4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
28 /* Portions Copyright 2007 Jeremy Teo */
29 /* Portions Copyright 2010 Robert Milkowski */
32 #include <sys/types.h>
33 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/sysmacros.h>
37 #include <sys/resource.h>
39 #include <sys/vfs_opreg.h>
43 #include <sys/taskq.h>
45 #include <sys/vmsystm.h>
46 #include <sys/atomic.h>
48 #include <sys/pathname.h>
49 #include <sys/cmn_err.h>
50 #include <sys/errno.h>
51 #include <sys/unistd.h>
52 #include <sys/zfs_dir.h>
53 #include <sys/zfs_acl.h>
54 #include <sys/zfs_ioctl.h>
55 #include <sys/fs/zfs.h>
57 #include <sys/dmu_objset.h>
63 #include <sys/dirent.h>
64 #include <sys/policy.h>
65 #include <sys/sunddi.h>
68 #include "fs/fs_subr.h"
69 #include <sys/zfs_ctldir.h>
70 #include <sys/zfs_fuid.h>
71 #include <sys/zfs_sa.h>
72 #include <sys/zfs_vnops.h>
74 #include <sys/zfs_rlock.h>
75 #include <sys/extdirent.h>
76 #include <sys/kidmap.h>
84 * Each vnode op performs some logical unit of work. To do this, the ZPL must
85 * properly lock its in-core state, create a DMU transaction, do the work,
86 * record this work in the intent log (ZIL), commit the DMU transaction,
87 * and wait for the intent log to commit if it is a synchronous operation.
88 * Moreover, the vnode ops must work in both normal and log replay context.
89 * The ordering of events is important to avoid deadlocks and references
90 * to freed memory. The example below illustrates the following Big Rules:
92 * (1) A check must be made in each zfs thread for a mounted file system.
93 * This is done avoiding races using ZFS_ENTER(zsb).
94 * A ZFS_EXIT(zsb) is needed before all returns. Any znodes
95 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
96 * can return EIO from the calling function.
98 * (2) iput() should always be the last thing except for zil_commit()
99 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
100 * First, if it's the last reference, the vnode/znode
101 * can be freed, so the zp may point to freed memory. Second, the last
102 * reference will call zfs_zinactive(), which may induce a lot of work --
103 * pushing cached pages (which acquires range locks) and syncing out
104 * cached atime changes. Third, zfs_zinactive() may require a new tx,
105 * which could deadlock the system if you were already holding one.
106 * If you must call iput() within a tx then use zfs_iput_async().
108 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
109 * as they can span dmu_tx_assign() calls.
111 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
112 * dmu_tx_assign(). This is critical because we don't want to block
113 * while holding locks.
115 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
116 * reduces lock contention and CPU usage when we must wait (note that if
117 * throughput is constrained by the storage, nearly every transaction
120 * Note, in particular, that if a lock is sometimes acquired before
121 * the tx assigns, and sometimes after (e.g. z_lock), then failing
122 * to use a non-blocking assign can deadlock the system. The scenario:
124 * Thread A has grabbed a lock before calling dmu_tx_assign().
125 * Thread B is in an already-assigned tx, and blocks for this lock.
126 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
127 * forever, because the previous txg can't quiesce until B's tx commits.
129 * If dmu_tx_assign() returns ERESTART and zsb->z_assign is TXG_NOWAIT,
130 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
131 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
132 * to indicate that this operation has already called dmu_tx_wait().
133 * This will ensure that we don't retry forever, waiting a short bit
136 * (5) If the operation succeeded, generate the intent log entry for it
137 * before dropping locks. This ensures that the ordering of events
138 * in the intent log matches the order in which they actually occurred.
139 * During ZIL replay the zfs_log_* functions will update the sequence
140 * number to indicate the zil transaction has replayed.
142 * (6) At the end of each vnode op, the DMU tx must always commit,
143 * regardless of whether there were any errors.
