4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
25 * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
26 * Copyright 2017 Nexenta Systems, Inc.
29 /* Portions Copyright 2007 Jeremy Teo */
30 /* Portions Copyright 2010 Robert Milkowski */
33 #include <sys/types.h>
34 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/sysmacros.h>
38 #include <sys/resource.h>
40 #include <sys/vfs_opreg.h>
44 #include <sys/taskq.h>
46 #include <sys/vmsystm.h>
47 #include <sys/atomic.h>
49 #include <sys/pathname.h>
50 #include <sys/cmn_err.h>
51 #include <sys/errno.h>
52 #include <sys/unistd.h>
53 #include <sys/zfs_dir.h>
54 #include <sys/zfs_acl.h>
55 #include <sys/zfs_ioctl.h>
56 #include <sys/fs/zfs.h>
58 #include <sys/dmu_objset.h>
64 #include <sys/dirent.h>
65 #include <sys/policy.h>
66 #include <sys/sunddi.h>
69 #include "fs/fs_subr.h"
70 #include <sys/zfs_ctldir.h>
71 #include <sys/zfs_fuid.h>
72 #include <sys/zfs_sa.h>
73 #include <sys/zfs_vnops.h>
75 #include <sys/zfs_rlock.h>
76 #include <sys/extdirent.h>
77 #include <sys/kidmap.h>
85 * Each vnode op performs some logical unit of work. To do this, the ZPL must
86 * properly lock its in-core state, create a DMU transaction, do the work,
87 * record this work in the intent log (ZIL), commit the DMU transaction,
88 * and wait for the intent log to commit if it is a synchronous operation.
89 * Moreover, the vnode ops must work in both normal and log replay context.
90 * The ordering of events is important to avoid deadlocks and references
91 * to freed memory. The example below illustrates the following Big Rules:
93 * (1) A check must be made in each zfs thread for a mounted file system.
94 * This is done avoiding races using ZFS_ENTER(zfsvfs).
95 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
96 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
97 * can return EIO from the calling function.
99 * (2) iput() should always be the last thing except for zil_commit()
100 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
101 * First, if it's the last reference, the vnode/znode
102 * can be freed, so the zp may point to freed memory. Second, the last
103 * reference will call zfs_zinactive(), which may induce a lot of work --
104 * pushing cached pages (which acquires range locks) and syncing out
105 * cached atime changes. Third, zfs_zinactive() may require a new tx,
106 * which could deadlock the system if you were already holding one.
107 * If you must call iput() within a tx then use zfs_iput_async().
109 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
110 * as they can span dmu_tx_assign() calls.
112 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
113 * dmu_tx_assign(). This is critical because we don't want to block
114 * while holding locks.
116 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
117 * reduces lock contention and CPU usage when we must wait (note that if
118 * throughput is constrained by the storage, nearly every transaction
121 * Note, in particular, that if a lock is sometimes acquired before
122 * the tx assigns, and sometimes after (e.g. z_lock), then failing
123 * to use a non-blocking assign can deadlock the system. The scenario:
125 * Thread A has grabbed a lock before calling dmu_tx_assign().
126 * Thread B is in an already-assigned tx, and blocks for this lock.
127 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
128 * forever, because the previous txg can't quiesce until B's tx commits.
130 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
131 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
132 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
133 * to indicate that this operation has already called dmu_tx_wait().
134 * This will ensure that we don't retry forever, waiting a short bit
137 * (5) If the operation succeeded, generate the intent log entry for it
138 * before dropping locks. This ensures that the ordering of events
139 * in the intent log matches the order in which they actually occurred.
140 * During ZIL replay the zfs_log_* functions will update the sequence
141 * number to indicate the zil transaction has replayed.
143 * (6) At the end of each vnode op, the DMU tx must always commit,
144 * regardless of whether there were any errors.
146 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
147 * to ensure that synchronous semantics are provided when necessary.
149 * In general, this is how things should be ordered in each vnode op:
151 * ZFS_ENTER(zfsvfs); // exit if unmounted
153 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
154 * rw_enter(...); // grab any other locks you need
155 * tx = dmu_tx_create(...); // get DMU tx
156 * dmu_tx_hold_*(); // hold each object you might modify
157 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
159 * rw_exit(...); // drop locks
160 * zfs_dirent_unlock(dl); // unlock directory entry
161 * iput(...); // release held vnodes
162 * if (error == ERESTART) {
168 * dmu_tx_abort(tx); // abort DMU tx
169 * ZFS_EXIT(zfsvfs); // finished in zfs
170 * return (error); // really out of space
172 * error = do_real_work(); // do whatever this VOP does
174 * zfs_log_*(...); // on success, make ZIL entry
175 * dmu_tx_commit(tx); // commit DMU tx -- error or not
176 * rw_exit(...); // drop locks
177 * zfs_dirent_unlock(dl); // unlock directory entry
178 * iput(...); // release held vnodes
179 * zil_commit(zilog, foid); // synchronous when necessary
180 * ZFS_EXIT(zfsvfs); // finished in zfs
181 * return (error); // done, report error
185 * Virus scanning is unsupported. It would be possible to add a hook
186 * here to performance the required virus scan. This could be done
187 * entirely in the kernel or potentially as an update to invoke a
191 zfs_vscan(struct inode *ip, cred_t *cr, int async)
198 zfs_open(struct inode *ip, int mode, int flag, cred_t *cr)
200 znode_t *zp = ITOZ(ip);
201 zfsvfs_t *zfsvfs = ITOZSB(ip);
206 /* Honor ZFS_APPENDONLY file attribute */
207 if ((mode & FMODE_WRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
208 ((flag & O_APPEND) == 0)) {
210 return (SET_ERROR(EPERM));
213 /* Virus scan eligible files on open */
214 if (!zfs_has_ctldir(zp) && zfsvfs->z_vscan && S_ISREG(ip->i_mode) &&
215 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
216 if (zfs_vscan(ip, cr, 0) != 0) {
218 return (SET_ERROR(EACCES));
222 /* Keep a count of the synchronous opens in the znode */
224 atomic_inc_32(&zp->z_sync_cnt);
232 zfs_close(struct inode *ip, int flag, cred_t *cr)
234 znode_t *zp = ITOZ(ip);
235 zfsvfs_t *zfsvfs = ITOZSB(ip);
240 /* Decrement the synchronous opens in the znode */
242 atomic_dec_32(&zp->z_sync_cnt);
244 if (!zfs_has_ctldir(zp) && zfsvfs->z_vscan && S_ISREG(ip->i_mode) &&
245 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
246 VERIFY(zfs_vscan(ip, cr, 1) == 0);
252 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
254 * Lseek support for finding holes (cmd == SEEK_HOLE) and
255 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
258 zfs_holey_common(struct inode *ip, int cmd, loff_t *off)
260 znode_t *zp = ITOZ(ip);
261 uint64_t noff = (uint64_t)*off; /* new offset */
266 file_sz = zp->z_size;
267 if (noff >= file_sz) {
268 return (SET_ERROR(ENXIO));
271 if (cmd == SEEK_HOLE)
276 error = dmu_offset_next(ZTOZSB(zp)->z_os, zp->z_id, hole, &noff);
279 return (SET_ERROR(ENXIO));
281 /* file was dirty, so fall back to using generic logic */
282 if (error == EBUSY) {
290 * We could find a hole that begins after the logical end-of-file,
291 * because dmu_offset_next() only works on whole blocks. If the
292 * EOF falls mid-block, then indicate that the "virtual hole"
293 * at the end of the file begins at the logical EOF, rather than
294 * at the end of the last block.
296 if (noff > file_sz) {
308 zfs_holey(struct inode *ip, int cmd, loff_t *off)
310 znode_t *zp = ITOZ(ip);
311 zfsvfs_t *zfsvfs = ITOZSB(ip);
317 error = zfs_holey_common(ip, cmd, off);
322 #endif /* SEEK_HOLE && SEEK_DATA */
326 * When a file is memory mapped, we must keep the IO data synchronized
327 * between the DMU cache and the memory mapped pages. What this means:
329 * On Write: If we find a memory mapped page, we write to *both*
330 * the page and the dmu buffer.
333 update_pages(struct inode *ip, int64_t start, int len,
334 objset_t *os, uint64_t oid)
336 struct address_space *mp = ip->i_mapping;
342 off = start & (PAGE_SIZE-1);
343 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
344 nbytes = MIN(PAGE_SIZE - off, len);
346 pp = find_lock_page(mp, start >> PAGE_SHIFT);
348 if (mapping_writably_mapped(mp))
349 flush_dcache_page(pp);
352 (void) dmu_read(os, oid, start+off, nbytes, pb+off,
356 if (mapping_writably_mapped(mp))
357 flush_dcache_page(pp);
359 mark_page_accessed(pp);
372 * When a file is memory mapped, we must keep the IO data synchronized
373 * between the DMU cache and the memory mapped pages. What this means:
375 * On Read: We "read" preferentially from memory mapped pages,
376 * else we default from the dmu buffer.
378 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
379 * the file is memory mapped.
382 mappedread(struct inode *ip, int nbytes, uio_t *uio)
384 struct address_space *mp = ip->i_mapping;
386 znode_t *zp = ITOZ(ip);
393 start = uio->uio_loffset;
394 off = start & (PAGE_SIZE-1);
395 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
396 bytes = MIN(PAGE_SIZE - off, len);
398 pp = find_lock_page(mp, start >> PAGE_SHIFT);
400 ASSERT(PageUptodate(pp));
403 error = uiomove(pb + off, bytes, UIO_READ, uio);
406 if (mapping_writably_mapped(mp))
407 flush_dcache_page(pp);
409 mark_page_accessed(pp);
413 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
426 unsigned long zfs_read_chunk_size = 1024 * 1024; /* Tunable */
427 unsigned long zfs_delete_blocks = DMU_MAX_DELETEBLKCNT;
430 * Read bytes from specified file into supplied buffer.
432 * IN: ip - inode of file to be read from.
433 * uio - structure supplying read location, range info,
435 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
436 * O_DIRECT flag; used to bypass page cache.
437 * cr - credentials of caller.
439 * OUT: uio - updated offset and range, buffer filled.
441 * RETURN: 0 on success, error code on failure.
444 * inode - atime updated if byte count > 0
448 zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
450 znode_t *zp = ITOZ(ip);
451 zfsvfs_t *zfsvfs = ITOZSB(ip);
455 #ifdef HAVE_UIO_ZEROCOPY
457 #endif /* HAVE_UIO_ZEROCOPY */
462 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
464 return (SET_ERROR(EACCES));
468 * Validate file offset
470 if (uio->uio_loffset < (offset_t)0) {
472 return (SET_ERROR(EINVAL));
476 * Fasttrack empty reads
478 if (uio->uio_resid == 0) {
484 * If we're in FRSYNC mode, sync out this znode before reading it.
486 if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
487 zil_commit(zfsvfs->z_log, zp->z_id);
490 * Lock the range against changes.
492 rl = zfs_range_lock(&zp->z_range_lock, uio->uio_loffset, uio->uio_resid,
496 * If we are reading past end-of-file we can skip
497 * to the end; but we might still need to set atime.
499 if (uio->uio_loffset >= zp->z_size) {
504 ASSERT(uio->uio_loffset < zp->z_size);
505 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
507 #ifdef HAVE_UIO_ZEROCOPY
508 if ((uio->uio_extflg == UIO_XUIO) &&
509 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
511 int blksz = zp->z_blksz;
512 uint64_t offset = uio->uio_loffset;
514 xuio = (xuio_t *)uio;
516 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
519 ASSERT(offset + n <= blksz);
522 (void) dmu_xuio_init(xuio, nblk);
524 if (vn_has_cached_data(ip)) {
526 * For simplicity, we always allocate a full buffer
527 * even if we only expect to read a portion of a block.
529 while (--nblk >= 0) {
530 (void) dmu_xuio_add(xuio,
531 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
536 #endif /* HAVE_UIO_ZEROCOPY */
539 nbytes = MIN(n, zfs_read_chunk_size -
540 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
542 if (zp->z_is_mapped && !(ioflag & O_DIRECT)) {
543 error = mappedread(ip, nbytes, uio);
545 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
550 /* convert checksum errors into IO errors */
552 error = SET_ERROR(EIO);
559 zfs_range_unlock(rl);
566 * Write the bytes to a file.
568 * IN: ip - inode of file to be written to.
569 * uio - structure supplying write location, range info,
571 * ioflag - FAPPEND flag set if in append mode.
