
827 if (zsb->z_replay && zsb->z_replay_eof != 0)
828 zp->z_size = zsb->z_replay_eof;
830 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
832 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
837 ASSERT(tx_bytes == nbytes);
841 uio_prefaultpages(MIN(n, max_blksz), uio);
844 zfs_range_unlock(rl);
847 * If we're in replay mode, or we made no progress, return error.
848 * Otherwise, it's at least a partial write, so it's successful.
850 if (zsb->z_replay || uio->uio_resid == start_resid) {
855 if (ioflag & (FSYNC | FDSYNC) ||
856 zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
857 zil_commit(zilog, zp->z_id);
859 zfs_inode_update(zp);
863 EXPORT_SYMBOL(zfs_write);
866 iput_async(struct inode *ip, taskq_t *taskq)
868 ASSERT(atomic_read(&ip->i_count) > 0);
869 if (atomic_read(&ip->i_count) == 1)
870 taskq_dispatch(taskq, (task_func_t *)iput, ip, TQ_SLEEP);
876 zfs_get_done(zgd_t *zgd, int error)
878 znode_t *zp = zgd->zgd_private;
879 objset_t *os = ZTOZSB(zp)->z_os;
882 dmu_buf_rele(zgd->zgd_db, zgd);
884 zfs_range_unlock(zgd->zgd_rl);
887 * Release the vnode asynchronously as we currently have the
888 * txg stopped from syncing.
890 iput_async(ZTOI(zp), dsl_pool_iput_taskq(dmu_objset_pool(os)));
892 if (error == 0 && zgd->zgd_bp)
893 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
895 kmem_free(zgd, sizeof (zgd_t));
899 static int zil_fault_io = 0;
903 * Get data to generate a TX_WRITE intent log record.
906 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
909 objset_t *os = zsb->z_os;
911 uint64_t object = lr->lr_foid;
912 uint64_t offset = lr->lr_offset;
913 uint64_t size = lr->lr_length;
914 blkptr_t *bp = &lr->lr_blkptr;
923 * Nothing to do if the file has been removed
925 if (zfs_zget(zsb, object, &zp) != 0)
927 if (zp->z_unlinked) {
929 * Release the vnode asynchronously as we currently have the
930 * txg stopped from syncing.
932 iput_async(ZTOI(zp), dsl_pool_iput_taskq(dmu_objset_pool(os)));
936 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
937 zgd->zgd_zilog = zsb->z_log;
938 zgd->zgd_private = zp;
941 * Write records come in two flavors: immediate and indirect.
942 * For small writes it's cheaper to store the data with the
943 * log record (immediate); for large writes it's cheaper to
944 * sync the data and get a pointer to it (indirect) so that
945 * we don't have to write the data twice.
947 if (buf != NULL) { /* immediate write */
948 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
949 /* test for truncation needs to be done while range locked */
950 if (offset >= zp->z_size) {
953 error = dmu_read(os, object, offset, size, buf,
954 DMU_READ_NO_PREFETCH);
956 ASSERT(error == 0 || error == ENOENT);
957 } else { /* indirect write */
959 * Have to lock the whole block to ensure when it's
960 * written out and it's checksum is being calculated
961 * that no one can change the data. We need to re-check
962 * blocksize after we get the lock in case it's changed!
967 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
969 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
971 if (zp->z_blksz == size)
974 zfs_range_unlock(zgd->zgd_rl);
976 /* test for truncation needs to be done while range locked */
977 if (lr->lr_offset >= zp->z_size)
986 error = dmu_buf_hold(os, object, offset, zgd, &db,
987 DMU_READ_NO_PREFETCH);
993 ASSERT(db->db_offset == offset);
994 ASSERT(db->db_size == size);
996 error = dmu_sync(zio, lr->lr_common.lrc_txg,
998 ASSERT(error || lr->lr_length <= zp->z_blksz);
1001 * On success, we need to wait for the write I/O
1002 * initiated by dmu_sync() to complete before we can
1003 * release this dbuf. We will finish everything up
1004 * in the zfs_get_done() callback.
1009 if (error == EALREADY) {
1010 lr->lr_common.lrc_txtype = TX_WRITE2;
1016 zfs_get_done(zgd, error);
1023 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
1025 znode_t *zp = ITOZ(ip);
1026 zfs_sb_t *zsb = ITOZSB(ip);
1032 if (flag & V_ACE_MASK)
1033 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1035 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1040 EXPORT_SYMBOL(zfs_access);
1043 * Lookup an entry in a directory, or an extended attribute directory.
1044 * If it exists, return a held inode reference for it.
1046 * IN: dip - inode of directory to search.
1047 * nm - name of entry to lookup.
1048 * flags - LOOKUP_XATTR set if looking for an attribute.
1049 * cr - credentials of caller.
1050 * direntflags - directory lookup flags
1051 * realpnp - returned pathname.
1053 * OUT: ipp - inode of located entry, NULL if not found.
1055 * RETURN: 0 if success
1056 * error code if failure
1063 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
1064 cred_t *cr, int *direntflags, pathname_t *realpnp)
1066 znode_t *zdp = ITOZ(dip);
1067 zfs_sb_t *zsb = ITOZSB(dip);
1071 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1073 if (!S_ISDIR(dip->i_mode)) {
1075 } else if (zdp->z_sa_hdl == NULL) {
1079 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1080 error = zfs_fastaccesschk_execute(zdp, cr);
1089 vnode_t *tvp = dnlc_lookup(dvp, nm);
1092 error = zfs_fastaccesschk_execute(zdp, cr);
1097 if (tvp == DNLC_NO_VNODE) {
1102 return (specvp_check(vpp, cr));
1105 #endif /* HAVE_DNLC */
1114 if (flags & LOOKUP_XATTR) {
1116 * We don't allow recursive attributes..
1117 * Maybe someday we will.
1119 if (zdp->z_pflags & ZFS_XATTR) {
1124 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1130 * Do we have permission to get into attribute directory?
1133 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1143 if (!S_ISDIR(dip->i_mode)) {
1149 * Check accessibility of directory.
1152 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1157 if (zsb->z_utf8 && u8_validate(nm, strlen(nm),
1158 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1163 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1164 if ((error == 0) && (*ipp))
1165 zfs_inode_update(ITOZ(*ipp));
1170 EXPORT_SYMBOL(zfs_lookup);
1173 * Attempt to create a new entry in a directory. If the entry
1174 * already exists, truncate the file if permissible, else return
1175 * an error. Return the ip of the created or trunc'd file.
1177 * IN: dip - inode of directory to put new file entry in.
1178 * name - name of new file entry.
1179 * vap - attributes of new file.
1180 * excl - flag indicating exclusive or non-exclusive mode.
1181 * mode - mode to open file with.
1182 * cr - credentials of caller.
1183 * flag - large file flag [UNUSED].
1184 * vsecp - ACL to be set
1186 * OUT: ipp - inode of created or trunc'd entry.
1188 * RETURN: 0 if success
1189 * error code if failure
1192 * dip - ctime|mtime updated if new entry created
1193 * ip - ctime|mtime always, atime if new
1198 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1199 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1201 znode_t *zp, *dzp = ITOZ(dip);
1202 zfs_sb_t *zsb = ITOZSB(dip);
1210 zfs_acl_ids_t acl_ids;
1211 boolean_t fuid_dirtied;
1212 boolean_t have_acl = B_FALSE;
1215 * If we have an ephemeral id, ACL, or XVATTR then
1216 * make sure file system is at proper version
1222 if (zsb->z_use_fuids == B_FALSE &&
1223 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1231 if (zsb->z_utf8 && u8_validate(name, strlen(name),
1232 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1237 if (vap->va_mask & ATTR_XVATTR) {
1238 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1239 crgetuid(cr), cr, vap->va_mode)) != 0) {
1247 if (*name == '\0') {
1249 * Null component name refers to the directory itself.
1256 /* possible igrab(zp) */
1259 if (flag & FIGNORECASE)
1262 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1266 zfs_acl_ids_free(&acl_ids);
1267 if (strcmp(name, "..") == 0)
1278 * Create a new file object and update the directory
1281 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1283 zfs_acl_ids_free(&acl_ids);
1288 * We only support the creation of regular files in
1289 * extended attribute directories.
