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>
86 * Each vnode op performs some logical unit of work. To do this, the ZPL must
87 * properly lock its in-core state, create a DMU transaction, do the work,
88 * record this work in the intent log (ZIL), commit the DMU transaction,
89 * and wait for the intent log to commit if it is a synchronous operation.
90 * Moreover, the vnode ops must work in both normal and log replay context.
91 * The ordering of events is important to avoid deadlocks and references
92 * to freed memory. The example below illustrates the following Big Rules:
94 * (1) A check must be made in each zfs thread for a mounted file system.
95 * This is done avoiding races using ZFS_ENTER(zfsvfs).
96 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
97 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
98 * can return EIO from the calling function.
100 * (2) iput() should always be the last thing except for zil_commit()
101 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
102 * First, if it's the last reference, the vnode/znode
103 * can be freed, so the zp may point to freed memory. Second, the last
104 * reference will call zfs_zinactive(), which may induce a lot of work --
105 * pushing cached pages (which acquires range locks) and syncing out
106 * cached atime changes. Third, zfs_zinactive() may require a new tx,
107 * which could deadlock the system if you were already holding one.
108 * If you must call iput() within a tx then use zfs_iput_async().
110 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
111 * as they can span dmu_tx_assign() calls.
113 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
114 * dmu_tx_assign(). This is critical because we don't want to block
115 * while holding locks.
117 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
118 * reduces lock contention and CPU usage when we must wait (note that if
119 * throughput is constrained by the storage, nearly every transaction
122 * Note, in particular, that if a lock is sometimes acquired before
123 * the tx assigns, and sometimes after (e.g. z_lock), then failing
124 * to use a non-blocking assign can deadlock the system. The scenario:
126 * Thread A has grabbed a lock before calling dmu_tx_assign().
127 * Thread B is in an already-assigned tx, and blocks for this lock.
128 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
129 * forever, because the previous txg can't quiesce until B's tx commits.
131 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
132 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
133 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT,
134 * to indicate that this operation has already called dmu_tx_wait().
135 * This will ensure that we don't retry forever, waiting a short bit
138 * (5) If the operation succeeded, generate the intent log entry for it
139 * before dropping locks. This ensures that the ordering of events
140 * in the intent log matches the order in which they actually occurred.
141 * During ZIL replay the zfs_log_* functions will update the sequence
142 * number to indicate the zil transaction has replayed.
144 * (6) At the end of each vnode op, the DMU tx must always commit,
145 * regardless of whether there were any errors.
147 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
148 * to ensure that synchronous semantics are provided when necessary.
150 * In general, this is how things should be ordered in each vnode op:
152 * ZFS_ENTER(zfsvfs); // exit if unmounted
154 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
155 * rw_enter(...); // grab any other locks you need
156 * tx = dmu_tx_create(...); // get DMU tx
157 * dmu_tx_hold_*(); // hold each object you might modify
158 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
160 * rw_exit(...); // drop locks
161 * zfs_dirent_unlock(dl); // unlock directory entry
162 * iput(...); // release held vnodes
163 * if (error == ERESTART) {
169 * dmu_tx_abort(tx); // abort DMU tx
170 * ZFS_EXIT(zfsvfs); // finished in zfs
171 * return (error); // really out of space
173 * error = do_real_work(); // do whatever this VOP does
175 * zfs_log_*(...); // on success, make ZIL entry
176 * dmu_tx_commit(tx); // commit DMU tx -- error or not
177 * rw_exit(...); // drop locks
178 * zfs_dirent_unlock(dl); // unlock directory entry
179 * iput(...); // release held vnodes
180 * zil_commit(zilog, foid); // synchronous when necessary
181 * ZFS_EXIT(zfsvfs); // finished in zfs
182 * return (error); // done, report error
186 * Virus scanning is unsupported. It would be possible to add a hook
187 * here to performance the required virus scan. This could be done
188 * entirely in the kernel or potentially as an update to invoke a
192 zfs_vscan(struct inode *ip, cred_t *cr, int async)
199 zfs_open(struct inode *ip, int mode, int flag, cred_t *cr)
201 znode_t *zp = ITOZ(ip);
202 zfsvfs_t *zfsvfs = ITOZSB(ip);
207 /* Honor ZFS_APPENDONLY file attribute */
208 if ((mode & FMODE_WRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
209 ((flag & O_APPEND) == 0)) {
211 return (SET_ERROR(EPERM));
214 /* Virus scan eligible files on open */
215 if (!zfs_has_ctldir(zp) && zfsvfs->z_vscan && S_ISREG(ip->i_mode) &&
216 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
217 if (zfs_vscan(ip, cr, 0) != 0) {
219 return (SET_ERROR(EACCES));
223 /* Keep a count of the synchronous opens in the znode */
225 atomic_inc_32(&zp->z_sync_cnt);
233 zfs_close(struct inode *ip, int flag, cred_t *cr)
235 znode_t *zp = ITOZ(ip);
236 zfsvfs_t *zfsvfs = ITOZSB(ip);
241 /* Decrement the synchronous opens in the znode */
243 atomic_dec_32(&zp->z_sync_cnt);
245 if (!zfs_has_ctldir(zp) && zfsvfs->z_vscan && S_ISREG(ip->i_mode) &&
246 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
247 VERIFY(zfs_vscan(ip, cr, 1) == 0);
253 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
255 * Lseek support for finding holes (cmd == SEEK_HOLE) and
256 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
259 zfs_holey_common(struct inode *ip, int cmd, loff_t *off)
261 znode_t *zp = ITOZ(ip);
262 uint64_t noff = (uint64_t)*off; /* new offset */
267 file_sz = zp->z_size;
268 if (noff >= file_sz) {
269 return (SET_ERROR(ENXIO));
272 if (cmd == SEEK_HOLE)
277 error = dmu_offset_next(ZTOZSB(zp)->z_os, zp->z_id, hole, &noff);
280 return (SET_ERROR(ENXIO));
282 /* file was dirty, so fall back to using generic logic */
283 if (error == EBUSY) {
291 * We could find a hole that begins after the logical end-of-file,
292 * because dmu_offset_next() only works on whole blocks. If the
293 * EOF falls mid-block, then indicate that the "virtual hole"
294 * at the end of the file begins at the logical EOF, rather than
295 * at the end of the last block.
297 if (noff > file_sz) {
309 zfs_holey(struct inode *ip, int cmd, loff_t *off)
311 znode_t *zp = ITOZ(ip);
312 zfsvfs_t *zfsvfs = ITOZSB(ip);
318 error = zfs_holey_common(ip, cmd, off);
323 #endif /* SEEK_HOLE && SEEK_DATA */
327 * When a file is memory mapped, we must keep the IO data synchronized
328 * between the DMU cache and the memory mapped pages. What this means:
330 * On Write: If we find a memory mapped page, we write to *both*
331 * the page and the dmu buffer.
334 update_pages(struct inode *ip, int64_t start, int len,
335 objset_t *os, uint64_t oid)
337 struct address_space *mp = ip->i_mapping;
343 off = start & (PAGE_SIZE-1);
344 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
345 nbytes = MIN(PAGE_SIZE - off, len);
347 pp = find_lock_page(mp, start >> PAGE_SHIFT);
349 if (mapping_writably_mapped(mp))
350 flush_dcache_page(pp);
353 (void) dmu_read(os, oid, start+off, nbytes, pb+off,
357 if (mapping_writably_mapped(mp))
358 flush_dcache_page(pp);
360 mark_page_accessed(pp);
373 * When a file is memory mapped, we must keep the IO data synchronized
374 * between the DMU cache and the memory mapped pages. What this means:
376 * On Read: We "read" preferentially from memory mapped pages,
377 * else we default from the dmu buffer.
379 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
380 * the file is memory mapped.
383 mappedread(struct inode *ip, int nbytes, uio_t *uio)
385 struct address_space *mp = ip->i_mapping;
387 znode_t *zp = ITOZ(ip);
394 start = uio->uio_loffset;
395 off = start & (PAGE_SIZE-1);
396 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
397 bytes = MIN(PAGE_SIZE - off, len);
399 pp = find_lock_page(mp, start >> PAGE_SHIFT);
401 ASSERT(PageUptodate(pp));
404 error = uiomove(pb + off, bytes, UIO_READ, uio);
407 if (mapping_writably_mapped(mp))
408 flush_dcache_page(pp);
410 mark_page_accessed(pp);
414 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
427 unsigned long zfs_read_chunk_size = 1024 * 1024; /* Tunable */
428 unsigned long zfs_delete_blocks = DMU_MAX_DELETEBLKCNT;
431 * Read bytes from specified file into supplied buffer.
433 * IN: ip - inode of file to be read from.
434 * uio - structure supplying read location, range info,
436 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
437 * O_DIRECT flag; used to bypass page cache.
438 * cr - credentials of caller.
440 * OUT: uio - updated offset and range, buffer filled.
442 * RETURN: 0 on success, error code on failure.
445 * inode - atime updated if byte count > 0
449 zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
451 znode_t *zp = ITOZ(ip);
452 zfsvfs_t *zfsvfs = ITOZSB(ip);
456 #ifdef HAVE_UIO_ZEROCOPY
458 #endif /* HAVE_UIO_ZEROCOPY */
463 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
465 return (SET_ERROR(EACCES));
469 * Validate file offset
471 if (uio->uio_loffset < (offset_t)0) {
473 return (SET_ERROR(EINVAL));
477 * Fasttrack empty reads
479 if (uio->uio_resid == 0) {
485 * If we're in FRSYNC mode, sync out this znode before reading it.
486 * Only do this for non-snapshots.
489 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
490 zil_commit(zfsvfs->z_log, zp->z_id);
493 * Lock the range against changes.
495 rl = zfs_range_lock(&zp->z_range_lock, uio->uio_loffset, uio->uio_resid,
499 * If we are reading past end-of-file we can skip
500 * to the end; but we might still need to set atime.
502 if (uio->uio_loffset >= zp->z_size) {
507 ASSERT(uio->uio_loffset < zp->z_size);
508 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
510 #ifdef HAVE_UIO_ZEROCOPY
511 if ((uio->uio_extflg == UIO_XUIO) &&
512 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
514 int blksz = zp->z_blksz;
515 uint64_t offset = uio->uio_loffset;
517 xuio = (xuio_t *)uio;
519 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
522 ASSERT(offset + n <= blksz);
525 (void) dmu_xuio_init(xuio, nblk);
527 if (vn_has_cached_data(ip)) {
529 * For simplicity, we always allocate a full buffer
530 * even if we only expect to read a portion of a block.
532 while (--nblk >= 0) {
533 (void) dmu_xuio_add(xuio,
534 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
539 #endif /* HAVE_UIO_ZEROCOPY */
542 nbytes = MIN(n, zfs_read_chunk_size -
543 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
545 if (zp->z_is_mapped && !(ioflag & O_DIRECT)) {
546 error = mappedread(ip, nbytes, uio);
548 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
553 /* convert checksum errors into IO errors */
555 error = SET_ERROR(EIO);
562 zfs_range_unlock(rl);
569 * Write the bytes to a file.
571 * IN: ip - inode of file to be written to.
572 * uio - structure supplying write location, range info,
574 * ioflag - FAPPEND flag set if in append mode.
575 * O_DIRECT flag; used to bypass page cache.
576 * cr - credentials of caller.
578 * OUT: uio - updated offset and range.
580 * RETURN: 0 if success
581 * error code if failure
584 * ip - ctime|mtime updated if byte count > 0
589 zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
591 znode_t *zp = ITOZ(ip);
592 rlim64_t limit = uio->uio_limit;
593 ssize_t start_resid = uio->uio_resid;
597 zfsvfs_t *zfsvfs = ZTOZSB(zp);
602 int max_blksz = zfsvfs->z_max_blksz;
605 const iovec_t *aiov = NULL;
609 sa_bulk_attr_t bulk[4];
610 uint64_t mtime[2], ctime[2];
612 #ifdef HAVE_UIO_ZEROCOPY
614 const iovec_t *iovp = uio->uio_iov;
615 ASSERTV(int iovcnt = uio->uio_iovcnt);
619 * Fasttrack empty write
625 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
631 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
632 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
633 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
635 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
639 * Callers might not be able to detect properly that we are read-only,
640 * so check it explicitly here.
