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, 2018 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/sysmacros.h>
41 #include <sys/taskq.h>
43 #include <sys/vmsystm.h>
44 #include <sys/atomic.h>
45 #include <sys/pathname.h>
46 #include <sys/cmn_err.h>
47 #include <sys/errno.h>
48 #include <sys/zfs_dir.h>
49 #include <sys/zfs_acl.h>
50 #include <sys/zfs_ioctl.h>
51 #include <sys/fs/zfs.h>
53 #include <sys/dmu_objset.h>
59 #include <sys/policy.h>
60 #include <sys/sunddi.h>
63 #include <sys/zfs_ctldir.h>
64 #include <sys/zfs_fuid.h>
65 #include <sys/zfs_sa.h>
66 #include <sys/zfs_vnops.h>
67 #include <sys/zfs_rlock.h>
71 #include <sys/sa_impl.h>
76 * Each vnode op performs some logical unit of work. To do this, the ZPL must
77 * properly lock its in-core state, create a DMU transaction, do the work,
78 * record this work in the intent log (ZIL), commit the DMU transaction,
79 * and wait for the intent log to commit if it is a synchronous operation.
80 * Moreover, the vnode ops must work in both normal and log replay context.
81 * The ordering of events is important to avoid deadlocks and references
82 * to freed memory. The example below illustrates the following Big Rules:
84 * (1) A check must be made in each zfs thread for a mounted file system.
85 * This is done avoiding races using ZFS_ENTER(zfsvfs).
86 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
87 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
88 * can return EIO from the calling function.
90 * (2) iput() should always be the last thing except for zil_commit()
91 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
92 * First, if it's the last reference, the vnode/znode
93 * can be freed, so the zp may point to freed memory. Second, the last
94 * reference will call zfs_zinactive(), which may induce a lot of work --
95 * pushing cached pages (which acquires range locks) and syncing out
96 * cached atime changes. Third, zfs_zinactive() may require a new tx,
97 * which could deadlock the system if you were already holding one.
98 * If you must call iput() within a tx then use zfs_iput_async().
100 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
101 * as they can span dmu_tx_assign() calls.
103 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
104 * dmu_tx_assign(). This is critical because we don't want to block
105 * while holding locks.
107 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
108 * reduces lock contention and CPU usage when we must wait (note that if
109 * throughput is constrained by the storage, nearly every transaction
112 * Note, in particular, that if a lock is sometimes acquired before
113 * the tx assigns, and sometimes after (e.g. z_lock), then failing
114 * to use a non-blocking assign can deadlock the system. The scenario:
116 * Thread A has grabbed a lock before calling dmu_tx_assign().
117 * Thread B is in an already-assigned tx, and blocks for this lock.
118 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
119 * forever, because the previous txg can't quiesce until B's tx commits.
121 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
122 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
123 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT,
124 * to indicate that this operation has already called dmu_tx_wait().
125 * This will ensure that we don't retry forever, waiting a short bit
128 * (5) If the operation succeeded, generate the intent log entry for it
129 * before dropping locks. This ensures that the ordering of events
130 * in the intent log matches the order in which they actually occurred.
131 * During ZIL replay the zfs_log_* functions will update the sequence
132 * number to indicate the zil transaction has replayed.
134 * (6) At the end of each vnode op, the DMU tx must always commit,
135 * regardless of whether there were any errors.
137 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
138 * to ensure that synchronous semantics are provided when necessary.
140 * In general, this is how things should be ordered in each vnode op:
142 * ZFS_ENTER(zfsvfs); // exit if unmounted
144 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
145 * rw_enter(...); // grab any other locks you need
146 * tx = dmu_tx_create(...); // get DMU tx
147 * dmu_tx_hold_*(); // hold each object you might modify
148 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
150 * rw_exit(...); // drop locks
151 * zfs_dirent_unlock(dl); // unlock directory entry
152 * iput(...); // release held vnodes
153 * if (error == ERESTART) {
159 * dmu_tx_abort(tx); // abort DMU tx
160 * ZFS_EXIT(zfsvfs); // finished in zfs
161 * return (error); // really out of space
163 * error = do_real_work(); // do whatever this VOP does
165 * zfs_log_*(...); // on success, make ZIL entry
166 * dmu_tx_commit(tx); // commit DMU tx -- error or not
167 * rw_exit(...); // drop locks
168 * zfs_dirent_unlock(dl); // unlock directory entry
169 * iput(...); // release held vnodes
170 * zil_commit(zilog, foid); // synchronous when necessary
171 * ZFS_EXIT(zfsvfs); // finished in zfs
172 * return (error); // done, report error
176 * Virus scanning is unsupported. It would be possible to add a hook
177 * here to performance the required virus scan. This could be done
178 * entirely in the kernel or potentially as an update to invoke a
182 zfs_vscan(struct inode *ip, cred_t *cr, int async)
189 zfs_open(struct inode *ip, int mode, int flag, cred_t *cr)
191 znode_t *zp = ITOZ(ip);
192 zfsvfs_t *zfsvfs = ITOZSB(ip);
197 /* Honor ZFS_APPENDONLY file attribute */
198 if ((mode & FMODE_WRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
199 ((flag & O_APPEND) == 0)) {
201 return (SET_ERROR(EPERM));
204 /* Virus scan eligible files on open */
205 if (!zfs_has_ctldir(zp) && zfsvfs->z_vscan && S_ISREG(ip->i_mode) &&
206 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
207 if (zfs_vscan(ip, cr, 0) != 0) {
209 return (SET_ERROR(EACCES));
213 /* Keep a count of the synchronous opens in the znode */
215 atomic_inc_32(&zp->z_sync_cnt);
223 zfs_close(struct inode *ip, int flag, cred_t *cr)
225 znode_t *zp = ITOZ(ip);
226 zfsvfs_t *zfsvfs = ITOZSB(ip);
231 /* Decrement the synchronous opens in the znode */
233 atomic_dec_32(&zp->z_sync_cnt);
235 if (!zfs_has_ctldir(zp) && zfsvfs->z_vscan && S_ISREG(ip->i_mode) &&
236 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
237 VERIFY(zfs_vscan(ip, cr, 1) == 0);
243 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
245 * Lseek support for finding holes (cmd == SEEK_HOLE) and
246 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
249 zfs_holey_common(struct inode *ip, int cmd, loff_t *off)
251 znode_t *zp = ITOZ(ip);
252 uint64_t noff = (uint64_t)*off; /* new offset */
257 file_sz = zp->z_size;
258 if (noff >= file_sz) {
259 return (SET_ERROR(ENXIO));
262 if (cmd == SEEK_HOLE)
267 error = dmu_offset_next(ZTOZSB(zp)->z_os, zp->z_id, hole, &noff);
270 return (SET_ERROR(ENXIO));
272 /* file was dirty, so fall back to using generic logic */
273 if (error == EBUSY) {
281 * We could find a hole that begins after the logical end-of-file,
282 * because dmu_offset_next() only works on whole blocks. If the
283 * EOF falls mid-block, then indicate that the "virtual hole"
284 * at the end of the file begins at the logical EOF, rather than
285 * at the end of the last block.
287 if (noff > file_sz) {
299 zfs_holey(struct inode *ip, int cmd, loff_t *off)
301 znode_t *zp = ITOZ(ip);
302 zfsvfs_t *zfsvfs = ITOZSB(ip);
308 error = zfs_holey_common(ip, cmd, off);
313 #endif /* SEEK_HOLE && SEEK_DATA */
317 * When a file is memory mapped, we must keep the IO data synchronized
318 * between the DMU cache and the memory mapped pages. What this means:
320 * On Write: If we find a memory mapped page, we write to *both*
321 * the page and the dmu buffer.
324 update_pages(struct inode *ip, int64_t start, int len,
325 objset_t *os, uint64_t oid)
327 struct address_space *mp = ip->i_mapping;
333 off = start & (PAGE_SIZE-1);
334 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
335 nbytes = MIN(PAGE_SIZE - off, len);
337 pp = find_lock_page(mp, start >> PAGE_SHIFT);
339 if (mapping_writably_mapped(mp))
340 flush_dcache_page(pp);
343 (void) dmu_read(os, oid, start+off, nbytes, pb+off,
347 if (mapping_writably_mapped(mp))
348 flush_dcache_page(pp);
350 mark_page_accessed(pp);
363 * When a file is memory mapped, we must keep the IO data synchronized
364 * between the DMU cache and the memory mapped pages. What this means:
366 * On Read: We "read" preferentially from memory mapped pages,
367 * else we default from the dmu buffer.
369 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
370 * the file is memory mapped.
373 mappedread(struct inode *ip, int nbytes, uio_t *uio)
375 struct address_space *mp = ip->i_mapping;
377 znode_t *zp = ITOZ(ip);
384 start = uio->uio_loffset;
385 off = start & (PAGE_SIZE-1);
386 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
387 bytes = MIN(PAGE_SIZE - off, len);
389 pp = find_lock_page(mp, start >> PAGE_SHIFT);
391 ASSERT(PageUptodate(pp));
395 error = uiomove(pb + off, bytes, UIO_READ, uio);
398 if (mapping_writably_mapped(mp))
399 flush_dcache_page(pp);
401 mark_page_accessed(pp);
404 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
417 unsigned long zfs_read_chunk_size = 1024 * 1024; /* Tunable */
418 unsigned long zfs_delete_blocks = DMU_MAX_DELETEBLKCNT;
421 * Read bytes from specified file into supplied buffer.
423 * IN: ip - inode of file to be read from.
424 * uio - structure supplying read location, range info,
426 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
427 * O_DIRECT flag; used to bypass page cache.
428 * cr - credentials of caller.
430 * OUT: uio - updated offset and range, buffer filled.
432 * RETURN: 0 on success, error code on failure.
435 * inode - atime updated if byte count > 0
439 zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
443 znode_t *zp = ITOZ(ip);
444 zfsvfs_t *zfsvfs = ITOZSB(ip);
448 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
450 return (SET_ERROR(EACCES));
454 * Validate file offset
456 if (uio->uio_loffset < (offset_t)0) {
458 return (SET_ERROR(EINVAL));
462 * Fasttrack empty reads
464 if (uio->uio_resid == 0) {
470 * If we're in FRSYNC mode, sync out this znode before reading it.
471 * Only do this for non-snapshots.
474 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
475 zil_commit(zfsvfs->z_log, zp->z_id);
478 * Lock the range against changes.
480 rl_t *rl = zfs_range_lock(&zp->z_range_lock,
481 uio->uio_loffset, uio->uio_resid, RL_READER);
484 * If we are reading past end-of-file we can skip
485 * to the end; but we might still need to set atime.
487 if (uio->uio_loffset >= zp->z_size) {
492 ASSERT(uio->uio_loffset < zp->z_size);
493 ssize_t n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
494 ssize_t start_resid = n;
496 #ifdef HAVE_UIO_ZEROCOPY
498 if ((uio->uio_extflg == UIO_XUIO) &&
499 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
501 int blksz = zp->z_blksz;
502 uint64_t offset = uio->uio_loffset;
504 xuio = (xuio_t *)uio;
506 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
509 ASSERT(offset + n <= blksz);
512 (void) dmu_xuio_init(xuio, nblk);
514 if (vn_has_cached_data(ip)) {
516 * For simplicity, we always allocate a full buffer
517 * even if we only expect to read a portion of a block.
519 while (--nblk >= 0) {
520 (void) dmu_xuio_add(xuio,
521 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
526 #endif /* HAVE_UIO_ZEROCOPY */
529 ssize_t nbytes = MIN(n, zfs_read_chunk_size -
530 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
532 if (zp->z_is_mapped && !(ioflag & O_DIRECT)) {
533 error = mappedread(ip, nbytes, uio);
535 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
540 /* convert checksum errors into IO errors */
542 error = SET_ERROR(EIO);
549 int64_t nread = start_resid - n;
550 dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, nread);
551 task_io_account_read(nread);
553 zfs_range_unlock(rl);
560 * Write the bytes to a file.
562 * IN: ip - inode of file to be written to.
563 * uio - structure supplying write location, range info,
565 * ioflag - FAPPEND flag set if in append mode.
566 * O_DIRECT flag; used to bypass page cache.
567 * cr - credentials of caller.
569 * OUT: uio - updated offset and range.
571 * RETURN: 0 if success
572 * error code if failure
575 * ip - ctime|mtime updated if byte count > 0
580 zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
583 ssize_t start_resid = uio->uio_resid;
586 * Fasttrack empty write
588 ssize_t n = start_resid;
592 rlim64_t limit = uio->uio_limit;
593 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
596 znode_t *zp = ITOZ(ip);
597 zfsvfs_t *zfsvfs = ZTOZSB(zp);
601 sa_bulk_attr_t bulk[4];
603 uint64_t mtime[2], ctime[2];
604 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
605 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
606 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
608 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
612 * Callers might not be able to detect properly that we are read-only,
613 * so check it explicitly here.
615 if (zfs_is_readonly(zfsvfs)) {
617 return (SET_ERROR(EROFS));
621 * If immutable or not appending then return EPERM
623 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
624 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
625 (uio->uio_loffset < zp->z_size))) {
627 return (SET_ERROR(EPERM));
631 * Validate file offset
633 offset_t woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
636 return (SET_ERROR(EINVAL));
639 int max_blksz = zfsvfs->z_max_blksz;
643 * Pre-fault the pages to ensure slow (eg NFS) pages
645 * Skip this if uio contains loaned arc_buf.