572 * O_DIRECT flag; used to bypass page cache.
573 * cr - credentials of caller.
575 * OUT: uio - updated offset and range.
577 * RETURN: 0 if success
578 * error code if failure
581 * ip - ctime|mtime updated if byte count > 0
586 zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
588 znode_t *zp = ITOZ(ip);
589 rlim64_t limit = uio->uio_limit;
590 ssize_t start_resid = uio->uio_resid;
594 zfsvfs_t *zfsvfs = ZTOZSB(zp);
599 int max_blksz = zfsvfs->z_max_blksz;
602 const iovec_t *aiov = NULL;
606 sa_bulk_attr_t bulk[4];
607 uint64_t mtime[2], ctime[2];
609 #ifdef HAVE_UIO_ZEROCOPY
611 const iovec_t *iovp = uio->uio_iov;
612 ASSERTV(int iovcnt = uio->uio_iovcnt);
616 * Fasttrack empty write
622 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
628 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
629 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
630 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
632 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
636 * Callers might not be able to detect properly that we are read-only,
637 * so check it explicitly here.
639 if (zfs_is_readonly(zfsvfs)) {
641 return (SET_ERROR(EROFS));
645 * If immutable or not appending then return EPERM
647 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
648 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
649 (uio->uio_loffset < zp->z_size))) {
651 return (SET_ERROR(EPERM));
654 zilog = zfsvfs->z_log;
657 * Validate file offset
659 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
662 return (SET_ERROR(EINVAL));
666 * Pre-fault the pages to ensure slow (eg NFS) pages
668 * Skip this if uio contains loaned arc_buf.
670 #ifdef HAVE_UIO_ZEROCOPY
671 if ((uio->uio_extflg == UIO_XUIO) &&
672 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
673 xuio = (xuio_t *)uio;
676 uio_prefaultpages(MIN(n, max_blksz), uio);
679 * If in append mode, set the io offset pointer to eof.
681 if (ioflag & FAPPEND) {
683 * Obtain an appending range lock to guarantee file append
684 * semantics. We reset the write offset once we have the lock.
686 rl = zfs_range_lock(&zp->z_range_lock, 0, n, RL_APPEND);
688 if (rl->r_len == UINT64_MAX) {
690 * We overlocked the file because this write will cause
691 * the file block size to increase.
692 * Note that zp_size cannot change with this lock held.
696 uio->uio_loffset = woff;
699 * Note that if the file block size will change as a result of
700 * this write, then this range lock will lock the entire file
701 * so that we can re-write the block safely.
703 rl = zfs_range_lock(&zp->z_range_lock, woff, n, RL_WRITER);
707 zfs_range_unlock(rl);
709 return (SET_ERROR(EFBIG));
712 if ((woff + n) > limit || woff > (limit - n))
715 /* Will this write extend the file length? */
716 write_eof = (woff + n > zp->z_size);
718 end_size = MAX(zp->z_size, woff + n);
721 * Write the file in reasonable size chunks. Each chunk is written
722 * in a separate transaction; this keeps the intent log records small
723 * and allows us to do more fine-grained space accounting.
727 woff = uio->uio_loffset;
728 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
729 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
731 dmu_return_arcbuf(abuf);
732 error = SET_ERROR(EDQUOT);
736 if (xuio && abuf == NULL) {
737 #ifdef HAVE_UIO_ZEROCOPY
738 ASSERT(i_iov < iovcnt);
739 ASSERT3U(uio->uio_segflg, !=, UIO_BVEC);
741 abuf = dmu_xuio_arcbuf(xuio, i_iov);
742 dmu_xuio_clear(xuio, i_iov);
743 ASSERT((aiov->iov_base == abuf->b_data) ||
744 ((char *)aiov->iov_base - (char *)abuf->b_data +
745 aiov->iov_len == arc_buf_size(abuf)));
748 } else if (abuf == NULL && n >= max_blksz &&
749 woff >= zp->z_size &&
750 P2PHASE(woff, max_blksz) == 0 &&
751 zp->z_blksz == max_blksz) {
753 * This write covers a full block. "Borrow" a buffer
754 * from the dmu so that we can fill it before we enter
755 * a transaction. This avoids the possibility of
756 * holding up the transaction if the data copy hangs
757 * up on a pagefault (e.g., from an NFS server mapping).
761 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
763 ASSERT(abuf != NULL);
764 ASSERT(arc_buf_size(abuf) == max_blksz);
765 if ((error = uiocopy(abuf->b_data, max_blksz,
766 UIO_WRITE, uio, &cbytes))) {
767 dmu_return_arcbuf(abuf);
770 ASSERT(cbytes == max_blksz);
774 * Start a transaction.
776 tx = dmu_tx_create(zfsvfs->z_os);
777 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
778 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
779 zfs_sa_upgrade_txholds(tx, zp);
780 error = dmu_tx_assign(tx, TXG_WAIT);
784 dmu_return_arcbuf(abuf);
789 * If zfs_range_lock() over-locked we grow the blocksize
790 * and then reduce the lock range. This will only happen
791 * on the first iteration since zfs_range_reduce() will
792 * shrink down r_len to the appropriate size.
794 if (rl->r_len == UINT64_MAX) {
797 if (zp->z_blksz > max_blksz) {
799 * File's blocksize is already larger than the
800 * "recordsize" property. Only let it grow to
801 * the next power of 2.
803 ASSERT(!ISP2(zp->z_blksz));
804 new_blksz = MIN(end_size,
805 1 << highbit64(zp->z_blksz));
807 new_blksz = MIN(end_size, max_blksz);
809 zfs_grow_blocksize(zp, new_blksz, tx);
810 zfs_range_reduce(rl, woff, n);
814 * XXX - should we really limit each write to z_max_blksz?
815 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
817 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
820 tx_bytes = uio->uio_resid;
821 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
823 tx_bytes -= uio->uio_resid;
826 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
828 * If this is not a full block write, but we are
829 * extending the file past EOF and this data starts
830 * block-aligned, use assign_arcbuf(). Otherwise,
831 * write via dmu_write().
833 if (tx_bytes < max_blksz && (!write_eof ||
834 aiov->iov_base != abuf->b_data)) {
836 dmu_write(zfsvfs->z_os, zp->z_id, woff,
837 // cppcheck-suppress nullPointer
838 aiov->iov_len, aiov->iov_base, tx);
839 dmu_return_arcbuf(abuf);
840 xuio_stat_wbuf_copied();
842 ASSERT(xuio || tx_bytes == max_blksz);
843 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
846 ASSERT(tx_bytes <= uio->uio_resid);
847 uioskip(uio, tx_bytes);
849 if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT)) {
850 update_pages(ip, woff,
851 tx_bytes, zfsvfs->z_os, zp->z_id);
855 * If we made no progress, we're done. If we made even
856 * partial progress, update the znode and ZIL accordingly.
859 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
860 (void *)&zp->z_size, sizeof (uint64_t), tx);
867 * Clear Set-UID/Set-GID bits on successful write if not
868 * privileged and at least one of the execute bits is set.
870 * It would be nice to to this after all writes have
871 * been done, but that would still expose the ISUID/ISGID
872 * to another app after the partial write is committed.
874 * Note: we don't call zfs_fuid_map_id() here because
875 * user 0 is not an ephemeral uid.
877 mutex_enter(&zp->z_acl_lock);
878 uid = KUID_TO_SUID(ip->i_uid);
879 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
880 (S_IXUSR >> 6))) != 0 &&
881 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
882 secpolicy_vnode_setid_retain(cr,
883 ((zp->z_mode & S_ISUID) != 0 && uid == 0)) != 0) {
885 zp->z_mode &= ~(S_ISUID | S_ISGID);
886 ip->i_mode = newmode = zp->z_mode;
887 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
888 (void *)&newmode, sizeof (uint64_t), tx);
890 mutex_exit(&zp->z_acl_lock);
892 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
895 * Update the file size (zp_size) if it has changed;
896 * account for possible concurrent updates.
898 while ((end_size = zp->z_size) < uio->uio_loffset) {
899 (void) atomic_cas_64(&zp->z_size, end_size,
904 * If we are replaying and eof is non zero then force
905 * the file size to the specified eof. Note, there's no
906 * concurrency during replay.
908 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
909 zp->z_size = zfsvfs->z_replay_eof;
911 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
913 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag,
919 ASSERT(tx_bytes == nbytes);
923 uio_prefaultpages(MIN(n, max_blksz), uio);
926 zfs_inode_update(zp);
927 zfs_range_unlock(rl);
930 * If we're in replay mode, or we made no progress, return error.
931 * Otherwise, it's at least a partial write, so it's successful.
933 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
938 if (ioflag & (FSYNC | FDSYNC) ||
939 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
940 zil_commit(zilog, zp->z_id);
947 * Drop a reference on the passed inode asynchronously. This ensures
948 * that the caller will never drop the last reference on an inode in
949 * the current context. Doing so while holding open a tx could result
950 * in a deadlock if iput_final() re-enters the filesystem code.
953 zfs_iput_async(struct inode *ip)
955 objset_t *os = ITOZSB(ip)->z_os;
957 ASSERT(atomic_read(&ip->i_count) > 0);
960 if (atomic_read(&ip->i_count) == 1)
961 VERIFY(taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os)),
962 (task_func_t *)iput, ip, TQ_SLEEP) != TASKQID_INVALID);
968 zfs_get_done(zgd_t *zgd, int error)
970 znode_t *zp = zgd->zgd_private;
973 dmu_buf_rele(zgd->zgd_db, zgd);
975 zfs_range_unlock(zgd->zgd_rl);
978 * Release the vnode asynchronously as we currently have the
979 * txg stopped from syncing.
981 zfs_iput_async(ZTOI(zp));
983 if (error == 0 && zgd->zgd_bp)
984 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
986 kmem_free(zgd, sizeof (zgd_t));
990 static int zil_fault_io = 0;
994 * Get data to generate a TX_WRITE intent log record.
997 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
999 zfsvfs_t *zfsvfs = arg;
1000 objset_t *os = zfsvfs->z_os;
1002 uint64_t object = lr->lr_foid;
1003 uint64_t offset = lr->lr_offset;
1004 uint64_t size = lr->lr_length;
1009 ASSERT(zio != NULL);
1013 * Nothing to do if the file has been removed
1015 if (zfs_zget(zfsvfs, object, &zp) != 0)
1016 return (SET_ERROR(ENOENT));
1017 if (zp->z_unlinked) {
1019 * Release the vnode asynchronously as we currently have the
1020 * txg stopped from syncing.
1022 zfs_iput_async(ZTOI(zp));
1023 return (SET_ERROR(ENOENT));
1026 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1027 zgd->zgd_zilog = zfsvfs->z_log;
1028 zgd->zgd_private = zp;
1031 * Write records come in two flavors: immediate and indirect.
1032 * For small writes it's cheaper to store the data with the
1033 * log record (immediate); for large writes it's cheaper to
1034 * sync the data and get a pointer to it (indirect) so that
1035 * we don't have to write the data twice.
1037 if (buf != NULL) { /* immediate write */
1038 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset, size,
1040 /* test for truncation needs to be done while range locked */
1041 if (offset >= zp->z_size) {
1042 error = SET_ERROR(ENOENT);
1044 error = dmu_read(os, object, offset, size, buf,
1045 DMU_READ_NO_PREFETCH);
1047 ASSERT(error == 0 || error == ENOENT);
1048 } else { /* indirect write */
1050 * Have to lock the whole block to ensure when it's
1051 * written out and it's checksum is being calculated
1052 * that no one can change the data. We need to re-check
1053 * blocksize after we get the lock in case it's changed!
1058 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1060 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset,
1062 if (zp->z_blksz == size)
1065 zfs_range_unlock(zgd->zgd_rl);
1067 /* test for truncation needs to be done while range locked */
1068 if (lr->lr_offset >= zp->z_size)
1069 error = SET_ERROR(ENOENT);
1072 error = SET_ERROR(EIO);
1077 error = dmu_buf_hold(os, object, offset, zgd, &db,
1078 DMU_READ_NO_PREFETCH);
1081 blkptr_t *bp = &lr->lr_blkptr;
1086 ASSERT(db->db_offset == offset);
1087 ASSERT(db->db_size == size);
1089 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1091 ASSERT(error || lr->lr_length <= size);
1094 * On success, we need to wait for the write I/O
1095 * initiated by dmu_sync() to complete before we can
1096 * release this dbuf. We will finish everything up
1097 * in the zfs_get_done() callback.