642 if (zfs_is_readonly(zfsvfs)) {
644 return (SET_ERROR(EROFS));
648 * If immutable or not appending then return EPERM
650 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
651 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
652 (uio->uio_loffset < zp->z_size))) {
654 return (SET_ERROR(EPERM));
657 zilog = zfsvfs->z_log;
660 * Validate file offset
662 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
665 return (SET_ERROR(EINVAL));
669 * Pre-fault the pages to ensure slow (eg NFS) pages
671 * Skip this if uio contains loaned arc_buf.
673 #ifdef HAVE_UIO_ZEROCOPY
674 if ((uio->uio_extflg == UIO_XUIO) &&
675 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
676 xuio = (xuio_t *)uio;
679 uio_prefaultpages(MIN(n, max_blksz), uio);
682 * If in append mode, set the io offset pointer to eof.
684 if (ioflag & FAPPEND) {
686 * Obtain an appending range lock to guarantee file append
687 * semantics. We reset the write offset once we have the lock.
689 rl = zfs_range_lock(&zp->z_range_lock, 0, n, RL_APPEND);
691 if (rl->r_len == UINT64_MAX) {
693 * We overlocked the file because this write will cause
694 * the file block size to increase.
695 * Note that zp_size cannot change with this lock held.
699 uio->uio_loffset = woff;
702 * Note that if the file block size will change as a result of
703 * this write, then this range lock will lock the entire file
704 * so that we can re-write the block safely.
706 rl = zfs_range_lock(&zp->z_range_lock, woff, n, RL_WRITER);
710 zfs_range_unlock(rl);
712 return (SET_ERROR(EFBIG));
715 if ((woff + n) > limit || woff > (limit - n))
718 /* Will this write extend the file length? */
719 write_eof = (woff + n > zp->z_size);
721 end_size = MAX(zp->z_size, woff + n);
724 * Write the file in reasonable size chunks. Each chunk is written
725 * in a separate transaction; this keeps the intent log records small
726 * and allows us to do more fine-grained space accounting.
730 woff = uio->uio_loffset;
731 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
732 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
734 dmu_return_arcbuf(abuf);
735 error = SET_ERROR(EDQUOT);
739 if (xuio && abuf == NULL) {
740 #ifdef HAVE_UIO_ZEROCOPY
741 ASSERT(i_iov < iovcnt);
742 ASSERT3U(uio->uio_segflg, !=, UIO_BVEC);
744 abuf = dmu_xuio_arcbuf(xuio, i_iov);
745 dmu_xuio_clear(xuio, i_iov);
746 ASSERT((aiov->iov_base == abuf->b_data) ||
747 ((char *)aiov->iov_base - (char *)abuf->b_data +
748 aiov->iov_len == arc_buf_size(abuf)));
751 } else if (abuf == NULL && n >= max_blksz &&
752 woff >= zp->z_size &&
753 P2PHASE(woff, max_blksz) == 0 &&
754 zp->z_blksz == max_blksz) {
756 * This write covers a full block. "Borrow" a buffer
757 * from the dmu so that we can fill it before we enter
758 * a transaction. This avoids the possibility of
759 * holding up the transaction if the data copy hangs
760 * up on a pagefault (e.g., from an NFS server mapping).
764 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
766 ASSERT(abuf != NULL);
767 ASSERT(arc_buf_size(abuf) == max_blksz);
768 if ((error = uiocopy(abuf->b_data, max_blksz,
769 UIO_WRITE, uio, &cbytes))) {
770 dmu_return_arcbuf(abuf);
773 ASSERT(cbytes == max_blksz);
777 * Start a transaction.
779 tx = dmu_tx_create(zfsvfs->z_os);
780 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
781 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
782 zfs_sa_upgrade_txholds(tx, zp);
783 error = dmu_tx_assign(tx, TXG_WAIT);
787 dmu_return_arcbuf(abuf);
792 * If zfs_range_lock() over-locked we grow the blocksize
793 * and then reduce the lock range. This will only happen
794 * on the first iteration since zfs_range_reduce() will
795 * shrink down r_len to the appropriate size.
797 if (rl->r_len == UINT64_MAX) {
800 if (zp->z_blksz > max_blksz) {
802 * File's blocksize is already larger than the
803 * "recordsize" property. Only let it grow to
804 * the next power of 2.
806 ASSERT(!ISP2(zp->z_blksz));
807 new_blksz = MIN(end_size,
808 1 << highbit64(zp->z_blksz));
810 new_blksz = MIN(end_size, max_blksz);
812 zfs_grow_blocksize(zp, new_blksz, tx);
813 zfs_range_reduce(rl, woff, n);
817 * XXX - should we really limit each write to z_max_blksz?
818 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
820 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
823 tx_bytes = uio->uio_resid;
824 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
826 tx_bytes -= uio->uio_resid;
829 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
831 * If this is not a full block write, but we are
832 * extending the file past EOF and this data starts
833 * block-aligned, use assign_arcbuf(). Otherwise,
834 * write via dmu_write().
836 if (tx_bytes < max_blksz && (!write_eof ||
837 aiov->iov_base != abuf->b_data)) {
839 dmu_write(zfsvfs->z_os, zp->z_id, woff,
840 /* cppcheck-suppress nullPointer */
841 aiov->iov_len, aiov->iov_base, tx);
842 dmu_return_arcbuf(abuf);
843 xuio_stat_wbuf_copied();
845 ASSERT(xuio || tx_bytes == max_blksz);
846 dmu_assign_arcbuf_by_dbuf(
847 sa_get_db(zp->z_sa_hdl), woff, abuf, tx);
849 ASSERT(tx_bytes <= uio->uio_resid);
850 uioskip(uio, tx_bytes);
852 if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT)) {
853 update_pages(ip, woff,
854 tx_bytes, zfsvfs->z_os, zp->z_id);
858 * If we made no progress, we're done. If we made even
859 * partial progress, update the znode and ZIL accordingly.
862 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
863 (void *)&zp->z_size, sizeof (uint64_t), tx);
870 * Clear Set-UID/Set-GID bits on successful write if not
871 * privileged and at least one of the execute bits is set.
873 * It would be nice to to this after all writes have
874 * been done, but that would still expose the ISUID/ISGID
875 * to another app after the partial write is committed.
877 * Note: we don't call zfs_fuid_map_id() here because
878 * user 0 is not an ephemeral uid.
880 mutex_enter(&zp->z_acl_lock);
881 uid = KUID_TO_SUID(ip->i_uid);
882 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
883 (S_IXUSR >> 6))) != 0 &&
884 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
885 secpolicy_vnode_setid_retain(cr,
886 ((zp->z_mode & S_ISUID) != 0 && uid == 0)) != 0) {
888 zp->z_mode &= ~(S_ISUID | S_ISGID);
889 ip->i_mode = newmode = zp->z_mode;
890 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
891 (void *)&newmode, sizeof (uint64_t), tx);
893 mutex_exit(&zp->z_acl_lock);
895 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
898 * Update the file size (zp_size) if it has changed;
899 * account for possible concurrent updates.
901 while ((end_size = zp->z_size) < uio->uio_loffset) {
902 (void) atomic_cas_64(&zp->z_size, end_size,
907 * If we are replaying and eof is non zero then force
908 * the file size to the specified eof. Note, there's no
909 * concurrency during replay.
911 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
912 zp->z_size = zfsvfs->z_replay_eof;
914 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
916 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag,
922 ASSERT(tx_bytes == nbytes);
926 uio_prefaultpages(MIN(n, max_blksz), uio);
929 zfs_inode_update(zp);
930 zfs_range_unlock(rl);
933 * If we're in replay mode, or we made no progress, return error.
934 * Otherwise, it's at least a partial write, so it's successful.
936 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
941 if (ioflag & (FSYNC | FDSYNC) ||
942 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
943 zil_commit(zilog, zp->z_id);
950 * Drop a reference on the passed inode asynchronously. This ensures
951 * that the caller will never drop the last reference on an inode in
952 * the current context. Doing so while holding open a tx could result
953 * in a deadlock if iput_final() re-enters the filesystem code.
956 zfs_iput_async(struct inode *ip)
958 objset_t *os = ITOZSB(ip)->z_os;
960 ASSERT(atomic_read(&ip->i_count) > 0);
963 if (atomic_read(&ip->i_count) == 1)
964 VERIFY(taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os)),
965 (task_func_t *)iput, ip, TQ_SLEEP) != TASKQID_INVALID);
971 zfs_get_done(zgd_t *zgd, int error)
973 znode_t *zp = zgd->zgd_private;
976 dmu_buf_rele(zgd->zgd_db, zgd);
978 zfs_range_unlock(zgd->zgd_rl);
981 * Release the vnode asynchronously as we currently have the
982 * txg stopped from syncing.
984 zfs_iput_async(ZTOI(zp));
986 if (error == 0 && zgd->zgd_bp)
987 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
989 kmem_free(zgd, sizeof (zgd_t));
993 static int zil_fault_io = 0;
997 * Get data to generate a TX_WRITE intent log record.
1000 zfs_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
1002 zfsvfs_t *zfsvfs = arg;
1003 objset_t *os = zfsvfs->z_os;
1005 uint64_t object = lr->lr_foid;
1006 uint64_t offset = lr->lr_offset;
1007 uint64_t size = lr->lr_length;
1012 ASSERT3P(lwb, !=, NULL);
1013 ASSERT3P(zio, !=, NULL);
1014 ASSERT3U(size, !=, 0);
1017 * Nothing to do if the file has been removed
1019 if (zfs_zget(zfsvfs, object, &zp) != 0)
1020 return (SET_ERROR(ENOENT));
1021 if (zp->z_unlinked) {
1023 * Release the vnode asynchronously as we currently have the
1024 * txg stopped from syncing.
1026 zfs_iput_async(ZTOI(zp));
1027 return (SET_ERROR(ENOENT));
1030 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1032 zgd->zgd_private = zp;
1035 * Write records come in two flavors: immediate and indirect.
1036 * For small writes it's cheaper to store the data with the
1037 * log record (immediate); for large writes it's cheaper to
1038 * sync the data and get a pointer to it (indirect) so that
1039 * we don't have to write the data twice.
1041 if (buf != NULL) { /* immediate write */
1042 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset, size,
1044 /* test for truncation needs to be done while range locked */
1045 if (offset >= zp->z_size) {
1046 error = SET_ERROR(ENOENT);
1048 error = dmu_read(os, object, offset, size, buf,
1049 DMU_READ_NO_PREFETCH);
1051 ASSERT(error == 0 || error == ENOENT);
1052 } else { /* indirect write */
1054 * Have to lock the whole block to ensure when it's
1055 * written out and its checksum is being calculated
1056 * that no one can change the data. We need to re-check
1057 * blocksize after we get the lock in case it's changed!
1062 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1064 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset,
1066 if (zp->z_blksz == size)
1069 zfs_range_unlock(zgd->zgd_rl);
1071 /* test for truncation needs to be done while range locked */
1072 if (lr->lr_offset >= zp->z_size)
1073 error = SET_ERROR(ENOENT);
1076 error = SET_ERROR(EIO);
1081 error = dmu_buf_hold(os, object, offset, zgd, &db,
1082 DMU_READ_NO_PREFETCH);
1085 blkptr_t *bp = &lr->lr_blkptr;
1090 ASSERT(db->db_offset == offset);
1091 ASSERT(db->db_size == size);
1093 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1095 ASSERT(error || lr->lr_length <= size);
1098 * On success, we need to wait for the write I/O
1099 * initiated by dmu_sync() to complete before we can
1100 * release this dbuf. We will finish everything up
1101 * in the zfs_get_done() callback.
1106 if (error == EALREADY) {
1107 lr->lr_common.lrc_txtype = TX_WRITE2;
1113 zfs_get_done(zgd, error);
1120 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
1122 znode_t *zp = ITOZ(ip);
1123 zfsvfs_t *zfsvfs = ITOZSB(ip);
1129 if (flag & V_ACE_MASK)
1130 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1132 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1139 * Lookup an entry in a directory, or an extended attribute directory.
1140 * If it exists, return a held inode reference for it.
1142 * IN: dip - inode of directory to search.
1143 * nm - name of entry to lookup.
1144 * flags - LOOKUP_XATTR set if looking for an attribute.
1145 * cr - credentials of caller.
1146 * direntflags - directory lookup flags
1147 * realpnp - returned pathname.
1149 * OUT: ipp - inode of located entry, NULL if not found.
1151 * RETURN: 0 on success, error code on failure.