647 #ifdef HAVE_UIO_ZEROCOPY
648 if ((uio->uio_extflg == UIO_XUIO) &&
649 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
650 xuio = (xuio_t *)uio;
653 uio_prefaultpages(MIN(n, max_blksz), uio);
658 * If in append mode, set the io offset pointer to eof.
660 if (ioflag & FAPPEND) {
662 * Obtain an appending range lock to guarantee file append
663 * semantics. We reset the write offset once we have the lock.
665 rl = zfs_range_lock(&zp->z_range_lock, 0, n, RL_APPEND);
667 if (rl->r_len == UINT64_MAX) {
669 * We overlocked the file because this write will cause
670 * the file block size to increase.
671 * Note that zp_size cannot change with this lock held.
675 uio->uio_loffset = woff;
678 * Note that if the file block size will change as a result of
679 * this write, then this range lock will lock the entire file
680 * so that we can re-write the block safely.
682 rl = zfs_range_lock(&zp->z_range_lock, woff, n, RL_WRITER);
686 zfs_range_unlock(rl);
688 return (SET_ERROR(EFBIG));
691 if ((woff + n) > limit || woff > (limit - n))
694 /* Will this write extend the file length? */
695 int write_eof = (woff + n > zp->z_size);
697 uint64_t end_size = MAX(zp->z_size, woff + n);
698 zilog_t *zilog = zfsvfs->z_log;
699 #ifdef HAVE_UIO_ZEROCOPY
701 const iovec_t *iovp = uio->uio_iov;
702 ASSERTV(int iovcnt = uio->uio_iovcnt);
707 * Write the file in reasonable size chunks. Each chunk is written
708 * in a separate transaction; this keeps the intent log records small
709 * and allows us to do more fine-grained space accounting.
712 woff = uio->uio_loffset;
714 if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT,
715 KUID_TO_SUID(ip->i_uid)) ||
716 zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT,
717 KGID_TO_SGID(ip->i_gid)) ||
718 (zp->z_projid != ZFS_DEFAULT_PROJID &&
719 zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT,
721 error = SET_ERROR(EDQUOT);
725 arc_buf_t *abuf = NULL;
726 const iovec_t *aiov = NULL;
728 #ifdef HAVE_UIO_ZEROCOPY
729 ASSERT(i_iov < iovcnt);
730 ASSERT3U(uio->uio_segflg, !=, UIO_BVEC);
732 abuf = dmu_xuio_arcbuf(xuio, i_iov);
733 dmu_xuio_clear(xuio, i_iov);
734 ASSERT((aiov->iov_base == abuf->b_data) ||
735 ((char *)aiov->iov_base - (char *)abuf->b_data +
736 aiov->iov_len == arc_buf_size(abuf)));
739 } else if (n >= max_blksz && woff >= zp->z_size &&
740 P2PHASE(woff, max_blksz) == 0 &&
741 zp->z_blksz == max_blksz) {
743 * This write covers a full block. "Borrow" a buffer
744 * from the dmu so that we can fill it before we enter
745 * a transaction. This avoids the possibility of
746 * holding up the transaction if the data copy hangs
747 * up on a pagefault (e.g., from an NFS server mapping).
751 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
753 ASSERT(abuf != NULL);
754 ASSERT(arc_buf_size(abuf) == max_blksz);
755 if ((error = uiocopy(abuf->b_data, max_blksz,
756 UIO_WRITE, uio, &cbytes))) {
757 dmu_return_arcbuf(abuf);
760 ASSERT(cbytes == max_blksz);
764 * Start a transaction.
766 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
767 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
768 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
769 zfs_sa_upgrade_txholds(tx, zp);
770 error = dmu_tx_assign(tx, TXG_WAIT);
774 dmu_return_arcbuf(abuf);
779 * If zfs_range_lock() over-locked we grow the blocksize
780 * and then reduce the lock range. This will only happen
781 * on the first iteration since zfs_range_reduce() will
782 * shrink down r_len to the appropriate size.
784 if (rl->r_len == UINT64_MAX) {
787 if (zp->z_blksz > max_blksz) {
789 * File's blocksize is already larger than the
790 * "recordsize" property. Only let it grow to
791 * the next power of 2.
793 ASSERT(!ISP2(zp->z_blksz));
794 new_blksz = MIN(end_size,
795 1 << highbit64(zp->z_blksz));
797 new_blksz = MIN(end_size, max_blksz);
799 zfs_grow_blocksize(zp, new_blksz, tx);
800 zfs_range_reduce(rl, woff, n);
804 * XXX - should we really limit each write to z_max_blksz?
805 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
807 ssize_t nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
811 tx_bytes = uio->uio_resid;
812 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
814 tx_bytes -= uio->uio_resid;
817 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
819 * If this is not a full block write, but we are
820 * extending the file past EOF and this data starts
821 * block-aligned, use assign_arcbuf(). Otherwise,
822 * write via dmu_write().
824 if (tx_bytes < max_blksz && (!write_eof ||
825 aiov->iov_base != abuf->b_data)) {
827 dmu_write(zfsvfs->z_os, zp->z_id, woff,
828 /* cppcheck-suppress nullPointer */
829 aiov->iov_len, aiov->iov_base, tx);
830 dmu_return_arcbuf(abuf);
831 xuio_stat_wbuf_copied();
833 ASSERT(xuio || tx_bytes == max_blksz);
834 dmu_assign_arcbuf_by_dbuf(
835 sa_get_db(zp->z_sa_hdl), woff, abuf, tx);
837 ASSERT(tx_bytes <= uio->uio_resid);
838 uioskip(uio, tx_bytes);
840 if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT)) {
841 update_pages(ip, woff,
842 tx_bytes, zfsvfs->z_os, zp->z_id);
846 * If we made no progress, we're done. If we made even
847 * partial progress, update the znode and ZIL accordingly.
850 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
851 (void *)&zp->z_size, sizeof (uint64_t), tx);
858 * Clear Set-UID/Set-GID bits on successful write if not
859 * privileged and at least one of the execute bits is set.
861 * It would be nice to to this after all writes have
862 * been done, but that would still expose the ISUID/ISGID
863 * to another app after the partial write is committed.
865 * Note: we don't call zfs_fuid_map_id() here because
866 * user 0 is not an ephemeral uid.
868 mutex_enter(&zp->z_acl_lock);
869 uint32_t uid = KUID_TO_SUID(ip->i_uid);
870 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
871 (S_IXUSR >> 6))) != 0 &&
872 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
873 secpolicy_vnode_setid_retain(cr,
874 ((zp->z_mode & S_ISUID) != 0 && uid == 0)) != 0) {
876 zp->z_mode &= ~(S_ISUID | S_ISGID);
877 ip->i_mode = newmode = zp->z_mode;
878 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
879 (void *)&newmode, sizeof (uint64_t), tx);
881 mutex_exit(&zp->z_acl_lock);
883 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
886 * Update the file size (zp_size) if it has changed;
887 * account for possible concurrent updates.
889 while ((end_size = zp->z_size) < uio->uio_loffset) {
890 (void) atomic_cas_64(&zp->z_size, end_size,
895 * If we are replaying and eof is non zero then force
896 * the file size to the specified eof. Note, there's no
897 * concurrency during replay.
899 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
900 zp->z_size = zfsvfs->z_replay_eof;
902 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
904 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag,
910 ASSERT(tx_bytes == nbytes);
914 uio_prefaultpages(MIN(n, max_blksz), uio);
917 zfs_inode_update(zp);
918 zfs_range_unlock(rl);
921 * If we're in replay mode, or we made no progress, return error.
922 * Otherwise, it's at least a partial write, so it's successful.
924 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
929 if (ioflag & (FSYNC | FDSYNC) ||
930 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
931 zil_commit(zilog, zp->z_id);
933 int64_t nwritten = start_resid - uio->uio_resid;
934 dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, nwritten);
935 task_io_account_write(nwritten);
942 * Drop a reference on the passed inode asynchronously. This ensures
943 * that the caller will never drop the last reference on an inode in
944 * the current context. Doing so while holding open a tx could result
945 * in a deadlock if iput_final() re-enters the filesystem code.
948 zfs_iput_async(struct inode *ip)
950 objset_t *os = ITOZSB(ip)->z_os;
952 ASSERT(atomic_read(&ip->i_count) > 0);
955 if (atomic_read(&ip->i_count) == 1)
956 VERIFY(taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os)),
957 (task_func_t *)iput, ip, TQ_SLEEP) != TASKQID_INVALID);
963 zfs_get_done(zgd_t *zgd, int error)
965 znode_t *zp = zgd->zgd_private;
968 dmu_buf_rele(zgd->zgd_db, zgd);
970 zfs_range_unlock(zgd->zgd_rl);
973 * Release the vnode asynchronously as we currently have the
974 * txg stopped from syncing.
976 zfs_iput_async(ZTOI(zp));
978 if (error == 0 && zgd->zgd_bp)
979 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
981 kmem_free(zgd, sizeof (zgd_t));
985 static int zil_fault_io = 0;
989 * Get data to generate a TX_WRITE intent log record.
992 zfs_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
994 zfsvfs_t *zfsvfs = arg;
995 objset_t *os = zfsvfs->z_os;
997 uint64_t object = lr->lr_foid;
998 uint64_t offset = lr->lr_offset;
999 uint64_t size = lr->lr_length;
1004 ASSERT3P(lwb, !=, NULL);
1005 ASSERT3P(zio, !=, NULL);
1006 ASSERT3U(size, !=, 0);
1009 * Nothing to do if the file has been removed
1011 if (zfs_zget(zfsvfs, object, &zp) != 0)
1012 return (SET_ERROR(ENOENT));
1013 if (zp->z_unlinked) {
1015 * Release the vnode asynchronously as we currently have the
1016 * txg stopped from syncing.
1018 zfs_iput_async(ZTOI(zp));
1019 return (SET_ERROR(ENOENT));
1022 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1024 zgd->zgd_private = zp;
1027 * Write records come in two flavors: immediate and indirect.
1028 * For small writes it's cheaper to store the data with the
1029 * log record (immediate); for large writes it's cheaper to
1030 * sync the data and get a pointer to it (indirect) so that
1031 * we don't have to write the data twice.
1033 if (buf != NULL) { /* immediate write */
1034 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset, size,
1036 /* test for truncation needs to be done while range locked */
1037 if (offset >= zp->z_size) {
1038 error = SET_ERROR(ENOENT);
1040 error = dmu_read(os, object, offset, size, buf,
1041 DMU_READ_NO_PREFETCH);
1043 ASSERT(error == 0 || error == ENOENT);
1044 } else { /* indirect write */
1046 * Have to lock the whole block to ensure when it's
1047 * written out and its checksum is being calculated
1048 * that no one can change the data. We need to re-check
1049 * blocksize after we get the lock in case it's changed!
1054 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1056 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset,
1058 if (zp->z_blksz == size)
1061 zfs_range_unlock(zgd->zgd_rl);
1063 /* test for truncation needs to be done while range locked */
1064 if (lr->lr_offset >= zp->z_size)
1065 error = SET_ERROR(ENOENT);
1068 error = SET_ERROR(EIO);
1073 error = dmu_buf_hold(os, object, offset, zgd, &db,
1074 DMU_READ_NO_PREFETCH);
1077 blkptr_t *bp = &lr->lr_blkptr;
1082 ASSERT(db->db_offset == offset);
1083 ASSERT(db->db_size == size);
1085 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1087 ASSERT(error || lr->lr_length <= size);
1090 * On success, we need to wait for the write I/O
1091 * initiated by dmu_sync() to complete before we can
1092 * release this dbuf. We will finish everything up
1093 * in the zfs_get_done() callback.
1098 if (error == EALREADY) {
1099 lr->lr_common.lrc_txtype = TX_WRITE2;
1101 * TX_WRITE2 relies on the data previously
1102 * written by the TX_WRITE that caused
1103 * EALREADY. We zero out the BP because
1104 * it is the old, currently-on-disk BP,
1105 * so there's no need to zio_flush() its
1106 * vdevs (flushing would needlesly hurt
1107 * performance, and doesn't work on
1117 zfs_get_done(zgd, error);
1124 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
1126 znode_t *zp = ITOZ(ip);
1127 zfsvfs_t *zfsvfs = ITOZSB(ip);
1133 if (flag & V_ACE_MASK)
1134 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1136 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1143 * Lookup an entry in a directory, or an extended attribute directory.
1144 * If it exists, return a held inode reference for it.
1146 * IN: dip - inode of directory to search.
1147 * nm - name of entry to lookup.
1148 * flags - LOOKUP_XATTR set if looking for an attribute.
1149 * cr - credentials of caller.
1150 * direntflags - directory lookup flags
1151 * realpnp - returned pathname.
1153 * OUT: ipp - inode of located entry, NULL if not found.
1155 * RETURN: 0 on success, error code on failure.
1162 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
1163 cred_t *cr, int *direntflags, pathname_t *realpnp)
1165 znode_t *zdp = ITOZ(dip);
1166 zfsvfs_t *zfsvfs = ITOZSB(dip);
1170 * Fast path lookup, however we must skip DNLC lookup
1171 * for case folding or normalizing lookups because the
1172 * DNLC code only stores the passed in name. This means
1173 * creating 'a' and removing 'A' on a case insensitive
1174 * file system would work, but DNLC still thinks 'a'
1175 * exists and won't let you create it again on the next
1176 * pass through fast path.