1102 if (error == EALREADY) {
1103 lr->lr_common.lrc_txtype = TX_WRITE2;
1109 zfs_get_done(zgd, error);
1116 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
1118 znode_t *zp = ITOZ(ip);
1119 zfsvfs_t *zfsvfs = ITOZSB(ip);
1125 if (flag & V_ACE_MASK)
1126 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1128 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1135 * Lookup an entry in a directory, or an extended attribute directory.
1136 * If it exists, return a held inode reference for it.
1138 * IN: dip - inode of directory to search.
1139 * nm - name of entry to lookup.
1140 * flags - LOOKUP_XATTR set if looking for an attribute.
1141 * cr - credentials of caller.
1142 * direntflags - directory lookup flags
1143 * realpnp - returned pathname.
1145 * OUT: ipp - inode of located entry, NULL if not found.
1147 * RETURN: 0 on success, error code on failure.
1154 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
1155 cred_t *cr, int *direntflags, pathname_t *realpnp)
1157 znode_t *zdp = ITOZ(dip);
1158 zfsvfs_t *zfsvfs = ITOZSB(dip);
1162 * Fast path lookup, however we must skip DNLC lookup
1163 * for case folding or normalizing lookups because the
1164 * DNLC code only stores the passed in name. This means
1165 * creating 'a' and removing 'A' on a case insensitive
1166 * file system would work, but DNLC still thinks 'a'
1167 * exists and won't let you create it again on the next
1168 * pass through fast path.
1170 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1172 if (!S_ISDIR(dip->i_mode)) {
1173 return (SET_ERROR(ENOTDIR));
1174 } else if (zdp->z_sa_hdl == NULL) {
1175 return (SET_ERROR(EIO));
1178 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1179 error = zfs_fastaccesschk_execute(zdp, cr);
1187 } else if (!zdp->z_zfsvfs->z_norm &&
1188 (zdp->z_zfsvfs->z_case == ZFS_CASE_SENSITIVE)) {
1190 vnode_t *tvp = dnlc_lookup(dvp, nm);
1193 error = zfs_fastaccesschk_execute(zdp, cr);
1198 if (tvp == DNLC_NO_VNODE) {
1200 return (SET_ERROR(ENOENT));
1203 return (specvp_check(vpp, cr));
1206 #endif /* HAVE_DNLC */
1215 if (flags & LOOKUP_XATTR) {
1217 * We don't allow recursive attributes..
1218 * Maybe someday we will.
1220 if (zdp->z_pflags & ZFS_XATTR) {
1222 return (SET_ERROR(EINVAL));
1225 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1231 * Do we have permission to get into attribute directory?
1234 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1244 if (!S_ISDIR(dip->i_mode)) {
1246 return (SET_ERROR(ENOTDIR));
1250 * Check accessibility of directory.
1253 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1258 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1259 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1261 return (SET_ERROR(EILSEQ));
1264 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1265 if ((error == 0) && (*ipp))
1266 zfs_inode_update(ITOZ(*ipp));
1273 * Attempt to create a new entry in a directory. If the entry
1274 * already exists, truncate the file if permissible, else return
1275 * an error. Return the ip of the created or trunc'd file.
1277 * IN: dip - inode of directory to put new file entry in.
1278 * name - name of new file entry.
1279 * vap - attributes of new file.
1280 * excl - flag indicating exclusive or non-exclusive mode.
1281 * mode - mode to open file with.
1282 * cr - credentials of caller.
1283 * flag - large file flag [UNUSED].
1284 * vsecp - ACL to be set
1286 * OUT: ipp - inode of created or trunc'd entry.
1288 * RETURN: 0 on success, error code on failure.
1291 * dip - ctime|mtime updated if new entry created
1292 * ip - ctime|mtime always, atime if new
1297 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1298 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1300 znode_t *zp, *dzp = ITOZ(dip);
1301 zfsvfs_t *zfsvfs = ITOZSB(dip);
1309 zfs_acl_ids_t acl_ids;
1310 boolean_t fuid_dirtied;
1311 boolean_t have_acl = B_FALSE;
1312 boolean_t waited = B_FALSE;
1315 * If we have an ephemeral id, ACL, or XVATTR then
1316 * make sure file system is at proper version
1322 if (zfsvfs->z_use_fuids == B_FALSE &&
1323 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1324 return (SET_ERROR(EINVAL));
1327 return (SET_ERROR(EINVAL));
1332 zilog = zfsvfs->z_log;
1334 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1335 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1337 return (SET_ERROR(EILSEQ));
1340 if (vap->va_mask & ATTR_XVATTR) {
1341 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1342 crgetuid(cr), cr, vap->va_mode)) != 0) {
1350 if (*name == '\0') {
1352 * Null component name refers to the directory itself.
1359 /* possible igrab(zp) */
1362 if (flag & FIGNORECASE)
1365 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1369 zfs_acl_ids_free(&acl_ids);
1370 if (strcmp(name, "..") == 0)
1371 error = SET_ERROR(EISDIR);
1381 * Create a new file object and update the directory
1384 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1386 zfs_acl_ids_free(&acl_ids);
1391 * We only support the creation of regular files in
1392 * extended attribute directories.
1395 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1397 zfs_acl_ids_free(&acl_ids);
1398 error = SET_ERROR(EINVAL);
1402 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1403 cr, vsecp, &acl_ids)) != 0)
1407 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1408 zfs_acl_ids_free(&acl_ids);
1409 error = SET_ERROR(EDQUOT);
1413 tx = dmu_tx_create(os);
1415 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1416 ZFS_SA_BASE_ATTR_SIZE);
1418 fuid_dirtied = zfsvfs->z_fuid_dirty;
1420 zfs_fuid_txhold(zfsvfs, tx);
1421 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1422 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1423 if (!zfsvfs->z_use_sa &&
1424 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1425 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1426 0, acl_ids.z_aclp->z_acl_bytes);
1428 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1430 zfs_dirent_unlock(dl);
1431 if (error == ERESTART) {
1437 zfs_acl_ids_free(&acl_ids);
1442 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1445 zfs_fuid_sync(zfsvfs, tx);
1447 (void) zfs_link_create(dl, zp, tx, ZNEW);
1448 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1449 if (flag & FIGNORECASE)
1451 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1452 vsecp, acl_ids.z_fuidp, vap);
1453 zfs_acl_ids_free(&acl_ids);
1456 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1459 zfs_acl_ids_free(&acl_ids);
1463 * A directory entry already exists for this name.
1466 * Can't truncate an existing file if in exclusive mode.
1469 error = SET_ERROR(EEXIST);
1473 * Can't open a directory for writing.
1475 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1476 error = SET_ERROR(EISDIR);
1480 * Verify requested access to file.
1482 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1486 mutex_enter(&dzp->z_lock);
1488 mutex_exit(&dzp->z_lock);
1491 * Truncate regular files if requested.
1493 if (S_ISREG(ZTOI(zp)->i_mode) &&
1494 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1495 /* we can't hold any locks when calling zfs_freesp() */
1497 zfs_dirent_unlock(dl);
1500 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1506 zfs_dirent_unlock(dl);
1512 zfs_inode_update(dzp);
1513 zfs_inode_update(zp);
1517 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1518 zil_commit(zilog, 0);
1526 zfs_tmpfile(struct inode *dip, vattr_t *vap, int excl,
1527 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1529 znode_t *zp = NULL, *dzp = ITOZ(dip);
1530 zfsvfs_t *zfsvfs = ITOZSB(dip);
1536 zfs_acl_ids_t acl_ids;
1537 boolean_t fuid_dirtied;
1538 boolean_t have_acl = B_FALSE;
1539 boolean_t waited = B_FALSE;
1542 * If we have an ephemeral id, ACL, or XVATTR then
1543 * make sure file system is at proper version
1549 if (zfsvfs->z_use_fuids == B_FALSE &&
1550 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1551 return (SET_ERROR(EINVAL));
1557 if (vap->va_mask & ATTR_XVATTR) {
1558 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1559 crgetuid(cr), cr, vap->va_mode)) != 0) {
1569 * Create a new file object and update the directory
1572 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1574 zfs_acl_ids_free(&acl_ids);
1578 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1579 cr, vsecp, &acl_ids)) != 0)
1583 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1584 zfs_acl_ids_free(&acl_ids);
1585 error = SET_ERROR(EDQUOT);
1589 tx = dmu_tx_create(os);
1591 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1592 ZFS_SA_BASE_ATTR_SIZE);
1593 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1595 fuid_dirtied = zfsvfs->z_fuid_dirty;
1597 zfs_fuid_txhold(zfsvfs, tx);
1598 if (!zfsvfs->z_use_sa &&
1599 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1600 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1601 0, acl_ids.z_aclp->z_acl_bytes);
1603 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1605 if (error == ERESTART) {
1611 zfs_acl_ids_free(&acl_ids);
1616 zfs_mknode(dzp, vap, tx, cr, IS_TMPFILE, &zp, &acl_ids);
1619 zfs_fuid_sync(zfsvfs, tx);
1621 /* Add to unlinked set */
1623 zfs_unlinked_add(zp, tx);
1624 zfs_acl_ids_free(&acl_ids);
1632 zfs_inode_update(dzp);
1633 zfs_inode_update(zp);
1642 * Remove an entry from a directory.
1644 * IN: dip - inode of directory to remove entry from.
1645 * name - name of entry to remove.
1646 * cr - credentials of caller.
1648 * RETURN: 0 if success
1649 * error code if failure
1653 * ip - ctime (if nlink > 0)
1656 uint64_t null_xattr = 0;
1660 zfs_remove(struct inode *dip, char *name, cred_t *cr, int flags)
1662 znode_t *zp, *dzp = ITOZ(dip);
1665 zfsvfs_t *zfsvfs = ITOZSB(dip);
1667 uint64_t acl_obj, xattr_obj;
1668 uint64_t xattr_obj_unlinked = 0;
1673 boolean_t may_delete_now, delete_now = FALSE;
1674 boolean_t unlinked, toobig = FALSE;
1676 pathname_t *realnmp = NULL;
1680 boolean_t waited = B_FALSE;
1683 return (SET_ERROR(EINVAL));
1687 zilog = zfsvfs->z_log;
1689 if (flags & FIGNORECASE) {
1699 * Attempt to lock directory; fail if entry doesn't exist.
1701 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1711 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1716 * Need to use rmdir for removing directories.
1718 if (S_ISDIR(ip->i_mode)) {
1719 error = SET_ERROR(EPERM);
1725 dnlc_remove(dvp, realnmp->pn_buf);
1727 dnlc_remove(dvp, name);
1728 #endif /* HAVE_DNLC */
1730 mutex_enter(&zp->z_lock);
1731 may_delete_now = atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped);
1732 mutex_exit(&zp->z_lock);
1735 * We may delete the znode now, or we may put it in the unlinked set;
1736 * it depends on whether we're the last link, and on whether there are
1737 * other holds on the inode. So we dmu_tx_hold() the right things to
1738 * allow for either case.
1741 tx = dmu_tx_create(zfsvfs->z_os);
1742 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1743 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1744 zfs_sa_upgrade_txholds(tx, zp);
1745 zfs_sa_upgrade_txholds(tx, dzp);
1746 if (may_delete_now) {
1747 toobig = zp->z_size > zp->z_blksz * zfs_delete_blocks;
1748 /* if the file is too big, only hold_free a token amount */
1749 dmu_tx_hold_free(tx, zp->z_id, 0,
1750 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1753 /* are there any extended attributes? */
1754 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1755 &xattr_obj, sizeof (xattr_obj));
1756 if (error == 0 && xattr_obj) {
1757 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1759 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1760 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1763 mutex_enter(&zp->z_lock);
1764 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1765 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1766 mutex_exit(&zp->z_lock);
1768 /* charge as an update -- would be nice not to charge at all */
1769 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1772 * Mark this transaction as typically resulting in a net free of space
1774 dmu_tx_mark_netfree(tx);
1776 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1778 zfs_dirent_unlock(dl);
1779 if (error == ERESTART) {
1799 * Remove the directory entry.