1292 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1294 zfs_acl_ids_free(&acl_ids);
1299 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1300 cr, vsecp, &acl_ids)) != 0)
1304 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1305 zfs_acl_ids_free(&acl_ids);
1310 tx = dmu_tx_create(os);
1312 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1313 ZFS_SA_BASE_ATTR_SIZE);
1315 fuid_dirtied = zsb->z_fuid_dirty;
1317 zfs_fuid_txhold(zsb, tx);
1318 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1319 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1320 if (!zsb->z_use_sa &&
1321 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1322 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1323 0, acl_ids.z_aclp->z_acl_bytes);
1325 error = dmu_tx_assign(tx, TXG_NOWAIT);
1327 zfs_dirent_unlock(dl);
1328 if (error == ERESTART) {
1333 zfs_acl_ids_free(&acl_ids);
1338 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1341 zfs_fuid_sync(zsb, tx);
1343 (void) zfs_link_create(dl, zp, tx, ZNEW);
1344 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1345 if (flag & FIGNORECASE)
1347 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1348 vsecp, acl_ids.z_fuidp, vap);
1349 zfs_acl_ids_free(&acl_ids);
1352 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1355 zfs_acl_ids_free(&acl_ids);
1359 * A directory entry already exists for this name.
1362 * Can't truncate an existing file if in exclusive mode.
1369 * Can't open a directory for writing.
1371 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1376 * Verify requested access to file.
1378 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1382 mutex_enter(&dzp->z_lock);
1384 mutex_exit(&dzp->z_lock);
1387 * Truncate regular files if requested.
1389 if (S_ISREG(ZTOI(zp)->i_mode) &&
1390 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1391 /* we can't hold any locks when calling zfs_freesp() */
1392 zfs_dirent_unlock(dl);
1394 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1400 zfs_dirent_unlock(dl);
1406 zfs_inode_update(dzp);
1407 zfs_inode_update(zp);
1411 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1412 zil_commit(zilog, 0);
1417 EXPORT_SYMBOL(zfs_create);
1420 * Remove an entry from a directory.
1422 * IN: dip - inode of directory to remove entry from.
1423 * name - name of entry to remove.
1424 * cr - credentials of caller.
1426 * RETURN: 0 if success
1427 * error code if failure
1431 * ip - ctime (if nlink > 0)
1434 uint64_t null_xattr = 0;
1438 zfs_remove(struct inode *dip, char *name, cred_t *cr)
1440 znode_t *zp, *dzp = ITOZ(dip);
1443 zfs_sb_t *zsb = ITOZSB(dip);
1446 uint64_t xattr_obj_unlinked = 0;
1452 pathname_t *realnmp = NULL;
1453 #ifdef HAVE_PN_UTILS
1455 #endif /* HAVE_PN_UTILS */
1463 #ifdef HAVE_PN_UTILS
1464 if (flags & FIGNORECASE) {
1469 #endif /* HAVE_PN_UTILS */
1475 * Attempt to lock directory; fail if entry doesn't exist.
1477 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1479 #ifdef HAVE_PN_UTILS
1482 #endif /* HAVE_PN_UTILS */
1489 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1494 * Need to use rmdir for removing directories.
1496 if (S_ISDIR(ip->i_mode)) {
1503 dnlc_remove(dvp, realnmp->pn_buf);
1505 dnlc_remove(dvp, name);
1506 #endif /* HAVE_DNLC */
1509 * We never delete the znode and always place it in the unlinked
1510 * set. The dentry cache will always hold the last reference and
1511 * is responsible for safely freeing the znode.
1514 tx = dmu_tx_create(zsb->z_os);
1515 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1516 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1517 zfs_sa_upgrade_txholds(tx, zp);
1518 zfs_sa_upgrade_txholds(tx, dzp);
1520 /* are there any extended attributes? */
1521 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1522 &xattr_obj, sizeof (xattr_obj));
1523 if (error == 0 && xattr_obj) {
1524 error = zfs_zget(zsb, xattr_obj, &xzp);
1525 ASSERT3U(error, ==, 0);
1526 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1527 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1530 /* charge as an update -- would be nice not to charge at all */
1531 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
1533 error = dmu_tx_assign(tx, TXG_NOWAIT);
1535 zfs_dirent_unlock(dl);
1539 if (error == ERESTART) {
1544 #ifdef HAVE_PN_UTILS
1547 #endif /* HAVE_PN_UTILS */
1554 * Remove the directory entry.
1556 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1565 * Hold z_lock so that we can make sure that the ACL obj
1566 * hasn't changed. Could have been deleted due to
1569 mutex_enter(&zp->z_lock);
1570 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1571 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1572 mutex_exit(&zp->z_lock);
1573 zfs_unlinked_add(zp, tx);
1577 #ifdef HAVE_PN_UTILS
1578 if (flags & FIGNORECASE)
1580 #endif /* HAVE_PN_UTILS */
1581 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1585 #ifdef HAVE_PN_UTILS
1588 #endif /* HAVE_PN_UTILS */
1590 zfs_dirent_unlock(dl);
1591 zfs_inode_update(dzp);
1592 zfs_inode_update(zp);
1594 zfs_inode_update(xzp);
1600 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1601 zil_commit(zilog, 0);
1606 EXPORT_SYMBOL(zfs_remove);
1609 * Create a new directory and insert it into dip using the name
1610 * provided. Return a pointer to the inserted directory.
1612 * IN: dip - inode of directory to add subdir to.
1613 * dirname - name of new directory.
1614 * vap - attributes of new directory.
1615 * cr - credentials of caller.
1616 * vsecp - ACL to be set
1618 * OUT: ipp - inode of created directory.
1620 * RETURN: 0 if success
1621 * error code if failure
1624 * dip - ctime|mtime updated
1625 * ipp - ctime|mtime|atime updated
1629 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1630 cred_t *cr, int flags, vsecattr_t *vsecp)
1632 znode_t *zp, *dzp = ITOZ(dip);
1633 zfs_sb_t *zsb = ITOZSB(dip);
1641 gid_t gid = crgetgid(cr);
1642 zfs_acl_ids_t acl_ids;
1643 boolean_t fuid_dirtied;
1645 ASSERT(S_ISDIR(vap->va_mode));
1648 * If we have an ephemeral id, ACL, or XVATTR then
1649 * make sure file system is at proper version
1653 if (zsb->z_use_fuids == B_FALSE &&
1654 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1661 if (dzp->z_pflags & ZFS_XATTR) {
1666 if (zsb->z_utf8 && u8_validate(dirname,
1667 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1671 if (flags & FIGNORECASE)
1674 if (vap->va_mask & ATTR_XVATTR) {
1675 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1676 crgetuid(cr), cr, vap->va_mode)) != 0) {
1682 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1683 vsecp, &acl_ids)) != 0) {
1688 * First make sure the new directory doesn't exist.
1690 * Existence is checked first to make sure we don't return
1691 * EACCES instead of EEXIST which can cause some applications
1697 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1699 zfs_acl_ids_free(&acl_ids);
1704 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
1705 zfs_acl_ids_free(&acl_ids);
1706 zfs_dirent_unlock(dl);
1711 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1712 zfs_acl_ids_free(&acl_ids);
1713 zfs_dirent_unlock(dl);
1719 * Add a new entry to the directory.
1721 tx = dmu_tx_create(zsb->z_os);
1722 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1723 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1724 fuid_dirtied = zsb->z_fuid_dirty;
1726 zfs_fuid_txhold(zsb, tx);
1727 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1728 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1729 acl_ids.z_aclp->z_acl_bytes);
1732 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1733 ZFS_SA_BASE_ATTR_SIZE);
1735 error = dmu_tx_assign(tx, TXG_NOWAIT);
1737 zfs_dirent_unlock(dl);
1738 if (error == ERESTART) {
1743 zfs_acl_ids_free(&acl_ids);
1752 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1755 zfs_fuid_sync(zsb, tx);
1758 * Now put new name in parent dir.
1760 (void) zfs_link_create(dl, zp, tx, ZNEW);
1764 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1765 if (flags & FIGNORECASE)
1767 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1768 acl_ids.z_fuidp, vap);
1770 zfs_acl_ids_free(&acl_ids);
1774 zfs_dirent_unlock(dl);
1776 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1777 zil_commit(zilog, 0);
1779 zfs_inode_update(dzp);
1780 zfs_inode_update(zp);
1784 EXPORT_SYMBOL(zfs_mkdir);
1787 * Remove a directory subdir entry. If the current working
1788 * directory is the same as the subdir to be removed, the
1791 * IN: dip - inode of directory to remove from.
1792 * name - name of directory to be removed.
1793 * cwd - inode of current working directory.
1794 * cr - credentials of caller.