1178 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1180 if (!S_ISDIR(dip->i_mode)) {
1181 return (SET_ERROR(ENOTDIR));
1182 } else if (zdp->z_sa_hdl == NULL) {
1183 return (SET_ERROR(EIO));
1186 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1187 error = zfs_fastaccesschk_execute(zdp, cr);
1195 } else if (!zdp->z_zfsvfs->z_norm &&
1196 (zdp->z_zfsvfs->z_case == ZFS_CASE_SENSITIVE)) {
1198 vnode_t *tvp = dnlc_lookup(dvp, nm);
1201 error = zfs_fastaccesschk_execute(zdp, cr);
1206 if (tvp == DNLC_NO_VNODE) {
1208 return (SET_ERROR(ENOENT));
1211 return (specvp_check(vpp, cr));
1214 #endif /* HAVE_DNLC */
1223 if (flags & LOOKUP_XATTR) {
1225 * We don't allow recursive attributes..
1226 * Maybe someday we will.
1228 if (zdp->z_pflags & ZFS_XATTR) {
1230 return (SET_ERROR(EINVAL));
1233 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1239 * Do we have permission to get into attribute directory?
1242 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1252 if (!S_ISDIR(dip->i_mode)) {
1254 return (SET_ERROR(ENOTDIR));
1258 * Check accessibility of directory.
1261 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1266 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1267 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1269 return (SET_ERROR(EILSEQ));
1272 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1273 if ((error == 0) && (*ipp))
1274 zfs_inode_update(ITOZ(*ipp));
1281 * Attempt to create a new entry in a directory. If the entry
1282 * already exists, truncate the file if permissible, else return
1283 * an error. Return the ip of the created or trunc'd file.
1285 * IN: dip - inode of directory to put new file entry in.
1286 * name - name of new file entry.
1287 * vap - attributes of new file.
1288 * excl - flag indicating exclusive or non-exclusive mode.
1289 * mode - mode to open file with.
1290 * cr - credentials of caller.
1291 * flag - large file flag [UNUSED].
1292 * vsecp - ACL to be set
1294 * OUT: ipp - inode of created or trunc'd entry.
1296 * RETURN: 0 on success, error code on failure.
1299 * dip - ctime|mtime updated if new entry created
1300 * ip - ctime|mtime always, atime if new
1305 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1306 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1308 znode_t *zp, *dzp = ITOZ(dip);
1309 zfsvfs_t *zfsvfs = ITOZSB(dip);
1317 zfs_acl_ids_t acl_ids;
1318 boolean_t fuid_dirtied;
1319 boolean_t have_acl = B_FALSE;
1320 boolean_t waited = B_FALSE;
1323 * If we have an ephemeral id, ACL, or XVATTR then
1324 * make sure file system is at proper version
1330 if (zfsvfs->z_use_fuids == B_FALSE &&
1331 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1332 return (SET_ERROR(EINVAL));
1335 return (SET_ERROR(EINVAL));
1340 zilog = zfsvfs->z_log;
1342 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1343 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1345 return (SET_ERROR(EILSEQ));
1348 if (vap->va_mask & ATTR_XVATTR) {
1349 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1350 crgetuid(cr), cr, vap->va_mode)) != 0) {
1358 if (*name == '\0') {
1360 * Null component name refers to the directory itself.
1367 /* possible igrab(zp) */
1370 if (flag & FIGNORECASE)
1373 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1377 zfs_acl_ids_free(&acl_ids);
1378 if (strcmp(name, "..") == 0)
1379 error = SET_ERROR(EISDIR);
1387 uint64_t projid = ZFS_DEFAULT_PROJID;
1390 * Create a new file object and update the directory
1393 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1395 zfs_acl_ids_free(&acl_ids);
1400 * We only support the creation of regular files in
1401 * extended attribute directories.
1404 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1406 zfs_acl_ids_free(&acl_ids);
1407 error = SET_ERROR(EINVAL);
1411 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1412 cr, vsecp, &acl_ids)) != 0)
1416 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode))
1417 projid = zfs_inherit_projid(dzp);
1418 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) {
1419 zfs_acl_ids_free(&acl_ids);
1420 error = SET_ERROR(EDQUOT);
1424 tx = dmu_tx_create(os);
1426 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1427 ZFS_SA_BASE_ATTR_SIZE);
1429 fuid_dirtied = zfsvfs->z_fuid_dirty;
1431 zfs_fuid_txhold(zfsvfs, tx);
1432 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1433 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1434 if (!zfsvfs->z_use_sa &&
1435 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1436 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1437 0, acl_ids.z_aclp->z_acl_bytes);
1440 error = dmu_tx_assign(tx,
1441 (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1443 zfs_dirent_unlock(dl);
1444 if (error == ERESTART) {
1450 zfs_acl_ids_free(&acl_ids);
1455 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1457 error = zfs_link_create(dl, zp, tx, ZNEW);
1460 * Since, we failed to add the directory entry for it,
1461 * delete the newly created dnode.
1463 zfs_znode_delete(zp, tx);
1464 remove_inode_hash(ZTOI(zp));
1465 zfs_acl_ids_free(&acl_ids);
1471 zfs_fuid_sync(zfsvfs, tx);
1473 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1474 if (flag & FIGNORECASE)
1476 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1477 vsecp, acl_ids.z_fuidp, vap);
1478 zfs_acl_ids_free(&acl_ids);
1481 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1484 zfs_acl_ids_free(&acl_ids);
1488 * A directory entry already exists for this name.
1491 * Can't truncate an existing file if in exclusive mode.
1494 error = SET_ERROR(EEXIST);
1498 * Can't open a directory for writing.
1500 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1501 error = SET_ERROR(EISDIR);
1505 * Verify requested access to file.
1507 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1511 mutex_enter(&dzp->z_lock);
1513 mutex_exit(&dzp->z_lock);
1516 * Truncate regular files if requested.
1518 if (S_ISREG(ZTOI(zp)->i_mode) &&
1519 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1520 /* we can't hold any locks when calling zfs_freesp() */
1522 zfs_dirent_unlock(dl);
1525 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1531 zfs_dirent_unlock(dl);
1537 zfs_inode_update(dzp);
1538 zfs_inode_update(zp);
1542 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1543 zil_commit(zilog, 0);
1551 zfs_tmpfile(struct inode *dip, vattr_t *vap, int excl,
1552 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1554 znode_t *zp = NULL, *dzp = ITOZ(dip);
1555 zfsvfs_t *zfsvfs = ITOZSB(dip);
1561 zfs_acl_ids_t acl_ids;
1562 uint64_t projid = ZFS_DEFAULT_PROJID;
1563 boolean_t fuid_dirtied;
1564 boolean_t have_acl = B_FALSE;
1565 boolean_t waited = B_FALSE;
1568 * If we have an ephemeral id, ACL, or XVATTR then
1569 * make sure file system is at proper version
1575 if (zfsvfs->z_use_fuids == B_FALSE &&
1576 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1577 return (SET_ERROR(EINVAL));
1583 if (vap->va_mask & ATTR_XVATTR) {
1584 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1585 crgetuid(cr), cr, vap->va_mode)) != 0) {
1595 * Create a new file object and update the directory
1598 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1600 zfs_acl_ids_free(&acl_ids);
1604 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1605 cr, vsecp, &acl_ids)) != 0)
1609 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode))
1610 projid = zfs_inherit_projid(dzp);
1611 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) {
1612 zfs_acl_ids_free(&acl_ids);
1613 error = SET_ERROR(EDQUOT);
1617 tx = dmu_tx_create(os);
1619 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1620 ZFS_SA_BASE_ATTR_SIZE);
1621 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1623 fuid_dirtied = zfsvfs->z_fuid_dirty;
1625 zfs_fuid_txhold(zfsvfs, tx);
1626 if (!zfsvfs->z_use_sa &&
1627 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1628 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1629 0, acl_ids.z_aclp->z_acl_bytes);
1631 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1633 if (error == ERESTART) {
1639 zfs_acl_ids_free(&acl_ids);
1644 zfs_mknode(dzp, vap, tx, cr, IS_TMPFILE, &zp, &acl_ids);
1647 zfs_fuid_sync(zfsvfs, tx);
1649 /* Add to unlinked set */
1651 zfs_unlinked_add(zp, tx);
1652 zfs_acl_ids_free(&acl_ids);
1660 zfs_inode_update(dzp);
1661 zfs_inode_update(zp);
1670 * Remove an entry from a directory.
1672 * IN: dip - inode of directory to remove entry from.
1673 * name - name of entry to remove.
1674 * cr - credentials of caller.
1676 * RETURN: 0 if success
1677 * error code if failure
1681 * ip - ctime (if nlink > 0)
1684 uint64_t null_xattr = 0;
1688 zfs_remove(struct inode *dip, char *name, cred_t *cr, int flags)
1690 znode_t *zp, *dzp = ITOZ(dip);
1693 zfsvfs_t *zfsvfs = ITOZSB(dip);
1695 uint64_t acl_obj, xattr_obj;
1696 uint64_t xattr_obj_unlinked = 0;
1701 boolean_t may_delete_now, delete_now = FALSE;
1702 boolean_t unlinked, toobig = FALSE;
1704 pathname_t *realnmp = NULL;
1708 boolean_t waited = B_FALSE;
1711 return (SET_ERROR(EINVAL));
1715 zilog = zfsvfs->z_log;
1717 if (flags & FIGNORECASE) {
1727 * Attempt to lock directory; fail if entry doesn't exist.
1729 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1739 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1744 * Need to use rmdir for removing directories.
1746 if (S_ISDIR(ip->i_mode)) {
1747 error = SET_ERROR(EPERM);
1753 dnlc_remove(dvp, realnmp->pn_buf);
1755 dnlc_remove(dvp, name);
1756 #endif /* HAVE_DNLC */
1758 mutex_enter(&zp->z_lock);
1759 may_delete_now = atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped);
1760 mutex_exit(&zp->z_lock);
1763 * We may delete the znode now, or we may put it in the unlinked set;
1764 * it depends on whether we're the last link, and on whether there are
1765 * other holds on the inode. So we dmu_tx_hold() the right things to
1766 * allow for either case.
1769 tx = dmu_tx_create(zfsvfs->z_os);
1770 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1771 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1772 zfs_sa_upgrade_txholds(tx, zp);
1773 zfs_sa_upgrade_txholds(tx, dzp);
1774 if (may_delete_now) {
1775 toobig = zp->z_size > zp->z_blksz * zfs_delete_blocks;
1776 /* if the file is too big, only hold_free a token amount */
1777 dmu_tx_hold_free(tx, zp->z_id, 0,
1778 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1781 /* are there any extended attributes? */
1782 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1783 &xattr_obj, sizeof (xattr_obj));
1784 if (error == 0 && xattr_obj) {
1785 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1787 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1788 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1791 mutex_enter(&zp->z_lock);
1792 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1793 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1794 mutex_exit(&zp->z_lock);
1796 /* charge as an update -- would be nice not to charge at all */
1797 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1800 * Mark this transaction as typically resulting in a net free of space
1802 dmu_tx_mark_netfree(tx);
1804 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1806 zfs_dirent_unlock(dl);
1807 if (error == ERESTART) {
1827 * Remove the directory entry.
1829 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1838 * Hold z_lock so that we can make sure that the ACL obj
1839 * hasn't changed. Could have been deleted due to
1842 mutex_enter(&zp->z_lock);
1843 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1844 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1845 delete_now = may_delete_now && !toobig &&
1846 atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped) &&
1847 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1852 if (xattr_obj_unlinked) {
1853 ASSERT3U(ZTOI(xzp)->i_nlink, ==, 2);
1854 mutex_enter(&xzp->z_lock);
1855 xzp->z_unlinked = 1;
1856 clear_nlink(ZTOI(xzp));
1858 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1859 &links, sizeof (links), tx);
1860 ASSERT3U(error, ==, 0);
1861 mutex_exit(&xzp->z_lock);
1862 zfs_unlinked_add(xzp, tx);
1865 error = sa_remove(zp->z_sa_hdl,
1866 SA_ZPL_XATTR(zfsvfs), tx);
1868 error = sa_update(zp->z_sa_hdl,
1869 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1870 sizeof (uint64_t), tx);
1874 * Add to the unlinked set because a new reference could be
1875 * taken concurrently resulting in a deferred destruction.
1877 zfs_unlinked_add(zp, tx);
1878 mutex_exit(&zp->z_lock);
1879 } else if (unlinked) {
1880 mutex_exit(&zp->z_lock);
1881 zfs_unlinked_add(zp, tx);
1885 if (flags & FIGNORECASE)
1887 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1894 zfs_dirent_unlock(dl);
1895 zfs_inode_update(dzp);
1896 zfs_inode_update(zp);
1904 zfs_inode_update(xzp);
1905 zfs_iput_async(ZTOI(xzp));
1908 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1909 zil_commit(zilog, 0);
1916 * Create a new directory and insert it into dip using the name
1917 * provided. Return a pointer to the inserted directory.
1919 * IN: dip - inode of directory to add subdir to.
1920 * dirname - name of new directory.
1921 * vap - attributes of new directory.
1922 * cr - credentials of caller.
1923 * vsecp - ACL to be set
1925 * OUT: ipp - inode of created directory.