1801 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1810 * Hold z_lock so that we can make sure that the ACL obj
1811 * hasn't changed. Could have been deleted due to
1814 mutex_enter(&zp->z_lock);
1815 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1816 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1817 delete_now = may_delete_now && !toobig &&
1818 atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped) &&
1819 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1824 if (xattr_obj_unlinked) {
1825 ASSERT3U(ZTOI(xzp)->i_nlink, ==, 2);
1826 mutex_enter(&xzp->z_lock);
1827 xzp->z_unlinked = 1;
1828 clear_nlink(ZTOI(xzp));
1830 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1831 &links, sizeof (links), tx);
1832 ASSERT3U(error, ==, 0);
1833 mutex_exit(&xzp->z_lock);
1834 zfs_unlinked_add(xzp, tx);
1837 error = sa_remove(zp->z_sa_hdl,
1838 SA_ZPL_XATTR(zfsvfs), tx);
1840 error = sa_update(zp->z_sa_hdl,
1841 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1842 sizeof (uint64_t), tx);
1846 * Add to the unlinked set because a new reference could be
1847 * taken concurrently resulting in a deferred destruction.
1849 zfs_unlinked_add(zp, tx);
1850 mutex_exit(&zp->z_lock);
1851 } else if (unlinked) {
1852 mutex_exit(&zp->z_lock);
1853 zfs_unlinked_add(zp, tx);
1857 if (flags & FIGNORECASE)
1859 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1866 zfs_dirent_unlock(dl);
1867 zfs_inode_update(dzp);
1868 zfs_inode_update(zp);
1876 zfs_inode_update(xzp);
1877 zfs_iput_async(ZTOI(xzp));
1880 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1881 zil_commit(zilog, 0);
1888 * Create a new directory and insert it into dip using the name
1889 * provided. Return a pointer to the inserted directory.
1891 * IN: dip - inode of directory to add subdir to.
1892 * dirname - name of new directory.
1893 * vap - attributes of new directory.
1894 * cr - credentials of caller.
1895 * vsecp - ACL to be set
1897 * OUT: ipp - inode of created directory.
1899 * RETURN: 0 if success
1900 * error code if failure
1903 * dip - ctime|mtime updated
1904 * ipp - ctime|mtime|atime updated
1908 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1909 cred_t *cr, int flags, vsecattr_t *vsecp)
1911 znode_t *zp, *dzp = ITOZ(dip);
1912 zfsvfs_t *zfsvfs = ITOZSB(dip);
1920 gid_t gid = crgetgid(cr);
1921 zfs_acl_ids_t acl_ids;
1922 boolean_t fuid_dirtied;
1923 boolean_t waited = B_FALSE;
1925 ASSERT(S_ISDIR(vap->va_mode));
1928 * If we have an ephemeral id, ACL, or XVATTR then
1929 * make sure file system is at proper version
1933 if (zfsvfs->z_use_fuids == B_FALSE &&
1934 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1935 return (SET_ERROR(EINVAL));
1937 if (dirname == NULL)
1938 return (SET_ERROR(EINVAL));
1942 zilog = zfsvfs->z_log;
1944 if (dzp->z_pflags & ZFS_XATTR) {
1946 return (SET_ERROR(EINVAL));
1949 if (zfsvfs->z_utf8 && u8_validate(dirname,
1950 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1952 return (SET_ERROR(EILSEQ));
1954 if (flags & FIGNORECASE)
1957 if (vap->va_mask & ATTR_XVATTR) {
1958 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1959 crgetuid(cr), cr, vap->va_mode)) != 0) {
1965 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1966 vsecp, &acl_ids)) != 0) {
1971 * First make sure the new directory doesn't exist.
1973 * Existence is checked first to make sure we don't return
1974 * EACCES instead of EEXIST which can cause some applications
1980 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1982 zfs_acl_ids_free(&acl_ids);
1987 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
1988 zfs_acl_ids_free(&acl_ids);
1989 zfs_dirent_unlock(dl);
1994 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1995 zfs_acl_ids_free(&acl_ids);
1996 zfs_dirent_unlock(dl);
1998 return (SET_ERROR(EDQUOT));
2002 * Add a new entry to the directory.
2004 tx = dmu_tx_create(zfsvfs->z_os);
2005 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2006 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2007 fuid_dirtied = zfsvfs->z_fuid_dirty;
2009 zfs_fuid_txhold(zfsvfs, tx);
2010 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2011 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2012 acl_ids.z_aclp->z_acl_bytes);
2015 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2016 ZFS_SA_BASE_ATTR_SIZE);
2018 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2020 zfs_dirent_unlock(dl);
2021 if (error == ERESTART) {
2027 zfs_acl_ids_free(&acl_ids);
2036 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2039 zfs_fuid_sync(zfsvfs, tx);
2042 * Now put new name in parent dir.
2044 (void) zfs_link_create(dl, zp, tx, ZNEW);
2048 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2049 if (flags & FIGNORECASE)
2051 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2052 acl_ids.z_fuidp, vap);
2054 zfs_acl_ids_free(&acl_ids);
2058 zfs_dirent_unlock(dl);
2060 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2061 zil_commit(zilog, 0);
2063 zfs_inode_update(dzp);
2064 zfs_inode_update(zp);
2070 * Remove a directory subdir entry. If the current working
2071 * directory is the same as the subdir to be removed, the
2074 * IN: dip - inode of directory to remove from.
2075 * name - name of directory to be removed.
2076 * cwd - inode of current working directory.
2077 * cr - credentials of caller.
2078 * flags - case flags
2080 * RETURN: 0 on success, error code on failure.
2083 * dip - ctime|mtime updated
2087 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
2090 znode_t *dzp = ITOZ(dip);
2093 zfsvfs_t *zfsvfs = ITOZSB(dip);
2099 boolean_t waited = B_FALSE;
2102 return (SET_ERROR(EINVAL));
2106 zilog = zfsvfs->z_log;
2108 if (flags & FIGNORECASE)
2114 * Attempt to lock directory; fail if entry doesn't exist.
2116 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2124 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
2128 if (!S_ISDIR(ip->i_mode)) {
2129 error = SET_ERROR(ENOTDIR);
2134 error = SET_ERROR(EINVAL);
2139 * Grab a lock on the directory to make sure that no one is
2140 * trying to add (or lookup) entries while we are removing it.
2142 rw_enter(&zp->z_name_lock, RW_WRITER);
2145 * Grab a lock on the parent pointer to make sure we play well
2146 * with the treewalk and directory rename code.
2148 rw_enter(&zp->z_parent_lock, RW_WRITER);
2150 tx = dmu_tx_create(zfsvfs->z_os);
2151 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2152 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2153 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2154 zfs_sa_upgrade_txholds(tx, zp);
2155 zfs_sa_upgrade_txholds(tx, dzp);
2156 dmu_tx_mark_netfree(tx);
2157 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2159 rw_exit(&zp->z_parent_lock);
2160 rw_exit(&zp->z_name_lock);
2161 zfs_dirent_unlock(dl);
2162 if (error == ERESTART) {
2175 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2178 uint64_t txtype = TX_RMDIR;
2179 if (flags & FIGNORECASE)
2181 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2186 rw_exit(&zp->z_parent_lock);
2187 rw_exit(&zp->z_name_lock);
2189 zfs_dirent_unlock(dl);
2191 zfs_inode_update(dzp);
2192 zfs_inode_update(zp);
2195 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2196 zil_commit(zilog, 0);
2203 * Read as many directory entries as will fit into the provided
2204 * dirent buffer from the given directory cursor position.
2206 * IN: ip - inode of directory to read.
2207 * dirent - buffer for directory entries.
2209 * OUT: dirent - filler buffer of directory entries.
2211 * RETURN: 0 if success
2212 * error code if failure
2215 * ip - atime updated
2217 * Note that the low 4 bits of the cookie returned by zap is always zero.
2218 * This allows us to use the low range for "special" directory entries:
2219 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2220 * we use the offset 2 for the '.zfs' directory.
2224 zfs_readdir(struct inode *ip, struct dir_context *ctx, cred_t *cr)
2226 znode_t *zp = ITOZ(ip);
2227 zfsvfs_t *zfsvfs = ITOZSB(ip);
2230 zap_attribute_t zap;
2236 uint64_t offset; /* must be unsigned; checks for < 1 */
2241 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2242 &parent, sizeof (parent))) != 0)
2246 * Quit if directory has been removed (posix)
2254 prefetch = zp->z_zn_prefetch;
2257 * Initialize the iterator cursor.
2261 * Start iteration from the beginning of the directory.
2263 zap_cursor_init(&zc, os, zp->z_id);
2266 * The offset is a serialized cursor.
2268 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2272 * Transform to file-system independent format
2277 * Special case `.', `..', and `.zfs'.
2280 (void) strcpy(zap.za_name, ".");
2281 zap.za_normalization_conflict = 0;
2284 } else if (offset == 1) {
2285 (void) strcpy(zap.za_name, "..");
2286 zap.za_normalization_conflict = 0;
2289 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2290 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2291 zap.za_normalization_conflict = 0;
2292 objnum = ZFSCTL_INO_ROOT;
2298 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2299 if (error == ENOENT)
2306 * Allow multiple entries provided the first entry is
2307 * the object id. Non-zpl consumers may safely make
2308 * use of the additional space.
2310 * XXX: This should be a feature flag for compatibility
2312 if (zap.za_integer_length != 8 ||
2313 zap.za_num_integers == 0) {
2314 cmn_err(CE_WARN, "zap_readdir: bad directory "
2315 "entry, obj = %lld, offset = %lld, "
2316 "length = %d, num = %lld\n",
2317 (u_longlong_t)zp->z_id,
2318 (u_longlong_t)offset,
2319 zap.za_integer_length,
2320 (u_longlong_t)zap.za_num_integers);
2321 error = SET_ERROR(ENXIO);
2325 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2326 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2329 done = !dir_emit(ctx, zap.za_name, strlen(zap.za_name),
2334 /* Prefetch znode */
2336 dmu_prefetch(os, objnum, 0, 0, 0,
2337 ZIO_PRIORITY_SYNC_READ);
2341 * Move to the next entry, fill in the previous offset.
2343 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2344 zap_cursor_advance(&zc);
2345 offset = zap_cursor_serialize(&zc);
2351 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2354 zap_cursor_fini(&zc);
2355 if (error == ENOENT)
2363 ulong_t zfs_fsync_sync_cnt = 4;
2366 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2368 znode_t *zp = ITOZ(ip);
2369 zfsvfs_t *zfsvfs = ITOZSB(ip);
2371 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2373 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2376 zil_commit(zfsvfs->z_log, zp->z_id);
2379 tsd_set(zfs_fsyncer_key, NULL);
2386 * Get the requested file attributes and place them in the provided
2389 * IN: ip - inode of file.
2390 * vap - va_mask identifies requested attributes.
2391 * If ATTR_XVATTR set, then optional attrs are requested
2392 * flags - ATTR_NOACLCHECK (CIFS server context)
2393 * cr - credentials of caller.
2395 * OUT: vap - attribute values.
2397 * RETURN: 0 (always succeeds)
2401 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2403 znode_t *zp = ITOZ(ip);
2404 zfsvfs_t *zfsvfs = ITOZSB(ip);
2407 uint64_t atime[2], mtime[2], ctime[2];
2408 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2409 xoptattr_t *xoap = NULL;
2410 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2411 sa_bulk_attr_t bulk[3];
2417 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2419 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
2420 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2421 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2423 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2429 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2430 * Also, if we are the owner don't bother, since owner should
2431 * always be allowed to read basic attributes of file.
2433 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2434 (vap->va_uid != crgetuid(cr))) {
2435 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2443 * Return all attributes. It's cheaper to provide the answer
2444 * than to determine whether we were asked the question.
2447 mutex_enter(&zp->z_lock);
2448 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2449 vap->va_mode = zp->z_mode;
2450 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2451 vap->va_nodeid = zp->z_id;
2452 if ((zp->z_id == zfsvfs->z_root) && zfs_show_ctldir(zp))
2453 links = ZTOI(zp)->i_nlink + 1;
2455 links = ZTOI(zp)->i_nlink;
2456 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2457 vap->va_size = i_size_read(ip);
2458 vap->va_rdev = ip->i_rdev;
2459 vap->va_seq = ip->i_generation;
2462 * Add in any requested optional attributes and the create time.
2463 * Also set the corresponding bits in the returned attribute bitmap.