145 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
146 * to ensure that synchronous semantics are provided when necessary.
148 * In general, this is how things should be ordered in each vnode op:
150 * ZFS_ENTER(zsb); // exit if unmounted
152 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
153 * rw_enter(...); // grab any other locks you need
154 * tx = dmu_tx_create(...); // get DMU tx
155 * dmu_tx_hold_*(); // hold each object you might modify
156 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
158 * rw_exit(...); // drop locks
159 * zfs_dirent_unlock(dl); // unlock directory entry
160 * iput(...); // release held vnodes
161 * if (error == ERESTART) {
167 * dmu_tx_abort(tx); // abort DMU tx
168 * ZFS_EXIT(zsb); // finished in zfs
169 * return (error); // really out of space
171 * error = do_real_work(); // do whatever this VOP does
173 * zfs_log_*(...); // on success, make ZIL entry
174 * dmu_tx_commit(tx); // commit DMU tx -- error or not
175 * rw_exit(...); // drop locks
176 * zfs_dirent_unlock(dl); // unlock directory entry
177 * iput(...); // release held vnodes
178 * zil_commit(zilog, foid); // synchronous when necessary
179 * ZFS_EXIT(zsb); // finished in zfs
180 * return (error); // done, report error
184 * Virus scanning is unsupported. It would be possible to add a hook
185 * here to performance the required virus scan. This could be done
186 * entirely in the kernel or potentially as an update to invoke a
190 zfs_vscan(struct inode *ip, cred_t *cr, int async)
197 zfs_open(struct inode *ip, int mode, int flag, cred_t *cr)
199 znode_t *zp = ITOZ(ip);
200 zfs_sb_t *zsb = ITOZSB(ip);
205 /* Honor ZFS_APPENDONLY file attribute */
206 if ((mode & FMODE_WRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
207 ((flag & O_APPEND) == 0)) {
209 return (SET_ERROR(EPERM));
212 /* Virus scan eligible files on open */
213 if (!zfs_has_ctldir(zp) && zsb->z_vscan && S_ISREG(ip->i_mode) &&
214 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
215 if (zfs_vscan(ip, cr, 0) != 0) {
217 return (SET_ERROR(EACCES));
221 /* Keep a count of the synchronous opens in the znode */
223 atomic_inc_32(&zp->z_sync_cnt);
228 EXPORT_SYMBOL(zfs_open);
232 zfs_close(struct inode *ip, int flag, cred_t *cr)
234 znode_t *zp = ITOZ(ip);
235 zfs_sb_t *zsb = ITOZSB(ip);
240 /* Decrement the synchronous opens in the znode */
242 atomic_dec_32(&zp->z_sync_cnt);
244 if (!zfs_has_ctldir(zp) && zsb->z_vscan && S_ISREG(ip->i_mode) &&
245 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
246 VERIFY(zfs_vscan(ip, cr, 1) == 0);
251 EXPORT_SYMBOL(zfs_close);
253 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
255 * Lseek support for finding holes (cmd == SEEK_HOLE) and
256 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
259 zfs_holey_common(struct inode *ip, int cmd, loff_t *off)
261 znode_t *zp = ITOZ(ip);
262 uint64_t noff = (uint64_t)*off; /* new offset */
267 file_sz = zp->z_size;
268 if (noff >= file_sz) {
269 return (SET_ERROR(ENXIO));
272 if (cmd == SEEK_HOLE)
277 error = dmu_offset_next(ZTOZSB(zp)->z_os, zp->z_id, hole, &noff);
280 return (SET_ERROR(ENXIO));
283 * We could find a hole that begins after the logical end-of-file,
284 * because dmu_offset_next() only works on whole blocks. If the
285 * EOF falls mid-block, then indicate that the "virtual hole"
286 * at the end of the file begins at the logical EOF, rather than
287 * at the end of the last block.