1927 * RETURN: 0 if success
1928 * error code if failure
1931 * dip - ctime|mtime updated
1932 * ipp - ctime|mtime|atime updated
1936 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1937 cred_t *cr, int flags, vsecattr_t *vsecp)
1939 znode_t *zp, *dzp = ITOZ(dip);
1940 zfsvfs_t *zfsvfs = ITOZSB(dip);
1948 gid_t gid = crgetgid(cr);
1949 zfs_acl_ids_t acl_ids;
1950 boolean_t fuid_dirtied;
1951 boolean_t waited = B_FALSE;
1953 ASSERT(S_ISDIR(vap->va_mode));
1956 * If we have an ephemeral id, ACL, or XVATTR then
1957 * make sure file system is at proper version
1961 if (zfsvfs->z_use_fuids == B_FALSE &&
1962 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1963 return (SET_ERROR(EINVAL));
1965 if (dirname == NULL)
1966 return (SET_ERROR(EINVAL));
1970 zilog = zfsvfs->z_log;
1972 if (dzp->z_pflags & ZFS_XATTR) {
1974 return (SET_ERROR(EINVAL));
1977 if (zfsvfs->z_utf8 && u8_validate(dirname,
1978 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1980 return (SET_ERROR(EILSEQ));
1982 if (flags & FIGNORECASE)
1985 if (vap->va_mask & ATTR_XVATTR) {
1986 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1987 crgetuid(cr), cr, vap->va_mode)) != 0) {
1993 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1994 vsecp, &acl_ids)) != 0) {
1999 * First make sure the new directory doesn't exist.
2001 * Existence is checked first to make sure we don't return
2002 * EACCES instead of EEXIST which can cause some applications
2008 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2010 zfs_acl_ids_free(&acl_ids);
2015 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
2016 zfs_acl_ids_free(&acl_ids);
2017 zfs_dirent_unlock(dl);
2022 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) {
2023 zfs_acl_ids_free(&acl_ids);
2024 zfs_dirent_unlock(dl);
2026 return (SET_ERROR(EDQUOT));
2030 * Add a new entry to the directory.
2032 tx = dmu_tx_create(zfsvfs->z_os);
2033 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2034 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2035 fuid_dirtied = zfsvfs->z_fuid_dirty;
2037 zfs_fuid_txhold(zfsvfs, tx);
2038 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2039 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2040 acl_ids.z_aclp->z_acl_bytes);
2043 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2044 ZFS_SA_BASE_ATTR_SIZE);
2046 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
2048 zfs_dirent_unlock(dl);
2049 if (error == ERESTART) {
2055 zfs_acl_ids_free(&acl_ids);
2064 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2067 * Now put new name in parent dir.
2069 error = zfs_link_create(dl, zp, tx, ZNEW);
2071 zfs_znode_delete(zp, tx);
2072 remove_inode_hash(ZTOI(zp));
2077 zfs_fuid_sync(zfsvfs, tx);
2081 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2082 if (flags & FIGNORECASE)
2084 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2085 acl_ids.z_fuidp, vap);
2088 zfs_acl_ids_free(&acl_ids);
2092 zfs_dirent_unlock(dl);
2094 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2095 zil_commit(zilog, 0);
2100 zfs_inode_update(dzp);
2101 zfs_inode_update(zp);
2108 * Remove a directory subdir entry. If the current working
2109 * directory is the same as the subdir to be removed, the
2112 * IN: dip - inode of directory to remove from.
2113 * name - name of directory to be removed.
2114 * cwd - inode of current working directory.
2115 * cr - credentials of caller.
2116 * flags - case flags
2118 * RETURN: 0 on success, error code on failure.
2121 * dip - ctime|mtime updated
2125 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
2128 znode_t *dzp = ITOZ(dip);
2131 zfsvfs_t *zfsvfs = ITOZSB(dip);
2137 boolean_t waited = B_FALSE;
2140 return (SET_ERROR(EINVAL));
2144 zilog = zfsvfs->z_log;
2146 if (flags & FIGNORECASE)
2152 * Attempt to lock directory; fail if entry doesn't exist.
2154 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2162 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
2166 if (!S_ISDIR(ip->i_mode)) {
2167 error = SET_ERROR(ENOTDIR);
2172 error = SET_ERROR(EINVAL);
2177 * Grab a lock on the directory to make sure that no one is
2178 * trying to add (or lookup) entries while we are removing it.
2180 rw_enter(&zp->z_name_lock, RW_WRITER);
2183 * Grab a lock on the parent pointer to make sure we play well
2184 * with the treewalk and directory rename code.
2186 rw_enter(&zp->z_parent_lock, RW_WRITER);
2188 tx = dmu_tx_create(zfsvfs->z_os);
2189 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2190 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2191 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2192 zfs_sa_upgrade_txholds(tx, zp);
2193 zfs_sa_upgrade_txholds(tx, dzp);
2194 dmu_tx_mark_netfree(tx);
2195 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
2197 rw_exit(&zp->z_parent_lock);
2198 rw_exit(&zp->z_name_lock);
2199 zfs_dirent_unlock(dl);
2200 if (error == ERESTART) {
2213 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2216 uint64_t txtype = TX_RMDIR;
2217 if (flags & FIGNORECASE)
2219 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2224 rw_exit(&zp->z_parent_lock);
2225 rw_exit(&zp->z_name_lock);
2227 zfs_dirent_unlock(dl);
2229 zfs_inode_update(dzp);
2230 zfs_inode_update(zp);
2233 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2234 zil_commit(zilog, 0);
2241 * Read as many directory entries as will fit into the provided
2242 * dirent buffer from the given directory cursor position.
2244 * IN: ip - inode of directory to read.
2245 * dirent - buffer for directory entries.
2247 * OUT: dirent - filler buffer of directory entries.
2249 * RETURN: 0 if success
2250 * error code if failure
2253 * ip - atime updated
2255 * Note that the low 4 bits of the cookie returned by zap is always zero.
2256 * This allows us to use the low range for "special" directory entries:
2257 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2258 * we use the offset 2 for the '.zfs' directory.
2262 zfs_readdir(struct inode *ip, zpl_dir_context_t *ctx, cred_t *cr)
2264 znode_t *zp = ITOZ(ip);
2265 zfsvfs_t *zfsvfs = ITOZSB(ip);
2268 zap_attribute_t zap;
2274 uint64_t offset; /* must be unsigned; checks for < 1 */
2279 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2280 &parent, sizeof (parent))) != 0)
2284 * Quit if directory has been removed (posix)
2292 prefetch = zp->z_zn_prefetch;
2295 * Initialize the iterator cursor.
2299 * Start iteration from the beginning of the directory.
2301 zap_cursor_init(&zc, os, zp->z_id);
2304 * The offset is a serialized cursor.
2306 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2310 * Transform to file-system independent format
2315 * Special case `.', `..', and `.zfs'.
2318 (void) strcpy(zap.za_name, ".");
2319 zap.za_normalization_conflict = 0;
2322 } else if (offset == 1) {
2323 (void) strcpy(zap.za_name, "..");
2324 zap.za_normalization_conflict = 0;
2327 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2328 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2329 zap.za_normalization_conflict = 0;
2330 objnum = ZFSCTL_INO_ROOT;
2336 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2337 if (error == ENOENT)
2344 * Allow multiple entries provided the first entry is
2345 * the object id. Non-zpl consumers may safely make
2346 * use of the additional space.
2348 * XXX: This should be a feature flag for compatibility
2350 if (zap.za_integer_length != 8 ||
2351 zap.za_num_integers == 0) {
2352 cmn_err(CE_WARN, "zap_readdir: bad directory "
2353 "entry, obj = %lld, offset = %lld, "
2354 "length = %d, num = %lld\n",
2355 (u_longlong_t)zp->z_id,
2356 (u_longlong_t)offset,
2357 zap.za_integer_length,
2358 (u_longlong_t)zap.za_num_integers);
2359 error = SET_ERROR(ENXIO);
2363 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2364 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2367 done = !zpl_dir_emit(ctx, zap.za_name, strlen(zap.za_name),
2372 /* Prefetch znode */
2374 dmu_prefetch(os, objnum, 0, 0, 0,
2375 ZIO_PRIORITY_SYNC_READ);
2379 * Move to the next entry, fill in the previous offset.
2381 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2382 zap_cursor_advance(&zc);
2383 offset = zap_cursor_serialize(&zc);
2389 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2392 zap_cursor_fini(&zc);
2393 if (error == ENOENT)
2401 ulong_t zfs_fsync_sync_cnt = 4;
2404 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2406 znode_t *zp = ITOZ(ip);
2407 zfsvfs_t *zfsvfs = ITOZSB(ip);
2409 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2411 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2414 zil_commit(zfsvfs->z_log, zp->z_id);
2417 tsd_set(zfs_fsyncer_key, NULL);
2424 * Get the requested file attributes and place them in the provided
2427 * IN: ip - inode of file.
2428 * vap - va_mask identifies requested attributes.
2429 * If ATTR_XVATTR set, then optional attrs are requested
2430 * flags - ATTR_NOACLCHECK (CIFS server context)
2431 * cr - credentials of caller.
2433 * OUT: vap - attribute values.
2435 * RETURN: 0 (always succeeds)
2439 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2441 znode_t *zp = ITOZ(ip);
2442 zfsvfs_t *zfsvfs = ITOZSB(ip);
2445 uint64_t atime[2], mtime[2], ctime[2];
2446 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2447 xoptattr_t *xoap = NULL;
2448 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2449 sa_bulk_attr_t bulk[3];
2455 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2457 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
2458 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2459 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2461 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2467 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2468 * Also, if we are the owner don't bother, since owner should
2469 * always be allowed to read basic attributes of file.
2471 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2472 (vap->va_uid != crgetuid(cr))) {
2473 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2481 * Return all attributes. It's cheaper to provide the answer
2482 * than to determine whether we were asked the question.
2485 mutex_enter(&zp->z_lock);
2486 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2487 vap->va_mode = zp->z_mode;
2488 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2489 vap->va_nodeid = zp->z_id;
2490 if ((zp->z_id == zfsvfs->z_root) && zfs_show_ctldir(zp))
2491 links = ZTOI(zp)->i_nlink + 1;
2493 links = ZTOI(zp)->i_nlink;
2494 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2495 vap->va_size = i_size_read(ip);
2496 vap->va_rdev = ip->i_rdev;
2497 vap->va_seq = ip->i_generation;
2500 * Add in any requested optional attributes and the create time.
2501 * Also set the corresponding bits in the returned attribute bitmap.
2503 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2504 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2506 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2507 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2510 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2511 xoap->xoa_readonly =
2512 ((zp->z_pflags & ZFS_READONLY) != 0);
2513 XVA_SET_RTN(xvap, XAT_READONLY);
2516 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2518 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2519 XVA_SET_RTN(xvap, XAT_SYSTEM);
2522 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2524 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2525 XVA_SET_RTN(xvap, XAT_HIDDEN);
2528 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2529 xoap->xoa_nounlink =
2530 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2531 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2534 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2535 xoap->xoa_immutable =
2536 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2537 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2540 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2541 xoap->xoa_appendonly =
2542 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2543 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2546 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2548 ((zp->z_pflags & ZFS_NODUMP) != 0);
2549 XVA_SET_RTN(xvap, XAT_NODUMP);
2552 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2554 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2555 XVA_SET_RTN(xvap, XAT_OPAQUE);
2558 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2559 xoap->xoa_av_quarantined =
2560 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2561 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2564 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2565 xoap->xoa_av_modified =
2566 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2567 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2570 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2571 S_ISREG(ip->i_mode)) {
2572 zfs_sa_get_scanstamp(zp, xvap);
2575 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2578 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2579 times, sizeof (times));
2580 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2581 XVA_SET_RTN(xvap, XAT_CREATETIME);
2584 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2585 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2586 XVA_SET_RTN(xvap, XAT_REPARSE);
2588 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2589 xoap->xoa_generation = ip->i_generation;
2590 XVA_SET_RTN(xvap, XAT_GEN);
2593 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2595 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2596 XVA_SET_RTN(xvap, XAT_OFFLINE);
2599 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2601 ((zp->z_pflags & ZFS_SPARSE) != 0);
2602 XVA_SET_RTN(xvap, XAT_SPARSE);
2605 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
2606 xoap->xoa_projinherit =
2607 ((zp->z_pflags & ZFS_PROJINHERIT) != 0);
2608 XVA_SET_RTN(xvap, XAT_PROJINHERIT);
2611 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
2612 xoap->xoa_projid = zp->z_projid;
2613 XVA_SET_RTN(xvap, XAT_PROJID);
2617 ZFS_TIME_DECODE(&vap->va_atime, atime);
2618 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2619 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2621 mutex_exit(&zp->z_lock);
2623 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2625 if (zp->z_blksz == 0) {
2627 * Block size hasn't been set; suggest maximal I/O transfers.
2629 vap->va_blksize = zfsvfs->z_max_blksz;
2637 * Get the basic file attributes and place them in the provided kstat
2638 * structure. The inode is assumed to be the authoritative source
2639 * for most of the attributes. However, the znode currently has the
2640 * authoritative atime, blksize, and block count.
2642 * IN: ip - inode of file.
2644 * OUT: sp - kstat values.
2646 * RETURN: 0 (always succeeds)
2650 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2652 znode_t *zp = ITOZ(ip);
2653 zfsvfs_t *zfsvfs = ITOZSB(ip);
2655 u_longlong_t nblocks;
2660 mutex_enter(&zp->z_lock);
2662 generic_fillattr(ip, sp);
2664 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2665 sp->blksize = blksize;
2666 sp->blocks = nblocks;
2668 if (unlikely(zp->z_blksz == 0)) {
2670 * Block size hasn't been set; suggest maximal I/O transfers.
2672 sp->blksize = zfsvfs->z_max_blksz;
2675 mutex_exit(&zp->z_lock);
2678 * Required to prevent NFS client from detecting different inode
2679 * numbers of snapshot root dentry before and after snapshot mount.