1158 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
1159 cred_t *cr, int *direntflags, pathname_t *realpnp)
1161 znode_t *zdp = ITOZ(dip);
1162 zfsvfs_t *zfsvfs = ITOZSB(dip);
1166 * Fast path lookup, however we must skip DNLC lookup
1167 * for case folding or normalizing lookups because the
1168 * DNLC code only stores the passed in name. This means
1169 * creating 'a' and removing 'A' on a case insensitive
1170 * file system would work, but DNLC still thinks 'a'
1171 * exists and won't let you create it again on the next
1172 * pass through fast path.
1174 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1176 if (!S_ISDIR(dip->i_mode)) {
1177 return (SET_ERROR(ENOTDIR));
1178 } else if (zdp->z_sa_hdl == NULL) {
1179 return (SET_ERROR(EIO));
1182 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1183 error = zfs_fastaccesschk_execute(zdp, cr);
1191 } else if (!zdp->z_zfsvfs->z_norm &&
1192 (zdp->z_zfsvfs->z_case == ZFS_CASE_SENSITIVE)) {
1194 vnode_t *tvp = dnlc_lookup(dvp, nm);
1197 error = zfs_fastaccesschk_execute(zdp, cr);
1202 if (tvp == DNLC_NO_VNODE) {
1204 return (SET_ERROR(ENOENT));
1207 return (specvp_check(vpp, cr));
1210 #endif /* HAVE_DNLC */
1219 if (flags & LOOKUP_XATTR) {
1221 * We don't allow recursive attributes..
1222 * Maybe someday we will.
1224 if (zdp->z_pflags & ZFS_XATTR) {
1226 return (SET_ERROR(EINVAL));
1229 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1235 * Do we have permission to get into attribute directory?
1238 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1248 if (!S_ISDIR(dip->i_mode)) {
1250 return (SET_ERROR(ENOTDIR));
1254 * Check accessibility of directory.
1257 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1262 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1263 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1265 return (SET_ERROR(EILSEQ));
1268 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1269 if ((error == 0) && (*ipp))
1270 zfs_inode_update(ITOZ(*ipp));
1277 * Attempt to create a new entry in a directory. If the entry
1278 * already exists, truncate the file if permissible, else return
1279 * an error. Return the ip of the created or trunc'd file.
1281 * IN: dip - inode of directory to put new file entry in.
1282 * name - name of new file entry.
1283 * vap - attributes of new file.
1284 * excl - flag indicating exclusive or non-exclusive mode.
1285 * mode - mode to open file with.
1286 * cr - credentials of caller.
1287 * flag - large file flag [UNUSED].
1288 * vsecp - ACL to be set
1290 * OUT: ipp - inode of created or trunc'd entry.
1292 * RETURN: 0 on success, error code on failure.
1295 * dip - ctime|mtime updated if new entry created
1296 * ip - ctime|mtime always, atime if new
1301 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1302 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1304 znode_t *zp, *dzp = ITOZ(dip);
1305 zfsvfs_t *zfsvfs = ITOZSB(dip);
1313 zfs_acl_ids_t acl_ids;
1314 boolean_t fuid_dirtied;
1315 boolean_t have_acl = B_FALSE;
1316 boolean_t waited = B_FALSE;
1319 * If we have an ephemeral id, ACL, or XVATTR then
1320 * make sure file system is at proper version
1326 if (zfsvfs->z_use_fuids == B_FALSE &&
1327 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1328 return (SET_ERROR(EINVAL));
1331 return (SET_ERROR(EINVAL));
1336 zilog = zfsvfs->z_log;
1338 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1339 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1341 return (SET_ERROR(EILSEQ));
1344 if (vap->va_mask & ATTR_XVATTR) {
1345 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1346 crgetuid(cr), cr, vap->va_mode)) != 0) {
1354 if (*name == '\0') {
1356 * Null component name refers to the directory itself.
1363 /* possible igrab(zp) */
1366 if (flag & FIGNORECASE)
1369 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1373 zfs_acl_ids_free(&acl_ids);
1374 if (strcmp(name, "..") == 0)
1375 error = SET_ERROR(EISDIR);
1385 * Create a new file object and update the directory
1388 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1390 zfs_acl_ids_free(&acl_ids);
1395 * We only support the creation of regular files in
1396 * extended attribute directories.
1399 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1401 zfs_acl_ids_free(&acl_ids);
1402 error = SET_ERROR(EINVAL);
1406 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1407 cr, vsecp, &acl_ids)) != 0)
1411 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1412 zfs_acl_ids_free(&acl_ids);
1413 error = SET_ERROR(EDQUOT);
1417 tx = dmu_tx_create(os);
1419 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1420 ZFS_SA_BASE_ATTR_SIZE);
1422 fuid_dirtied = zfsvfs->z_fuid_dirty;
1424 zfs_fuid_txhold(zfsvfs, tx);
1425 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1426 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1427 if (!zfsvfs->z_use_sa &&
1428 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1429 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1430 0, acl_ids.z_aclp->z_acl_bytes);
1433 error = dmu_tx_assign(tx,
1434 (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1436 zfs_dirent_unlock(dl);
1437 if (error == ERESTART) {
1443 zfs_acl_ids_free(&acl_ids);
1448 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1450 error = zfs_link_create(dl, zp, tx, ZNEW);
1453 * Since, we failed to add the directory entry for it,
1454 * delete the newly created dnode.
1456 zfs_znode_delete(zp, tx);
1457 remove_inode_hash(ZTOI(zp));
1458 zfs_acl_ids_free(&acl_ids);
1464 zfs_fuid_sync(zfsvfs, tx);
1466 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1467 if (flag & FIGNORECASE)
1469 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1470 vsecp, acl_ids.z_fuidp, vap);
1471 zfs_acl_ids_free(&acl_ids);
1474 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1477 zfs_acl_ids_free(&acl_ids);
1481 * A directory entry already exists for this name.
1484 * Can't truncate an existing file if in exclusive mode.
1487 error = SET_ERROR(EEXIST);
1491 * Can't open a directory for writing.
1493 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1494 error = SET_ERROR(EISDIR);
1498 * Verify requested access to file.
1500 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1504 mutex_enter(&dzp->z_lock);
1506 mutex_exit(&dzp->z_lock);
1509 * Truncate regular files if requested.
1511 if (S_ISREG(ZTOI(zp)->i_mode) &&
1512 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1513 /* we can't hold any locks when calling zfs_freesp() */
1515 zfs_dirent_unlock(dl);
1518 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1524 zfs_dirent_unlock(dl);
1530 zfs_inode_update(dzp);
1531 zfs_inode_update(zp);
1535 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1536 zil_commit(zilog, 0);
1544 zfs_tmpfile(struct inode *dip, vattr_t *vap, int excl,
1545 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1547 znode_t *zp = NULL, *dzp = ITOZ(dip);
1548 zfsvfs_t *zfsvfs = ITOZSB(dip);
1554 zfs_acl_ids_t acl_ids;
1555 boolean_t fuid_dirtied;
1556 boolean_t have_acl = B_FALSE;
1557 boolean_t waited = B_FALSE;
1560 * If we have an ephemeral id, ACL, or XVATTR then
1561 * make sure file system is at proper version
1567 if (zfsvfs->z_use_fuids == B_FALSE &&
1568 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1569 return (SET_ERROR(EINVAL));
1575 if (vap->va_mask & ATTR_XVATTR) {
1576 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1577 crgetuid(cr), cr, vap->va_mode)) != 0) {
1587 * Create a new file object and update the directory
1590 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1592 zfs_acl_ids_free(&acl_ids);
1596 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1597 cr, vsecp, &acl_ids)) != 0)
1601 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1602 zfs_acl_ids_free(&acl_ids);
1603 error = SET_ERROR(EDQUOT);
1607 tx = dmu_tx_create(os);
1609 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1610 ZFS_SA_BASE_ATTR_SIZE);
1611 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1613 fuid_dirtied = zfsvfs->z_fuid_dirty;
1615 zfs_fuid_txhold(zfsvfs, tx);
1616 if (!zfsvfs->z_use_sa &&
1617 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1618 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1619 0, acl_ids.z_aclp->z_acl_bytes);
1621 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1623 if (error == ERESTART) {
1629 zfs_acl_ids_free(&acl_ids);
1634 zfs_mknode(dzp, vap, tx, cr, IS_TMPFILE, &zp, &acl_ids);
1637 zfs_fuid_sync(zfsvfs, tx);
1639 /* Add to unlinked set */
1641 zfs_unlinked_add(zp, tx);
1642 zfs_acl_ids_free(&acl_ids);
1650 zfs_inode_update(dzp);
1651 zfs_inode_update(zp);
1660 * Remove an entry from a directory.
1662 * IN: dip - inode of directory to remove entry from.
1663 * name - name of entry to remove.
1664 * cr - credentials of caller.
1666 * RETURN: 0 if success
1667 * error code if failure
1671 * ip - ctime (if nlink > 0)
1674 uint64_t null_xattr = 0;
1678 zfs_remove(struct inode *dip, char *name, cred_t *cr, int flags)
1680 znode_t *zp, *dzp = ITOZ(dip);
1683 zfsvfs_t *zfsvfs = ITOZSB(dip);
1685 uint64_t acl_obj, xattr_obj;
1686 uint64_t xattr_obj_unlinked = 0;
1691 boolean_t may_delete_now, delete_now = FALSE;
1692 boolean_t unlinked, toobig = FALSE;
1694 pathname_t *realnmp = NULL;
1698 boolean_t waited = B_FALSE;
1701 return (SET_ERROR(EINVAL));
1705 zilog = zfsvfs->z_log;
1707 if (flags & FIGNORECASE) {
1717 * Attempt to lock directory; fail if entry doesn't exist.
1719 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1729 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1734 * Need to use rmdir for removing directories.
1736 if (S_ISDIR(ip->i_mode)) {
1737 error = SET_ERROR(EPERM);
1743 dnlc_remove(dvp, realnmp->pn_buf);
1745 dnlc_remove(dvp, name);
1746 #endif /* HAVE_DNLC */
1748 mutex_enter(&zp->z_lock);
1749 may_delete_now = atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped);
1750 mutex_exit(&zp->z_lock);
1753 * We may delete the znode now, or we may put it in the unlinked set;
1754 * it depends on whether we're the last link, and on whether there are
1755 * other holds on the inode. So we dmu_tx_hold() the right things to
1756 * allow for either case.
1759 tx = dmu_tx_create(zfsvfs->z_os);
1760 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1761 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1762 zfs_sa_upgrade_txholds(tx, zp);
1763 zfs_sa_upgrade_txholds(tx, dzp);
1764 if (may_delete_now) {
1765 toobig = zp->z_size > zp->z_blksz * zfs_delete_blocks;
1766 /* if the file is too big, only hold_free a token amount */
1767 dmu_tx_hold_free(tx, zp->z_id, 0,
1768 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1771 /* are there any extended attributes? */
1772 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1773 &xattr_obj, sizeof (xattr_obj));
1774 if (error == 0 && xattr_obj) {
1775 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1777 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1778 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1781 mutex_enter(&zp->z_lock);
1782 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1783 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1784 mutex_exit(&zp->z_lock);
1786 /* charge as an update -- would be nice not to charge at all */
1787 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1790 * Mark this transaction as typically resulting in a net free of space
1792 dmu_tx_mark_netfree(tx);
1794 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1796 zfs_dirent_unlock(dl);
1797 if (error == ERESTART) {
1817 * Remove the directory entry.
1819 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1828 * Hold z_lock so that we can make sure that the ACL obj
1829 * hasn't changed. Could have been deleted due to
1832 mutex_enter(&zp->z_lock);
1833 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1834 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1835 delete_now = may_delete_now && !toobig &&
1836 atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped) &&
1837 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1842 if (xattr_obj_unlinked) {
1843 ASSERT3U(ZTOI(xzp)->i_nlink, ==, 2);
1844 mutex_enter(&xzp->z_lock);
1845 xzp->z_unlinked = 1;
1846 clear_nlink(ZTOI(xzp));
1848 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1849 &links, sizeof (links), tx);
1850 ASSERT3U(error, ==, 0);
1851 mutex_exit(&xzp->z_lock);
1852 zfs_unlinked_add(xzp, tx);
1855 error = sa_remove(zp->z_sa_hdl,
1856 SA_ZPL_XATTR(zfsvfs), tx);
1858 error = sa_update(zp->z_sa_hdl,
1859 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1860 sizeof (uint64_t), tx);
1864 * Add to the unlinked set because a new reference could be
1865 * taken concurrently resulting in a deferred destruction.