1795 * flags - case flags
1797 * RETURN: 0 if success
1798 * error code if failure
1801 * dip - ctime|mtime updated
1805 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
1808 znode_t *dzp = ITOZ(dip);
1811 zfs_sb_t *zsb = ITOZSB(dip);
1822 if (flags & FIGNORECASE)
1828 * Attempt to lock directory; fail if entry doesn't exist.
1830 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1838 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1842 if (!S_ISDIR(ip->i_mode)) {
1853 * Grab a lock on the directory to make sure that noone is
1854 * trying to add (or lookup) entries while we are removing it.
1856 rw_enter(&zp->z_name_lock, RW_WRITER);
1859 * Grab a lock on the parent pointer to make sure we play well
1860 * with the treewalk and directory rename code.
1862 rw_enter(&zp->z_parent_lock, RW_WRITER);
1864 tx = dmu_tx_create(zsb->z_os);
1865 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1866 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1867 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
1868 zfs_sa_upgrade_txholds(tx, zp);
1869 zfs_sa_upgrade_txholds(tx, dzp);
1870 error = dmu_tx_assign(tx, TXG_NOWAIT);
1872 rw_exit(&zp->z_parent_lock);
1873 rw_exit(&zp->z_name_lock);
1874 zfs_dirent_unlock(dl);
1876 if (error == ERESTART) {
1886 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
1889 uint64_t txtype = TX_RMDIR;
1890 if (flags & FIGNORECASE)
1892 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
1897 rw_exit(&zp->z_parent_lock);
1898 rw_exit(&zp->z_name_lock);
1900 zfs_dirent_unlock(dl);
1904 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1905 zil_commit(zilog, 0);
1907 zfs_inode_update(dzp);
1908 zfs_inode_update(zp);
1912 EXPORT_SYMBOL(zfs_rmdir);
1915 * Read as many directory entries as will fit into the provided
1916 * dirent buffer from the given directory cursor position.
1918 * IN: ip - inode of directory to read.
1919 * dirent - buffer for directory entries.
1921 * OUT: dirent - filler buffer of directory entries.
1923 * RETURN: 0 if success
1924 * error code if failure
1927 * ip - atime updated
1929 * Note that the low 4 bits of the cookie returned by zap is always zero.
1930 * This allows us to use the low range for "special" directory entries:
1931 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
1932 * we use the offset 2 for the '.zfs' directory.
1936 zfs_readdir(struct inode *ip, void *dirent, filldir_t filldir,
1937 loff_t *pos, cred_t *cr)
1939 znode_t *zp = ITOZ(ip);
1940 zfs_sb_t *zsb = ITOZSB(ip);
1943 zap_attribute_t zap;
1953 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zsb),
1954 &parent, sizeof (parent))) != 0)
1958 * Quit if directory has been removed (posix)
1965 prefetch = zp->z_zn_prefetch;
1968 * Initialize the iterator cursor.
1972 * Start iteration from the beginning of the directory.
1974 zap_cursor_init(&zc, os, zp->z_id);
1977 * The offset is a serialized cursor.
1979 zap_cursor_init_serialized(&zc, os, zp->z_id, *pos);
1983 * Transform to file-system independent format
1990 * Special case `.', `..', and `.zfs'.
1993 (void) strcpy(zap.za_name, ".");
1994 zap.za_normalization_conflict = 0;
1996 } else if (*pos == 1) {
1997 (void) strcpy(zap.za_name, "..");
1998 zap.za_normalization_conflict = 0;
2000 } else if (*pos == 2 && zfs_show_ctldir(zp)) {
2001 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2002 zap.za_normalization_conflict = 0;
2003 objnum = ZFSCTL_INO_ROOT;
2008 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2009 if (error == ENOENT)
2016 * Allow multiple entries provided the first entry is
2017 * the object id. Non-zpl consumers may safely make
2018 * use of the additional space.
2020 * XXX: This should be a feature flag for compatibility
2022 if (zap.za_integer_length != 8 ||
2023 zap.za_num_integers == 0) {
2024 cmn_err(CE_WARN, "zap_readdir: bad directory "
2025 "entry, obj = %lld, offset = %lld, "
2026 "length = %d, num = %lld\n",
2027 (u_longlong_t)zp->z_id,
2029 zap.za_integer_length,
2030 (u_longlong_t)zap.za_num_integers);
2035 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2037 done = filldir(dirent, zap.za_name, strlen(zap.za_name),
2038 zap_cursor_serialize(&zc), objnum, 0);
2043 /* Prefetch znode */
2045 dmu_prefetch(os, objnum, 0, 0);
2049 zap_cursor_advance(&zc);
2050 *pos = zap_cursor_serialize(&zc);
2055 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2058 zap_cursor_fini(&zc);
2059 if (error == ENOENT)
2062 ZFS_ACCESSTIME_STAMP(zsb, zp);
2063 zfs_inode_update(zp);
2070 EXPORT_SYMBOL(zfs_readdir);
2072 ulong_t zfs_fsync_sync_cnt = 4;
2075 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2077 znode_t *zp = ITOZ(ip);
2078 zfs_sb_t *zsb = ITOZSB(ip);
2080 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2082 if (zsb->z_os->os_sync != ZFS_SYNC_DISABLED) {
2085 zil_commit(zsb->z_log, zp->z_id);
2090 EXPORT_SYMBOL(zfs_fsync);
2094 * Get the requested file attributes and place them in the provided
2097 * IN: ip - inode of file.
2098 * vap - va_mask identifies requested attributes.
2099 * If ATTR_XVATTR set, then optional attrs are requested
2100 * flags - ATTR_NOACLCHECK (CIFS server context)
2101 * cr - credentials of caller.
2103 * OUT: vap - attribute values.
2105 * RETURN: 0 (always succeeds)
2109 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2111 znode_t *zp = ITOZ(ip);
2112 zfs_sb_t *zsb = ITOZSB(ip);
2115 uint64_t mtime[2], ctime[2];
2116 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2117 xoptattr_t *xoap = NULL;
2118 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2119 sa_bulk_attr_t bulk[2];
2125 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2127 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
2128 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
2130 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2136 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2137 * Also, if we are the owner don't bother, since owner should
2138 * always be allowed to read basic attributes of file.
2140 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2141 (vap->va_uid != crgetuid(cr))) {
2142 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2150 * Return all attributes. It's cheaper to provide the answer
2151 * than to determine whether we were asked the question.
2154 mutex_enter(&zp->z_lock);
2155 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2156 vap->va_mode = zp->z_mode;
2157 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2158 vap->va_nodeid = zp->z_id;
2159 if ((zp->z_id == zsb->z_root) && zfs_show_ctldir(zp))
2160 links = zp->z_links + 1;
2162 links = zp->z_links;
2163 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2164 vap->va_size = i_size_read(ip);
2165 vap->va_rdev = ip->i_rdev;
2166 vap->va_seq = ip->i_generation;
2169 * Add in any requested optional attributes and the create time.
2170 * Also set the corresponding bits in the returned attribute bitmap.
2172 if ((xoap = xva_getxoptattr(xvap)) != NULL && zsb->z_use_fuids) {
2173 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2175 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2176 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2179 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2180 xoap->xoa_readonly =
2181 ((zp->z_pflags & ZFS_READONLY) != 0);
2182 XVA_SET_RTN(xvap, XAT_READONLY);
2185 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2187 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2188 XVA_SET_RTN(xvap, XAT_SYSTEM);
2191 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2193 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2194 XVA_SET_RTN(xvap, XAT_HIDDEN);
2197 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2198 xoap->xoa_nounlink =
2199 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2200 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2203 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2204 xoap->xoa_immutable =
2205 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2206 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2209 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2210 xoap->xoa_appendonly =
2211 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2212 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2215 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2217 ((zp->z_pflags & ZFS_NODUMP) != 0);
2218 XVA_SET_RTN(xvap, XAT_NODUMP);
2221 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2223 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2224 XVA_SET_RTN(xvap, XAT_OPAQUE);
2227 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2228 xoap->xoa_av_quarantined =
2229 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2230 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2233 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2234 xoap->xoa_av_modified =
2235 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2236 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2239 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2240 S_ISREG(ip->i_mode)) {
2241 zfs_sa_get_scanstamp(zp, xvap);
2244 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2247 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zsb),
2248 times, sizeof (times));
2249 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2250 XVA_SET_RTN(xvap, XAT_CREATETIME);
2253 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2254 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2255 XVA_SET_RTN(xvap, XAT_REPARSE);
2257 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2258 xoap->xoa_generation = zp->z_gen;
2259 XVA_SET_RTN(xvap, XAT_GEN);
2262 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2264 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2265 XVA_SET_RTN(xvap, XAT_OFFLINE);
2268 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2270 ((zp->z_pflags & ZFS_SPARSE) != 0);
2271 XVA_SET_RTN(xvap, XAT_SPARSE);
2275 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2276 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2277 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2279 mutex_exit(&zp->z_lock);
2281 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2283 if (zp->z_blksz == 0) {
2285 * Block size hasn't been set; suggest maximal I/O transfers.