2465 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2466 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2468 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2469 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2472 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2473 xoap->xoa_readonly =
2474 ((zp->z_pflags & ZFS_READONLY) != 0);
2475 XVA_SET_RTN(xvap, XAT_READONLY);
2478 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2480 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2481 XVA_SET_RTN(xvap, XAT_SYSTEM);
2484 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2486 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2487 XVA_SET_RTN(xvap, XAT_HIDDEN);
2490 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2491 xoap->xoa_nounlink =
2492 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2493 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2496 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2497 xoap->xoa_immutable =
2498 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2499 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2502 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2503 xoap->xoa_appendonly =
2504 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2505 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2508 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2510 ((zp->z_pflags & ZFS_NODUMP) != 0);
2511 XVA_SET_RTN(xvap, XAT_NODUMP);
2514 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2516 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2517 XVA_SET_RTN(xvap, XAT_OPAQUE);
2520 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2521 xoap->xoa_av_quarantined =
2522 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2523 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2526 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2527 xoap->xoa_av_modified =
2528 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2529 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2532 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2533 S_ISREG(ip->i_mode)) {
2534 zfs_sa_get_scanstamp(zp, xvap);
2537 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2540 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2541 times, sizeof (times));
2542 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2543 XVA_SET_RTN(xvap, XAT_CREATETIME);
2546 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2547 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2548 XVA_SET_RTN(xvap, XAT_REPARSE);
2550 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2551 xoap->xoa_generation = ip->i_generation;
2552 XVA_SET_RTN(xvap, XAT_GEN);
2555 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2557 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2558 XVA_SET_RTN(xvap, XAT_OFFLINE);
2561 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2563 ((zp->z_pflags & ZFS_SPARSE) != 0);
2564 XVA_SET_RTN(xvap, XAT_SPARSE);
2568 ZFS_TIME_DECODE(&vap->va_atime, atime);
2569 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2570 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2572 mutex_exit(&zp->z_lock);
2574 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2576 if (zp->z_blksz == 0) {
2578 * Block size hasn't been set; suggest maximal I/O transfers.
2580 vap->va_blksize = zfsvfs->z_max_blksz;
2588 * Get the basic file attributes and place them in the provided kstat
2589 * structure. The inode is assumed to be the authoritative source
2590 * for most of the attributes. However, the znode currently has the
2591 * authoritative atime, blksize, and block count.
2593 * IN: ip - inode of file.
2595 * OUT: sp - kstat values.
2597 * RETURN: 0 (always succeeds)
2601 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2603 znode_t *zp = ITOZ(ip);
2604 zfsvfs_t *zfsvfs = ITOZSB(ip);
2606 u_longlong_t nblocks;
2611 mutex_enter(&zp->z_lock);
2613 generic_fillattr(ip, sp);
2615 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2616 sp->blksize = blksize;
2617 sp->blocks = nblocks;
2619 if (unlikely(zp->z_blksz == 0)) {
2621 * Block size hasn't been set; suggest maximal I/O transfers.
2623 sp->blksize = zfsvfs->z_max_blksz;
2626 mutex_exit(&zp->z_lock);
2629 * Required to prevent NFS client from detecting different inode
2630 * numbers of snapshot root dentry before and after snapshot mount.
2632 if (zfsvfs->z_issnap) {
2633 if (ip->i_sb->s_root->d_inode == ip)
2634 sp->ino = ZFSCTL_INO_SNAPDIRS -
2635 dmu_objset_id(zfsvfs->z_os);
2644 * Set the file attributes to the values contained in the
2647 * IN: ip - inode of file to be modified.
2648 * vap - new attribute values.
2649 * If ATTR_XVATTR set, then optional attrs are being set
2650 * flags - ATTR_UTIME set if non-default time values provided.
2651 * - ATTR_NOACLCHECK (CIFS context only).
2652 * cr - credentials of caller.
2654 * RETURN: 0 if success
2655 * error code if failure
2658 * ip - ctime updated, mtime updated if size changed.
2662 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2664 znode_t *zp = ITOZ(ip);
2665 zfsvfs_t *zfsvfs = ITOZSB(ip);
2669 xvattr_t *tmpxvattr;
2670 uint_t mask = vap->va_mask;
2671 uint_t saved_mask = 0;
2674 uint64_t new_kuid = 0, new_kgid = 0, new_uid, new_gid;
2676 uint64_t mtime[2], ctime[2], atime[2];
2678 int need_policy = FALSE;
2680 zfs_fuid_info_t *fuidp = NULL;
2681 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2684 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2685 boolean_t fuid_dirtied = B_FALSE;
2686 sa_bulk_attr_t *bulk, *xattr_bulk;
2687 int count = 0, xattr_count = 0;
2695 zilog = zfsvfs->z_log;
2698 * Make sure that if we have ephemeral uid/gid or xvattr specified
2699 * that file system is at proper version level
2702 if (zfsvfs->z_use_fuids == B_FALSE &&
2703 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2704 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2705 (mask & ATTR_XVATTR))) {
2707 return (SET_ERROR(EINVAL));
2710 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2712 return (SET_ERROR(EISDIR));
2715 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2717 return (SET_ERROR(EINVAL));
2721 * If this is an xvattr_t, then get a pointer to the structure of
2722 * optional attributes. If this is NULL, then we have a vattr_t.
2724 xoap = xva_getxoptattr(xvap);
2726 tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP);
2727 xva_init(tmpxvattr);
2729 bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2730 xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2733 * Immutable files can only alter immutable bit and atime
2735 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2736 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2737 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2738 err = SET_ERROR(EPERM);
2742 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2743 err = SET_ERROR(EPERM);
2748 * Verify timestamps doesn't overflow 32 bits.
2749 * ZFS can handle large timestamps, but 32bit syscalls can't
2750 * handle times greater than 2039. This check should be removed
2751 * once large timestamps are fully supported.
2753 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2754 if (((mask & ATTR_ATIME) &&
2755 TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2756 ((mask & ATTR_MTIME) &&
2757 TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2758 err = SET_ERROR(EOVERFLOW);
2767 /* Can this be moved to before the top label? */
2768 if (zfs_is_readonly(zfsvfs)) {
2769 err = SET_ERROR(EROFS);
2774 * First validate permissions
2777 if (mask & ATTR_SIZE) {
2778 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2783 * XXX - Note, we are not providing any open
2784 * mode flags here (like FNDELAY), so we may
2785 * block if there are locks present... this
2786 * should be addressed in openat().
2788 /* XXX - would it be OK to generate a log record here? */
2789 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2794 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2795 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2796 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2797 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2798 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2799 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2800 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2801 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2802 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2806 if (mask & (ATTR_UID|ATTR_GID)) {
2807 int idmask = (mask & (ATTR_UID|ATTR_GID));
2812 * NOTE: even if a new mode is being set,
2813 * we may clear S_ISUID/S_ISGID bits.
2816 if (!(mask & ATTR_MODE))
2817 vap->va_mode = zp->z_mode;
2820 * Take ownership or chgrp to group we are a member of
2823 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
2824 take_group = (mask & ATTR_GID) &&
2825 zfs_groupmember(zfsvfs, vap->va_gid, cr);
2828 * If both ATTR_UID and ATTR_GID are set then take_owner and
2829 * take_group must both be set in order to allow taking
2832 * Otherwise, send the check through secpolicy_vnode_setattr()
2836 if (((idmask == (ATTR_UID|ATTR_GID)) &&
2837 take_owner && take_group) ||
2838 ((idmask == ATTR_UID) && take_owner) ||
2839 ((idmask == ATTR_GID) && take_group)) {
2840 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2841 skipaclchk, cr) == 0) {
2843 * Remove setuid/setgid for non-privileged users
2845 (void) secpolicy_setid_clear(vap, cr);
2846 trim_mask = (mask & (ATTR_UID|ATTR_GID));
2855 mutex_enter(&zp->z_lock);
2856 oldva.va_mode = zp->z_mode;
2857 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2858 if (mask & ATTR_XVATTR) {
2860 * Update xvattr mask to include only those attributes
2861 * that are actually changing.
2863 * the bits will be restored prior to actually setting
2864 * the attributes so the caller thinks they were set.
2866 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2867 if (xoap->xoa_appendonly !=
2868 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2871 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2872 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
2876 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2877 if (xoap->xoa_nounlink !=
2878 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2881 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2882 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
2886 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2887 if (xoap->xoa_immutable !=
2888 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2891 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2892 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
2896 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2897 if (xoap->xoa_nodump !=
2898 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2901 XVA_CLR_REQ(xvap, XAT_NODUMP);
2902 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
2906 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2907 if (xoap->xoa_av_modified !=
2908 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2911 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2912 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
2916 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2917 if ((!S_ISREG(ip->i_mode) &&
2918 xoap->xoa_av_quarantined) ||
2919 xoap->xoa_av_quarantined !=
2920 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2923 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2924 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
2928 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2929 mutex_exit(&zp->z_lock);
2930 err = SET_ERROR(EPERM);
2934 if (need_policy == FALSE &&
2935 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2936 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2941 mutex_exit(&zp->z_lock);
2943 if (mask & ATTR_MODE) {
2944 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2945 err = secpolicy_setid_setsticky_clear(ip, vap,
2950 trim_mask |= ATTR_MODE;
2958 * If trim_mask is set then take ownership
2959 * has been granted or write_acl is present and user
2960 * has the ability to modify mode. In that case remove
2961 * UID|GID and or MODE from mask so that
2962 * secpolicy_vnode_setattr() doesn't revoke it.
2966 saved_mask = vap->va_mask;
2967 vap->va_mask &= ~trim_mask;
2969 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
2970 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2975 vap->va_mask |= saved_mask;
2979 * secpolicy_vnode_setattr, or take ownership may have
2982 mask = vap->va_mask;
2984 if ((mask & (ATTR_UID | ATTR_GID))) {
2985 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2986 &xattr_obj, sizeof (xattr_obj));
2988 if (err == 0 && xattr_obj) {
2989 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
2993 if (mask & ATTR_UID) {
2994 new_kuid = zfs_fuid_create(zfsvfs,
2995 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2996 if (new_kuid != KUID_TO_SUID(ZTOI(zp)->i_uid) &&
2997 zfs_fuid_overquota(zfsvfs, B_FALSE, new_kuid)) {
3000 err = SET_ERROR(EDQUOT);
3005 if (mask & ATTR_GID) {
3006 new_kgid = zfs_fuid_create(zfsvfs,
3007 (uint64_t)vap->va_gid, cr, ZFS_GROUP, &fuidp);
3008 if (new_kgid != KGID_TO_SGID(ZTOI(zp)->i_gid) &&
3009 zfs_fuid_overquota(zfsvfs, B_TRUE, new_kgid)) {
3012 err = SET_ERROR(EDQUOT);
3017 tx = dmu_tx_create(zfsvfs->z_os);
3019 if (mask & ATTR_MODE) {
3020 uint64_t pmode = zp->z_mode;
3022 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3024 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
3026 mutex_enter(&zp->z_lock);
3027 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3029 * Are we upgrading ACL from old V0 format
3032 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3033 zfs_znode_acl_version(zp) ==
3034 ZFS_ACL_VERSION_INITIAL) {
3035 dmu_tx_hold_free(tx, acl_obj, 0,
3037 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3038 0, aclp->z_acl_bytes);
3040 dmu_tx_hold_write(tx, acl_obj, 0,
3043 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3044 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3045 0, aclp->z_acl_bytes);
3047 mutex_exit(&zp->z_lock);
3048 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3050 if ((mask & ATTR_XVATTR) &&
3051 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3052 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3054 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3058 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3061 fuid_dirtied = zfsvfs->z_fuid_dirty;
3063 zfs_fuid_txhold(zfsvfs, tx);
3065 zfs_sa_upgrade_txholds(tx, zp);
3067 err = dmu_tx_assign(tx, TXG_WAIT);
3073 * Set each attribute requested.
3074 * We group settings according to the locks they need to acquire.
3076 * Note: you cannot set ctime directly, although it will be
3077 * updated as a side-effect of calling this function.