2681 if (zfsvfs->z_issnap) {
2682 if (ip->i_sb->s_root->d_inode == ip)
2683 sp->ino = ZFSCTL_INO_SNAPDIRS -
2684 dmu_objset_id(zfsvfs->z_os);
2693 * For the operation of changing file's user/group/project, we need to
2694 * handle not only the main object that is assigned to the file directly,
2695 * but also the ones that are used by the file via hidden xattr directory.
2697 * Because the xattr directory may contains many EA entries, as to it may
2698 * be impossible to change all of them via the transaction of changing the
2699 * main object's user/group/project attributes. Then we have to change them
2700 * via other multiple independent transactions one by one. It may be not good
2701 * solution, but we have no better idea yet.
2704 zfs_setattr_dir(znode_t *dzp)
2706 struct inode *dxip = ZTOI(dzp);
2707 struct inode *xip = NULL;
2708 zfsvfs_t *zfsvfs = ITOZSB(dxip);
2709 objset_t *os = zfsvfs->z_os;
2711 zap_attribute_t zap;
2714 dmu_tx_t *tx = NULL;
2716 sa_bulk_attr_t bulk[4];
2720 zap_cursor_init(&zc, os, dzp->z_id);
2721 while ((err = zap_cursor_retrieve(&zc, &zap)) == 0) {
2722 if (zap.za_integer_length != 8 || zap.za_num_integers != 1) {
2727 err = zfs_dirent_lock(&dl, dzp, (char *)zap.za_name, &zp,
2728 ZEXISTS, NULL, NULL);
2735 if (KUID_TO_SUID(xip->i_uid) == KUID_TO_SUID(dxip->i_uid) &&
2736 KGID_TO_SGID(xip->i_gid) == KGID_TO_SGID(dxip->i_gid) &&
2737 zp->z_projid == dzp->z_projid)
2740 tx = dmu_tx_create(os);
2741 if (!(zp->z_pflags & ZFS_PROJID))
2742 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2744 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2746 err = dmu_tx_assign(tx, TXG_WAIT);
2750 mutex_enter(&dzp->z_lock);
2752 if (KUID_TO_SUID(xip->i_uid) != KUID_TO_SUID(dxip->i_uid)) {
2753 xip->i_uid = dxip->i_uid;
2754 uid = zfs_uid_read(dxip);
2755 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
2756 &uid, sizeof (uid));
2759 if (KGID_TO_SGID(xip->i_gid) != KGID_TO_SGID(dxip->i_gid)) {
2760 xip->i_gid = dxip->i_gid;
2761 gid = zfs_gid_read(dxip);
2762 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
2763 &gid, sizeof (gid));
2766 if (zp->z_projid != dzp->z_projid) {
2767 if (!(zp->z_pflags & ZFS_PROJID)) {
2768 zp->z_pflags |= ZFS_PROJID;
2769 SA_ADD_BULK_ATTR(bulk, count,
2770 SA_ZPL_FLAGS(zfsvfs), NULL, &zp->z_pflags,
2771 sizeof (zp->z_pflags));
2774 zp->z_projid = dzp->z_projid;
2775 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PROJID(zfsvfs),
2776 NULL, &zp->z_projid, sizeof (zp->z_projid));
2779 mutex_exit(&dzp->z_lock);
2781 if (likely(count > 0)) {
2782 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
2788 if (err != 0 && err != ENOENT)
2795 zfs_dirent_unlock(dl);
2797 zap_cursor_advance(&zc);
2804 zfs_dirent_unlock(dl);
2806 zap_cursor_fini(&zc);
2808 return (err == ENOENT ? 0 : err);
2812 * Set the file attributes to the values contained in the
2815 * IN: ip - inode of file to be modified.
2816 * vap - new attribute values.
2817 * If ATTR_XVATTR set, then optional attrs are being set
2818 * flags - ATTR_UTIME set if non-default time values provided.
2819 * - ATTR_NOACLCHECK (CIFS context only).
2820 * cr - credentials of caller.
2822 * RETURN: 0 if success
2823 * error code if failure
2826 * ip - ctime updated, mtime updated if size changed.
2830 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2832 znode_t *zp = ITOZ(ip);
2833 zfsvfs_t *zfsvfs = ITOZSB(ip);
2834 objset_t *os = zfsvfs->z_os;
2838 xvattr_t *tmpxvattr;
2839 uint_t mask = vap->va_mask;
2840 uint_t saved_mask = 0;
2843 uint64_t new_kuid = 0, new_kgid = 0, new_uid, new_gid;
2845 uint64_t mtime[2], ctime[2], atime[2];
2846 uint64_t projid = ZFS_INVALID_PROJID;
2848 int need_policy = FALSE;
2850 zfs_fuid_info_t *fuidp = NULL;
2851 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2854 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2855 boolean_t fuid_dirtied = B_FALSE;
2856 boolean_t handle_eadir = B_FALSE;
2857 sa_bulk_attr_t *bulk, *xattr_bulk;
2858 int count = 0, xattr_count = 0, bulks = 8;
2867 * If this is a xvattr_t, then get a pointer to the structure of
2868 * optional attributes. If this is NULL, then we have a vattr_t.
2870 xoap = xva_getxoptattr(xvap);
2871 if (xoap != NULL && (mask & ATTR_XVATTR)) {
2872 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
2873 if (!dmu_objset_projectquota_enabled(os) ||
2874 (!S_ISREG(ip->i_mode) && !S_ISDIR(ip->i_mode))) {
2876 return (SET_ERROR(ENOTSUP));
2879 projid = xoap->xoa_projid;
2880 if (unlikely(projid == ZFS_INVALID_PROJID)) {
2882 return (SET_ERROR(EINVAL));
2885 if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID)
2886 projid = ZFS_INVALID_PROJID;
2891 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) &&
2892 (xoap->xoa_projinherit !=
2893 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) &&
2894 (!dmu_objset_projectquota_enabled(os) ||
2895 (!S_ISREG(ip->i_mode) && !S_ISDIR(ip->i_mode)))) {
2897 return (SET_ERROR(ENOTSUP));
2901 zilog = zfsvfs->z_log;
2904 * Make sure that if we have ephemeral uid/gid or xvattr specified
2905 * that file system is at proper version level
2908 if (zfsvfs->z_use_fuids == B_FALSE &&
2909 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2910 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2911 (mask & ATTR_XVATTR))) {
2913 return (SET_ERROR(EINVAL));
2916 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2918 return (SET_ERROR(EISDIR));
2921 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2923 return (SET_ERROR(EINVAL));
2926 tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP);
2927 xva_init(tmpxvattr);
2929 bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * bulks, KM_SLEEP);
2930 xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * bulks, KM_SLEEP);
2933 * Immutable files can only alter immutable bit and atime
2935 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2936 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2937 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2938 err = SET_ERROR(EPERM);
2942 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2943 err = SET_ERROR(EPERM);
2948 * Verify timestamps doesn't overflow 32 bits.
2949 * ZFS can handle large timestamps, but 32bit syscalls can't
2950 * handle times greater than 2039. This check should be removed
2951 * once large timestamps are fully supported.
2953 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2954 if (((mask & ATTR_ATIME) &&
2955 TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2956 ((mask & ATTR_MTIME) &&
2957 TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2958 err = SET_ERROR(EOVERFLOW);
2967 /* Can this be moved to before the top label? */
2968 if (zfs_is_readonly(zfsvfs)) {
2969 err = SET_ERROR(EROFS);
2974 * First validate permissions
2977 if (mask & ATTR_SIZE) {
2978 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2983 * XXX - Note, we are not providing any open
2984 * mode flags here (like FNDELAY), so we may
2985 * block if there are locks present... this
2986 * should be addressed in openat().
2988 /* XXX - would it be OK to generate a log record here? */
2989 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2994 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2995 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2996 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2997 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2998 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2999 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3000 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3001 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3002 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3006 if (mask & (ATTR_UID|ATTR_GID)) {
3007 int idmask = (mask & (ATTR_UID|ATTR_GID));
3012 * NOTE: even if a new mode is being set,
3013 * we may clear S_ISUID/S_ISGID bits.
3016 if (!(mask & ATTR_MODE))
3017 vap->va_mode = zp->z_mode;
3020 * Take ownership or chgrp to group we are a member of
3023 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
3024 take_group = (mask & ATTR_GID) &&
3025 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3028 * If both ATTR_UID and ATTR_GID are set then take_owner and
3029 * take_group must both be set in order to allow taking
3032 * Otherwise, send the check through secpolicy_vnode_setattr()
3036 if (((idmask == (ATTR_UID|ATTR_GID)) &&
3037 take_owner && take_group) ||
3038 ((idmask == ATTR_UID) && take_owner) ||
3039 ((idmask == ATTR_GID) && take_group)) {
3040 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3041 skipaclchk, cr) == 0) {
3043 * Remove setuid/setgid for non-privileged users
3045 (void) secpolicy_setid_clear(vap, cr);
3046 trim_mask = (mask & (ATTR_UID|ATTR_GID));
3055 mutex_enter(&zp->z_lock);
3056 oldva.va_mode = zp->z_mode;
3057 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3058 if (mask & ATTR_XVATTR) {
3060 * Update xvattr mask to include only those attributes
3061 * that are actually changing.
3063 * the bits will be restored prior to actually setting
3064 * the attributes so the caller thinks they were set.
3066 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3067 if (xoap->xoa_appendonly !=
3068 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3071 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3072 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
3076 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
3077 if (xoap->xoa_projinherit !=
3078 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) {
3081 XVA_CLR_REQ(xvap, XAT_PROJINHERIT);
3082 XVA_SET_REQ(tmpxvattr, XAT_PROJINHERIT);
3086 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3087 if (xoap->xoa_nounlink !=
3088 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3091 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3092 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
3096 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3097 if (xoap->xoa_immutable !=
3098 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3101 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3102 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
3106 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3107 if (xoap->xoa_nodump !=
3108 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3111 XVA_CLR_REQ(xvap, XAT_NODUMP);
3112 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
3116 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3117 if (xoap->xoa_av_modified !=
3118 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3121 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3122 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
3126 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3127 if ((!S_ISREG(ip->i_mode) &&
3128 xoap->xoa_av_quarantined) ||
3129 xoap->xoa_av_quarantined !=
3130 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3133 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3134 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
3138 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3139 mutex_exit(&zp->z_lock);
3140 err = SET_ERROR(EPERM);
3144 if (need_policy == FALSE &&
3145 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3146 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3151 mutex_exit(&zp->z_lock);
3153 if (mask & ATTR_MODE) {
3154 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3155 err = secpolicy_setid_setsticky_clear(ip, vap,
3160 trim_mask |= ATTR_MODE;
3168 * If trim_mask is set then take ownership
3169 * has been granted or write_acl is present and user
3170 * has the ability to modify mode. In that case remove
3171 * UID|GID and or MODE from mask so that
3172 * secpolicy_vnode_setattr() doesn't revoke it.
3176 saved_mask = vap->va_mask;
3177 vap->va_mask &= ~trim_mask;
3179 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
3180 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3185 vap->va_mask |= saved_mask;
3189 * secpolicy_vnode_setattr, or take ownership may have
3192 mask = vap->va_mask;
3194 if ((mask & (ATTR_UID | ATTR_GID)) || projid != ZFS_INVALID_PROJID) {
3195 handle_eadir = B_TRUE;
3196 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3197 &xattr_obj, sizeof (xattr_obj));
3199 if (err == 0 && xattr_obj) {
3200 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
3204 if (mask & ATTR_UID) {
3205 new_kuid = zfs_fuid_create(zfsvfs,
3206 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3207 if (new_kuid != KUID_TO_SUID(ZTOI(zp)->i_uid) &&
3208 zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT,
3212 err = SET_ERROR(EDQUOT);
3217 if (mask & ATTR_GID) {
3218 new_kgid = zfs_fuid_create(zfsvfs,
3219 (uint64_t)vap->va_gid, cr, ZFS_GROUP, &fuidp);
3220 if (new_kgid != KGID_TO_SGID(ZTOI(zp)->i_gid) &&
3221 zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT,
3225 err = SET_ERROR(EDQUOT);
3230 if (projid != ZFS_INVALID_PROJID &&
3231 zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) {
3238 tx = dmu_tx_create(os);
3240 if (mask & ATTR_MODE) {
3241 uint64_t pmode = zp->z_mode;
3243 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3245 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
3247 mutex_enter(&zp->z_lock);
3248 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3250 * Are we upgrading ACL from old V0 format
3253 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3254 zfs_znode_acl_version(zp) ==
3255 ZFS_ACL_VERSION_INITIAL) {
3256 dmu_tx_hold_free(tx, acl_obj, 0,
3258 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3259 0, aclp->z_acl_bytes);
3261 dmu_tx_hold_write(tx, acl_obj, 0,
3264 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3265 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3266 0, aclp->z_acl_bytes);
3268 mutex_exit(&zp->z_lock);
3269 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3271 if (((mask & ATTR_XVATTR) &&
3272 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ||
3273 (projid != ZFS_INVALID_PROJID &&
3274 !(zp->z_pflags & ZFS_PROJID)))
3275 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3277 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3281 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3284 fuid_dirtied = zfsvfs->z_fuid_dirty;
3286 zfs_fuid_txhold(zfsvfs, tx);
3288 zfs_sa_upgrade_txholds(tx, zp);
3290 err = dmu_tx_assign(tx, TXG_WAIT);
3296 * Set each attribute requested.
3297 * We group settings according to the locks they need to acquire.
3299 * Note: you cannot set ctime directly, although it will be
3300 * updated as a side-effect of calling this function.
3303 if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) {
3305 * For the existed object that is upgraded from old system,
3306 * its on-disk layout has no slot for the project ID attribute.