289 if (noff > file_sz) {
301 zfs_holey(struct inode *ip, int cmd, loff_t *off)
303 znode_t *zp = ITOZ(ip);
304 zfs_sb_t *zsb = ITOZSB(ip);
310 error = zfs_holey_common(ip, cmd, off);
315 EXPORT_SYMBOL(zfs_holey);
316 #endif /* SEEK_HOLE && SEEK_DATA */
320 * When a file is memory mapped, we must keep the IO data synchronized
321 * between the DMU cache and the memory mapped pages. What this means:
323 * On Write: If we find a memory mapped page, we write to *both*
324 * the page and the dmu buffer.
327 update_pages(struct inode *ip, int64_t start, int len,
328 objset_t *os, uint64_t oid)
330 struct address_space *mp = ip->i_mapping;
336 off = start & (PAGE_SIZE-1);
337 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
338 nbytes = MIN(PAGE_SIZE - off, len);
340 pp = find_lock_page(mp, start >> PAGE_SHIFT);
342 if (mapping_writably_mapped(mp))
343 flush_dcache_page(pp);
346 (void) dmu_read(os, oid, start+off, nbytes, pb+off,
350 if (mapping_writably_mapped(mp))
351 flush_dcache_page(pp);
353 mark_page_accessed(pp);
366 * When a file is memory mapped, we must keep the IO data synchronized
367 * between the DMU cache and the memory mapped pages. What this means:
369 * On Read: We "read" preferentially from memory mapped pages,
370 * else we default from the dmu buffer.
372 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
373 * the file is memory mapped.
376 mappedread(struct inode *ip, int nbytes, uio_t *uio)
378 struct address_space *mp = ip->i_mapping;
380 znode_t *zp = ITOZ(ip);
387 start = uio->uio_loffset;
388 off = start & (PAGE_SIZE-1);
389 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
390 bytes = MIN(PAGE_SIZE - off, len);
392 pp = find_lock_page(mp, start >> PAGE_SHIFT);
394 ASSERT(PageUptodate(pp));
397 error = uiomove(pb + off, bytes, UIO_READ, uio);
400 if (mapping_writably_mapped(mp))
401 flush_dcache_page(pp);
403 mark_page_accessed(pp);
407 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
420 unsigned long zfs_read_chunk_size = 1024 * 1024; /* Tunable */
421 unsigned long zfs_delete_blocks = DMU_MAX_DELETEBLKCNT;
424 * Read bytes from specified file into supplied buffer.
426 * IN: ip - inode of file to be read from.
427 * uio - structure supplying read location, range info,
429 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
430 * O_DIRECT flag; used to bypass page cache.
431 * cr - credentials of caller.
433 * OUT: uio - updated offset and range, buffer filled.
435 * RETURN: 0 on success, error code on failure.
438 * inode - atime updated if byte count > 0
442 zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
444 znode_t *zp = ITOZ(ip);
445 zfs_sb_t *zsb = ITOZSB(ip);
449 #ifdef HAVE_UIO_ZEROCOPY
451 #endif /* HAVE_UIO_ZEROCOPY */
456 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
458 return (SET_ERROR(EACCES));
462 * Validate file offset
464 if (uio->uio_loffset < (offset_t)0) {
466 return (SET_ERROR(EINVAL));
470 * Fasttrack empty reads
472 if (uio->uio_resid == 0) {
478 * If we're in FRSYNC mode, sync out this znode before reading it.
480 if (ioflag & FRSYNC || zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
481 zil_commit(zsb->z_log, zp->z_id);
484 * Lock the range against changes.
486 rl = zfs_range_lock(&zp->z_range_lock, uio->uio_loffset, uio->uio_resid,
490 * If we are reading past end-of-file we can skip
491 * to the end; but we might still need to set atime.