3081 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3082 mutex_enter(&zp->z_acl_lock);
3083 mutex_enter(&zp->z_lock);
3085 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3086 &zp->z_pflags, sizeof (zp->z_pflags));
3089 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3090 mutex_enter(&attrzp->z_acl_lock);
3091 mutex_enter(&attrzp->z_lock);
3092 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3093 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3094 sizeof (attrzp->z_pflags));
3097 if (mask & (ATTR_UID|ATTR_GID)) {
3099 if (mask & ATTR_UID) {
3100 ZTOI(zp)->i_uid = SUID_TO_KUID(new_kuid);
3101 new_uid = zfs_uid_read(ZTOI(zp));
3102 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3103 &new_uid, sizeof (new_uid));
3105 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3106 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3108 ZTOI(attrzp)->i_uid = SUID_TO_KUID(new_uid);
3112 if (mask & ATTR_GID) {
3113 ZTOI(zp)->i_gid = SGID_TO_KGID(new_kgid);
3114 new_gid = zfs_gid_read(ZTOI(zp));
3115 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3116 NULL, &new_gid, sizeof (new_gid));
3118 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3119 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3121 ZTOI(attrzp)->i_gid = SGID_TO_KGID(new_kgid);
3124 if (!(mask & ATTR_MODE)) {
3125 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3126 NULL, &new_mode, sizeof (new_mode));
3127 new_mode = zp->z_mode;
3129 err = zfs_acl_chown_setattr(zp);
3132 err = zfs_acl_chown_setattr(attrzp);
3137 if (mask & ATTR_MODE) {
3138 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3139 &new_mode, sizeof (new_mode));
3140 zp->z_mode = ZTOI(zp)->i_mode = new_mode;
3141 ASSERT3P(aclp, !=, NULL);
3142 err = zfs_aclset_common(zp, aclp, cr, tx);
3144 if (zp->z_acl_cached)
3145 zfs_acl_free(zp->z_acl_cached);
3146 zp->z_acl_cached = aclp;
3150 if ((mask & ATTR_ATIME) || zp->z_atime_dirty) {
3151 zp->z_atime_dirty = 0;
3152 ZFS_TIME_ENCODE(&ip->i_atime, atime);
3153 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3154 &atime, sizeof (atime));
3157 if (mask & ATTR_MTIME) {
3158 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3159 ZTOI(zp)->i_mtime = timespec_trunc(vap->va_mtime,
3160 ZTOI(zp)->i_sb->s_time_gran);
3162 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3163 mtime, sizeof (mtime));
3166 if (mask & ATTR_CTIME) {
3167 ZFS_TIME_ENCODE(&vap->va_ctime, ctime);
3168 ZTOI(zp)->i_ctime = timespec_trunc(vap->va_ctime,
3169 ZTOI(zp)->i_sb->s_time_gran);
3170 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3171 ctime, sizeof (ctime));
3174 if (attrzp && mask) {
3175 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3176 SA_ZPL_CTIME(zfsvfs), NULL, &ctime,
3181 * Do this after setting timestamps to prevent timestamp
3182 * update from toggling bit
3185 if (xoap && (mask & ATTR_XVATTR)) {
3188 * restore trimmed off masks
3189 * so that return masks can be set for caller.
3192 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
3193 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3195 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
3196 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3198 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
3199 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3201 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
3202 XVA_SET_REQ(xvap, XAT_NODUMP);
3204 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
3205 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3207 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
3208 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3211 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3212 ASSERT(S_ISREG(ip->i_mode));
3214 zfs_xvattr_set(zp, xvap, tx);
3218 zfs_fuid_sync(zfsvfs, tx);
3221 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3223 mutex_exit(&zp->z_lock);
3224 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3225 mutex_exit(&zp->z_acl_lock);
3228 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3229 mutex_exit(&attrzp->z_acl_lock);
3230 mutex_exit(&attrzp->z_lock);
3233 if (err == 0 && attrzp) {
3234 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3243 zfs_fuid_info_free(fuidp);
3251 if (err == ERESTART)
3254 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3258 zfs_inode_update(zp);
3262 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3263 zil_commit(zilog, 0);
3266 kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * 7);
3267 kmem_free(bulk, sizeof (sa_bulk_attr_t) * 7);
3268 kmem_free(tmpxvattr, sizeof (xvattr_t));
3273 typedef struct zfs_zlock {
3274 krwlock_t *zl_rwlock; /* lock we acquired */
3275 znode_t *zl_znode; /* znode we held */
3276 struct zfs_zlock *zl_next; /* next in list */
3280 * Drop locks and release vnodes that were held by zfs_rename_lock().
3283 zfs_rename_unlock(zfs_zlock_t **zlpp)
3287 while ((zl = *zlpp) != NULL) {
3288 if (zl->zl_znode != NULL)
3289 zfs_iput_async(ZTOI(zl->zl_znode));
3290 rw_exit(zl->zl_rwlock);
3291 *zlpp = zl->zl_next;
3292 kmem_free(zl, sizeof (*zl));
3297 * Search back through the directory tree, using the ".." entries.
3298 * Lock each directory in the chain to prevent concurrent renames.
3299 * Fail any attempt to move a directory into one of its own descendants.
3300 * XXX - z_parent_lock can overlap with map or grow locks
3303 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3307 uint64_t rootid = ZTOZSB(zp)->z_root;
3308 uint64_t oidp = zp->z_id;
3309 krwlock_t *rwlp = &szp->z_parent_lock;
3310 krw_t rw = RW_WRITER;
3313 * First pass write-locks szp and compares to zp->z_id.
3314 * Later passes read-lock zp and compare to zp->z_parent.
3317 if (!rw_tryenter(rwlp, rw)) {
3319 * Another thread is renaming in this path.
3320 * Note that if we are a WRITER, we don't have any
3321 * parent_locks held yet.
3323 if (rw == RW_READER && zp->z_id > szp->z_id) {
3325 * Drop our locks and restart
3327 zfs_rename_unlock(&zl);
3331 rwlp = &szp->z_parent_lock;
3336 * Wait for other thread to drop its locks
3342 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3343 zl->zl_rwlock = rwlp;
3344 zl->zl_znode = NULL;
3345 zl->zl_next = *zlpp;
3348 if (oidp == szp->z_id) /* We're a descendant of szp */
3349 return (SET_ERROR(EINVAL));
3351 if (oidp == rootid) /* We've hit the top */
3354 if (rw == RW_READER) { /* i.e. not the first pass */
3355 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3360 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3361 &oidp, sizeof (oidp));
3362 rwlp = &zp->z_parent_lock;
3365 } while (zp->z_id != sdzp->z_id);
3371 * Move an entry from the provided source directory to the target
3372 * directory. Change the entry name as indicated.
3374 * IN: sdip - Source directory containing the "old entry".
3375 * snm - Old entry name.
3376 * tdip - Target directory to contain the "new entry".
3377 * tnm - New entry name.
3378 * cr - credentials of caller.
3379 * flags - case flags
3381 * RETURN: 0 on success, error code on failure.
3384 * sdip,tdip - ctime|mtime updated
3388 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3389 cred_t *cr, int flags)
3391 znode_t *tdzp, *szp, *tzp;
3392 znode_t *sdzp = ITOZ(sdip);
3393 zfsvfs_t *zfsvfs = ITOZSB(sdip);
3395 zfs_dirlock_t *sdl, *tdl;
3398 int cmp, serr, terr;
3401 boolean_t waited = B_FALSE;
3403 if (snm == NULL || tnm == NULL)
3404 return (SET_ERROR(EINVAL));
3407 ZFS_VERIFY_ZP(sdzp);
3408 zilog = zfsvfs->z_log;
3411 ZFS_VERIFY_ZP(tdzp);
3414 * We check i_sb because snapshots and the ctldir must have different
3417 if (tdip->i_sb != sdip->i_sb || zfsctl_is_node(tdip)) {
3419 return (SET_ERROR(EXDEV));
3422 if (zfsvfs->z_utf8 && u8_validate(tnm,
3423 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3425 return (SET_ERROR(EILSEQ));
3428 if (flags & FIGNORECASE)
3437 * This is to prevent the creation of links into attribute space
3438 * by renaming a linked file into/outof an attribute directory.
3439 * See the comment in zfs_link() for why this is considered bad.
3441 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3443 return (SET_ERROR(EINVAL));
3447 * Lock source and target directory entries. To prevent deadlock,
3448 * a lock ordering must be defined. We lock the directory with
3449 * the smallest object id first, or if it's a tie, the one with
3450 * the lexically first name.
3452 if (sdzp->z_id < tdzp->z_id) {
3454 } else if (sdzp->z_id > tdzp->z_id) {
3458 * First compare the two name arguments without
3459 * considering any case folding.
3461 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3463 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3464 ASSERT(error == 0 || !zfsvfs->z_utf8);
3467 * POSIX: "If the old argument and the new argument
3468 * both refer to links to the same existing file,
3469 * the rename() function shall return successfully
3470 * and perform no other action."
3476 * If the file system is case-folding, then we may
3477 * have some more checking to do. A case-folding file
3478 * system is either supporting mixed case sensitivity
3479 * access or is completely case-insensitive. Note
3480 * that the file system is always case preserving.
3482 * In mixed sensitivity mode case sensitive behavior
3483 * is the default. FIGNORECASE must be used to
3484 * explicitly request case insensitive behavior.
3486 * If the source and target names provided differ only
3487 * by case (e.g., a request to rename 'tim' to 'Tim'),
3488 * we will treat this as a special case in the
3489 * case-insensitive mode: as long as the source name
3490 * is an exact match, we will allow this to proceed as
3491 * a name-change request.
3493 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3494 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3495 flags & FIGNORECASE)) &&
3496 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3499 * case preserving rename request, require exact
3508 * If the source and destination directories are the same, we should
3509 * grab the z_name_lock of that directory only once.
3513 rw_enter(&sdzp->z_name_lock, RW_READER);
3517 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3518 ZEXISTS | zflg, NULL, NULL);
3519 terr = zfs_dirent_lock(&tdl,
3520 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3522 terr = zfs_dirent_lock(&tdl,
3523 tdzp, tnm, &tzp, zflg, NULL, NULL);
3524 serr = zfs_dirent_lock(&sdl,
3525 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3531 * Source entry invalid or not there.
3534 zfs_dirent_unlock(tdl);
3540 rw_exit(&sdzp->z_name_lock);
3542 if (strcmp(snm, "..") == 0)
3548 zfs_dirent_unlock(sdl);
3552 rw_exit(&sdzp->z_name_lock);
3554 if (strcmp(tnm, "..") == 0)
3561 * Must have write access at the source to remove the old entry
3562 * and write access at the target to create the new entry.
3563 * Note that if target and source are the same, this can be
3564 * done in a single check.
3567 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3570 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3572 * Check to make sure rename is valid.
3573 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3575 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3580 * Does target exist?
3584 * Source and target must be the same type.
3586 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3587 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3588 error = SET_ERROR(ENOTDIR);
3592 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3593 error = SET_ERROR(EISDIR);
3598 * POSIX dictates that when the source and target
3599 * entries refer to the same file object, rename
3600 * must do nothing and exit without error.
3602 if (szp->z_id == tzp->z_id) {
3608 tx = dmu_tx_create(zfsvfs->z_os);
3609 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3610 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3611 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3612 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3614 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3615 zfs_sa_upgrade_txholds(tx, tdzp);
3618 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3619 zfs_sa_upgrade_txholds(tx, tzp);
3622 zfs_sa_upgrade_txholds(tx, szp);
3623 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3624 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3627 zfs_rename_unlock(&zl);
3628 zfs_dirent_unlock(sdl);
3629 zfs_dirent_unlock(tdl);
3632 rw_exit(&sdzp->z_name_lock);
3634 if (error == ERESTART) {
3651 if (tzp) /* Attempt to remove the existing target */
3652 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3655 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3657 szp->z_pflags |= ZFS_AV_MODIFIED;
3659 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3660 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3663 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3665 zfs_log_rename(zilog, tx, TX_RENAME |
3666 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3667 sdl->dl_name, tdzp, tdl->dl_name, szp);
3670 * At this point, we have successfully created
3671 * the target name, but have failed to remove
3672 * the source name. Since the create was done
3673 * with the ZRENAMING flag, there are
3674 * complications; for one, the link count is
3675 * wrong. The easiest way to deal with this
3676 * is to remove the newly created target, and
3677 * return the original error. This must
3678 * succeed; fortunately, it is very unlikely to
3679 * fail, since we just created it.
3681 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3682 ZRENAMING, NULL), ==, 0);
3690 zfs_rename_unlock(&zl);
3692 zfs_dirent_unlock(sdl);
3693 zfs_dirent_unlock(tdl);
3695 zfs_inode_update(sdzp);
3697 rw_exit(&sdzp->z_name_lock);
3700 zfs_inode_update(tdzp);
3702 zfs_inode_update(szp);
3705 zfs_inode_update(tzp);
3709 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3710 zil_commit(zilog, 0);
3717 * Insert the indicated symbolic reference entry into the directory.