1867 zfs_unlinked_add(zp, tx);
1868 mutex_exit(&zp->z_lock);
1869 } else if (unlinked) {
1870 mutex_exit(&zp->z_lock);
1871 zfs_unlinked_add(zp, tx);
1875 if (flags & FIGNORECASE)
1877 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1884 zfs_dirent_unlock(dl);
1885 zfs_inode_update(dzp);
1886 zfs_inode_update(zp);
1894 zfs_inode_update(xzp);
1895 zfs_iput_async(ZTOI(xzp));
1898 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1899 zil_commit(zilog, 0);
1906 * Create a new directory and insert it into dip using the name
1907 * provided. Return a pointer to the inserted directory.
1909 * IN: dip - inode of directory to add subdir to.
1910 * dirname - name of new directory.
1911 * vap - attributes of new directory.
1912 * cr - credentials of caller.
1913 * vsecp - ACL to be set
1915 * OUT: ipp - inode of created directory.
1917 * RETURN: 0 if success
1918 * error code if failure
1921 * dip - ctime|mtime updated
1922 * ipp - ctime|mtime|atime updated
1926 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1927 cred_t *cr, int flags, vsecattr_t *vsecp)
1929 znode_t *zp, *dzp = ITOZ(dip);
1930 zfsvfs_t *zfsvfs = ITOZSB(dip);
1938 gid_t gid = crgetgid(cr);
1939 zfs_acl_ids_t acl_ids;
1940 boolean_t fuid_dirtied;
1941 boolean_t waited = B_FALSE;
1943 ASSERT(S_ISDIR(vap->va_mode));
1946 * If we have an ephemeral id, ACL, or XVATTR then
1947 * make sure file system is at proper version
1951 if (zfsvfs->z_use_fuids == B_FALSE &&
1952 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1953 return (SET_ERROR(EINVAL));
1955 if (dirname == NULL)
1956 return (SET_ERROR(EINVAL));
1960 zilog = zfsvfs->z_log;
1962 if (dzp->z_pflags & ZFS_XATTR) {
1964 return (SET_ERROR(EINVAL));
1967 if (zfsvfs->z_utf8 && u8_validate(dirname,
1968 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1970 return (SET_ERROR(EILSEQ));
1972 if (flags & FIGNORECASE)
1975 if (vap->va_mask & ATTR_XVATTR) {
1976 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1977 crgetuid(cr), cr, vap->va_mode)) != 0) {
1983 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1984 vsecp, &acl_ids)) != 0) {
1989 * First make sure the new directory doesn't exist.
1991 * Existence is checked first to make sure we don't return
1992 * EACCES instead of EEXIST which can cause some applications
1998 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2000 zfs_acl_ids_free(&acl_ids);
2005 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
2006 zfs_acl_ids_free(&acl_ids);
2007 zfs_dirent_unlock(dl);
2012 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2013 zfs_acl_ids_free(&acl_ids);
2014 zfs_dirent_unlock(dl);
2016 return (SET_ERROR(EDQUOT));
2020 * Add a new entry to the directory.
2022 tx = dmu_tx_create(zfsvfs->z_os);
2023 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2024 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2025 fuid_dirtied = zfsvfs->z_fuid_dirty;
2027 zfs_fuid_txhold(zfsvfs, tx);
2028 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2029 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2030 acl_ids.z_aclp->z_acl_bytes);
2033 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2034 ZFS_SA_BASE_ATTR_SIZE);
2036 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
2038 zfs_dirent_unlock(dl);
2039 if (error == ERESTART) {
2045 zfs_acl_ids_free(&acl_ids);
2054 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2057 * Now put new name in parent dir.
2059 error = zfs_link_create(dl, zp, tx, ZNEW);
2061 zfs_znode_delete(zp, tx);
2062 remove_inode_hash(ZTOI(zp));
2067 zfs_fuid_sync(zfsvfs, tx);
2071 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2072 if (flags & FIGNORECASE)
2074 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2075 acl_ids.z_fuidp, vap);
2078 zfs_acl_ids_free(&acl_ids);
2082 zfs_dirent_unlock(dl);
2084 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2085 zil_commit(zilog, 0);
2090 zfs_inode_update(dzp);
2091 zfs_inode_update(zp);
2098 * Remove a directory subdir entry. If the current working
2099 * directory is the same as the subdir to be removed, the
2102 * IN: dip - inode of directory to remove from.
2103 * name - name of directory to be removed.
2104 * cwd - inode of current working directory.
2105 * cr - credentials of caller.
2106 * flags - case flags
2108 * RETURN: 0 on success, error code on failure.
2111 * dip - ctime|mtime updated
2115 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
2118 znode_t *dzp = ITOZ(dip);
2121 zfsvfs_t *zfsvfs = ITOZSB(dip);
2127 boolean_t waited = B_FALSE;
2130 return (SET_ERROR(EINVAL));
2134 zilog = zfsvfs->z_log;
2136 if (flags & FIGNORECASE)
2142 * Attempt to lock directory; fail if entry doesn't exist.
2144 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2152 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
2156 if (!S_ISDIR(ip->i_mode)) {
2157 error = SET_ERROR(ENOTDIR);
2162 error = SET_ERROR(EINVAL);
2167 * Grab a lock on the directory to make sure that no one is
2168 * trying to add (or lookup) entries while we are removing it.
2170 rw_enter(&zp->z_name_lock, RW_WRITER);
2173 * Grab a lock on the parent pointer to make sure we play well
2174 * with the treewalk and directory rename code.
2176 rw_enter(&zp->z_parent_lock, RW_WRITER);
2178 tx = dmu_tx_create(zfsvfs->z_os);
2179 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2180 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2181 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2182 zfs_sa_upgrade_txholds(tx, zp);
2183 zfs_sa_upgrade_txholds(tx, dzp);
2184 dmu_tx_mark_netfree(tx);
2185 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
2187 rw_exit(&zp->z_parent_lock);
2188 rw_exit(&zp->z_name_lock);
2189 zfs_dirent_unlock(dl);
2190 if (error == ERESTART) {
2203 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2206 uint64_t txtype = TX_RMDIR;
2207 if (flags & FIGNORECASE)
2209 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2214 rw_exit(&zp->z_parent_lock);
2215 rw_exit(&zp->z_name_lock);
2217 zfs_dirent_unlock(dl);
2219 zfs_inode_update(dzp);
2220 zfs_inode_update(zp);
2223 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2224 zil_commit(zilog, 0);
2231 * Read as many directory entries as will fit into the provided
2232 * dirent buffer from the given directory cursor position.
2234 * IN: ip - inode of directory to read.
2235 * dirent - buffer for directory entries.
2237 * OUT: dirent - filler buffer of directory entries.
2239 * RETURN: 0 if success
2240 * error code if failure
2243 * ip - atime updated
2245 * Note that the low 4 bits of the cookie returned by zap is always zero.
2246 * This allows us to use the low range for "special" directory entries:
2247 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2248 * we use the offset 2 for the '.zfs' directory.
2252 zfs_readdir(struct inode *ip, struct dir_context *ctx, cred_t *cr)
2254 znode_t *zp = ITOZ(ip);
2255 zfsvfs_t *zfsvfs = ITOZSB(ip);
2258 zap_attribute_t zap;
2264 uint64_t offset; /* must be unsigned; checks for < 1 */
2269 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2270 &parent, sizeof (parent))) != 0)
2274 * Quit if directory has been removed (posix)
2282 prefetch = zp->z_zn_prefetch;
2285 * Initialize the iterator cursor.
2289 * Start iteration from the beginning of the directory.
2291 zap_cursor_init(&zc, os, zp->z_id);
2294 * The offset is a serialized cursor.
2296 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2300 * Transform to file-system independent format
2305 * Special case `.', `..', and `.zfs'.
2308 (void) strcpy(zap.za_name, ".");
2309 zap.za_normalization_conflict = 0;
2312 } else if (offset == 1) {
2313 (void) strcpy(zap.za_name, "..");
2314 zap.za_normalization_conflict = 0;
2317 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2318 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2319 zap.za_normalization_conflict = 0;
2320 objnum = ZFSCTL_INO_ROOT;
2326 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2327 if (error == ENOENT)
2334 * Allow multiple entries provided the first entry is
2335 * the object id. Non-zpl consumers may safely make
2336 * use of the additional space.
2338 * XXX: This should be a feature flag for compatibility
2340 if (zap.za_integer_length != 8 ||
2341 zap.za_num_integers == 0) {
2342 cmn_err(CE_WARN, "zap_readdir: bad directory "
2343 "entry, obj = %lld, offset = %lld, "
2344 "length = %d, num = %lld\n",
2345 (u_longlong_t)zp->z_id,
2346 (u_longlong_t)offset,
2347 zap.za_integer_length,
2348 (u_longlong_t)zap.za_num_integers);
2349 error = SET_ERROR(ENXIO);
2353 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2354 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2357 done = !dir_emit(ctx, zap.za_name, strlen(zap.za_name),
2362 /* Prefetch znode */
2364 dmu_prefetch(os, objnum, 0, 0, 0,
2365 ZIO_PRIORITY_SYNC_READ);
2369 * Move to the next entry, fill in the previous offset.
2371 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2372 zap_cursor_advance(&zc);
2373 offset = zap_cursor_serialize(&zc);
2379 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2382 zap_cursor_fini(&zc);
2383 if (error == ENOENT)
2391 ulong_t zfs_fsync_sync_cnt = 4;
2394 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2396 znode_t *zp = ITOZ(ip);
2397 zfsvfs_t *zfsvfs = ITOZSB(ip);
2399 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2401 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2404 zil_commit(zfsvfs->z_log, zp->z_id);
2407 tsd_set(zfs_fsyncer_key, NULL);
2414 * Get the requested file attributes and place them in the provided
2417 * IN: ip - inode of file.
2418 * vap - va_mask identifies requested attributes.
2419 * If ATTR_XVATTR set, then optional attrs are requested
2420 * flags - ATTR_NOACLCHECK (CIFS server context)
2421 * cr - credentials of caller.
2423 * OUT: vap - attribute values.
2425 * RETURN: 0 (always succeeds)
2429 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2431 znode_t *zp = ITOZ(ip);
2432 zfsvfs_t *zfsvfs = ITOZSB(ip);
2435 uint64_t atime[2], mtime[2], ctime[2];
2436 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2437 xoptattr_t *xoap = NULL;
2438 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2439 sa_bulk_attr_t bulk[3];
2445 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2447 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
2448 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2449 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2451 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2457 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2458 * Also, if we are the owner don't bother, since owner should
2459 * always be allowed to read basic attributes of file.
2461 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2462 (vap->va_uid != crgetuid(cr))) {
2463 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2471 * Return all attributes. It's cheaper to provide the answer
2472 * than to determine whether we were asked the question.
2475 mutex_enter(&zp->z_lock);
2476 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2477 vap->va_mode = zp->z_mode;
2478 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2479 vap->va_nodeid = zp->z_id;
2480 if ((zp->z_id == zfsvfs->z_root) && zfs_show_ctldir(zp))
2481 links = ZTOI(zp)->i_nlink + 1;
2483 links = ZTOI(zp)->i_nlink;
2484 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2485 vap->va_size = i_size_read(ip);
2486 vap->va_rdev = ip->i_rdev;
2487 vap->va_seq = ip->i_generation;
2490 * Add in any requested optional attributes and the create time.
2491 * Also set the corresponding bits in the returned attribute bitmap.