2287 vap->va_blksize = zsb->z_max_blksz;
2293 EXPORT_SYMBOL(zfs_getattr);
2296 * Get the basic file attributes and place them in the provided kstat
2297 * structure. The inode is assumed to be the authoritative source
2298 * for most of the attributes. However, the znode currently has the
2299 * authoritative atime, blksize, and block count.
2301 * IN: ip - inode of file.
2303 * OUT: sp - kstat values.
2305 * RETURN: 0 (always succeeds)
2309 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2311 znode_t *zp = ITOZ(ip);
2312 zfs_sb_t *zsb = ITOZSB(ip);
2317 mutex_enter(&zp->z_lock);
2319 generic_fillattr(ip, sp);
2320 ZFS_TIME_DECODE(&sp->atime, zp->z_atime);
2322 sa_object_size(zp->z_sa_hdl, (uint32_t *)&sp->blksize, &sp->blocks);
2323 if (unlikely(zp->z_blksz == 0)) {
2325 * Block size hasn't been set; suggest maximal I/O transfers.
2327 sp->blksize = zsb->z_max_blksz;
2330 mutex_exit(&zp->z_lock);
2336 EXPORT_SYMBOL(zfs_getattr_fast);
2339 * Set the file attributes to the values contained in the
2342 * IN: ip - inode of file to be modified.
2343 * vap - new attribute values.
2344 * If ATTR_XVATTR set, then optional attrs are being set
2345 * flags - ATTR_UTIME set if non-default time values provided.
2346 * - ATTR_NOACLCHECK (CIFS context only).
2347 * cr - credentials of caller.
2349 * RETURN: 0 if success
2350 * error code if failure
2353 * ip - ctime updated, mtime updated if size changed.
2357 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2359 znode_t *zp = ITOZ(ip);
2360 zfs_sb_t *zsb = ITOZSB(ip);
2364 xvattr_t *tmpxvattr;
2365 uint_t mask = vap->va_mask;
2369 uint64_t new_uid, new_gid;
2371 uint64_t mtime[2], ctime[2];
2373 int need_policy = FALSE;
2375 zfs_fuid_info_t *fuidp = NULL;
2376 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2379 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2380 boolean_t fuid_dirtied = B_FALSE;
2381 sa_bulk_attr_t *bulk, *xattr_bulk;
2382 int count = 0, xattr_count = 0;
2393 * Make sure that if we have ephemeral uid/gid or xvattr specified
2394 * that file system is at proper version level
2397 if (zsb->z_use_fuids == B_FALSE &&
2398 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2399 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2400 (mask & ATTR_XVATTR))) {
2405 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2410 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2416 * If this is an xvattr_t, then get a pointer to the structure of
2417 * optional attributes. If this is NULL, then we have a vattr_t.
2419 xoap = xva_getxoptattr(xvap);
2421 tmpxvattr = kmem_alloc(sizeof(xvattr_t), KM_SLEEP);
2422 xva_init(tmpxvattr);
2424 bulk = kmem_alloc(sizeof(sa_bulk_attr_t) * 7, KM_SLEEP);
2425 xattr_bulk = kmem_alloc(sizeof(sa_bulk_attr_t) * 7, KM_SLEEP);
2428 * Immutable files can only alter immutable bit and atime
2430 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2431 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2432 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2437 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2443 * Verify timestamps doesn't overflow 32 bits.
2444 * ZFS can handle large timestamps, but 32bit syscalls can't
2445 * handle times greater than 2039. This check should be removed
2446 * once large timestamps are fully supported.
2448 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2449 if (((mask & ATTR_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2450 ((mask & ATTR_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2460 /* Can this be moved to before the top label? */
2461 if (zfs_is_readonly(zsb)) {
2467 * First validate permissions
2470 if (mask & ATTR_SIZE) {
2471 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2475 truncate_setsize(ip, vap->va_size);
2478 * XXX - Note, we are not providing any open
2479 * mode flags here (like FNDELAY), so we may
2480 * block if there are locks present... this
2481 * should be addressed in openat().
2483 /* XXX - would it be OK to generate a log record here? */
2484 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2489 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2490 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2491 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2492 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2493 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2494 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2495 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2496 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2497 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2501 if (mask & (ATTR_UID|ATTR_GID)) {
2502 int idmask = (mask & (ATTR_UID|ATTR_GID));
2507 * NOTE: even if a new mode is being set,
2508 * we may clear S_ISUID/S_ISGID bits.
2511 if (!(mask & ATTR_MODE))
2512 vap->va_mode = zp->z_mode;
2515 * Take ownership or chgrp to group we are a member of
2518 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
2519 take_group = (mask & ATTR_GID) &&
2520 zfs_groupmember(zsb, vap->va_gid, cr);
2523 * If both ATTR_UID and ATTR_GID are set then take_owner and
2524 * take_group must both be set in order to allow taking
2527 * Otherwise, send the check through secpolicy_vnode_setattr()
2531 if (((idmask == (ATTR_UID|ATTR_GID)) &&
2532 take_owner && take_group) ||
2533 ((idmask == ATTR_UID) && take_owner) ||
2534 ((idmask == ATTR_GID) && take_group)) {
2535 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2536 skipaclchk, cr) == 0) {
2538 * Remove setuid/setgid for non-privileged users
2540 (void) secpolicy_setid_clear(vap, cr);
2541 trim_mask = (mask & (ATTR_UID|ATTR_GID));
2550 mutex_enter(&zp->z_lock);
2551 oldva.va_mode = zp->z_mode;
2552 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2553 if (mask & ATTR_XVATTR) {
2555 * Update xvattr mask to include only those attributes
2556 * that are actually changing.
2558 * the bits will be restored prior to actually setting
2559 * the attributes so the caller thinks they were set.
2561 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2562 if (xoap->xoa_appendonly !=
2563 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2566 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2567 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
2571 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2572 if (xoap->xoa_nounlink !=
2573 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2576 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2577 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
2581 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2582 if (xoap->xoa_immutable !=
2583 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2586 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2587 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
2591 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2592 if (xoap->xoa_nodump !=
2593 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2596 XVA_CLR_REQ(xvap, XAT_NODUMP);
2597 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
2601 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2602 if (xoap->xoa_av_modified !=
2603 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2606 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2607 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
2611 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2612 if ((!S_ISREG(ip->i_mode) &&
2613 xoap->xoa_av_quarantined) ||
2614 xoap->xoa_av_quarantined !=
2615 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2618 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2619 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
2623 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2624 mutex_exit(&zp->z_lock);
2629 if (need_policy == FALSE &&
2630 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2631 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2636 mutex_exit(&zp->z_lock);
2638 if (mask & ATTR_MODE) {
2639 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2640 err = secpolicy_setid_setsticky_clear(ip, vap,
2645 trim_mask |= ATTR_MODE;
2653 * If trim_mask is set then take ownership
2654 * has been granted or write_acl is present and user
2655 * has the ability to modify mode. In that case remove
2656 * UID|GID and or MODE from mask so that
2657 * secpolicy_vnode_setattr() doesn't revoke it.