3719 * IN: dip - Directory to contain new symbolic link.
3720 * link - Name for new symlink entry.
3721 * vap - Attributes of new entry.
3722 * target - Target path of new symlink.
3724 * cr - credentials of caller.
3725 * flags - case flags
3727 * RETURN: 0 on success, error code on failure.
3730 * dip - ctime|mtime updated
3734 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
3735 struct inode **ipp, cred_t *cr, int flags)
3737 znode_t *zp, *dzp = ITOZ(dip);
3740 zfsvfs_t *zfsvfs = ITOZSB(dip);
3742 uint64_t len = strlen(link);
3745 zfs_acl_ids_t acl_ids;
3746 boolean_t fuid_dirtied;
3747 uint64_t txtype = TX_SYMLINK;
3748 boolean_t waited = B_FALSE;
3750 ASSERT(S_ISLNK(vap->va_mode));
3753 return (SET_ERROR(EINVAL));
3757 zilog = zfsvfs->z_log;
3759 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3760 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3762 return (SET_ERROR(EILSEQ));
3764 if (flags & FIGNORECASE)
3767 if (len > MAXPATHLEN) {
3769 return (SET_ERROR(ENAMETOOLONG));
3772 if ((error = zfs_acl_ids_create(dzp, 0,
3773 vap, cr, NULL, &acl_ids)) != 0) {
3781 * Attempt to lock directory; fail if entry already exists.
3783 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3785 zfs_acl_ids_free(&acl_ids);
3790 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3791 zfs_acl_ids_free(&acl_ids);
3792 zfs_dirent_unlock(dl);
3797 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3798 zfs_acl_ids_free(&acl_ids);
3799 zfs_dirent_unlock(dl);
3801 return (SET_ERROR(EDQUOT));
3803 tx = dmu_tx_create(zfsvfs->z_os);
3804 fuid_dirtied = zfsvfs->z_fuid_dirty;
3805 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3806 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3807 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3808 ZFS_SA_BASE_ATTR_SIZE + len);
3809 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3810 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3811 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3812 acl_ids.z_aclp->z_acl_bytes);
3815 zfs_fuid_txhold(zfsvfs, tx);
3816 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3818 zfs_dirent_unlock(dl);
3819 if (error == ERESTART) {
3825 zfs_acl_ids_free(&acl_ids);
3832 * Create a new object for the symlink.
3833 * for version 4 ZPL datsets the symlink will be an SA attribute
3835 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3838 zfs_fuid_sync(zfsvfs, tx);
3840 mutex_enter(&zp->z_lock);
3842 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3845 zfs_sa_symlink(zp, link, len, tx);
3846 mutex_exit(&zp->z_lock);
3849 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3850 &zp->z_size, sizeof (zp->z_size), tx);
3852 * Insert the new object into the directory.
3854 (void) zfs_link_create(dl, zp, tx, ZNEW);
3856 if (flags & FIGNORECASE)
3858 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3860 zfs_inode_update(dzp);
3861 zfs_inode_update(zp);
3863 zfs_acl_ids_free(&acl_ids);
3867 zfs_dirent_unlock(dl);
3871 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3872 zil_commit(zilog, 0);
3879 * Return, in the buffer contained in the provided uio structure,
3880 * the symbolic path referred to by ip.
3882 * IN: ip - inode of symbolic link
3883 * uio - structure to contain the link path.
3884 * cr - credentials of caller.
3886 * RETURN: 0 if success
3887 * error code if failure
3890 * ip - atime updated
3894 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
3896 znode_t *zp = ITOZ(ip);
3897 zfsvfs_t *zfsvfs = ITOZSB(ip);
3903 mutex_enter(&zp->z_lock);
3905 error = sa_lookup_uio(zp->z_sa_hdl,
3906 SA_ZPL_SYMLINK(zfsvfs), uio);
3908 error = zfs_sa_readlink(zp, uio);
3909 mutex_exit(&zp->z_lock);
3916 * Insert a new entry into directory tdip referencing sip.
3918 * IN: tdip - Directory to contain new entry.
3919 * sip - inode of new entry.
3920 * name - name of new entry.
3921 * cr - credentials of caller.
3923 * RETURN: 0 if success
3924 * error code if failure
3927 * tdip - ctime|mtime updated
3928 * sip - ctime updated
3932 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr,
3935 znode_t *dzp = ITOZ(tdip);
3937 zfsvfs_t *zfsvfs = ITOZSB(tdip);
3945 boolean_t waited = B_FALSE;
3946 boolean_t is_tmpfile = 0;
3949 is_tmpfile = (sip->i_nlink == 0 && (sip->i_state & I_LINKABLE));
3951 ASSERT(S_ISDIR(tdip->i_mode));
3954 return (SET_ERROR(EINVAL));
3958 zilog = zfsvfs->z_log;
3961 * POSIX dictates that we return EPERM here.
3962 * Better choices include ENOTSUP or EISDIR.
3964 if (S_ISDIR(sip->i_mode)) {
3966 return (SET_ERROR(EPERM));
3973 * We check i_sb because snapshots and the ctldir must have different
3976 if (sip->i_sb != tdip->i_sb || zfsctl_is_node(sip)) {
3978 return (SET_ERROR(EXDEV));
3981 /* Prevent links to .zfs/shares files */
3983 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3984 &parent, sizeof (uint64_t))) != 0) {
3988 if (parent == zfsvfs->z_shares_dir) {
3990 return (SET_ERROR(EPERM));
3993 if (zfsvfs->z_utf8 && u8_validate(name,
3994 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3996 return (SET_ERROR(EILSEQ));
3998 if (flags & FIGNORECASE)
4002 * We do not support links between attributes and non-attributes
4003 * because of the potential security risk of creating links
4004 * into "normal" file space in order to circumvent restrictions
4005 * imposed in attribute space.
4007 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4009 return (SET_ERROR(EINVAL));
4012 owner = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(sip->i_uid),
4014 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4016 return (SET_ERROR(EPERM));
4019 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
4026 * Attempt to lock directory; fail if entry already exists.
4028 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4034 tx = dmu_tx_create(zfsvfs->z_os);
4035 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4036 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4038 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
4040 zfs_sa_upgrade_txholds(tx, szp);
4041 zfs_sa_upgrade_txholds(tx, dzp);
4042 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4044 zfs_dirent_unlock(dl);
4045 if (error == ERESTART) {
4055 /* unmark z_unlinked so zfs_link_create will not reject */
4057 szp->z_unlinked = 0;
4058 error = zfs_link_create(dl, szp, tx, 0);
4061 uint64_t txtype = TX_LINK;
4063 * tmpfile is created to be in z_unlinkedobj, so remove it.
4064 * Also, we don't log in ZIL, be cause all previous file
4065 * operation on the tmpfile are ignored by ZIL. Instead we
4066 * always wait for txg to sync to make sure all previous
4067 * operation are sync safe.
4070 VERIFY(zap_remove_int(zfsvfs->z_os,
4071 zfsvfs->z_unlinkedobj, szp->z_id, tx) == 0);
4073 if (flags & FIGNORECASE)
4075 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4077 } else if (is_tmpfile) {
4078 /* restore z_unlinked since when linking failed */
4079 szp->z_unlinked = 1;
4081 txg = dmu_tx_get_txg(tx);
4084 zfs_dirent_unlock(dl);
4086 if (!is_tmpfile && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4087 zil_commit(zilog, 0);
4090 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), txg);
4092 zfs_inode_update(dzp);
4093 zfs_inode_update(szp);
4099 zfs_putpage_commit_cb(void *arg)
4101 struct page *pp = arg;
4104 end_page_writeback(pp);
4108 * Push a page out to disk, once the page is on stable storage the
4109 * registered commit callback will be run as notification of completion.
4111 * IN: ip - page mapped for inode.
4112 * pp - page to push (page is locked)
4113 * wbc - writeback control data
4115 * RETURN: 0 if success
4116 * error code if failure
4119 * ip - ctime|mtime updated
4123 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
4125 znode_t *zp = ITOZ(ip);
4126 zfsvfs_t *zfsvfs = ITOZSB(ip);
4134 uint64_t mtime[2], ctime[2];
4135 sa_bulk_attr_t bulk[3];
4137 struct address_space *mapping;
4142 ASSERT(PageLocked(pp));
4144 pgoff = page_offset(pp); /* Page byte-offset in file */
4145 offset = i_size_read(ip); /* File length in bytes */
4146 pglen = MIN(PAGE_SIZE, /* Page length in bytes */
4147 P2ROUNDUP(offset, PAGE_SIZE)-pgoff);
4149 /* Page is beyond end of file */
4150 if (pgoff >= offset) {
4156 /* Truncate page length to end of file */
4157 if (pgoff + pglen > offset)
4158 pglen = offset - pgoff;
4162 * FIXME: Allow mmap writes past its quota. The correct fix
4163 * is to register a page_mkwrite() handler to count the page
4164 * against its quota when it is about to be dirtied.
4166 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4167 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4173 * The ordering here is critical and must adhere to the following
4174 * rules in order to avoid deadlocking in either zfs_read() or
4175 * zfs_free_range() due to a lock inversion.
4177 * 1) The page must be unlocked prior to acquiring the range lock.
4178 * This is critical because zfs_read() calls find_lock_page()
4179 * which may block on the page lock while holding the range lock.
4181 * 2) Before setting or clearing write back on a page the range lock
4182 * must be held in order to prevent a lock inversion with the
4183 * zfs_free_range() function.
4185 * This presents a problem because upon entering this function the
4186 * page lock is already held. To safely acquire the range lock the
4187 * page lock must be dropped. This creates a window where another
4188 * process could truncate, invalidate, dirty, or write out the page.
4190 * Therefore, after successfully reacquiring the range and page locks
4191 * the current page state is checked. In the common case everything
4192 * will be as is expected and it can be written out. However, if
4193 * the page state has changed it must be handled accordingly.
4195 mapping = pp->mapping;
4196 redirty_page_for_writepage(wbc, pp);
4199 rl = zfs_range_lock(&zp->z_range_lock, pgoff, pglen, RL_WRITER);
4202 /* Page mapping changed or it was no longer dirty, we're done */
4203 if (unlikely((mapping != pp->mapping) || !PageDirty(pp))) {
4205 zfs_range_unlock(rl);
4210 /* Another process started write block if required */
4211 if (PageWriteback(pp)) {
4213 zfs_range_unlock(rl);
4215 if (wbc->sync_mode != WB_SYNC_NONE)
4216 wait_on_page_writeback(pp);
4222 /* Clear the dirty flag the required locks are held */
4223 if (!clear_page_dirty_for_io(pp)) {
4225 zfs_range_unlock(rl);
4231 * Counterpart for redirty_page_for_writepage() above. This page
4232 * was in fact not skipped and should not be counted as if it were.
4234 wbc->pages_skipped--;
4235 set_page_writeback(pp);
4238 tx = dmu_tx_create(zfsvfs->z_os);
4239 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
4240 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4241 zfs_sa_upgrade_txholds(tx, zp);
4243 err = dmu_tx_assign(tx, TXG_NOWAIT);
4245 if (err == ERESTART)
4249 __set_page_dirty_nobuffers(pp);
4251 end_page_writeback(pp);
4252 zfs_range_unlock(rl);
4258 ASSERT3U(pglen, <=, PAGE_SIZE);
4259 dmu_write(zfsvfs->z_os, zp->z_id, pgoff, pglen, va, tx);
4262 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
4263 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
4264 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zfsvfs), NULL,
4267 /* Preserve the mtime and ctime provided by the inode */
4268 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4269 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4270 zp->z_atime_dirty = 0;
4273 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4275 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0,
4276 zfs_putpage_commit_cb, pp);
4279 zfs_range_unlock(rl);
4281 if (wbc->sync_mode != WB_SYNC_NONE) {
4283 * Note that this is rarely called under writepages(), because
4284 * writepages() normally handles the entire commit for
4285 * performance reasons.
4287 zil_commit(zfsvfs->z_log, zp->z_id);
4295 * Update the system attributes when the inode has been dirtied. For the
4296 * moment we only update the mode, atime, mtime, and ctime.
4299 zfs_dirty_inode(struct inode *ip, int flags)
4301 znode_t *zp = ITOZ(ip);
4302 zfsvfs_t *zfsvfs = ITOZSB(ip);
4304 uint64_t mode, atime[2], mtime[2], ctime[2];
4305 sa_bulk_attr_t bulk[4];
4309 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
4317 * This is the lazytime semantic indroduced in Linux 4.0
4318 * This flag will only be called from update_time when lazytime is set.