493 if (uio->uio_loffset >= zp->z_size) {
498 ASSERT(uio->uio_loffset < zp->z_size);
499 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
501 #ifdef HAVE_UIO_ZEROCOPY
502 if ((uio->uio_extflg == UIO_XUIO) &&
503 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
505 int blksz = zp->z_blksz;
506 uint64_t offset = uio->uio_loffset;
508 xuio = (xuio_t *)uio;
510 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
513 ASSERT(offset + n <= blksz);
516 (void) dmu_xuio_init(xuio, nblk);
518 if (vn_has_cached_data(ip)) {
520 * For simplicity, we always allocate a full buffer
521 * even if we only expect to read a portion of a block.
523 while (--nblk >= 0) {
524 (void) dmu_xuio_add(xuio,
525 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
530 #endif /* HAVE_UIO_ZEROCOPY */
533 nbytes = MIN(n, zfs_read_chunk_size -
534 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
536 if (zp->z_is_mapped && !(ioflag & O_DIRECT)) {
537 error = mappedread(ip, nbytes, uio);
539 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
544 /* convert checksum errors into IO errors */
546 error = SET_ERROR(EIO);
553 zfs_range_unlock(rl);
558 EXPORT_SYMBOL(zfs_read);
561 * Write the bytes to a file.
563 * IN: ip - inode of file to be written to.
564 * uio - structure supplying write location, range info,
566 * ioflag - FAPPEND flag set if in append mode.
567 * O_DIRECT flag; used to bypass page cache.
568 * cr - credentials of caller.
570 * OUT: uio - updated offset and range.
572 * RETURN: 0 if success
573 * error code if failure
576 * ip - ctime|mtime updated if byte count > 0
581 zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
583 znode_t *zp = ITOZ(ip);
584 rlim64_t limit = uio->uio_limit;
585 ssize_t start_resid = uio->uio_resid;
589 zfs_sb_t *zsb = ZTOZSB(zp);
594 int max_blksz = zsb->z_max_blksz;
597 const iovec_t *aiov = NULL;
600 const iovec_t *iovp = uio->uio_iov;
603 sa_bulk_attr_t bulk[4];
604 uint64_t mtime[2], ctime[2];
605 ASSERTV(int iovcnt = uio->uio_iovcnt);
608 * Fasttrack empty write
614 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
620 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
621 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
622 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb), NULL, &zp->z_size, 8);
623 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
627 * Callers might not be able to detect properly that we are read-only,
628 * so check it explicitly here.
630 if (zfs_is_readonly(zsb)) {
632 return (SET_ERROR(EROFS));
636 * If immutable or not appending then return EPERM
638 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
639 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
640 (uio->uio_loffset < zp->z_size))) {
642 return (SET_ERROR(EPERM));
648 * Validate file offset
650 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
653 return (SET_ERROR(EINVAL));
657 * Pre-fault the pages to ensure slow (eg NFS) pages
659 * Skip this if uio contains loaned arc_buf.
661 #ifdef HAVE_UIO_ZEROCOPY
662 if ((uio->uio_extflg == UIO_XUIO) &&
663 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
664 xuio = (xuio_t *)uio;
667 uio_prefaultpages(MIN(n, max_blksz), uio);
670 * If in append mode, set the io offset pointer to eof.
672 if (ioflag & FAPPEND) {
674 * Obtain an appending range lock to guarantee file append
675 * semantics. We reset the write offset once we have the lock.
677 rl = zfs_range_lock(&zp->z_range_lock, 0, n, RL_APPEND);
679 if (rl->r_len == UINT64_MAX) {
681 * We overlocked the file because this write will cause
682 * the file block size to increase.
683 * Note that zp_size cannot change with this lock held.
687 uio->uio_loffset = woff;
690 * Note that if the file block size will change as a result of
691 * this write, then this range lock will lock the entire file
692 * so that we can re-write the block safely.
694 rl = zfs_range_lock(&zp->z_range_lock, woff, n, RL_WRITER);
698 zfs_range_unlock(rl);
700 return (SET_ERROR(EFBIG));
703 if ((woff + n) > limit || woff > (limit - n))
706 /* Will this write extend the file length? */
707 write_eof = (woff + n > zp->z_size);
709 end_size = MAX(zp->z_size, woff + n);
712 * Write the file in reasonable size chunks. Each chunk is written
713 * in a separate transaction; this keeps the intent log records small
714 * and allows us to do more fine-grained space accounting.