2661 saved_mask = vap->va_mask;
2662 vap->va_mask &= ~trim_mask;
2664 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
2665 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2670 vap->va_mask |= saved_mask;
2674 * secpolicy_vnode_setattr, or take ownership may have
2677 mask = vap->va_mask;
2679 if ((mask & (ATTR_UID | ATTR_GID))) {
2680 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
2681 &xattr_obj, sizeof (xattr_obj));
2683 if (err == 0 && xattr_obj) {
2684 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
2688 if (mask & ATTR_UID) {
2689 new_uid = zfs_fuid_create(zsb,
2690 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2691 if (new_uid != zp->z_uid &&
2692 zfs_fuid_overquota(zsb, B_FALSE, new_uid)) {
2700 if (mask & ATTR_GID) {
2701 new_gid = zfs_fuid_create(zsb, (uint64_t)vap->va_gid,
2702 cr, ZFS_GROUP, &fuidp);
2703 if (new_gid != zp->z_gid &&
2704 zfs_fuid_overquota(zsb, B_TRUE, new_gid)) {
2712 tx = dmu_tx_create(zsb->z_os);
2714 if (mask & ATTR_MODE) {
2715 uint64_t pmode = zp->z_mode;
2717 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2719 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
2721 mutex_enter(&zp->z_lock);
2722 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2724 * Are we upgrading ACL from old V0 format
2727 if (zsb->z_version >= ZPL_VERSION_FUID &&
2728 zfs_znode_acl_version(zp) ==
2729 ZFS_ACL_VERSION_INITIAL) {
2730 dmu_tx_hold_free(tx, acl_obj, 0,
2732 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2733 0, aclp->z_acl_bytes);
2735 dmu_tx_hold_write(tx, acl_obj, 0,
2738 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2739 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2740 0, aclp->z_acl_bytes);
2742 mutex_exit(&zp->z_lock);
2743 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2745 if ((mask & ATTR_XVATTR) &&
2746 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2747 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2749 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2753 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
2756 fuid_dirtied = zsb->z_fuid_dirty;
2758 zfs_fuid_txhold(zsb, tx);
2760 zfs_sa_upgrade_txholds(tx, zp);
2762 err = dmu_tx_assign(tx, TXG_NOWAIT);
2764 if (err == ERESTART)
2771 * Set each attribute requested.
2772 * We group settings according to the locks they need to acquire.
2774 * Note: you cannot set ctime directly, although it will be
2775 * updated as a side-effect of calling this function.
2779 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2780 mutex_enter(&zp->z_acl_lock);
2781 mutex_enter(&zp->z_lock);
2783 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
2784 &zp->z_pflags, sizeof (zp->z_pflags));
2787 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2788 mutex_enter(&attrzp->z_acl_lock);
2789 mutex_enter(&attrzp->z_lock);
2790 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2791 SA_ZPL_FLAGS(zsb), NULL, &attrzp->z_pflags,
2792 sizeof (attrzp->z_pflags));
2795 if (mask & (ATTR_UID|ATTR_GID)) {
2797 if (mask & ATTR_UID) {
2798 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zsb), NULL,
2799 &new_uid, sizeof (new_uid));
2800 zp->z_uid = new_uid;
2802 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2803 SA_ZPL_UID(zsb), NULL, &new_uid,
2805 attrzp->z_uid = new_uid;
2809 if (mask & ATTR_GID) {
2810 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zsb),
2811 NULL, &new_gid, sizeof (new_gid));
2812 zp->z_gid = new_gid;
2814 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2815 SA_ZPL_GID(zsb), NULL, &new_gid,
2817 attrzp->z_gid = new_gid;
2820 if (!(mask & ATTR_MODE)) {
2821 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb),
2822 NULL, &new_mode, sizeof (new_mode));
2823 new_mode = zp->z_mode;
2825 err = zfs_acl_chown_setattr(zp);
2828 err = zfs_acl_chown_setattr(attrzp);
2833 if (mask & ATTR_MODE) {
2834 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb), NULL,
2835 &new_mode, sizeof (new_mode));
2836 zp->z_mode = new_mode;
2837 ASSERT3P(aclp, !=, NULL);
2838 err = zfs_aclset_common(zp, aclp, cr, tx);
2839 ASSERT3U(err, ==, 0);
2840 if (zp->z_acl_cached)
2841 zfs_acl_free(zp->z_acl_cached);
2842 zp->z_acl_cached = aclp;
2847 if (mask & ATTR_ATIME) {
2848 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
2849 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zsb), NULL,
2850 &zp->z_atime, sizeof (zp->z_atime));
2853 if (mask & ATTR_MTIME) {
2854 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
2855 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL,
2856 mtime, sizeof (mtime));
2859 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2860 if (mask & ATTR_SIZE && !(mask & ATTR_MTIME)) {
2861 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb),
2862 NULL, mtime, sizeof (mtime));
2863 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
2864 &ctime, sizeof (ctime));
2865 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
2867 } else if (mask != 0) {
2868 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
2869 &ctime, sizeof (ctime));
2870 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
2873 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2874 SA_ZPL_CTIME(zsb), NULL,
2875 &ctime, sizeof (ctime));
2876 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
2877 mtime, ctime, B_TRUE);
2881 * Do this after setting timestamps to prevent timestamp
2882 * update from toggling bit
2885 if (xoap && (mask & ATTR_XVATTR)) {
2888 * restore trimmed off masks
2889 * so that return masks can be set for caller.
2892 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
2893 XVA_SET_REQ(xvap, XAT_APPENDONLY);
2895 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
2896 XVA_SET_REQ(xvap, XAT_NOUNLINK);
2898 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
2899 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
2901 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
2902 XVA_SET_REQ(xvap, XAT_NODUMP);
2904 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
2905 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
2907 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
2908 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
2911 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2912 ASSERT(S_ISREG(ip->i_mode));
2914 zfs_xvattr_set(zp, xvap, tx);
2918 zfs_fuid_sync(zsb, tx);
2921 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2923 mutex_exit(&zp->z_lock);
2924 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2925 mutex_exit(&zp->z_acl_lock);
2928 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2929 mutex_exit(&attrzp->z_acl_lock);
2930 mutex_exit(&attrzp->z_lock);
2933 if (err == 0 && attrzp) {
2934 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
2945 zfs_fuid_info_free(fuidp);
2951 if (err == ERESTART)
2954 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
2956 zfs_inode_update(zp);
2960 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
2961 zil_commit(zilog, 0);
2964 kmem_free(xattr_bulk, sizeof(sa_bulk_attr_t) * 7);
2965 kmem_free(bulk, sizeof(sa_bulk_attr_t) * 7);
2966 kmem_free(tmpxvattr, sizeof(xvattr_t));
2970 EXPORT_SYMBOL(zfs_setattr);
2972 typedef struct zfs_zlock {
2973 krwlock_t *zl_rwlock; /* lock we acquired */
2974 znode_t *zl_znode; /* znode we held */
2975 struct zfs_zlock *zl_next; /* next in list */
2979 * Drop locks and release vnodes that were held by zfs_rename_lock().
2982 zfs_rename_unlock(zfs_zlock_t **zlpp)
2986 while ((zl = *zlpp) != NULL) {
2987 if (zl->zl_znode != NULL)
2988 iput(ZTOI(zl->zl_znode));
2989 rw_exit(zl->zl_rwlock);
2990 *zlpp = zl->zl_next;
2991 kmem_free(zl, sizeof (*zl));
2996 * Search back through the directory tree, using the ".." entries.
2997 * Lock each directory in the chain to prevent concurrent renames.
2998 * Fail any attempt to move a directory into one of its own descendants.
2999 * XXX - z_parent_lock can overlap with map or grow locks
3002 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3006 uint64_t rootid = ZTOZSB(zp)->z_root;
3007 uint64_t oidp = zp->z_id;
3008 krwlock_t *rwlp = &szp->z_parent_lock;
3009 krw_t rw = RW_WRITER;
3012 * First pass write-locks szp and compares to zp->z_id.
3013 * Later passes read-lock zp and compare to zp->z_parent.
3016 if (!rw_tryenter(rwlp, rw)) {
3018 * Another thread is renaming in this path.
3019 * Note that if we are a WRITER, we don't have any
3020 * parent_locks held yet.
3022 if (rw == RW_READER && zp->z_id > szp->z_id) {
3024 * Drop our locks and restart
3026 zfs_rename_unlock(&zl);
3030 rwlp = &szp->z_parent_lock;
3035 * Wait for other thread to drop its locks
3041 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3042 zl->zl_rwlock = rwlp;
3043 zl->zl_znode = NULL;
3044 zl->zl_next = *zlpp;
3047 if (oidp == szp->z_id) /* We're a descendant of szp */
3050 if (oidp == rootid) /* We've hit the top */
3053 if (rw == RW_READER) { /* i.e. not the first pass */
3054 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3059 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3060 &oidp, sizeof (oidp));
3061 rwlp = &zp->z_parent_lock;
3064 } while (zp->z_id != sdzp->z_id);
3070 * Move an entry from the provided source directory to the target
3071 * directory. Change the entry name as indicated.
3073 * IN: sdip - Source directory containing the "old entry".
3074 * snm - Old entry name.
3075 * tdip - Target directory to contain the "new entry".
3076 * tnm - New entry name.