2493 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2494 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2496 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2497 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2500 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2501 xoap->xoa_readonly =
2502 ((zp->z_pflags & ZFS_READONLY) != 0);
2503 XVA_SET_RTN(xvap, XAT_READONLY);
2506 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2508 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2509 XVA_SET_RTN(xvap, XAT_SYSTEM);
2512 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2514 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2515 XVA_SET_RTN(xvap, XAT_HIDDEN);
2518 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2519 xoap->xoa_nounlink =
2520 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2521 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2524 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2525 xoap->xoa_immutable =
2526 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2527 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2530 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2531 xoap->xoa_appendonly =
2532 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2533 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2536 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2538 ((zp->z_pflags & ZFS_NODUMP) != 0);
2539 XVA_SET_RTN(xvap, XAT_NODUMP);
2542 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2544 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2545 XVA_SET_RTN(xvap, XAT_OPAQUE);
2548 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2549 xoap->xoa_av_quarantined =
2550 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2551 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2554 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2555 xoap->xoa_av_modified =
2556 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2557 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2560 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2561 S_ISREG(ip->i_mode)) {
2562 zfs_sa_get_scanstamp(zp, xvap);
2565 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2568 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2569 times, sizeof (times));
2570 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2571 XVA_SET_RTN(xvap, XAT_CREATETIME);
2574 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2575 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2576 XVA_SET_RTN(xvap, XAT_REPARSE);
2578 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2579 xoap->xoa_generation = ip->i_generation;
2580 XVA_SET_RTN(xvap, XAT_GEN);
2583 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2585 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2586 XVA_SET_RTN(xvap, XAT_OFFLINE);
2589 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2591 ((zp->z_pflags & ZFS_SPARSE) != 0);
2592 XVA_SET_RTN(xvap, XAT_SPARSE);
2596 ZFS_TIME_DECODE(&vap->va_atime, atime);
2597 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2598 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2600 mutex_exit(&zp->z_lock);
2602 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2604 if (zp->z_blksz == 0) {
2606 * Block size hasn't been set; suggest maximal I/O transfers.
2608 vap->va_blksize = zfsvfs->z_max_blksz;
2616 * Get the basic file attributes and place them in the provided kstat
2617 * structure. The inode is assumed to be the authoritative source
2618 * for most of the attributes. However, the znode currently has the
2619 * authoritative atime, blksize, and block count.
2621 * IN: ip - inode of file.
2623 * OUT: sp - kstat values.
2625 * RETURN: 0 (always succeeds)
2629 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2631 znode_t *zp = ITOZ(ip);
2632 zfsvfs_t *zfsvfs = ITOZSB(ip);
2634 u_longlong_t nblocks;
2639 mutex_enter(&zp->z_lock);
2641 generic_fillattr(ip, sp);
2643 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2644 sp->blksize = blksize;
2645 sp->blocks = nblocks;
2647 if (unlikely(zp->z_blksz == 0)) {
2649 * Block size hasn't been set; suggest maximal I/O transfers.
2651 sp->blksize = zfsvfs->z_max_blksz;
2654 mutex_exit(&zp->z_lock);
2657 * Required to prevent NFS client from detecting different inode
2658 * numbers of snapshot root dentry before and after snapshot mount.
2660 if (zfsvfs->z_issnap) {
2661 if (ip->i_sb->s_root->d_inode == ip)
2662 sp->ino = ZFSCTL_INO_SNAPDIRS -
2663 dmu_objset_id(zfsvfs->z_os);
2672 * Set the file attributes to the values contained in the
2675 * IN: ip - inode of file to be modified.
2676 * vap - new attribute values.
2677 * If ATTR_XVATTR set, then optional attrs are being set
2678 * flags - ATTR_UTIME set if non-default time values provided.
2679 * - ATTR_NOACLCHECK (CIFS context only).
2680 * cr - credentials of caller.
2682 * RETURN: 0 if success
2683 * error code if failure
2686 * ip - ctime updated, mtime updated if size changed.
2690 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2692 znode_t *zp = ITOZ(ip);
2693 zfsvfs_t *zfsvfs = ITOZSB(ip);
2697 xvattr_t *tmpxvattr;
2698 uint_t mask = vap->va_mask;
2699 uint_t saved_mask = 0;
2702 uint64_t new_kuid = 0, new_kgid = 0, new_uid, new_gid;
2704 uint64_t mtime[2], ctime[2], atime[2];
2706 int need_policy = FALSE;
2708 zfs_fuid_info_t *fuidp = NULL;
2709 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2712 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2713 boolean_t fuid_dirtied = B_FALSE;
2714 sa_bulk_attr_t *bulk, *xattr_bulk;
2715 int count = 0, xattr_count = 0;
2723 zilog = zfsvfs->z_log;
2726 * Make sure that if we have ephemeral uid/gid or xvattr specified
2727 * that file system is at proper version level
2730 if (zfsvfs->z_use_fuids == B_FALSE &&
2731 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2732 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2733 (mask & ATTR_XVATTR))) {
2735 return (SET_ERROR(EINVAL));
2738 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2740 return (SET_ERROR(EISDIR));
2743 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2745 return (SET_ERROR(EINVAL));
2749 * If this is an xvattr_t, then get a pointer to the structure of
2750 * optional attributes. If this is NULL, then we have a vattr_t.
2752 xoap = xva_getxoptattr(xvap);
2754 tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP);
2755 xva_init(tmpxvattr);
2757 bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2758 xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2761 * Immutable files can only alter immutable bit and atime
2763 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2764 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2765 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2766 err = SET_ERROR(EPERM);
2770 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2771 err = SET_ERROR(EPERM);
2776 * Verify timestamps doesn't overflow 32 bits.
2777 * ZFS can handle large timestamps, but 32bit syscalls can't
2778 * handle times greater than 2039. This check should be removed
2779 * once large timestamps are fully supported.
2781 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2782 if (((mask & ATTR_ATIME) &&
2783 TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2784 ((mask & ATTR_MTIME) &&
2785 TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2786 err = SET_ERROR(EOVERFLOW);
2795 /* Can this be moved to before the top label? */
2796 if (zfs_is_readonly(zfsvfs)) {
2797 err = SET_ERROR(EROFS);
2802 * First validate permissions
2805 if (mask & ATTR_SIZE) {
2806 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2811 * XXX - Note, we are not providing any open
2812 * mode flags here (like FNDELAY), so we may
2813 * block if there are locks present... this
2814 * should be addressed in openat().
2816 /* XXX - would it be OK to generate a log record here? */
2817 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2822 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2823 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2824 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2825 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2826 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2827 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2828 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2829 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2830 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2834 if (mask & (ATTR_UID|ATTR_GID)) {
2835 int idmask = (mask & (ATTR_UID|ATTR_GID));
2840 * NOTE: even if a new mode is being set,
2841 * we may clear S_ISUID/S_ISGID bits.
2844 if (!(mask & ATTR_MODE))
2845 vap->va_mode = zp->z_mode;
2848 * Take ownership or chgrp to group we are a member of
2851 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
2852 take_group = (mask & ATTR_GID) &&
2853 zfs_groupmember(zfsvfs, vap->va_gid, cr);
2856 * If both ATTR_UID and ATTR_GID are set then take_owner and
2857 * take_group must both be set in order to allow taking
2860 * Otherwise, send the check through secpolicy_vnode_setattr()
2864 if (((idmask == (ATTR_UID|ATTR_GID)) &&
2865 take_owner && take_group) ||
2866 ((idmask == ATTR_UID) && take_owner) ||
2867 ((idmask == ATTR_GID) && take_group)) {
2868 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2869 skipaclchk, cr) == 0) {
2871 * Remove setuid/setgid for non-privileged users
2873 (void) secpolicy_setid_clear(vap, cr);
2874 trim_mask = (mask & (ATTR_UID|ATTR_GID));
2883 mutex_enter(&zp->z_lock);
2884 oldva.va_mode = zp->z_mode;
2885 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2886 if (mask & ATTR_XVATTR) {
2888 * Update xvattr mask to include only those attributes
2889 * that are actually changing.
2891 * the bits will be restored prior to actually setting
2892 * the attributes so the caller thinks they were set.
2894 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2895 if (xoap->xoa_appendonly !=
2896 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2899 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2900 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
2904 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2905 if (xoap->xoa_nounlink !=
2906 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2909 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2910 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
2914 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2915 if (xoap->xoa_immutable !=
2916 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2919 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2920 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
2924 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2925 if (xoap->xoa_nodump !=
2926 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2929 XVA_CLR_REQ(xvap, XAT_NODUMP);
2930 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
2934 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2935 if (xoap->xoa_av_modified !=
2936 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2939 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2940 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
2944 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2945 if ((!S_ISREG(ip->i_mode) &&
2946 xoap->xoa_av_quarantined) ||
2947 xoap->xoa_av_quarantined !=
2948 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2951 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2952 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
2956 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2957 mutex_exit(&zp->z_lock);
2958 err = SET_ERROR(EPERM);
2962 if (need_policy == FALSE &&
2963 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2964 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2969 mutex_exit(&zp->z_lock);
2971 if (mask & ATTR_MODE) {
2972 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2973 err = secpolicy_setid_setsticky_clear(ip, vap,
2978 trim_mask |= ATTR_MODE;
2986 * If trim_mask is set then take ownership
2987 * has been granted or write_acl is present and user
2988 * has the ability to modify mode. In that case remove
2989 * UID|GID and or MODE from mask so that
2990 * secpolicy_vnode_setattr() doesn't revoke it.
2994 saved_mask = vap->va_mask;
2995 vap->va_mask &= ~trim_mask;
2997 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
2998 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3003 vap->va_mask |= saved_mask;
3007 * secpolicy_vnode_setattr, or take ownership may have
3010 mask = vap->va_mask;
3012 if ((mask & (ATTR_UID | ATTR_GID))) {
3013 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3014 &xattr_obj, sizeof (xattr_obj));
3016 if (err == 0 && xattr_obj) {
3017 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
3021 if (mask & ATTR_UID) {
3022 new_kuid = zfs_fuid_create(zfsvfs,
3023 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3024 if (new_kuid != KUID_TO_SUID(ZTOI(zp)->i_uid) &&
3025 zfs_fuid_overquota(zfsvfs, B_FALSE, new_kuid)) {
3028 err = SET_ERROR(EDQUOT);
3033 if (mask & ATTR_GID) {
3034 new_kgid = zfs_fuid_create(zfsvfs,
3035 (uint64_t)vap->va_gid, cr, ZFS_GROUP, &fuidp);
3036 if (new_kgid != KGID_TO_SGID(ZTOI(zp)->i_gid) &&
3037 zfs_fuid_overquota(zfsvfs, B_TRUE, new_kgid)) {
3040 err = SET_ERROR(EDQUOT);
3045 tx = dmu_tx_create(zfsvfs->z_os);
3047 if (mask & ATTR_MODE) {
3048 uint64_t pmode = zp->z_mode;
3050 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3052 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
3054 mutex_enter(&zp->z_lock);
3055 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3057 * Are we upgrading ACL from old V0 format
3060 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3061 zfs_znode_acl_version(zp) ==
3062 ZFS_ACL_VERSION_INITIAL) {
3063 dmu_tx_hold_free(tx, acl_obj, 0,
3065 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3066 0, aclp->z_acl_bytes);
3068 dmu_tx_hold_write(tx, acl_obj, 0,
3071 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3072 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3073 0, aclp->z_acl_bytes);
3075 mutex_exit(&zp->z_lock);
3076 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3078 if ((mask & ATTR_XVATTR) &&
3079 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3080 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3082 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3086 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3089 fuid_dirtied = zfsvfs->z_fuid_dirty;
3091 zfs_fuid_txhold(zfsvfs, tx);
3093 zfs_sa_upgrade_txholds(tx, zp);
3095 err = dmu_tx_assign(tx, TXG_WAIT);
3101 * Set each attribute requested.
3102 * We group settings according to the locks they need to acquire.
3104 * Note: you cannot set ctime directly, although it will be
3105 * updated as a side-effect of calling this function.