718 woff = uio->uio_loffset;
719 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
720 zfs_owner_overquota(zsb, zp, B_TRUE)) {
722 dmu_return_arcbuf(abuf);
723 error = SET_ERROR(EDQUOT);
727 if (xuio && abuf == NULL) {
728 ASSERT(i_iov < iovcnt);
729 ASSERT3U(uio->uio_segflg, !=, UIO_BVEC);
731 abuf = dmu_xuio_arcbuf(xuio, i_iov);
732 dmu_xuio_clear(xuio, i_iov);
733 ASSERT((aiov->iov_base == abuf->b_data) ||
734 ((char *)aiov->iov_base - (char *)abuf->b_data +
735 aiov->iov_len == arc_buf_size(abuf)));
737 } else if (abuf == NULL && n >= max_blksz &&
738 woff >= zp->z_size &&
739 P2PHASE(woff, max_blksz) == 0 &&
740 zp->z_blksz == max_blksz) {
742 * This write covers a full block. "Borrow" a buffer
743 * from the dmu so that we can fill it before we enter
744 * a transaction. This avoids the possibility of
745 * holding up the transaction if the data copy hangs
746 * up on a pagefault (e.g., from an NFS server mapping).
750 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
752 ASSERT(abuf != NULL);
753 ASSERT(arc_buf_size(abuf) == max_blksz);
754 if ((error = uiocopy(abuf->b_data, max_blksz,
755 UIO_WRITE, uio, &cbytes))) {
756 dmu_return_arcbuf(abuf);
759 ASSERT(cbytes == max_blksz);
763 * Start a transaction.
765 tx = dmu_tx_create(zsb->z_os);
766 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
767 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
768 zfs_sa_upgrade_txholds(tx, zp);
769 error = dmu_tx_assign(tx, TXG_WAIT);
773 dmu_return_arcbuf(abuf);
778 * If zfs_range_lock() over-locked we grow the blocksize
779 * and then reduce the lock range. This will only happen
780 * on the first iteration since zfs_range_reduce() will
781 * shrink down r_len to the appropriate size.
783 if (rl->r_len == UINT64_MAX) {
786 if (zp->z_blksz > max_blksz) {
788 * File's blocksize is already larger than the
789 * "recordsize" property. Only let it grow to
790 * the next power of 2.
792 ASSERT(!ISP2(zp->z_blksz));
793 new_blksz = MIN(end_size,
794 1 << highbit64(zp->z_blksz));
796 new_blksz = MIN(end_size, max_blksz);
798 zfs_grow_blocksize(zp, new_blksz, tx);
799 zfs_range_reduce(rl, woff, n);
803 * XXX - should we really limit each write to z_max_blksz?
804 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
806 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
809 tx_bytes = uio->uio_resid;
810 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
812 tx_bytes -= uio->uio_resid;
815 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
817 * If this is not a full block write, but we are
818 * extending the file past EOF and this data starts
819 * block-aligned, use assign_arcbuf(). Otherwise,
820 * write via dmu_write().
822 if (tx_bytes < max_blksz && (!write_eof ||
823 aiov->iov_base != abuf->b_data)) {
825 dmu_write(zsb->z_os, zp->z_id, woff,
826 aiov->iov_len, aiov->iov_base, tx);
827 dmu_return_arcbuf(abuf);
828 xuio_stat_wbuf_copied();
830 ASSERT(xuio || tx_bytes == max_blksz);
831 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
834 ASSERT(tx_bytes <= uio->uio_resid);
835 uioskip(uio, tx_bytes);
838 if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT))
839 update_pages(ip, woff, tx_bytes, zsb->z_os, zp->z_id);
842 * If we made no progress, we're done. If we made even
843 * partial progress, update the znode and ZIL accordingly.