3077 * cr - credentials of caller.
3078 * flags - case flags
3080 * RETURN: 0 if success
3081 * error code if failure
3084 * sdip,tdip - ctime|mtime updated
3088 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3089 cred_t *cr, int flags)
3091 znode_t *tdzp, *szp, *tzp;
3092 znode_t *sdzp = ITOZ(sdip);
3093 zfs_sb_t *zsb = ITOZSB(sdip);
3095 zfs_dirlock_t *sdl, *tdl;
3098 int cmp, serr, terr;
3103 ZFS_VERIFY_ZP(sdzp);
3106 if (tdip->i_sb != sdip->i_sb) {
3112 ZFS_VERIFY_ZP(tdzp);
3113 if (zsb->z_utf8 && u8_validate(tnm,
3114 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3119 if (flags & FIGNORECASE)
3128 * This is to prevent the creation of links into attribute space
3129 * by renaming a linked file into/outof an attribute directory.
3130 * See the comment in zfs_link() for why this is considered bad.
3132 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3138 * Lock source and target directory entries. To prevent deadlock,
3139 * a lock ordering must be defined. We lock the directory with
3140 * the smallest object id first, or if it's a tie, the one with
3141 * the lexically first name.
3143 if (sdzp->z_id < tdzp->z_id) {
3145 } else if (sdzp->z_id > tdzp->z_id) {
3149 * First compare the two name arguments without
3150 * considering any case folding.
3152 int nofold = (zsb->z_norm & ~U8_TEXTPREP_TOUPPER);
3154 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3155 ASSERT(error == 0 || !zsb->z_utf8);
3158 * POSIX: "If the old argument and the new argument
3159 * both refer to links to the same existing file,
3160 * the rename() function shall return successfully
3161 * and perform no other action."
3167 * If the file system is case-folding, then we may
3168 * have some more checking to do. A case-folding file
3169 * system is either supporting mixed case sensitivity
3170 * access or is completely case-insensitive. Note
3171 * that the file system is always case preserving.
3173 * In mixed sensitivity mode case sensitive behavior
3174 * is the default. FIGNORECASE must be used to
3175 * explicitly request case insensitive behavior.
3177 * If the source and target names provided differ only
3178 * by case (e.g., a request to rename 'tim' to 'Tim'),
3179 * we will treat this as a special case in the
3180 * case-insensitive mode: as long as the source name
3181 * is an exact match, we will allow this to proceed as
3182 * a name-change request.
3184 if ((zsb->z_case == ZFS_CASE_INSENSITIVE ||
3185 (zsb->z_case == ZFS_CASE_MIXED &&
3186 flags & FIGNORECASE)) &&
3187 u8_strcmp(snm, tnm, 0, zsb->z_norm, U8_UNICODE_LATEST,
3190 * case preserving rename request, require exact
3199 * If the source and destination directories are the same, we should
3200 * grab the z_name_lock of that directory only once.
3204 rw_enter(&sdzp->z_name_lock, RW_READER);
3208 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3209 ZEXISTS | zflg, NULL, NULL);
3210 terr = zfs_dirent_lock(&tdl,
3211 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3213 terr = zfs_dirent_lock(&tdl,
3214 tdzp, tnm, &tzp, zflg, NULL, NULL);
3215 serr = zfs_dirent_lock(&sdl,
3216 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3222 * Source entry invalid or not there.
3225 zfs_dirent_unlock(tdl);
3231 rw_exit(&sdzp->z_name_lock);
3233 if (strcmp(snm, "..") == 0)
3239 zfs_dirent_unlock(sdl);
3243 rw_exit(&sdzp->z_name_lock);
3245 if (strcmp(tnm, "..") == 0)
3252 * Must have write access at the source to remove the old entry
3253 * and write access at the target to create the new entry.
3254 * Note that if target and source are the same, this can be
3255 * done in a single check.
3258 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3261 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3263 * Check to make sure rename is valid.
3264 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3266 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3271 * Does target exist?
3275 * Source and target must be the same type.
3277 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3278 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3283 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3289 * POSIX dictates that when the source and target
3290 * entries refer to the same file object, rename
3291 * must do nothing and exit without error.
3293 if (szp->z_id == tzp->z_id) {
3299 tx = dmu_tx_create(zsb->z_os);
3300 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3301 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3302 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3303 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3305 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3306 zfs_sa_upgrade_txholds(tx, tdzp);
3309 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3310 zfs_sa_upgrade_txholds(tx, tzp);
3313 zfs_sa_upgrade_txholds(tx, szp);
3314 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
3315 error = dmu_tx_assign(tx, TXG_NOWAIT);
3318 zfs_rename_unlock(&zl);
3319 zfs_dirent_unlock(sdl);
3320 zfs_dirent_unlock(tdl);
3323 rw_exit(&sdzp->z_name_lock);
3328 if (error == ERESTART) {
3338 if (tzp) /* Attempt to remove the existing target */
3339 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3342 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3344 szp->z_pflags |= ZFS_AV_MODIFIED;
3346 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zsb),
3347 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3348 ASSERT3U(error, ==, 0);
3350 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3352 zfs_log_rename(zilog, tx, TX_RENAME |
3353 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3354 sdl->dl_name, tdzp, tdl->dl_name, szp);
3357 * At this point, we have successfully created
3358 * the target name, but have failed to remove
3359 * the source name. Since the create was done
3360 * with the ZRENAMING flag, there are
3361 * complications; for one, the link count is
3362 * wrong. The easiest way to deal with this
3363 * is to remove the newly created target, and
3364 * return the original error. This must
3365 * succeed; fortunately, it is very unlikely to
3366 * fail, since we just created it.
3368 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3369 ZRENAMING, NULL), ==, 0);
3377 zfs_rename_unlock(&zl);
3379 zfs_dirent_unlock(sdl);
3380 zfs_dirent_unlock(tdl);
3382 zfs_inode_update(sdzp);
3384 rw_exit(&sdzp->z_name_lock);
3387 zfs_inode_update(tdzp);
3389 zfs_inode_update(szp);
3392 zfs_inode_update(tzp);
3396 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3397 zil_commit(zilog, 0);
3402 EXPORT_SYMBOL(zfs_rename);
3405 * Insert the indicated symbolic reference entry into the directory.
3407 * IN: dip - Directory to contain new symbolic link.
3408 * link - Name for new symlink entry.
3409 * vap - Attributes of new entry.
3410 * target - Target path of new symlink.
3412 * cr - credentials of caller.
3413 * flags - case flags
3415 * RETURN: 0 if success
3416 * error code if failure
3419 * dip - ctime|mtime updated
3423 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
3424 struct inode **ipp, cred_t *cr, int flags)
3426 znode_t *zp, *dzp = ITOZ(dip);
3429 zfs_sb_t *zsb = ITOZSB(dip);
3431 uint64_t len = strlen(link);
3434 zfs_acl_ids_t acl_ids;
3435 boolean_t fuid_dirtied;
3436 uint64_t txtype = TX_SYMLINK;
3438 ASSERT(S_ISLNK(vap->va_mode));
3444 if (zsb->z_utf8 && u8_validate(name, strlen(name),
3445 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3449 if (flags & FIGNORECASE)
3452 if (len > MAXPATHLEN) {
3454 return (ENAMETOOLONG);
3457 if ((error = zfs_acl_ids_create(dzp, 0,
3458 vap, cr, NULL, &acl_ids)) != 0) {
3466 * Attempt to lock directory; fail if entry already exists.
3468 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3470 zfs_acl_ids_free(&acl_ids);
3475 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3476 zfs_acl_ids_free(&acl_ids);
3477 zfs_dirent_unlock(dl);
3482 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
3483 zfs_acl_ids_free(&acl_ids);
3484 zfs_dirent_unlock(dl);
3488 tx = dmu_tx_create(zsb->z_os);
3489 fuid_dirtied = zsb->z_fuid_dirty;
3490 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3491 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3492 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3493 ZFS_SA_BASE_ATTR_SIZE + len);
3494 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3495 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3496 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3497 acl_ids.z_aclp->z_acl_bytes);
3500 zfs_fuid_txhold(zsb, tx);
3501 error = dmu_tx_assign(tx, TXG_NOWAIT);
3503 zfs_dirent_unlock(dl);
3504 if (error == ERESTART) {
3509 zfs_acl_ids_free(&acl_ids);
3516 * Create a new object for the symlink.