3109 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3110 mutex_enter(&zp->z_acl_lock);
3111 mutex_enter(&zp->z_lock);
3113 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3114 &zp->z_pflags, sizeof (zp->z_pflags));
3117 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3118 mutex_enter(&attrzp->z_acl_lock);
3119 mutex_enter(&attrzp->z_lock);
3120 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3121 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3122 sizeof (attrzp->z_pflags));
3125 if (mask & (ATTR_UID|ATTR_GID)) {
3127 if (mask & ATTR_UID) {
3128 ZTOI(zp)->i_uid = SUID_TO_KUID(new_kuid);
3129 new_uid = zfs_uid_read(ZTOI(zp));
3130 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3131 &new_uid, sizeof (new_uid));
3133 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3134 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3136 ZTOI(attrzp)->i_uid = SUID_TO_KUID(new_uid);
3140 if (mask & ATTR_GID) {
3141 ZTOI(zp)->i_gid = SGID_TO_KGID(new_kgid);
3142 new_gid = zfs_gid_read(ZTOI(zp));
3143 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3144 NULL, &new_gid, sizeof (new_gid));
3146 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3147 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3149 ZTOI(attrzp)->i_gid = SGID_TO_KGID(new_kgid);
3152 if (!(mask & ATTR_MODE)) {
3153 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3154 NULL, &new_mode, sizeof (new_mode));
3155 new_mode = zp->z_mode;
3157 err = zfs_acl_chown_setattr(zp);
3160 err = zfs_acl_chown_setattr(attrzp);
3165 if (mask & ATTR_MODE) {
3166 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3167 &new_mode, sizeof (new_mode));
3168 zp->z_mode = ZTOI(zp)->i_mode = new_mode;
3169 ASSERT3P(aclp, !=, NULL);
3170 err = zfs_aclset_common(zp, aclp, cr, tx);
3172 if (zp->z_acl_cached)
3173 zfs_acl_free(zp->z_acl_cached);
3174 zp->z_acl_cached = aclp;
3178 if ((mask & ATTR_ATIME) || zp->z_atime_dirty) {
3179 zp->z_atime_dirty = 0;
3180 ZFS_TIME_ENCODE(&ip->i_atime, atime);
3181 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3182 &atime, sizeof (atime));
3185 if (mask & (ATTR_MTIME | ATTR_SIZE)) {
3186 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3187 ZTOI(zp)->i_mtime = timespec_trunc(vap->va_mtime,
3188 ZTOI(zp)->i_sb->s_time_gran);
3190 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3191 mtime, sizeof (mtime));
3194 if (mask & (ATTR_CTIME | ATTR_SIZE)) {
3195 ZFS_TIME_ENCODE(&vap->va_ctime, ctime);
3196 ZTOI(zp)->i_ctime = timespec_trunc(vap->va_ctime,
3197 ZTOI(zp)->i_sb->s_time_gran);
3198 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3199 ctime, sizeof (ctime));
3202 if (attrzp && mask) {
3203 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3204 SA_ZPL_CTIME(zfsvfs), NULL, &ctime,
3209 * Do this after setting timestamps to prevent timestamp
3210 * update from toggling bit
3213 if (xoap && (mask & ATTR_XVATTR)) {
3216 * restore trimmed off masks
3217 * so that return masks can be set for caller.
3220 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
3221 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3223 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
3224 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3226 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
3227 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3229 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
3230 XVA_SET_REQ(xvap, XAT_NODUMP);
3232 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
3233 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3235 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
3236 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3239 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3240 ASSERT(S_ISREG(ip->i_mode));
3242 zfs_xvattr_set(zp, xvap, tx);
3246 zfs_fuid_sync(zfsvfs, tx);
3249 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3251 mutex_exit(&zp->z_lock);
3252 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3253 mutex_exit(&zp->z_acl_lock);
3256 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3257 mutex_exit(&attrzp->z_acl_lock);
3258 mutex_exit(&attrzp->z_lock);
3261 if (err == 0 && attrzp) {
3262 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3271 zfs_fuid_info_free(fuidp);
3279 if (err == ERESTART)
3282 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3286 zfs_inode_update(zp);
3290 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3291 zil_commit(zilog, 0);
3294 kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * 7);
3295 kmem_free(bulk, sizeof (sa_bulk_attr_t) * 7);
3296 kmem_free(tmpxvattr, sizeof (xvattr_t));
3301 typedef struct zfs_zlock {
3302 krwlock_t *zl_rwlock; /* lock we acquired */
3303 znode_t *zl_znode; /* znode we held */
3304 struct zfs_zlock *zl_next; /* next in list */
3308 * Drop locks and release vnodes that were held by zfs_rename_lock().
3311 zfs_rename_unlock(zfs_zlock_t **zlpp)
3315 while ((zl = *zlpp) != NULL) {
3316 if (zl->zl_znode != NULL)
3317 zfs_iput_async(ZTOI(zl->zl_znode));
3318 rw_exit(zl->zl_rwlock);
3319 *zlpp = zl->zl_next;
3320 kmem_free(zl, sizeof (*zl));
3325 * Search back through the directory tree, using the ".." entries.
3326 * Lock each directory in the chain to prevent concurrent renames.
3327 * Fail any attempt to move a directory into one of its own descendants.
3328 * XXX - z_parent_lock can overlap with map or grow locks
3331 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3335 uint64_t rootid = ZTOZSB(zp)->z_root;
3336 uint64_t oidp = zp->z_id;
3337 krwlock_t *rwlp = &szp->z_parent_lock;
3338 krw_t rw = RW_WRITER;
3341 * First pass write-locks szp and compares to zp->z_id.
3342 * Later passes read-lock zp and compare to zp->z_parent.
3345 if (!rw_tryenter(rwlp, rw)) {
3347 * Another thread is renaming in this path.
3348 * Note that if we are a WRITER, we don't have any
3349 * parent_locks held yet.
3351 if (rw == RW_READER && zp->z_id > szp->z_id) {
3353 * Drop our locks and restart
3355 zfs_rename_unlock(&zl);
3359 rwlp = &szp->z_parent_lock;
3364 * Wait for other thread to drop its locks
3370 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3371 zl->zl_rwlock = rwlp;
3372 zl->zl_znode = NULL;
3373 zl->zl_next = *zlpp;
3376 if (oidp == szp->z_id) /* We're a descendant of szp */
3377 return (SET_ERROR(EINVAL));
3379 if (oidp == rootid) /* We've hit the top */
3382 if (rw == RW_READER) { /* i.e. not the first pass */
3383 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3388 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3389 &oidp, sizeof (oidp));
3390 rwlp = &zp->z_parent_lock;
3393 } while (zp->z_id != sdzp->z_id);
3399 * Move an entry from the provided source directory to the target
3400 * directory. Change the entry name as indicated.
3402 * IN: sdip - Source directory containing the "old entry".
3403 * snm - Old entry name.
3404 * tdip - Target directory to contain the "new entry".
3405 * tnm - New entry name.
3406 * cr - credentials of caller.
3407 * flags - case flags
3409 * RETURN: 0 on success, error code on failure.
3412 * sdip,tdip - ctime|mtime updated
3416 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3417 cred_t *cr, int flags)
3419 znode_t *tdzp, *szp, *tzp;
3420 znode_t *sdzp = ITOZ(sdip);
3421 zfsvfs_t *zfsvfs = ITOZSB(sdip);
3423 zfs_dirlock_t *sdl, *tdl;
3426 int cmp, serr, terr;
3429 boolean_t waited = B_FALSE;
3431 if (snm == NULL || tnm == NULL)
3432 return (SET_ERROR(EINVAL));
3435 ZFS_VERIFY_ZP(sdzp);
3436 zilog = zfsvfs->z_log;
3439 ZFS_VERIFY_ZP(tdzp);
3442 * We check i_sb because snapshots and the ctldir must have different
3445 if (tdip->i_sb != sdip->i_sb || zfsctl_is_node(tdip)) {
3447 return (SET_ERROR(EXDEV));
3450 if (zfsvfs->z_utf8 && u8_validate(tnm,
3451 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3453 return (SET_ERROR(EILSEQ));
3456 if (flags & FIGNORECASE)
3465 * This is to prevent the creation of links into attribute space
3466 * by renaming a linked file into/outof an attribute directory.
3467 * See the comment in zfs_link() for why this is considered bad.
3469 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3471 return (SET_ERROR(EINVAL));
3475 * Lock source and target directory entries. To prevent deadlock,
3476 * a lock ordering must be defined. We lock the directory with
3477 * the smallest object id first, or if it's a tie, the one with
3478 * the lexically first name.
3480 if (sdzp->z_id < tdzp->z_id) {
3482 } else if (sdzp->z_id > tdzp->z_id) {
3486 * First compare the two name arguments without
3487 * considering any case folding.
3489 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3491 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3492 ASSERT(error == 0 || !zfsvfs->z_utf8);
3495 * POSIX: "If the old argument and the new argument
3496 * both refer to links to the same existing file,
3497 * the rename() function shall return successfully
3498 * and perform no other action."
3504 * If the file system is case-folding, then we may
3505 * have some more checking to do. A case-folding file
3506 * system is either supporting mixed case sensitivity
3507 * access or is completely case-insensitive. Note
3508 * that the file system is always case preserving.
3510 * In mixed sensitivity mode case sensitive behavior
3511 * is the default. FIGNORECASE must be used to
3512 * explicitly request case insensitive behavior.
3514 * If the source and target names provided differ only
3515 * by case (e.g., a request to rename 'tim' to 'Tim'),
3516 * we will treat this as a special case in the
3517 * case-insensitive mode: as long as the source name
3518 * is an exact match, we will allow this to proceed as
3519 * a name-change request.
3521 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3522 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3523 flags & FIGNORECASE)) &&
3524 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3527 * case preserving rename request, require exact
3536 * If the source and destination directories are the same, we should
3537 * grab the z_name_lock of that directory only once.
3541 rw_enter(&sdzp->z_name_lock, RW_READER);
3545 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3546 ZEXISTS | zflg, NULL, NULL);
3547 terr = zfs_dirent_lock(&tdl,
3548 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3550 terr = zfs_dirent_lock(&tdl,
3551 tdzp, tnm, &tzp, zflg, NULL, NULL);
3552 serr = zfs_dirent_lock(&sdl,
3553 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3559 * Source entry invalid or not there.
3562 zfs_dirent_unlock(tdl);
3568 rw_exit(&sdzp->z_name_lock);
3570 if (strcmp(snm, "..") == 0)
3576 zfs_dirent_unlock(sdl);
3580 rw_exit(&sdzp->z_name_lock);
3582 if (strcmp(tnm, "..") == 0)
3589 * Must have write access at the source to remove the old entry
3590 * and write access at the target to create the new entry.
3591 * Note that if target and source are the same, this can be
3592 * done in a single check.
3595 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3598 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3600 * Check to make sure rename is valid.
3601 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3603 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3608 * Does target exist?
3612 * Source and target must be the same type.
3614 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3615 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3616 error = SET_ERROR(ENOTDIR);
3620 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3621 error = SET_ERROR(EISDIR);
3626 * POSIX dictates that when the source and target
3627 * entries refer to the same file object, rename
3628 * must do nothing and exit without error.
3630 if (szp->z_id == tzp->z_id) {
3636 tx = dmu_tx_create(zfsvfs->z_os);
3637 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3638 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3639 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3640 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3642 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3643 zfs_sa_upgrade_txholds(tx, tdzp);
3646 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3647 zfs_sa_upgrade_txholds(tx, tzp);
3650 zfs_sa_upgrade_txholds(tx, szp);
3651 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3652 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
3655 zfs_rename_unlock(&zl);
3656 zfs_dirent_unlock(sdl);
3657 zfs_dirent_unlock(tdl);
3660 rw_exit(&sdzp->z_name_lock);
3662 if (error == ERESTART) {
3679 if (tzp) /* Attempt to remove the existing target */
3680 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3683 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3685 szp->z_pflags |= ZFS_AV_MODIFIED;
3687 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3688 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3691 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3693 zfs_log_rename(zilog, tx, TX_RENAME |
3694 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3695 sdl->dl_name, tdzp, tdl->dl_name, szp);
3698 * At this point, we have successfully created
3699 * the target name, but have failed to remove
3700 * the source name. Since the create was done
3701 * with the ZRENAMING flag, there are
3702 * complications; for one, the link count is
3703 * wrong. The easiest way to deal with this
3704 * is to remove the newly created target, and
3705 * return the original error. This must
3706 * succeed; fortunately, it is very unlikely to
3707 * fail, since we just created it.
3709 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3710 ZRENAMING, NULL), ==, 0);
3714 * If we had removed the existing target, subsequent
3715 * call to zfs_link_create() to add back the same entry
3716 * but, the new dnode (szp) should not fail.
3718 ASSERT(tzp == NULL);
3725 zfs_rename_unlock(&zl);
3727 zfs_dirent_unlock(sdl);
3728 zfs_dirent_unlock(tdl);
3730 zfs_inode_update(sdzp);
3732 rw_exit(&sdzp->z_name_lock);
3735 zfs_inode_update(tdzp);
3737 zfs_inode_update(szp);
3740 zfs_inode_update(tzp);
3744 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3745 zil_commit(zilog, 0);
3752 * Insert the indicated symbolic reference entry into the directory.
3754 * IN: dip - Directory to contain new symbolic link.
3755 * link - Name for new symlink entry.
3756 * vap - Attributes of new entry.
3757 * target - Target path of new symlink.