846 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
847 (void *)&zp->z_size, sizeof (uint64_t), tx);
854 * Clear Set-UID/Set-GID bits on successful write if not
855 * privileged and at least one of the excute bits is set.
857 * It would be nice to to this after all writes have
858 * been done, but that would still expose the ISUID/ISGID
859 * to another app after the partial write is committed.
861 * Note: we don't call zfs_fuid_map_id() here because
862 * user 0 is not an ephemeral uid.
864 mutex_enter(&zp->z_acl_lock);
865 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
866 (S_IXUSR >> 6))) != 0 &&
867 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
868 secpolicy_vnode_setid_retain(cr,
869 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
871 zp->z_mode &= ~(S_ISUID | S_ISGID);
872 newmode = zp->z_mode;
873 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zsb),
874 (void *)&newmode, sizeof (uint64_t), tx);
876 mutex_exit(&zp->z_acl_lock);
878 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
881 * Update the file size (zp_size) if it has changed;
882 * account for possible concurrent updates.
884 while ((end_size = zp->z_size) < uio->uio_loffset) {
885 (void) atomic_cas_64(&zp->z_size, end_size,
890 * If we are replaying and eof is non zero then force
891 * the file size to the specified eof. Note, there's no
892 * concurrency during replay.
894 if (zsb->z_replay && zsb->z_replay_eof != 0)
895 zp->z_size = zsb->z_replay_eof;
897 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
899 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag,
905 ASSERT(tx_bytes == nbytes);
909 uio_prefaultpages(MIN(n, max_blksz), uio);
912 zfs_inode_update(zp);
913 zfs_range_unlock(rl);
916 * If we're in replay mode, or we made no progress, return error.
917 * Otherwise, it's at least a partial write, so it's successful.
919 if (zsb->z_replay || uio->uio_resid == start_resid) {
924 if (ioflag & (FSYNC | FDSYNC) ||
925 zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
926 zil_commit(zilog, zp->z_id);
931 EXPORT_SYMBOL(zfs_write);
934 zfs_iput_async(struct inode *ip)
936 objset_t *os = ITOZSB(ip)->z_os;
938 ASSERT(atomic_read(&ip->i_count) > 0);
941 if (atomic_read(&ip->i_count) == 1)
942 taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os)),
943 (task_func_t *)iput, ip, TQ_SLEEP);
949 zfs_get_done(zgd_t *zgd, int error)
951 znode_t *zp = zgd->zgd_private;
954 dmu_buf_rele(zgd->zgd_db, zgd);
956 zfs_range_unlock(zgd->zgd_rl);
959 * Release the vnode asynchronously as we currently have the
960 * txg stopped from syncing.
962 zfs_iput_async(ZTOI(zp));
964 if (error == 0 && zgd->zgd_bp)
965 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
967 kmem_free(zgd, sizeof (zgd_t));
971 static int zil_fault_io = 0;
975 * Get data to generate a TX_WRITE intent log record.
978 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
981 objset_t *os = zsb->z_os;
983 uint64_t object = lr->lr_foid;
984 uint64_t offset = lr->lr_offset;
985 uint64_t size = lr->lr_length;
986 blkptr_t *bp = &lr->lr_blkptr;
995 * Nothing to do if the file has been removed
997 if (zfs_zget(zsb, object, &zp) != 0)
998 return (SET_ERROR(ENOENT));
999 if (zp->z_unlinked) {
1001 * Release the vnode asynchronously as we currently have the
1002 * txg stopped from syncing.
1004 zfs_iput_async(ZTOI(zp));
1005 return (SET_ERROR(ENOENT));
1008 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1009 zgd->zgd_zilog = zsb->z_log;
1010 zgd->zgd_private = zp;
1013 * Write records come in two flavors: immediate and indirect.