3517 * for version 4 ZPL datsets the symlink will be an SA attribute
3519 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3522 zfs_fuid_sync(zsb, tx);
3524 mutex_enter(&zp->z_lock);
3526 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zsb),
3529 zfs_sa_symlink(zp, link, len, tx);
3530 mutex_exit(&zp->z_lock);
3533 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
3534 &zp->z_size, sizeof (zp->z_size), tx);
3536 * Insert the new object into the directory.
3538 (void) zfs_link_create(dl, zp, tx, ZNEW);
3540 if (flags & FIGNORECASE)
3542 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3544 zfs_inode_update(dzp);
3545 zfs_inode_update(zp);
3547 zfs_acl_ids_free(&acl_ids);
3551 zfs_dirent_unlock(dl);
3555 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3556 zil_commit(zilog, 0);
3561 EXPORT_SYMBOL(zfs_symlink);
3564 * Return, in the buffer contained in the provided uio structure,
3565 * the symbolic path referred to by ip.
3567 * IN: ip - inode of symbolic link
3568 * uio - structure to contain the link path.
3569 * cr - credentials of caller.
3571 * RETURN: 0 if success
3572 * error code if failure
3575 * ip - atime updated
3579 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
3581 znode_t *zp = ITOZ(ip);
3582 zfs_sb_t *zsb = ITOZSB(ip);
3588 mutex_enter(&zp->z_lock);
3590 error = sa_lookup_uio(zp->z_sa_hdl,
3591 SA_ZPL_SYMLINK(zsb), uio);
3593 error = zfs_sa_readlink(zp, uio);
3594 mutex_exit(&zp->z_lock);
3596 ZFS_ACCESSTIME_STAMP(zsb, zp);
3597 zfs_inode_update(zp);
3601 EXPORT_SYMBOL(zfs_readlink);
3604 * Insert a new entry into directory tdip referencing sip.
3606 * IN: tdip - Directory to contain new entry.
3607 * sip - inode of new entry.
3608 * name - name of new entry.
3609 * cr - credentials of caller.
3611 * RETURN: 0 if success
3612 * error code if failure
3615 * tdip - ctime|mtime updated
3616 * sip - ctime updated
3620 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr)
3622 znode_t *dzp = ITOZ(tdip);
3624 zfs_sb_t *zsb = ITOZSB(tdip);
3633 ASSERT(S_ISDIR(tdip->i_mode));
3640 * POSIX dictates that we return EPERM here.
3641 * Better choices include ENOTSUP or EISDIR.
3643 if (S_ISDIR(sip->i_mode)) {
3648 if (sip->i_sb != tdip->i_sb) {
3656 /* Prevent links to .zfs/shares files */
3658 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zsb),
3659 &parent, sizeof (uint64_t))) != 0) {
3663 if (parent == zsb->z_shares_dir) {
3668 if (zsb->z_utf8 && u8_validate(name,
3669 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3673 #ifdef HAVE_PN_UTILS
3674 if (flags & FIGNORECASE)
3676 #endif /* HAVE_PN_UTILS */
3679 * We do not support links between attributes and non-attributes
3680 * because of the potential security risk of creating links
3681 * into "normal" file space in order to circumvent restrictions
3682 * imposed in attribute space.
3684 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3689 owner = zfs_fuid_map_id(zsb, szp->z_uid, cr, ZFS_OWNER);
3690 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3695 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3702 * Attempt to lock directory; fail if entry already exists.
3704 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3710 tx = dmu_tx_create(zsb->z_os);
3711 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3712 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3713 zfs_sa_upgrade_txholds(tx, szp);
3714 zfs_sa_upgrade_txholds(tx, dzp);
3715 error = dmu_tx_assign(tx, TXG_NOWAIT);
3717 zfs_dirent_unlock(dl);
3718 if (error == ERESTART) {
3728 error = zfs_link_create(dl, szp, tx, 0);
3731 uint64_t txtype = TX_LINK;
3732 #ifdef HAVE_PN_UTILS
3733 if (flags & FIGNORECASE)
3735 #endif /* HAVE_PN_UTILS */
3736 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3741 zfs_dirent_unlock(dl);
3743 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3744 zil_commit(zilog, 0);
3746 zfs_inode_update(dzp);
3747 zfs_inode_update(szp);
3751 EXPORT_SYMBOL(zfs_link);
3754 zfs_putpage_commit_cb(void *arg, int error)
3756 struct page *pp = arg;
3759 __set_page_dirty_nobuffers(pp);
3761 if (error != ECANCELED)
3767 end_page_writeback(pp);
3771 * Push a page out to disk, once the page is on stable storage the
3772 * registered commit callback will be run as notification of completion.
3774 * IN: ip - page mapped for inode.
3775 * pp - page to push (page is locked)
3776 * wbc - writeback control data
3778 * RETURN: 0 if success
3779 * error code if failure
3782 * ip - ctime|mtime updated
3786 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
3788 znode_t *zp = ITOZ(ip);
3789 zfs_sb_t *zsb = ITOZSB(ip);
3796 uint64_t mtime[2], ctime[2];
3797 sa_bulk_attr_t bulk[3];
3801 ASSERT(PageLocked(pp));
3803 pgoff = page_offset(pp); /* Page byte-offset in file */
3804 offset = i_size_read(ip); /* File length in bytes */
3805 pglen = MIN(PAGE_CACHE_SIZE, /* Page length in bytes */
3806 P2ROUNDUP(offset, PAGE_CACHE_SIZE)-pgoff);
3808 /* Page is beyond end of file */
3809 if (pgoff >= offset) {
3814 /* Truncate page length to end of file */
3815 if (pgoff + pglen > offset)
3816 pglen = offset - pgoff;
3820 * FIXME: Allow mmap writes past its quota. The correct fix
3821 * is to register a page_mkwrite() handler to count the page
3822 * against its quota when it is about to be dirtied.
3824 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
3825 zfs_owner_overquota(zsb, zp, B_TRUE)) {
3830 set_page_writeback(pp);
3833 tx = dmu_tx_create(zsb->z_os);
3835 dmu_tx_callback_register(tx, zfs_putpage_commit_cb, pp);
3837 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
3839 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3840 zfs_sa_upgrade_txholds(tx, zp);
3841 err = dmu_tx_assign(tx, TXG_NOWAIT);
3843 if (err == ERESTART)
3851 ASSERT3U(pglen, <=, PAGE_CACHE_SIZE);
3852 dmu_write(zsb->z_os, zp->z_id, pgoff, pglen, va, tx);
3855 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
3856 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
3857 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zsb), NULL, &zp->z_pflags, 8);
3858 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
3859 zfs_log_write(zsb->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0);
3862 ASSERT3S(err, ==, 0);
3864 if ((zsb->z_os->os_sync == ZFS_SYNC_ALWAYS) ||
3865 (wbc->sync_mode == WB_SYNC_ALL))
3866 zil_commit(zsb->z_log, zp->z_id);
3873 zfs_inactive(struct inode *ip)
3875 znode_t *zp = ITOZ(ip);
3876 zfs_sb_t *zsb = ITOZSB(ip);
3879 #ifdef HAVE_SNAPSHOT
3880 /* Early return for snapshot inode? */
3881 #endif /* HAVE_SNAPSHOT */
3883 rw_enter(&zsb->z_teardown_inactive_lock, RW_READER);
3884 if (zp->z_sa_hdl == NULL) {
3885 rw_exit(&zsb->z_teardown_inactive_lock);
3889 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3890 dmu_tx_t *tx = dmu_tx_create(zsb->z_os);
3892 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3893 zfs_sa_upgrade_txholds(tx, zp);
3894 error = dmu_tx_assign(tx, TXG_WAIT);
3898 mutex_enter(&zp->z_lock);
3899 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zsb),
3900 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
3901 zp->z_atime_dirty = 0;
3902 mutex_exit(&zp->z_lock);
3908 rw_exit(&zsb->z_teardown_inactive_lock);
3910 EXPORT_SYMBOL(zfs_inactive);
3913 * Bounds-check the seek operation.
3915 * IN: ip - inode seeking within
3916 * ooff - old file offset
3917 * noffp - pointer to new file offset
3918 * ct - caller context
3920 * RETURN: 0 if success
3921 * EINVAL if new offset invalid
3925 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
3927 if (S_ISDIR(ip->i_mode))
3929 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3931 EXPORT_SYMBOL(zfs_seek);
3934 * Fill pages with data from the disk.