4319 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4320 * Fortunately mtime and ctime are managed within ZFS itself, so we
4321 * only need to dirty atime.
4323 if (flags == I_DIRTY_TIME) {
4324 zp->z_atime_dirty = 1;
4329 tx = dmu_tx_create(zfsvfs->z_os);
4331 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4332 zfs_sa_upgrade_txholds(tx, zp);
4334 error = dmu_tx_assign(tx, TXG_WAIT);
4340 mutex_enter(&zp->z_lock);
4341 zp->z_atime_dirty = 0;
4343 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
4344 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
4345 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
4346 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
4348 /* Preserve the mode, mtime and ctime provided by the inode */
4349 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4350 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4351 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4356 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4357 mutex_exit(&zp->z_lock);
4367 zfs_inactive(struct inode *ip)
4369 znode_t *zp = ITOZ(ip);
4370 zfsvfs_t *zfsvfs = ITOZSB(ip);
4373 int need_unlock = 0;
4375 /* Only read lock if we haven't already write locked, e.g. rollback */
4376 if (!RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)) {
4378 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4380 if (zp->z_sa_hdl == NULL) {
4382 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4386 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4387 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4389 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4390 zfs_sa_upgrade_txholds(tx, zp);
4391 error = dmu_tx_assign(tx, TXG_WAIT);
4395 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4396 mutex_enter(&zp->z_lock);
4397 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4398 (void *)&atime, sizeof (atime), tx);
4399 zp->z_atime_dirty = 0;
4400 mutex_exit(&zp->z_lock);
4407 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4411 * Bounds-check the seek operation.
4413 * IN: ip - inode seeking within
4414 * ooff - old file offset
4415 * noffp - pointer to new file offset
4416 * ct - caller context
4418 * RETURN: 0 if success
4419 * EINVAL if new offset invalid
4423 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
4425 if (S_ISDIR(ip->i_mode))
4427 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4431 * Fill pages with data from the disk.
4434 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
4436 znode_t *zp = ITOZ(ip);
4437 zfsvfs_t *zfsvfs = ITOZSB(ip);
4439 struct page *cur_pp;
4440 u_offset_t io_off, total;
4447 io_len = nr_pages << PAGE_SHIFT;
4448 i_size = i_size_read(ip);
4449 io_off = page_offset(pl[0]);
4451 if (io_off + io_len > i_size)
4452 io_len = i_size - io_off;
4455 * Iterate over list of pages and read each page individually.
4458 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4461 cur_pp = pl[page_idx++];
4463 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4467 /* convert checksum errors into IO errors */
4469 err = SET_ERROR(EIO);
4478 * Uses zfs_fillpage to read data from the file and fill the pages.
4480 * IN: ip - inode of file to get data from.
4481 * pl - list of pages to read
4482 * nr_pages - number of pages to read
4484 * RETURN: 0 on success, error code on failure.
4487 * vp - atime updated
4491 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
4493 znode_t *zp = ITOZ(ip);
4494 zfsvfs_t *zfsvfs = ITOZSB(ip);
4503 err = zfs_fillpage(ip, pl, nr_pages);
4510 * Check ZFS specific permissions to memory map a section of a file.
4512 * IN: ip - inode of the file to mmap
4514 * addrp - start address in memory region
4515 * len - length of memory region
4516 * vm_flags- address flags
4518 * RETURN: 0 if success
4519 * error code if failure
4523 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4524 unsigned long vm_flags)
4526 znode_t *zp = ITOZ(ip);
4527 zfsvfs_t *zfsvfs = ITOZSB(ip);
4532 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4533 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4535 return (SET_ERROR(EPERM));
4538 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4539 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4541 return (SET_ERROR(EACCES));
4544 if (off < 0 || len > MAXOFFSET_T - off) {
4546 return (SET_ERROR(ENXIO));
4554 * convoff - converts the given data (start, whence) to the
4558 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4563 if ((lckdat->l_whence == 2) || (whence == 2)) {
4564 if ((error = zfs_getattr(ip, &vap, 0, CRED()) != 0))
4568 switch (lckdat->l_whence) {
4570 lckdat->l_start += offset;
4573 lckdat->l_start += vap.va_size;
4578 return (SET_ERROR(EINVAL));
4581 if (lckdat->l_start < 0)
4582 return (SET_ERROR(EINVAL));
4586 lckdat->l_start -= offset;
4589 lckdat->l_start -= vap.va_size;
4594 return (SET_ERROR(EINVAL));
4597 lckdat->l_whence = (short)whence;
4602 * Free or allocate space in a file. Currently, this function only
4603 * supports the `F_FREESP' command. However, this command is somewhat
4604 * misnamed, as its functionality includes the ability to allocate as
4605 * well as free space.
4607 * IN: ip - inode of file to free data in.
4608 * cmd - action to take (only F_FREESP supported).
4609 * bfp - section of file to free/alloc.
4610 * flag - current file open mode flags.
4611 * offset - current file offset.
4612 * cr - credentials of caller [UNUSED].
4614 * RETURN: 0 on success, error code on failure.
4617 * ip - ctime|mtime updated
4621 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4622 offset_t offset, cred_t *cr)
4624 znode_t *zp = ITOZ(ip);
4625 zfsvfs_t *zfsvfs = ITOZSB(ip);
4632 if (cmd != F_FREESP) {
4634 return (SET_ERROR(EINVAL));
4638 * Callers might not be able to detect properly that we are read-only,
4639 * so check it explicitly here.
4641 if (zfs_is_readonly(zfsvfs)) {
4643 return (SET_ERROR(EROFS));
4646 if ((error = convoff(ip, bfp, 0, offset))) {
4651 if (bfp->l_len < 0) {
4653 return (SET_ERROR(EINVAL));
4657 * Permissions aren't checked on Solaris because on this OS
4658 * zfs_space() can only be called with an opened file handle.
4659 * On Linux we can get here through truncate_range() which
4660 * operates directly on inodes, so we need to check access rights.
4662 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4668 len = bfp->l_len; /* 0 means from off to end of file */
4670 error = zfs_freesp(zp, off, len, flag, TRUE);
4678 zfs_fid(struct inode *ip, fid_t *fidp)
4680 znode_t *zp = ITOZ(ip);
4681 zfsvfs_t *zfsvfs = ITOZSB(ip);
4684 uint64_t object = zp->z_id;
4691 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4692 &gen64, sizeof (uint64_t))) != 0) {
4697 gen = (uint32_t)gen64;
4699 size = SHORT_FID_LEN;
4701 zfid = (zfid_short_t *)fidp;
4703 zfid->zf_len = size;
4705 for (i = 0; i < sizeof (zfid->zf_object); i++)
4706 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4708 /* Must have a non-zero generation number to distinguish from .zfs */
4711 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4712 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4720 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4722 znode_t *zp = ITOZ(ip);
4723 zfsvfs_t *zfsvfs = ITOZSB(ip);
4725 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4729 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4737 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4739 znode_t *zp = ITOZ(ip);
4740 zfsvfs_t *zfsvfs = ITOZSB(ip);
4742 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4743 zilog_t *zilog = zfsvfs->z_log;
4748 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4750 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4751 zil_commit(zilog, 0);
4757 #ifdef HAVE_UIO_ZEROCOPY
4759 * Tunable, both must be a power of 2.
4761 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4762 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4763 * an arcbuf for a partial block read
4765 int zcr_blksz_min = (1 << 10); /* 1K */
4766 int zcr_blksz_max = (1 << 17); /* 128K */
4770 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
4772 znode_t *zp = ITOZ(ip);
4773 zfsvfs_t *zfsvfs = ITOZSB(ip);
4774 int max_blksz = zfsvfs->z_max_blksz;
4775 uio_t *uio = &xuio->xu_uio;
4776 ssize_t size = uio->uio_resid;
4777 offset_t offset = uio->uio_loffset;
4782 int preamble, postamble;
4784 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4785 return (SET_ERROR(EINVAL));
4792 * Loan out an arc_buf for write if write size is bigger than
4793 * max_blksz, and the file's block size is also max_blksz.
4796 if (size < blksz || zp->z_blksz != blksz) {
4798 return (SET_ERROR(EINVAL));
4801 * Caller requests buffers for write before knowing where the
4802 * write offset might be (e.g. NFS TCP write).
4807 preamble = P2PHASE(offset, blksz);
4809 preamble = blksz - preamble;
4814 postamble = P2PHASE(size, blksz);
4817 fullblk = size / blksz;
4818 (void) dmu_xuio_init(xuio,
4819 (preamble != 0) + fullblk + (postamble != 0));
4822 * Have to fix iov base/len for partial buffers. They
4823 * currently represent full arc_buf's.
4826 /* data begins in the middle of the arc_buf */
4827 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4830 (void) dmu_xuio_add(xuio, abuf,
4831 blksz - preamble, preamble);
4834 for (i = 0; i < fullblk; i++) {
4835 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4838 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
4842 /* data ends in the middle of the arc_buf */
4843 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4846 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
4851 * Loan out an arc_buf for read if the read size is larger than
4852 * the current file block size. Block alignment is not
4853 * considered. Partial arc_buf will be loaned out for read.
4855 blksz = zp->z_blksz;
4856 if (blksz < zcr_blksz_min)
4857 blksz = zcr_blksz_min;
4858 if (blksz > zcr_blksz_max)
4859 blksz = zcr_blksz_max;
4860 /* avoid potential complexity of dealing with it */
4861 if (blksz > max_blksz) {
4863 return (SET_ERROR(EINVAL));
4866 maxsize = zp->z_size - uio->uio_loffset;
4872 return (SET_ERROR(EINVAL));
4877 return (SET_ERROR(EINVAL));
4880 uio->uio_extflg = UIO_XUIO;
4881 XUIO_XUZC_RW(xuio) = ioflag;
4888 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
4892 int ioflag = XUIO_XUZC_RW(xuio);
4894 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
4896 i = dmu_xuio_cnt(xuio);
4898 abuf = dmu_xuio_arcbuf(xuio, i);
4900 * if abuf == NULL, it must be a write buffer
4901 * that has been returned in zfs_write().
4904 dmu_return_arcbuf(abuf);
4905 ASSERT(abuf || ioflag == UIO_WRITE);
4908 dmu_xuio_fini(xuio);
4911 #endif /* HAVE_UIO_ZEROCOPY */
4913 #if defined(_KERNEL) && defined(HAVE_SPL)
4914 EXPORT_SYMBOL(zfs_open);
4915 EXPORT_SYMBOL(zfs_close);
4916 EXPORT_SYMBOL(zfs_read);
4917 EXPORT_SYMBOL(zfs_write);
4918 EXPORT_SYMBOL(zfs_access);
4919 EXPORT_SYMBOL(zfs_lookup);
4920 EXPORT_SYMBOL(zfs_create);
4921 EXPORT_SYMBOL(zfs_tmpfile);
4922 EXPORT_SYMBOL(zfs_remove);
4923 EXPORT_SYMBOL(zfs_mkdir);
4924 EXPORT_SYMBOL(zfs_rmdir);
4925 EXPORT_SYMBOL(zfs_readdir);
4926 EXPORT_SYMBOL(zfs_fsync);
4927 EXPORT_SYMBOL(zfs_getattr);
4928 EXPORT_SYMBOL(zfs_getattr_fast);
4929 EXPORT_SYMBOL(zfs_setattr);
4930 EXPORT_SYMBOL(zfs_rename);
4931 EXPORT_SYMBOL(zfs_symlink);
4932 EXPORT_SYMBOL(zfs_readlink);
4933 EXPORT_SYMBOL(zfs_link);
4934 EXPORT_SYMBOL(zfs_inactive);
4935 EXPORT_SYMBOL(zfs_space);
4936 EXPORT_SYMBOL(zfs_fid);
4937 EXPORT_SYMBOL(zfs_getsecattr);
4938 EXPORT_SYMBOL(zfs_setsecattr);
4939 EXPORT_SYMBOL(zfs_getpage);
4940 EXPORT_SYMBOL(zfs_putpage);
4941 EXPORT_SYMBOL(zfs_dirty_inode);
4942 EXPORT_SYMBOL(zfs_map);
4945 module_param(zfs_delete_blocks, ulong, 0644);
4946 MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async");
4947 module_param(zfs_read_chunk_size, long, 0644);
4948 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");