3307 * But quota accounting logic needs to access related slots by
3308 * offset directly. So we need to adjust old objects' layout
3309 * to make the project ID to some unified and fixed offset.
3312 err = sa_add_projid(attrzp->z_sa_hdl, tx, projid);
3314 err = sa_add_projid(zp->z_sa_hdl, tx, projid);
3316 if (unlikely(err == EEXIST))
3321 projid = ZFS_INVALID_PROJID;
3324 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3325 mutex_enter(&zp->z_acl_lock);
3326 mutex_enter(&zp->z_lock);
3328 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3329 &zp->z_pflags, sizeof (zp->z_pflags));
3332 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3333 mutex_enter(&attrzp->z_acl_lock);
3334 mutex_enter(&attrzp->z_lock);
3335 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3336 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3337 sizeof (attrzp->z_pflags));
3338 if (projid != ZFS_INVALID_PROJID) {
3339 attrzp->z_projid = projid;
3340 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3341 SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid,
3342 sizeof (attrzp->z_projid));
3346 if (mask & (ATTR_UID|ATTR_GID)) {
3348 if (mask & ATTR_UID) {
3349 ZTOI(zp)->i_uid = SUID_TO_KUID(new_kuid);
3350 new_uid = zfs_uid_read(ZTOI(zp));
3351 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3352 &new_uid, sizeof (new_uid));
3354 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3355 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3357 ZTOI(attrzp)->i_uid = SUID_TO_KUID(new_uid);
3361 if (mask & ATTR_GID) {
3362 ZTOI(zp)->i_gid = SGID_TO_KGID(new_kgid);
3363 new_gid = zfs_gid_read(ZTOI(zp));
3364 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3365 NULL, &new_gid, sizeof (new_gid));
3367 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3368 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3370 ZTOI(attrzp)->i_gid = SGID_TO_KGID(new_kgid);
3373 if (!(mask & ATTR_MODE)) {
3374 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3375 NULL, &new_mode, sizeof (new_mode));
3376 new_mode = zp->z_mode;
3378 err = zfs_acl_chown_setattr(zp);
3381 err = zfs_acl_chown_setattr(attrzp);
3386 if (mask & ATTR_MODE) {
3387 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3388 &new_mode, sizeof (new_mode));
3389 zp->z_mode = ZTOI(zp)->i_mode = new_mode;
3390 ASSERT3P(aclp, !=, NULL);
3391 err = zfs_aclset_common(zp, aclp, cr, tx);
3393 if (zp->z_acl_cached)
3394 zfs_acl_free(zp->z_acl_cached);
3395 zp->z_acl_cached = aclp;
3399 if ((mask & ATTR_ATIME) || zp->z_atime_dirty) {
3400 zp->z_atime_dirty = 0;
3401 ZFS_TIME_ENCODE(&ip->i_atime, atime);
3402 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3403 &atime, sizeof (atime));
3406 if (mask & (ATTR_MTIME | ATTR_SIZE)) {
3407 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3408 ZTOI(zp)->i_mtime = zpl_inode_timespec_trunc(vap->va_mtime,
3409 ZTOI(zp)->i_sb->s_time_gran);
3411 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3412 mtime, sizeof (mtime));
3415 if (mask & (ATTR_CTIME | ATTR_SIZE)) {
3416 ZFS_TIME_ENCODE(&vap->va_ctime, ctime);
3417 ZTOI(zp)->i_ctime = zpl_inode_timespec_trunc(vap->va_ctime,
3418 ZTOI(zp)->i_sb->s_time_gran);
3419 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3420 ctime, sizeof (ctime));
3423 if (projid != ZFS_INVALID_PROJID) {
3424 zp->z_projid = projid;
3425 SA_ADD_BULK_ATTR(bulk, count,
3426 SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid,
3427 sizeof (zp->z_projid));
3430 if (attrzp && mask) {
3431 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3432 SA_ZPL_CTIME(zfsvfs), NULL, &ctime,
3437 * Do this after setting timestamps to prevent timestamp
3438 * update from toggling bit
3441 if (xoap && (mask & ATTR_XVATTR)) {
3444 * restore trimmed off masks
3445 * so that return masks can be set for caller.
3448 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
3449 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3451 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
3452 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3454 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
3455 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3457 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
3458 XVA_SET_REQ(xvap, XAT_NODUMP);
3460 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
3461 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3463 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
3464 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3466 if (XVA_ISSET_REQ(tmpxvattr, XAT_PROJINHERIT)) {
3467 XVA_SET_REQ(xvap, XAT_PROJINHERIT);
3470 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3471 ASSERT(S_ISREG(ip->i_mode));
3473 zfs_xvattr_set(zp, xvap, tx);
3477 zfs_fuid_sync(zfsvfs, tx);
3480 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3482 mutex_exit(&zp->z_lock);
3483 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3484 mutex_exit(&zp->z_acl_lock);
3487 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3488 mutex_exit(&attrzp->z_acl_lock);
3489 mutex_exit(&attrzp->z_lock);
3492 if (err == 0 && xattr_count > 0) {
3493 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3502 zfs_fuid_info_free(fuidp);
3510 if (err == ERESTART)
3514 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3517 if (err2 == 0 && handle_eadir)
3518 err2 = zfs_setattr_dir(attrzp);
3521 zfs_inode_update(zp);
3525 if (os->os_sync == ZFS_SYNC_ALWAYS)
3526 zil_commit(zilog, 0);
3529 kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * bulks);
3530 kmem_free(bulk, sizeof (sa_bulk_attr_t) * bulks);
3531 kmem_free(tmpxvattr, sizeof (xvattr_t));
3536 typedef struct zfs_zlock {
3537 krwlock_t *zl_rwlock; /* lock we acquired */
3538 znode_t *zl_znode; /* znode we held */
3539 struct zfs_zlock *zl_next; /* next in list */
3543 * Drop locks and release vnodes that were held by zfs_rename_lock().
3546 zfs_rename_unlock(zfs_zlock_t **zlpp)
3550 while ((zl = *zlpp) != NULL) {
3551 if (zl->zl_znode != NULL)
3552 zfs_iput_async(ZTOI(zl->zl_znode));
3553 rw_exit(zl->zl_rwlock);
3554 *zlpp = zl->zl_next;
3555 kmem_free(zl, sizeof (*zl));
3560 * Search back through the directory tree, using the ".." entries.
3561 * Lock each directory in the chain to prevent concurrent renames.
3562 * Fail any attempt to move a directory into one of its own descendants.
3563 * XXX - z_parent_lock can overlap with map or grow locks
3566 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3570 uint64_t rootid = ZTOZSB(zp)->z_root;
3571 uint64_t oidp = zp->z_id;
3572 krwlock_t *rwlp = &szp->z_parent_lock;
3573 krw_t rw = RW_WRITER;
3576 * First pass write-locks szp and compares to zp->z_id.
3577 * Later passes read-lock zp and compare to zp->z_parent.
3580 if (!rw_tryenter(rwlp, rw)) {
3582 * Another thread is renaming in this path.
3583 * Note that if we are a WRITER, we don't have any
3584 * parent_locks held yet.
3586 if (rw == RW_READER && zp->z_id > szp->z_id) {
3588 * Drop our locks and restart
3590 zfs_rename_unlock(&zl);
3594 rwlp = &szp->z_parent_lock;
3599 * Wait for other thread to drop its locks
3605 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3606 zl->zl_rwlock = rwlp;
3607 zl->zl_znode = NULL;
3608 zl->zl_next = *zlpp;
3611 if (oidp == szp->z_id) /* We're a descendant of szp */
3612 return (SET_ERROR(EINVAL));
3614 if (oidp == rootid) /* We've hit the top */
3617 if (rw == RW_READER) { /* i.e. not the first pass */
3618 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3623 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3624 &oidp, sizeof (oidp));
3625 rwlp = &zp->z_parent_lock;
3628 } while (zp->z_id != sdzp->z_id);
3634 * Move an entry from the provided source directory to the target
3635 * directory. Change the entry name as indicated.
3637 * IN: sdip - Source directory containing the "old entry".
3638 * snm - Old entry name.
3639 * tdip - Target directory to contain the "new entry".
3640 * tnm - New entry name.
3641 * cr - credentials of caller.
3642 * flags - case flags
3644 * RETURN: 0 on success, error code on failure.
3647 * sdip,tdip - ctime|mtime updated
3651 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3652 cred_t *cr, int flags)
3654 znode_t *tdzp, *szp, *tzp;
3655 znode_t *sdzp = ITOZ(sdip);
3656 zfsvfs_t *zfsvfs = ITOZSB(sdip);
3658 zfs_dirlock_t *sdl, *tdl;
3661 int cmp, serr, terr;
3664 boolean_t waited = B_FALSE;
3666 if (snm == NULL || tnm == NULL)
3667 return (SET_ERROR(EINVAL));
3670 ZFS_VERIFY_ZP(sdzp);
3671 zilog = zfsvfs->z_log;
3674 ZFS_VERIFY_ZP(tdzp);
3677 * We check i_sb because snapshots and the ctldir must have different
3680 if (tdip->i_sb != sdip->i_sb || zfsctl_is_node(tdip)) {
3682 return (SET_ERROR(EXDEV));
3685 if (zfsvfs->z_utf8 && u8_validate(tnm,
3686 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3688 return (SET_ERROR(EILSEQ));
3691 if (flags & FIGNORECASE)
3700 * This is to prevent the creation of links into attribute space
3701 * by renaming a linked file into/outof an attribute directory.
3702 * See the comment in zfs_link() for why this is considered bad.
3704 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3706 return (SET_ERROR(EINVAL));
3710 * Lock source and target directory entries. To prevent deadlock,
3711 * a lock ordering must be defined. We lock the directory with
3712 * the smallest object id first, or if it's a tie, the one with
3713 * the lexically first name.
3715 if (sdzp->z_id < tdzp->z_id) {
3717 } else if (sdzp->z_id > tdzp->z_id) {
3721 * First compare the two name arguments without
3722 * considering any case folding.
3724 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3726 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3727 ASSERT(error == 0 || !zfsvfs->z_utf8);
3730 * POSIX: "If the old argument and the new argument
3731 * both refer to links to the same existing file,
3732 * the rename() function shall return successfully
3733 * and perform no other action."
3739 * If the file system is case-folding, then we may
3740 * have some more checking to do. A case-folding file
3741 * system is either supporting mixed case sensitivity
3742 * access or is completely case-insensitive. Note
3743 * that the file system is always case preserving.
3745 * In mixed sensitivity mode case sensitive behavior
3746 * is the default. FIGNORECASE must be used to
3747 * explicitly request case insensitive behavior.
3749 * If the source and target names provided differ only
3750 * by case (e.g., a request to rename 'tim' to 'Tim'),
3751 * we will treat this as a special case in the
3752 * case-insensitive mode: as long as the source name
3753 * is an exact match, we will allow this to proceed as
3754 * a name-change request.
3756 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3757 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3758 flags & FIGNORECASE)) &&
3759 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3762 * case preserving rename request, require exact
3771 * If the source and destination directories are the same, we should
3772 * grab the z_name_lock of that directory only once.
3776 rw_enter(&sdzp->z_name_lock, RW_READER);
3780 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3781 ZEXISTS | zflg, NULL, NULL);
3782 terr = zfs_dirent_lock(&tdl,
3783 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3785 terr = zfs_dirent_lock(&tdl,
3786 tdzp, tnm, &tzp, zflg, NULL, NULL);
3787 serr = zfs_dirent_lock(&sdl,
3788 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3794 * Source entry invalid or not there.
3797 zfs_dirent_unlock(tdl);
3803 rw_exit(&sdzp->z_name_lock);
3805 if (strcmp(snm, "..") == 0)
3811 zfs_dirent_unlock(sdl);
3815 rw_exit(&sdzp->z_name_lock);
3817 if (strcmp(tnm, "..") == 0)
3824 * If we are using project inheritance, means if the directory has
3825 * ZFS_PROJINHERIT set, then its descendant directories will inherit
3826 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
3827 * such case, we only allow renames into our tree when the project
3830 if (tdzp->z_pflags & ZFS_PROJINHERIT &&
3831 tdzp->z_projid != szp->z_projid) {
3832 error = SET_ERROR(EXDEV);
3837 * Must have write access at the source to remove the old entry
3838 * and write access at the target to create the new entry.
3839 * Note that if target and source are the same, this can be
3840 * done in a single check.
3843 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3846 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3848 * Check to make sure rename is valid.
3849 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3851 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3856 * Does target exist?
3860 * Source and target must be the same type.
3862 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3863 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3864 error = SET_ERROR(ENOTDIR);
3868 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3869 error = SET_ERROR(EISDIR);
3874 * POSIX dictates that when the source and target
3875 * entries refer to the same file object, rename
3876 * must do nothing and exit without error.