3937 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
3939 znode_t *zp = ITOZ(ip);
3940 zfs_sb_t *zsb = ITOZSB(ip);
3942 struct page *cur_pp;
3943 u_offset_t io_off, total;
3950 io_len = nr_pages << PAGE_CACHE_SHIFT;
3951 i_size = i_size_read(ip);
3952 io_off = page_offset(pl[0]);
3954 if (io_off + io_len > i_size)
3955 io_len = i_size - io_off;
3958 * Iterate over list of pages and read each page individually.
3962 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3966 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
3970 /* convert checksum errors into IO errors */
3975 cur_pp = pl[++page_idx];
3982 * Uses zfs_fillpage to read data from the file and fill the pages.
3984 * IN: ip - inode of file to get data from.
3985 * pl - list of pages to read
3986 * nr_pages - number of pages to read
3988 * RETURN: 0 if success
3989 * error code if failure
3992 * vp - atime updated
3996 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
3998 znode_t *zp = ITOZ(ip);
3999 zfs_sb_t *zsb = ITOZSB(ip);
4008 err = zfs_fillpage(ip, pl, nr_pages);
4011 ZFS_ACCESSTIME_STAMP(zsb, zp);
4016 EXPORT_SYMBOL(zfs_getpage);
4019 * Check ZFS specific permissions to memory map a section of a file.
4021 * IN: ip - inode of the file to mmap
4023 * addrp - start address in memory region
4024 * len - length of memory region
4025 * vm_flags- address flags
4027 * RETURN: 0 if success
4028 * error code if failure
4032 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4033 unsigned long vm_flags)
4035 znode_t *zp = ITOZ(ip);
4036 zfs_sb_t *zsb = ITOZSB(ip);
4041 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4042 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4047 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4048 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4053 if (off < 0 || len > MAXOFFSET_T - off) {
4061 EXPORT_SYMBOL(zfs_map);
4064 * convoff - converts the given data (start, whence) to the
4068 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4073 if ((lckdat->l_whence == 2) || (whence == 2)) {
4074 if ((error = zfs_getattr(ip, &vap, 0, CRED()) != 0))
4078 switch (lckdat->l_whence) {
4080 lckdat->l_start += offset;
4083 lckdat->l_start += vap.va_size;
4091 if (lckdat->l_start < 0)
4096 lckdat->l_start -= offset;
4099 lckdat->l_start -= vap.va_size;
4107 lckdat->l_whence = (short)whence;
4112 * Free or allocate space in a file. Currently, this function only
4113 * supports the `F_FREESP' command. However, this command is somewhat
4114 * misnamed, as its functionality includes the ability to allocate as
4115 * well as free space.
4117 * IN: ip - inode of file to free data in.
4118 * cmd - action to take (only F_FREESP supported).
4119 * bfp - section of file to free/alloc.
4120 * flag - current file open mode flags.
4121 * offset - current file offset.
4122 * cr - credentials of caller [UNUSED].
4124 * RETURN: 0 if success
4125 * error code if failure
4128 * ip - ctime|mtime updated
4132 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4133 offset_t offset, cred_t *cr)
4135 znode_t *zp = ITOZ(ip);
4136 zfs_sb_t *zsb = ITOZSB(ip);
4143 if (cmd != F_FREESP) {
4148 if ((error = convoff(ip, bfp, 0, offset))) {
4153 if (bfp->l_len < 0) {
4159 * Permissions aren't checked on Solaris because on this OS
4160 * zfs_space() can only be called with an opened file handle.
4161 * On Linux we can get here through truncate_range() which
4162 * operates directly on inodes, so we need to check access rights.
4164 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4170 len = bfp->l_len; /* 0 means from off to end of file */
4172 error = zfs_freesp(zp, off, len, flag, TRUE);
4177 EXPORT_SYMBOL(zfs_space);
4181 zfs_fid(struct inode *ip, fid_t *fidp)
4183 znode_t *zp = ITOZ(ip);
4184 zfs_sb_t *zsb = ITOZSB(ip);
4187 uint64_t object = zp->z_id;
4194 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zsb),
4195 &gen64, sizeof (uint64_t))) != 0) {
4200 gen = (uint32_t)gen64;
4202 size = (zsb->z_parent != zsb) ? LONG_FID_LEN : SHORT_FID_LEN;
4203 if (fidp->fid_len < size) {
4204 fidp->fid_len = size;
4209 zfid = (zfid_short_t *)fidp;
4211 zfid->zf_len = size;
4213 for (i = 0; i < sizeof (zfid->zf_object); i++)
4214 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4216 /* Must have a non-zero generation number to distinguish from .zfs */
4219 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4220 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4222 if (size == LONG_FID_LEN) {
4223 uint64_t objsetid = dmu_objset_id(zsb->z_os);
4226 zlfid = (zfid_long_t *)fidp;
4228 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4229 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4231 /* XXX - this should be the generation number for the objset */
4232 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4233 zlfid->zf_setgen[i] = 0;
4239 EXPORT_SYMBOL(zfs_fid);
4243 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4245 znode_t *zp = ITOZ(ip);
4246 zfs_sb_t *zsb = ITOZSB(ip);
4248 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4252 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4257 EXPORT_SYMBOL(zfs_getsecattr);
4261 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4263 znode_t *zp = ITOZ(ip);
4264 zfs_sb_t *zsb = ITOZSB(ip);
4266 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4267 zilog_t *zilog = zsb->z_log;
4272 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4274 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
4275 zil_commit(zilog, 0);
4280 EXPORT_SYMBOL(zfs_setsecattr);
4282 #ifdef HAVE_UIO_ZEROCOPY
4284 * Tunable, both must be a power of 2.
4286 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4287 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4288 * an arcbuf for a partial block read
4290 int zcr_blksz_min = (1 << 10); /* 1K */
4291 int zcr_blksz_max = (1 << 17); /* 128K */
4295 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
4297 znode_t *zp = ITOZ(ip);
4298 zfs_sb_t *zsb = ITOZSB(ip);
4299 int max_blksz = zsb->z_max_blksz;
4300 uio_t *uio = &xuio->xu_uio;
4301 ssize_t size = uio->uio_resid;
4302 offset_t offset = uio->uio_loffset;
4307 int preamble, postamble;
4309 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4317 * Loan out an arc_buf for write if write size is bigger than
4318 * max_blksz, and the file's block size is also max_blksz.
4321 if (size < blksz || zp->z_blksz != blksz) {
4326 * Caller requests buffers for write before knowing where the
4327 * write offset might be (e.g. NFS TCP write).
4332 preamble = P2PHASE(offset, blksz);
4334 preamble = blksz - preamble;
4339 postamble = P2PHASE(size, blksz);
4342 fullblk = size / blksz;
4343 (void) dmu_xuio_init(xuio,
4344 (preamble != 0) + fullblk + (postamble != 0));
4347 * Have to fix iov base/len for partial buffers. They
4348 * currently represent full arc_buf's.
4351 /* data begins in the middle of the arc_buf */
4352 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4355 (void) dmu_xuio_add(xuio, abuf,
4356 blksz - preamble, preamble);
4359 for (i = 0; i < fullblk; i++) {
4360 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4363 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
4367 /* data ends in the middle of the arc_buf */
4368 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4371 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
4376 * Loan out an arc_buf for read if the read size is larger than
4377 * the current file block size. Block alignment is not
4378 * considered. Partial arc_buf will be loaned out for read.
4380 blksz = zp->z_blksz;
4381 if (blksz < zcr_blksz_min)
4382 blksz = zcr_blksz_min;
4383 if (blksz > zcr_blksz_max)
4384 blksz = zcr_blksz_max;
4385 /* avoid potential complexity of dealing with it */
4386 if (blksz > max_blksz) {
4391 maxsize = zp->z_size - uio->uio_loffset;
4405 uio->uio_extflg = UIO_XUIO;
4406 XUIO_XUZC_RW(xuio) = ioflag;
4413 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
4417 int ioflag = XUIO_XUZC_RW(xuio);
4419 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
4421 i = dmu_xuio_cnt(xuio);
4423 abuf = dmu_xuio_arcbuf(xuio, i);
4425 * if abuf == NULL, it must be a write buffer
4426 * that has been returned in zfs_write().
4429 dmu_return_arcbuf(abuf);
4430 ASSERT(abuf || ioflag == UIO_WRITE);
4433 dmu_xuio_fini(xuio);
4436 #endif /* HAVE_UIO_ZEROCOPY */
4438 #if defined(_KERNEL) && defined(HAVE_SPL)
4439 module_param(zfs_read_chunk_size, long, 0644);
4440 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");