3759 * cr - credentials of caller.
3760 * flags - case flags
3762 * RETURN: 0 on success, error code on failure.
3765 * dip - ctime|mtime updated
3769 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
3770 struct inode **ipp, cred_t *cr, int flags)
3772 znode_t *zp, *dzp = ITOZ(dip);
3775 zfsvfs_t *zfsvfs = ITOZSB(dip);
3777 uint64_t len = strlen(link);
3780 zfs_acl_ids_t acl_ids;
3781 boolean_t fuid_dirtied;
3782 uint64_t txtype = TX_SYMLINK;
3783 boolean_t waited = B_FALSE;
3785 ASSERT(S_ISLNK(vap->va_mode));
3788 return (SET_ERROR(EINVAL));
3792 zilog = zfsvfs->z_log;
3794 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3795 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3797 return (SET_ERROR(EILSEQ));
3799 if (flags & FIGNORECASE)
3802 if (len > MAXPATHLEN) {
3804 return (SET_ERROR(ENAMETOOLONG));
3807 if ((error = zfs_acl_ids_create(dzp, 0,
3808 vap, cr, NULL, &acl_ids)) != 0) {
3816 * Attempt to lock directory; fail if entry already exists.
3818 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3820 zfs_acl_ids_free(&acl_ids);
3825 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3826 zfs_acl_ids_free(&acl_ids);
3827 zfs_dirent_unlock(dl);
3832 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3833 zfs_acl_ids_free(&acl_ids);
3834 zfs_dirent_unlock(dl);
3836 return (SET_ERROR(EDQUOT));
3838 tx = dmu_tx_create(zfsvfs->z_os);
3839 fuid_dirtied = zfsvfs->z_fuid_dirty;
3840 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3841 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3842 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3843 ZFS_SA_BASE_ATTR_SIZE + len);
3844 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3845 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3846 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3847 acl_ids.z_aclp->z_acl_bytes);
3850 zfs_fuid_txhold(zfsvfs, tx);
3851 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
3853 zfs_dirent_unlock(dl);
3854 if (error == ERESTART) {
3860 zfs_acl_ids_free(&acl_ids);
3867 * Create a new object for the symlink.
3868 * for version 4 ZPL datsets the symlink will be an SA attribute
3870 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3873 zfs_fuid_sync(zfsvfs, tx);
3875 mutex_enter(&zp->z_lock);
3877 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3880 zfs_sa_symlink(zp, link, len, tx);
3881 mutex_exit(&zp->z_lock);
3884 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3885 &zp->z_size, sizeof (zp->z_size), tx);
3887 * Insert the new object into the directory.
3889 error = zfs_link_create(dl, zp, tx, ZNEW);
3891 zfs_znode_delete(zp, tx);
3892 remove_inode_hash(ZTOI(zp));
3894 if (flags & FIGNORECASE)
3896 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3898 zfs_inode_update(dzp);
3899 zfs_inode_update(zp);
3902 zfs_acl_ids_free(&acl_ids);
3906 zfs_dirent_unlock(dl);
3911 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3912 zil_commit(zilog, 0);
3922 * Return, in the buffer contained in the provided uio structure,
3923 * the symbolic path referred to by ip.
3925 * IN: ip - inode of symbolic link
3926 * uio - structure to contain the link path.
3927 * cr - credentials of caller.
3929 * RETURN: 0 if success
3930 * error code if failure
3933 * ip - atime updated
3937 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
3939 znode_t *zp = ITOZ(ip);
3940 zfsvfs_t *zfsvfs = ITOZSB(ip);
3946 mutex_enter(&zp->z_lock);
3948 error = sa_lookup_uio(zp->z_sa_hdl,
3949 SA_ZPL_SYMLINK(zfsvfs), uio);
3951 error = zfs_sa_readlink(zp, uio);
3952 mutex_exit(&zp->z_lock);
3959 * Insert a new entry into directory tdip referencing sip.
3961 * IN: tdip - Directory to contain new entry.
3962 * sip - inode of new entry.
3963 * name - name of new entry.
3964 * cr - credentials of caller.
3966 * RETURN: 0 if success
3967 * error code if failure
3970 * tdip - ctime|mtime updated
3971 * sip - ctime updated
3975 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr,
3978 znode_t *dzp = ITOZ(tdip);
3980 zfsvfs_t *zfsvfs = ITOZSB(tdip);
3988 boolean_t waited = B_FALSE;
3989 boolean_t is_tmpfile = 0;
3992 is_tmpfile = (sip->i_nlink == 0 && (sip->i_state & I_LINKABLE));
3994 ASSERT(S_ISDIR(tdip->i_mode));
3997 return (SET_ERROR(EINVAL));
4001 zilog = zfsvfs->z_log;
4004 * POSIX dictates that we return EPERM here.
4005 * Better choices include ENOTSUP or EISDIR.
4007 if (S_ISDIR(sip->i_mode)) {
4009 return (SET_ERROR(EPERM));
4016 * We check i_sb because snapshots and the ctldir must have different
4019 if (sip->i_sb != tdip->i_sb || zfsctl_is_node(sip)) {
4021 return (SET_ERROR(EXDEV));
4024 /* Prevent links to .zfs/shares files */
4026 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4027 &parent, sizeof (uint64_t))) != 0) {
4031 if (parent == zfsvfs->z_shares_dir) {
4033 return (SET_ERROR(EPERM));
4036 if (zfsvfs->z_utf8 && u8_validate(name,
4037 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4039 return (SET_ERROR(EILSEQ));
4041 if (flags & FIGNORECASE)
4045 * We do not support links between attributes and non-attributes
4046 * because of the potential security risk of creating links
4047 * into "normal" file space in order to circumvent restrictions
4048 * imposed in attribute space.
4050 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4052 return (SET_ERROR(EINVAL));
4055 owner = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(sip->i_uid),
4057 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4059 return (SET_ERROR(EPERM));
4062 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
4069 * Attempt to lock directory; fail if entry already exists.
4071 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4077 tx = dmu_tx_create(zfsvfs->z_os);
4078 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4079 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4081 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
4083 zfs_sa_upgrade_txholds(tx, szp);
4084 zfs_sa_upgrade_txholds(tx, dzp);
4085 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
4087 zfs_dirent_unlock(dl);
4088 if (error == ERESTART) {
4098 /* unmark z_unlinked so zfs_link_create will not reject */
4100 szp->z_unlinked = 0;
4101 error = zfs_link_create(dl, szp, tx, 0);
4104 uint64_t txtype = TX_LINK;
4106 * tmpfile is created to be in z_unlinkedobj, so remove it.
4107 * Also, we don't log in ZIL, be cause all previous file
4108 * operation on the tmpfile are ignored by ZIL. Instead we
4109 * always wait for txg to sync to make sure all previous
4110 * operation are sync safe.
4113 VERIFY(zap_remove_int(zfsvfs->z_os,
4114 zfsvfs->z_unlinkedobj, szp->z_id, tx) == 0);
4116 if (flags & FIGNORECASE)
4118 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4120 } else if (is_tmpfile) {
4121 /* restore z_unlinked since when linking failed */
4122 szp->z_unlinked = 1;
4124 txg = dmu_tx_get_txg(tx);
4127 zfs_dirent_unlock(dl);
4129 if (!is_tmpfile && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4130 zil_commit(zilog, 0);
4133 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), txg);
4135 zfs_inode_update(dzp);
4136 zfs_inode_update(szp);
4142 zfs_putpage_commit_cb(void *arg)
4144 struct page *pp = arg;
4147 end_page_writeback(pp);
4151 * Push a page out to disk, once the page is on stable storage the
4152 * registered commit callback will be run as notification of completion.
4154 * IN: ip - page mapped for inode.
4155 * pp - page to push (page is locked)
4156 * wbc - writeback control data
4158 * RETURN: 0 if success
4159 * error code if failure
4162 * ip - ctime|mtime updated
4166 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
4168 znode_t *zp = ITOZ(ip);
4169 zfsvfs_t *zfsvfs = ITOZSB(ip);
4177 uint64_t mtime[2], ctime[2];
4178 sa_bulk_attr_t bulk[3];
4180 struct address_space *mapping;
4185 ASSERT(PageLocked(pp));
4187 pgoff = page_offset(pp); /* Page byte-offset in file */
4188 offset = i_size_read(ip); /* File length in bytes */
4189 pglen = MIN(PAGE_SIZE, /* Page length in bytes */
4190 P2ROUNDUP(offset, PAGE_SIZE)-pgoff);
4192 /* Page is beyond end of file */
4193 if (pgoff >= offset) {
4199 /* Truncate page length to end of file */
4200 if (pgoff + pglen > offset)
4201 pglen = offset - pgoff;
4205 * FIXME: Allow mmap writes past its quota. The correct fix
4206 * is to register a page_mkwrite() handler to count the page
4207 * against its quota when it is about to be dirtied.
4209 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4210 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4216 * The ordering here is critical and must adhere to the following
4217 * rules in order to avoid deadlocking in either zfs_read() or
4218 * zfs_free_range() due to a lock inversion.
4220 * 1) The page must be unlocked prior to acquiring the range lock.
4221 * This is critical because zfs_read() calls find_lock_page()
4222 * which may block on the page lock while holding the range lock.
4224 * 2) Before setting or clearing write back on a page the range lock
4225 * must be held in order to prevent a lock inversion with the
4226 * zfs_free_range() function.
4228 * This presents a problem because upon entering this function the
4229 * page lock is already held. To safely acquire the range lock the
4230 * page lock must be dropped. This creates a window where another
4231 * process could truncate, invalidate, dirty, or write out the page.
4233 * Therefore, after successfully reacquiring the range and page locks
4234 * the current page state is checked. In the common case everything
4235 * will be as is expected and it can be written out. However, if
4236 * the page state has changed it must be handled accordingly.
4238 mapping = pp->mapping;
4239 redirty_page_for_writepage(wbc, pp);
4242 rl = zfs_range_lock(&zp->z_range_lock, pgoff, pglen, RL_WRITER);
4245 /* Page mapping changed or it was no longer dirty, we're done */
4246 if (unlikely((mapping != pp->mapping) || !PageDirty(pp))) {
4248 zfs_range_unlock(rl);
4253 /* Another process started write block if required */
4254 if (PageWriteback(pp)) {
4256 zfs_range_unlock(rl);
4258 if (wbc->sync_mode != WB_SYNC_NONE)
4259 wait_on_page_writeback(pp);
4265 /* Clear the dirty flag the required locks are held */
4266 if (!clear_page_dirty_for_io(pp)) {
4268 zfs_range_unlock(rl);
4274 * Counterpart for redirty_page_for_writepage() above. This page
4275 * was in fact not skipped and should not be counted as if it were.
4277 wbc->pages_skipped--;
4278 set_page_writeback(pp);
4281 tx = dmu_tx_create(zfsvfs->z_os);
4282 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
4283 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4284 zfs_sa_upgrade_txholds(tx, zp);
4286 err = dmu_tx_assign(tx, TXG_NOWAIT);
4288 if (err == ERESTART)
4292 __set_page_dirty_nobuffers(pp);
4294 end_page_writeback(pp);
4295 zfs_range_unlock(rl);
4301 ASSERT3U(pglen, <=, PAGE_SIZE);
4302 dmu_write(zfsvfs->z_os, zp->z_id, pgoff, pglen, va, tx);
4305 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
4306 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
4307 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zfsvfs), NULL,
4310 /* Preserve the mtime and ctime provided by the inode */
4311 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4312 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4313 zp->z_atime_dirty = 0;
4316 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4318 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0,
4319 zfs_putpage_commit_cb, pp);
4322 zfs_range_unlock(rl);
4324 if (wbc->sync_mode != WB_SYNC_NONE) {
4326 * Note that this is rarely called under writepages(), because
4327 * writepages() normally handles the entire commit for
4328 * performance reasons.
4330 zil_commit(zfsvfs->z_log, zp->z_id);
4338 * Update the system attributes when the inode has been dirtied. For the
4339 * moment we only update the mode, atime, mtime, and ctime.
4342 zfs_dirty_inode(struct inode *ip, int flags)
4344 znode_t *zp = ITOZ(ip);
4345 zfsvfs_t *zfsvfs = ITOZSB(ip);
4347 uint64_t mode, atime[2], mtime[2], ctime[2];
4348 sa_bulk_attr_t bulk[4];
4352 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
4360 * This is the lazytime semantic indroduced in Linux 4.0
4361 * This flag will only be called from update_time when lazytime is set.