3878 if (szp->z_id == tzp->z_id) {
3884 tx = dmu_tx_create(zfsvfs->z_os);
3885 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3886 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3887 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3888 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3890 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3891 zfs_sa_upgrade_txholds(tx, tdzp);
3894 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3895 zfs_sa_upgrade_txholds(tx, tzp);
3898 zfs_sa_upgrade_txholds(tx, szp);
3899 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3900 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
3903 zfs_rename_unlock(&zl);
3904 zfs_dirent_unlock(sdl);
3905 zfs_dirent_unlock(tdl);
3908 rw_exit(&sdzp->z_name_lock);
3910 if (error == ERESTART) {
3927 if (tzp) /* Attempt to remove the existing target */
3928 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3931 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3933 szp->z_pflags |= ZFS_AV_MODIFIED;
3934 if (tdzp->z_pflags & ZFS_PROJINHERIT)
3935 szp->z_pflags |= ZFS_PROJINHERIT;
3937 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3938 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3941 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3943 zfs_log_rename(zilog, tx, TX_RENAME |
3944 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3945 sdl->dl_name, tdzp, tdl->dl_name, szp);
3948 * At this point, we have successfully created
3949 * the target name, but have failed to remove
3950 * the source name. Since the create was done
3951 * with the ZRENAMING flag, there are
3952 * complications; for one, the link count is
3953 * wrong. The easiest way to deal with this
3954 * is to remove the newly created target, and
3955 * return the original error. This must
3956 * succeed; fortunately, it is very unlikely to
3957 * fail, since we just created it.
3959 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3960 ZRENAMING, NULL), ==, 0);
3964 * If we had removed the existing target, subsequent
3965 * call to zfs_link_create() to add back the same entry
3966 * but, the new dnode (szp) should not fail.
3968 ASSERT(tzp == NULL);
3975 zfs_rename_unlock(&zl);
3977 zfs_dirent_unlock(sdl);
3978 zfs_dirent_unlock(tdl);
3980 zfs_inode_update(sdzp);
3982 rw_exit(&sdzp->z_name_lock);
3985 zfs_inode_update(tdzp);
3987 zfs_inode_update(szp);
3990 zfs_inode_update(tzp);
3994 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3995 zil_commit(zilog, 0);
4002 * Insert the indicated symbolic reference entry into the directory.
4004 * IN: dip - Directory to contain new symbolic link.
4005 * link - Name for new symlink entry.
4006 * vap - Attributes of new entry.
4007 * target - Target path of new symlink.
4009 * cr - credentials of caller.
4010 * flags - case flags
4012 * RETURN: 0 on success, error code on failure.
4015 * dip - ctime|mtime updated
4019 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
4020 struct inode **ipp, cred_t *cr, int flags)
4022 znode_t *zp, *dzp = ITOZ(dip);
4025 zfsvfs_t *zfsvfs = ITOZSB(dip);
4027 uint64_t len = strlen(link);
4030 zfs_acl_ids_t acl_ids;
4031 boolean_t fuid_dirtied;
4032 uint64_t txtype = TX_SYMLINK;
4033 boolean_t waited = B_FALSE;
4035 ASSERT(S_ISLNK(vap->va_mode));
4038 return (SET_ERROR(EINVAL));
4042 zilog = zfsvfs->z_log;
4044 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4045 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4047 return (SET_ERROR(EILSEQ));
4049 if (flags & FIGNORECASE)
4052 if (len > MAXPATHLEN) {
4054 return (SET_ERROR(ENAMETOOLONG));
4057 if ((error = zfs_acl_ids_create(dzp, 0,
4058 vap, cr, NULL, &acl_ids)) != 0) {
4066 * Attempt to lock directory; fail if entry already exists.
4068 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4070 zfs_acl_ids_free(&acl_ids);
4075 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
4076 zfs_acl_ids_free(&acl_ids);
4077 zfs_dirent_unlock(dl);
4082 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, ZFS_DEFAULT_PROJID)) {
4083 zfs_acl_ids_free(&acl_ids);
4084 zfs_dirent_unlock(dl);
4086 return (SET_ERROR(EDQUOT));
4088 tx = dmu_tx_create(zfsvfs->z_os);
4089 fuid_dirtied = zfsvfs->z_fuid_dirty;
4090 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4091 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4092 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4093 ZFS_SA_BASE_ATTR_SIZE + len);
4094 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4095 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4096 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4097 acl_ids.z_aclp->z_acl_bytes);
4100 zfs_fuid_txhold(zfsvfs, tx);
4101 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
4103 zfs_dirent_unlock(dl);
4104 if (error == ERESTART) {
4110 zfs_acl_ids_free(&acl_ids);
4117 * Create a new object for the symlink.
4118 * for version 4 ZPL datsets the symlink will be an SA attribute
4120 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4123 zfs_fuid_sync(zfsvfs, tx);
4125 mutex_enter(&zp->z_lock);
4127 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4130 zfs_sa_symlink(zp, link, len, tx);
4131 mutex_exit(&zp->z_lock);
4134 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4135 &zp->z_size, sizeof (zp->z_size), tx);
4137 * Insert the new object into the directory.
4139 error = zfs_link_create(dl, zp, tx, ZNEW);
4141 zfs_znode_delete(zp, tx);
4142 remove_inode_hash(ZTOI(zp));
4144 if (flags & FIGNORECASE)
4146 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4148 zfs_inode_update(dzp);
4149 zfs_inode_update(zp);
4152 zfs_acl_ids_free(&acl_ids);
4156 zfs_dirent_unlock(dl);
4161 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4162 zil_commit(zilog, 0);
4172 * Return, in the buffer contained in the provided uio structure,
4173 * the symbolic path referred to by ip.
4175 * IN: ip - inode of symbolic link
4176 * uio - structure to contain the link path.
4177 * cr - credentials of caller.
4179 * RETURN: 0 if success
4180 * error code if failure
4183 * ip - atime updated
4187 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
4189 znode_t *zp = ITOZ(ip);
4190 zfsvfs_t *zfsvfs = ITOZSB(ip);
4196 mutex_enter(&zp->z_lock);
4198 error = sa_lookup_uio(zp->z_sa_hdl,
4199 SA_ZPL_SYMLINK(zfsvfs), uio);
4201 error = zfs_sa_readlink(zp, uio);
4202 mutex_exit(&zp->z_lock);
4209 * Insert a new entry into directory tdip referencing sip.
4211 * IN: tdip - Directory to contain new entry.
4212 * sip - inode of new entry.
4213 * name - name of new entry.
4214 * cr - credentials of caller.
4216 * RETURN: 0 if success
4217 * error code if failure
4220 * tdip - ctime|mtime updated
4221 * sip - ctime updated
4225 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr,
4228 znode_t *dzp = ITOZ(tdip);
4230 zfsvfs_t *zfsvfs = ITOZSB(tdip);
4238 boolean_t waited = B_FALSE;
4239 boolean_t is_tmpfile = 0;
4242 is_tmpfile = (sip->i_nlink == 0 && (sip->i_state & I_LINKABLE));
4244 ASSERT(S_ISDIR(tdip->i_mode));
4247 return (SET_ERROR(EINVAL));
4251 zilog = zfsvfs->z_log;
4254 * POSIX dictates that we return EPERM here.
4255 * Better choices include ENOTSUP or EISDIR.
4257 if (S_ISDIR(sip->i_mode)) {
4259 return (SET_ERROR(EPERM));
4266 * If we are using project inheritance, means if the directory has
4267 * ZFS_PROJINHERIT set, then its descendant directories will inherit
4268 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
4269 * such case, we only allow hard link creation in our tree when the
4270 * project IDs are the same.
4272 if (dzp->z_pflags & ZFS_PROJINHERIT && dzp->z_projid != szp->z_projid) {
4274 return (SET_ERROR(EXDEV));
4278 * We check i_sb because snapshots and the ctldir must have different
4281 if (sip->i_sb != tdip->i_sb || zfsctl_is_node(sip)) {
4283 return (SET_ERROR(EXDEV));
4286 /* Prevent links to .zfs/shares files */
4288 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4289 &parent, sizeof (uint64_t))) != 0) {
4293 if (parent == zfsvfs->z_shares_dir) {
4295 return (SET_ERROR(EPERM));
4298 if (zfsvfs->z_utf8 && u8_validate(name,
4299 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4301 return (SET_ERROR(EILSEQ));
4303 if (flags & FIGNORECASE)
4307 * We do not support links between attributes and non-attributes
4308 * because of the potential security risk of creating links
4309 * into "normal" file space in order to circumvent restrictions
4310 * imposed in attribute space.
4312 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4314 return (SET_ERROR(EINVAL));
4317 owner = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(sip->i_uid),
4319 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4321 return (SET_ERROR(EPERM));
4324 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
4331 * Attempt to lock directory; fail if entry already exists.
4333 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4339 tx = dmu_tx_create(zfsvfs->z_os);
4340 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4341 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4343 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
4345 zfs_sa_upgrade_txholds(tx, szp);
4346 zfs_sa_upgrade_txholds(tx, dzp);
4347 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
4349 zfs_dirent_unlock(dl);
4350 if (error == ERESTART) {
4360 /* unmark z_unlinked so zfs_link_create will not reject */
4362 szp->z_unlinked = 0;
4363 error = zfs_link_create(dl, szp, tx, 0);
4366 uint64_t txtype = TX_LINK;
4368 * tmpfile is created to be in z_unlinkedobj, so remove it.
4369 * Also, we don't log in ZIL, be cause all previous file
4370 * operation on the tmpfile are ignored by ZIL. Instead we
4371 * always wait for txg to sync to make sure all previous
4372 * operation are sync safe.
4375 VERIFY(zap_remove_int(zfsvfs->z_os,
4376 zfsvfs->z_unlinkedobj, szp->z_id, tx) == 0);
4378 if (flags & FIGNORECASE)
4380 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4382 } else if (is_tmpfile) {
4383 /* restore z_unlinked since when linking failed */
4384 szp->z_unlinked = 1;
4386 txg = dmu_tx_get_txg(tx);
4389 zfs_dirent_unlock(dl);
4391 if (!is_tmpfile && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4392 zil_commit(zilog, 0);
4395 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), txg);
4397 zfs_inode_update(dzp);
4398 zfs_inode_update(szp);
4404 zfs_putpage_commit_cb(void *arg)
4406 struct page *pp = arg;
4409 end_page_writeback(pp);
4413 * Push a page out to disk, once the page is on stable storage the
4414 * registered commit callback will be run as notification of completion.
4416 * IN: ip - page mapped for inode.
4417 * pp - page to push (page is locked)
4418 * wbc - writeback control data
4420 * RETURN: 0 if success
4421 * error code if failure
4424 * ip - ctime|mtime updated
4428 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
4430 znode_t *zp = ITOZ(ip);
4431 zfsvfs_t *zfsvfs = ITOZSB(ip);
4439 uint64_t mtime[2], ctime[2];
4440 sa_bulk_attr_t bulk[3];
4442 struct address_space *mapping;
4447 ASSERT(PageLocked(pp));
4449 pgoff = page_offset(pp); /* Page byte-offset in file */
4450 offset = i_size_read(ip); /* File length in bytes */
4451 pglen = MIN(PAGE_SIZE, /* Page length in bytes */
4452 P2ROUNDUP(offset, PAGE_SIZE)-pgoff);
4454 /* Page is beyond end of file */
4455 if (pgoff >= offset) {
4461 /* Truncate page length to end of file */
4462 if (pgoff + pglen > offset)
4463 pglen = offset - pgoff;
4467 * FIXME: Allow mmap writes past its quota. The correct fix
4468 * is to register a page_mkwrite() handler to count the page
4469 * against its quota when it is about to be dirtied.
4471 if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT,
4472 KUID_TO_SUID(ip->i_uid)) ||
4473 zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT,
4474 KGID_TO_SGID(ip->i_gid)) ||
4475 (zp->z_projid != ZFS_DEFAULT_PROJID &&
4476 zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT,
4483 * The ordering here is critical and must adhere to the following
4484 * rules in order to avoid deadlocking in either zfs_read() or
4485 * zfs_free_range() due to a lock inversion.
4487 * 1) The page must be unlocked prior to acquiring the range lock.
4488 * This is critical because zfs_read() calls find_lock_page()
4489 * which may block on the page lock while holding the range lock.
4491 * 2) Before setting or clearing write back on a page the range lock
4492 * must be held in order to prevent a lock inversion with the
4493 * zfs_free_range() function.
4495 * This presents a problem because upon entering this function the
4496 * page lock is already held. To safely acquire the range lock the
4497 * page lock must be dropped. This creates a window where another
4498 * process could truncate, invalidate, dirty, or write out the page.
4500 * Therefore, after successfully reacquiring the range and page locks
4501 * the current page state is checked. In the common case everything
4502 * will be as is expected and it can be written out. However, if
4503 * the page state has changed it must be handled accordingly.
4505 mapping = pp->mapping;
4506 redirty_page_for_writepage(wbc, pp);
4509 rl = zfs_range_lock(&zp->z_range_lock, pgoff, pglen, RL_WRITER);
4512 /* Page mapping changed or it was no longer dirty, we're done */
4513 if (unlikely((mapping != pp->mapping) || !PageDirty(pp))) {
4515 zfs_range_unlock(rl);
4520 /* Another process started write block if required */
4521 if (PageWriteback(pp)) {
4523 zfs_range_unlock(rl);
4525 if (wbc->sync_mode != WB_SYNC_NONE)
4526 wait_on_page_writeback(pp);
4532 /* Clear the dirty flag the required locks are held */
4533 if (!clear_page_dirty_for_io(pp)) {
4535 zfs_range_unlock(rl);
4541 * Counterpart for redirty_page_for_writepage() above. This page
4542 * was in fact not skipped and should not be counted as if it were.