4362 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4363 * Fortunately mtime and ctime are managed within ZFS itself, so we
4364 * only need to dirty atime.
4366 if (flags == I_DIRTY_TIME) {
4367 zp->z_atime_dirty = 1;
4372 tx = dmu_tx_create(zfsvfs->z_os);
4374 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4375 zfs_sa_upgrade_txholds(tx, zp);
4377 error = dmu_tx_assign(tx, TXG_WAIT);
4383 mutex_enter(&zp->z_lock);
4384 zp->z_atime_dirty = 0;
4386 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
4387 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
4388 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
4389 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
4391 /* Preserve the mode, mtime and ctime provided by the inode */
4392 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4393 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4394 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4399 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4400 mutex_exit(&zp->z_lock);
4410 zfs_inactive(struct inode *ip)
4412 znode_t *zp = ITOZ(ip);
4413 zfsvfs_t *zfsvfs = ITOZSB(ip);
4416 int need_unlock = 0;
4418 /* Only read lock if we haven't already write locked, e.g. rollback */
4419 if (!RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)) {
4421 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4423 if (zp->z_sa_hdl == NULL) {
4425 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4429 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4430 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4432 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4433 zfs_sa_upgrade_txholds(tx, zp);
4434 error = dmu_tx_assign(tx, TXG_WAIT);
4438 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4439 mutex_enter(&zp->z_lock);
4440 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4441 (void *)&atime, sizeof (atime), tx);
4442 zp->z_atime_dirty = 0;
4443 mutex_exit(&zp->z_lock);
4450 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4454 * Bounds-check the seek operation.
4456 * IN: ip - inode seeking within
4457 * ooff - old file offset
4458 * noffp - pointer to new file offset
4459 * ct - caller context
4461 * RETURN: 0 if success
4462 * EINVAL if new offset invalid
4466 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
4468 if (S_ISDIR(ip->i_mode))
4470 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4474 * Fill pages with data from the disk.
4477 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
4479 znode_t *zp = ITOZ(ip);
4480 zfsvfs_t *zfsvfs = ITOZSB(ip);
4482 struct page *cur_pp;
4483 u_offset_t io_off, total;
4490 io_len = nr_pages << PAGE_SHIFT;
4491 i_size = i_size_read(ip);
4492 io_off = page_offset(pl[0]);
4494 if (io_off + io_len > i_size)
4495 io_len = i_size - io_off;
4498 * Iterate over list of pages and read each page individually.
4501 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4504 cur_pp = pl[page_idx++];
4506 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4510 /* convert checksum errors into IO errors */
4512 err = SET_ERROR(EIO);
4521 * Uses zfs_fillpage to read data from the file and fill the pages.
4523 * IN: ip - inode of file to get data from.
4524 * pl - list of pages to read
4525 * nr_pages - number of pages to read
4527 * RETURN: 0 on success, error code on failure.
4530 * vp - atime updated
4534 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
4536 znode_t *zp = ITOZ(ip);
4537 zfsvfs_t *zfsvfs = ITOZSB(ip);
4546 err = zfs_fillpage(ip, pl, nr_pages);
4553 * Check ZFS specific permissions to memory map a section of a file.
4555 * IN: ip - inode of the file to mmap
4557 * addrp - start address in memory region
4558 * len - length of memory region
4559 * vm_flags- address flags
4561 * RETURN: 0 if success
4562 * error code if failure
4566 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4567 unsigned long vm_flags)
4569 znode_t *zp = ITOZ(ip);
4570 zfsvfs_t *zfsvfs = ITOZSB(ip);
4575 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4576 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4578 return (SET_ERROR(EPERM));
4581 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4582 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4584 return (SET_ERROR(EACCES));
4587 if (off < 0 || len > MAXOFFSET_T - off) {
4589 return (SET_ERROR(ENXIO));
4597 * convoff - converts the given data (start, whence) to the
4601 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4606 if ((lckdat->l_whence == 2) || (whence == 2)) {
4607 if ((error = zfs_getattr(ip, &vap, 0, CRED())))
4611 switch (lckdat->l_whence) {
4613 lckdat->l_start += offset;
4616 lckdat->l_start += vap.va_size;
4621 return (SET_ERROR(EINVAL));
4624 if (lckdat->l_start < 0)
4625 return (SET_ERROR(EINVAL));
4629 lckdat->l_start -= offset;
4632 lckdat->l_start -= vap.va_size;
4637 return (SET_ERROR(EINVAL));
4640 lckdat->l_whence = (short)whence;
4645 * Free or allocate space in a file. Currently, this function only
4646 * supports the `F_FREESP' command. However, this command is somewhat
4647 * misnamed, as its functionality includes the ability to allocate as
4648 * well as free space.
4650 * IN: ip - inode of file to free data in.
4651 * cmd - action to take (only F_FREESP supported).
4652 * bfp - section of file to free/alloc.
4653 * flag - current file open mode flags.
4654 * offset - current file offset.
4655 * cr - credentials of caller [UNUSED].
4657 * RETURN: 0 on success, error code on failure.
4660 * ip - ctime|mtime updated
4664 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4665 offset_t offset, cred_t *cr)
4667 znode_t *zp = ITOZ(ip);
4668 zfsvfs_t *zfsvfs = ITOZSB(ip);
4675 if (cmd != F_FREESP) {
4677 return (SET_ERROR(EINVAL));
4681 * Callers might not be able to detect properly that we are read-only,
4682 * so check it explicitly here.
4684 if (zfs_is_readonly(zfsvfs)) {
4686 return (SET_ERROR(EROFS));
4689 if ((error = convoff(ip, bfp, 0, offset))) {
4694 if (bfp->l_len < 0) {
4696 return (SET_ERROR(EINVAL));
4700 * Permissions aren't checked on Solaris because on this OS
4701 * zfs_space() can only be called with an opened file handle.
4702 * On Linux we can get here through truncate_range() which
4703 * operates directly on inodes, so we need to check access rights.
4705 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4711 len = bfp->l_len; /* 0 means from off to end of file */
4713 error = zfs_freesp(zp, off, len, flag, TRUE);
4721 zfs_fid(struct inode *ip, fid_t *fidp)
4723 znode_t *zp = ITOZ(ip);
4724 zfsvfs_t *zfsvfs = ITOZSB(ip);
4727 uint64_t object = zp->z_id;
4734 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4735 &gen64, sizeof (uint64_t))) != 0) {
4740 gen = (uint32_t)gen64;
4742 size = SHORT_FID_LEN;
4744 zfid = (zfid_short_t *)fidp;
4746 zfid->zf_len = size;
4748 for (i = 0; i < sizeof (zfid->zf_object); i++)
4749 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4751 /* Must have a non-zero generation number to distinguish from .zfs */
4754 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4755 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4763 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4765 znode_t *zp = ITOZ(ip);
4766 zfsvfs_t *zfsvfs = ITOZSB(ip);
4768 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4772 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4780 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4782 znode_t *zp = ITOZ(ip);
4783 zfsvfs_t *zfsvfs = ITOZSB(ip);
4785 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4786 zilog_t *zilog = zfsvfs->z_log;
4791 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4793 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4794 zil_commit(zilog, 0);
4800 #ifdef HAVE_UIO_ZEROCOPY
4802 * Tunable, both must be a power of 2.
4804 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4805 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4806 * an arcbuf for a partial block read
4808 int zcr_blksz_min = (1 << 10); /* 1K */
4809 int zcr_blksz_max = (1 << 17); /* 128K */
4813 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
4815 znode_t *zp = ITOZ(ip);
4816 zfsvfs_t *zfsvfs = ITOZSB(ip);
4817 int max_blksz = zfsvfs->z_max_blksz;
4818 uio_t *uio = &xuio->xu_uio;
4819 ssize_t size = uio->uio_resid;
4820 offset_t offset = uio->uio_loffset;
4825 int preamble, postamble;
4827 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4828 return (SET_ERROR(EINVAL));
4835 * Loan out an arc_buf for write if write size is bigger than
4836 * max_blksz, and the file's block size is also max_blksz.
4839 if (size < blksz || zp->z_blksz != blksz) {
4841 return (SET_ERROR(EINVAL));
4844 * Caller requests buffers for write before knowing where the
4845 * write offset might be (e.g. NFS TCP write).
4850 preamble = P2PHASE(offset, blksz);
4852 preamble = blksz - preamble;
4857 postamble = P2PHASE(size, blksz);
4860 fullblk = size / blksz;
4861 (void) dmu_xuio_init(xuio,
4862 (preamble != 0) + fullblk + (postamble != 0));
4865 * Have to fix iov base/len for partial buffers. They
4866 * currently represent full arc_buf's.
4869 /* data begins in the middle of the arc_buf */
4870 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4873 (void) dmu_xuio_add(xuio, abuf,
4874 blksz - preamble, preamble);
4877 for (i = 0; i < fullblk; i++) {
4878 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4881 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
4885 /* data ends in the middle of the arc_buf */
4886 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4889 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
4894 * Loan out an arc_buf for read if the read size is larger than
4895 * the current file block size. Block alignment is not
4896 * considered. Partial arc_buf will be loaned out for read.
4898 blksz = zp->z_blksz;
4899 if (blksz < zcr_blksz_min)
4900 blksz = zcr_blksz_min;
4901 if (blksz > zcr_blksz_max)
4902 blksz = zcr_blksz_max;
4903 /* avoid potential complexity of dealing with it */
4904 if (blksz > max_blksz) {
4906 return (SET_ERROR(EINVAL));
4909 maxsize = zp->z_size - uio->uio_loffset;
4915 return (SET_ERROR(EINVAL));
4920 return (SET_ERROR(EINVAL));
4923 uio->uio_extflg = UIO_XUIO;
4924 XUIO_XUZC_RW(xuio) = ioflag;
4931 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
4935 int ioflag = XUIO_XUZC_RW(xuio);
4937 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
4939 i = dmu_xuio_cnt(xuio);
4941 abuf = dmu_xuio_arcbuf(xuio, i);
4943 * if abuf == NULL, it must be a write buffer
4944 * that has been returned in zfs_write().
4947 dmu_return_arcbuf(abuf);
4948 ASSERT(abuf || ioflag == UIO_WRITE);
4951 dmu_xuio_fini(xuio);
4954 #endif /* HAVE_UIO_ZEROCOPY */
4956 #if defined(_KERNEL) && defined(HAVE_SPL)
4957 EXPORT_SYMBOL(zfs_open);
4958 EXPORT_SYMBOL(zfs_close);
4959 EXPORT_SYMBOL(zfs_read);
4960 EXPORT_SYMBOL(zfs_write);
4961 EXPORT_SYMBOL(zfs_access);
4962 EXPORT_SYMBOL(zfs_lookup);
4963 EXPORT_SYMBOL(zfs_create);
4964 EXPORT_SYMBOL(zfs_tmpfile);
4965 EXPORT_SYMBOL(zfs_remove);
4966 EXPORT_SYMBOL(zfs_mkdir);
4967 EXPORT_SYMBOL(zfs_rmdir);
4968 EXPORT_SYMBOL(zfs_readdir);
4969 EXPORT_SYMBOL(zfs_fsync);
4970 EXPORT_SYMBOL(zfs_getattr);
4971 EXPORT_SYMBOL(zfs_getattr_fast);
4972 EXPORT_SYMBOL(zfs_setattr);
4973 EXPORT_SYMBOL(zfs_rename);
4974 EXPORT_SYMBOL(zfs_symlink);
4975 EXPORT_SYMBOL(zfs_readlink);
4976 EXPORT_SYMBOL(zfs_link);
4977 EXPORT_SYMBOL(zfs_inactive);
4978 EXPORT_SYMBOL(zfs_space);
4979 EXPORT_SYMBOL(zfs_fid);
4980 EXPORT_SYMBOL(zfs_getsecattr);
4981 EXPORT_SYMBOL(zfs_setsecattr);
4982 EXPORT_SYMBOL(zfs_getpage);
4983 EXPORT_SYMBOL(zfs_putpage);
4984 EXPORT_SYMBOL(zfs_dirty_inode);
4985 EXPORT_SYMBOL(zfs_map);
4988 module_param(zfs_delete_blocks, ulong, 0644);
4989 MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async");
4990 module_param(zfs_read_chunk_size, long, 0644);
4991 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");