4544 wbc->pages_skipped--;
4545 set_page_writeback(pp);
4548 tx = dmu_tx_create(zfsvfs->z_os);
4549 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
4550 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4551 zfs_sa_upgrade_txholds(tx, zp);
4553 err = dmu_tx_assign(tx, TXG_NOWAIT);
4555 if (err == ERESTART)
4559 __set_page_dirty_nobuffers(pp);
4561 end_page_writeback(pp);
4562 zfs_range_unlock(rl);
4568 ASSERT3U(pglen, <=, PAGE_SIZE);
4569 dmu_write(zfsvfs->z_os, zp->z_id, pgoff, pglen, va, tx);
4572 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
4573 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
4574 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zfsvfs), NULL,
4577 /* Preserve the mtime and ctime provided by the inode */
4578 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4579 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4580 zp->z_atime_dirty = 0;
4583 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4585 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0,
4586 zfs_putpage_commit_cb, pp);
4589 zfs_range_unlock(rl);
4591 if (wbc->sync_mode != WB_SYNC_NONE) {
4593 * Note that this is rarely called under writepages(), because
4594 * writepages() normally handles the entire commit for
4595 * performance reasons.
4597 zil_commit(zfsvfs->z_log, zp->z_id);
4605 * Update the system attributes when the inode has been dirtied. For the
4606 * moment we only update the mode, atime, mtime, and ctime.
4609 zfs_dirty_inode(struct inode *ip, int flags)
4611 znode_t *zp = ITOZ(ip);
4612 zfsvfs_t *zfsvfs = ITOZSB(ip);
4614 uint64_t mode, atime[2], mtime[2], ctime[2];
4615 sa_bulk_attr_t bulk[4];
4619 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
4627 * This is the lazytime semantic indroduced in Linux 4.0
4628 * This flag will only be called from update_time when lazytime is set.
4629 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4630 * Fortunately mtime and ctime are managed within ZFS itself, so we
4631 * only need to dirty atime.
4633 if (flags == I_DIRTY_TIME) {
4634 zp->z_atime_dirty = 1;
4639 tx = dmu_tx_create(zfsvfs->z_os);
4641 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4642 zfs_sa_upgrade_txholds(tx, zp);
4644 error = dmu_tx_assign(tx, TXG_WAIT);
4650 mutex_enter(&zp->z_lock);
4651 zp->z_atime_dirty = 0;
4653 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
4654 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
4655 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
4656 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
4658 /* Preserve the mode, mtime and ctime provided by the inode */
4659 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4660 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4661 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4666 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4667 mutex_exit(&zp->z_lock);
4677 zfs_inactive(struct inode *ip)
4679 znode_t *zp = ITOZ(ip);
4680 zfsvfs_t *zfsvfs = ITOZSB(ip);
4683 int need_unlock = 0;
4685 /* Only read lock if we haven't already write locked, e.g. rollback */
4686 if (!RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)) {
4688 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4690 if (zp->z_sa_hdl == NULL) {
4692 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4696 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4697 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4699 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4700 zfs_sa_upgrade_txholds(tx, zp);
4701 error = dmu_tx_assign(tx, TXG_WAIT);
4705 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4706 mutex_enter(&zp->z_lock);
4707 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4708 (void *)&atime, sizeof (atime), tx);
4709 zp->z_atime_dirty = 0;
4710 mutex_exit(&zp->z_lock);
4717 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4721 * Bounds-check the seek operation.
4723 * IN: ip - inode seeking within
4724 * ooff - old file offset
4725 * noffp - pointer to new file offset
4726 * ct - caller context
4728 * RETURN: 0 if success
4729 * EINVAL if new offset invalid
4733 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
4735 if (S_ISDIR(ip->i_mode))
4737 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4741 * Fill pages with data from the disk.
4744 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
4746 znode_t *zp = ITOZ(ip);
4747 zfsvfs_t *zfsvfs = ITOZSB(ip);
4749 struct page *cur_pp;
4750 u_offset_t io_off, total;
4757 io_len = nr_pages << PAGE_SHIFT;
4758 i_size = i_size_read(ip);
4759 io_off = page_offset(pl[0]);
4761 if (io_off + io_len > i_size)
4762 io_len = i_size - io_off;
4765 * Iterate over list of pages and read each page individually.
4768 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4771 cur_pp = pl[page_idx++];
4773 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4777 /* convert checksum errors into IO errors */
4779 err = SET_ERROR(EIO);
4788 * Uses zfs_fillpage to read data from the file and fill the pages.
4790 * IN: ip - inode of file to get data from.
4791 * pl - list of pages to read
4792 * nr_pages - number of pages to read
4794 * RETURN: 0 on success, error code on failure.
4797 * vp - atime updated
4801 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
4803 znode_t *zp = ITOZ(ip);
4804 zfsvfs_t *zfsvfs = ITOZSB(ip);
4813 err = zfs_fillpage(ip, pl, nr_pages);
4820 * Check ZFS specific permissions to memory map a section of a file.
4822 * IN: ip - inode of the file to mmap
4824 * addrp - start address in memory region
4825 * len - length of memory region
4826 * vm_flags- address flags
4828 * RETURN: 0 if success
4829 * error code if failure
4833 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4834 unsigned long vm_flags)
4836 znode_t *zp = ITOZ(ip);
4837 zfsvfs_t *zfsvfs = ITOZSB(ip);
4842 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4843 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4845 return (SET_ERROR(EPERM));
4848 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4849 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4851 return (SET_ERROR(EACCES));
4854 if (off < 0 || len > MAXOFFSET_T - off) {
4856 return (SET_ERROR(ENXIO));
4864 * convoff - converts the given data (start, whence) to the
4868 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4873 if ((lckdat->l_whence == 2) || (whence == 2)) {
4874 if ((error = zfs_getattr(ip, &vap, 0, CRED())))
4878 switch (lckdat->l_whence) {
4880 lckdat->l_start += offset;
4883 lckdat->l_start += vap.va_size;
4888 return (SET_ERROR(EINVAL));
4891 if (lckdat->l_start < 0)
4892 return (SET_ERROR(EINVAL));
4896 lckdat->l_start -= offset;
4899 lckdat->l_start -= vap.va_size;
4904 return (SET_ERROR(EINVAL));
4907 lckdat->l_whence = (short)whence;
4912 * Free or allocate space in a file. Currently, this function only
4913 * supports the `F_FREESP' command. However, this command is somewhat
4914 * misnamed, as its functionality includes the ability to allocate as
4915 * well as free space.
4917 * IN: ip - inode of file to free data in.
4918 * cmd - action to take (only F_FREESP supported).
4919 * bfp - section of file to free/alloc.
4920 * flag - current file open mode flags.
4921 * offset - current file offset.
4922 * cr - credentials of caller [UNUSED].
4924 * RETURN: 0 on success, error code on failure.
4927 * ip - ctime|mtime updated
4931 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4932 offset_t offset, cred_t *cr)
4934 znode_t *zp = ITOZ(ip);
4935 zfsvfs_t *zfsvfs = ITOZSB(ip);
4942 if (cmd != F_FREESP) {
4944 return (SET_ERROR(EINVAL));
4948 * Callers might not be able to detect properly that we are read-only,
4949 * so check it explicitly here.
4951 if (zfs_is_readonly(zfsvfs)) {
4953 return (SET_ERROR(EROFS));
4956 if ((error = convoff(ip, bfp, 0, offset))) {
4961 if (bfp->l_len < 0) {
4963 return (SET_ERROR(EINVAL));
4967 * Permissions aren't checked on Solaris because on this OS
4968 * zfs_space() can only be called with an opened file handle.
4969 * On Linux we can get here through truncate_range() which
4970 * operates directly on inodes, so we need to check access rights.
4972 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4978 len = bfp->l_len; /* 0 means from off to end of file */
4980 error = zfs_freesp(zp, off, len, flag, TRUE);
4988 zfs_fid(struct inode *ip, fid_t *fidp)
4990 znode_t *zp = ITOZ(ip);
4991 zfsvfs_t *zfsvfs = ITOZSB(ip);
4994 uint64_t object = zp->z_id;
5001 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5002 &gen64, sizeof (uint64_t))) != 0) {
5007 gen = (uint32_t)gen64;
5009 size = SHORT_FID_LEN;
5011 zfid = (zfid_short_t *)fidp;
5013 zfid->zf_len = size;
5015 for (i = 0; i < sizeof (zfid->zf_object); i++)
5016 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5018 /* Must have a non-zero generation number to distinguish from .zfs */
5021 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5022 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5030 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
5032 znode_t *zp = ITOZ(ip);
5033 zfsvfs_t *zfsvfs = ITOZSB(ip);
5035 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5039 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5047 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
5049 znode_t *zp = ITOZ(ip);
5050 zfsvfs_t *zfsvfs = ITOZSB(ip);
5052 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5053 zilog_t *zilog = zfsvfs->z_log;
5058 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5060 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5061 zil_commit(zilog, 0);
5067 #ifdef HAVE_UIO_ZEROCOPY
5069 * Tunable, both must be a power of 2.
5071 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
5072 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
5073 * an arcbuf for a partial block read
5075 int zcr_blksz_min = (1 << 10); /* 1K */
5076 int zcr_blksz_max = (1 << 17); /* 128K */
5080 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
5082 znode_t *zp = ITOZ(ip);
5083 zfsvfs_t *zfsvfs = ITOZSB(ip);
5084 int max_blksz = zfsvfs->z_max_blksz;
5085 uio_t *uio = &xuio->xu_uio;
5086 ssize_t size = uio->uio_resid;
5087 offset_t offset = uio->uio_loffset;
5092 int preamble, postamble;
5094 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5095 return (SET_ERROR(EINVAL));
5102 * Loan out an arc_buf for write if write size is bigger than
5103 * max_blksz, and the file's block size is also max_blksz.
5106 if (size < blksz || zp->z_blksz != blksz) {
5108 return (SET_ERROR(EINVAL));
5111 * Caller requests buffers for write before knowing where the
5112 * write offset might be (e.g. NFS TCP write).
5117 preamble = P2PHASE(offset, blksz);
5119 preamble = blksz - preamble;
5124 postamble = P2PHASE(size, blksz);
5127 fullblk = size / blksz;
5128 (void) dmu_xuio_init(xuio,
5129 (preamble != 0) + fullblk + (postamble != 0));
5132 * Have to fix iov base/len for partial buffers. They
5133 * currently represent full arc_buf's.
5136 /* data begins in the middle of the arc_buf */
5137 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5140 (void) dmu_xuio_add(xuio, abuf,
5141 blksz - preamble, preamble);
5144 for (i = 0; i < fullblk; i++) {
5145 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5148 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5152 /* data ends in the middle of the arc_buf */
5153 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5156 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5161 * Loan out an arc_buf for read if the read size is larger than
5162 * the current file block size. Block alignment is not
5163 * considered. Partial arc_buf will be loaned out for read.
5165 blksz = zp->z_blksz;
5166 if (blksz < zcr_blksz_min)
5167 blksz = zcr_blksz_min;
5168 if (blksz > zcr_blksz_max)
5169 blksz = zcr_blksz_max;
5170 /* avoid potential complexity of dealing with it */
5171 if (blksz > max_blksz) {
5173 return (SET_ERROR(EINVAL));
5176 maxsize = zp->z_size - uio->uio_loffset;
5182 return (SET_ERROR(EINVAL));
5187 return (SET_ERROR(EINVAL));
5190 uio->uio_extflg = UIO_XUIO;
5191 XUIO_XUZC_RW(xuio) = ioflag;
5198 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
5202 int ioflag = XUIO_XUZC_RW(xuio);
5204 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5206 i = dmu_xuio_cnt(xuio);
5208 abuf = dmu_xuio_arcbuf(xuio, i);
5210 * if abuf == NULL, it must be a write buffer
5211 * that has been returned in zfs_write().
5214 dmu_return_arcbuf(abuf);
5215 ASSERT(abuf || ioflag == UIO_WRITE);
5218 dmu_xuio_fini(xuio);
5221 #endif /* HAVE_UIO_ZEROCOPY */
5223 #if defined(_KERNEL)
5224 EXPORT_SYMBOL(zfs_open);
5225 EXPORT_SYMBOL(zfs_close);
5226 EXPORT_SYMBOL(zfs_read);
5227 EXPORT_SYMBOL(zfs_write);
5228 EXPORT_SYMBOL(zfs_access);
5229 EXPORT_SYMBOL(zfs_lookup);
5230 EXPORT_SYMBOL(zfs_create);
5231 EXPORT_SYMBOL(zfs_tmpfile);
5232 EXPORT_SYMBOL(zfs_remove);
5233 EXPORT_SYMBOL(zfs_mkdir);
5234 EXPORT_SYMBOL(zfs_rmdir);
5235 EXPORT_SYMBOL(zfs_readdir);
5236 EXPORT_SYMBOL(zfs_fsync);
5237 EXPORT_SYMBOL(zfs_getattr);
5238 EXPORT_SYMBOL(zfs_getattr_fast);
5239 EXPORT_SYMBOL(zfs_setattr);
5240 EXPORT_SYMBOL(zfs_rename);
5241 EXPORT_SYMBOL(zfs_symlink);
5242 EXPORT_SYMBOL(zfs_readlink);
5243 EXPORT_SYMBOL(zfs_link);
5244 EXPORT_SYMBOL(zfs_inactive);
5245 EXPORT_SYMBOL(zfs_space);
5246 EXPORT_SYMBOL(zfs_fid);
5247 EXPORT_SYMBOL(zfs_getsecattr);
5248 EXPORT_SYMBOL(zfs_setsecattr);
5249 EXPORT_SYMBOL(zfs_getpage);
5250 EXPORT_SYMBOL(zfs_putpage);
5251 EXPORT_SYMBOL(zfs_dirty_inode);
5252 EXPORT_SYMBOL(zfs_map);
5255 module_param(zfs_delete_blocks, ulong, 0644);
5256 MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async");
5257 module_param(zfs_read_chunk_size, long, 0644);
5258 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");