1014 * For small writes it's cheaper to store the data with the
1015 * log record (immediate); for large writes it's cheaper to
1016 * sync the data and get a pointer to it (indirect) so that
1017 * we don't have to write the data twice.
1019 if (buf != NULL) { /* immediate write */
1020 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset, size,
1022 /* test for truncation needs to be done while range locked */
1023 if (offset >= zp->z_size) {
1024 error = SET_ERROR(ENOENT);
1026 error = dmu_read(os, object, offset, size, buf,
1027 DMU_READ_NO_PREFETCH);
1029 ASSERT(error == 0 || error == ENOENT);
1030 } else { /* indirect write */
1032 * Have to lock the whole block to ensure when it's
1033 * written out and it's checksum is being calculated
1034 * that no one can change the data. We need to re-check
1035 * blocksize after we get the lock in case it's changed!
1040 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1042 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset,
1044 if (zp->z_blksz == size)
1047 zfs_range_unlock(zgd->zgd_rl);
1049 /* test for truncation needs to be done while range locked */
1050 if (lr->lr_offset >= zp->z_size)
1051 error = SET_ERROR(ENOENT);
1054 error = SET_ERROR(EIO);
1059 error = dmu_buf_hold(os, object, offset, zgd, &db,
1060 DMU_READ_NO_PREFETCH);
1063 blkptr_t *obp = dmu_buf_get_blkptr(db);
1065 ASSERT(BP_IS_HOLE(bp));
1072 ASSERT(db->db_offset == offset);
1073 ASSERT(db->db_size == size);
1075 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1077 ASSERT(error || lr->lr_length <= zp->z_blksz);
1080 * On success, we need to wait for the write I/O
1081 * initiated by dmu_sync() to complete before we can
1082 * release this dbuf. We will finish everything up
1083 * in the zfs_get_done() callback.
1088 if (error == EALREADY) {
1089 lr->lr_common.lrc_txtype = TX_WRITE2;
1095 zfs_get_done(zgd, error);
1102 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
1104 znode_t *zp = ITOZ(ip);
1105 zfs_sb_t *zsb = ITOZSB(ip);
1111 if (flag & V_ACE_MASK)
1112 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1114 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1119 EXPORT_SYMBOL(zfs_access);
1122 * Lookup an entry in a directory, or an extended attribute directory.
1123 * If it exists, return a held inode reference for it.
1125 * IN: dip - inode of directory to search.
1126 * nm - name of entry to lookup.
1127 * flags - LOOKUP_XATTR set if looking for an attribute.
1128 * cr - credentials of caller.
1129 * direntflags - directory lookup flags
1130 * realpnp - returned pathname.
1132 * OUT: ipp - inode of located entry, NULL if not found.
1134 * RETURN: 0 on success, error code on failure.
1141 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
1142 cred_t *cr, int *direntflags, pathname_t *realpnp)
1144 znode_t *zdp = ITOZ(dip);
1145 zfs_sb_t *zsb = ITOZSB(dip);
1149 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1151 if (!S_ISDIR(dip->i_mode)) {
1152 return (SET_ERROR(ENOTDIR));
1153 } else if (zdp->z_sa_hdl == NULL) {
1154 return (SET_ERROR(EIO));
1157 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1158 error = zfs_fastaccesschk_execute(zdp, cr);
1167 vnode_t *tvp = dnlc_lookup(dvp, nm);
1170 error = zfs_fastaccesschk_execute(zdp, cr);
1175 if (tvp == DNLC_NO_VNODE) {
1177 return (SET_ERROR(ENOENT));
1180 return (specvp_check(vpp, cr));
1183 #endif /* HAVE_DNLC */
1192 if (flags & LOOKUP_XATTR) {
1194 * We don't allow recursive attributes..
1195 * Maybe someday we will.
1197 if (zdp->z_pflags & ZFS_XATTR) {
1199 return (SET_ERROR(EINVAL));
1202 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1208 * Do we have permission to get into attribute directory?
1211 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1221 if (!S_ISDIR(dip->i_mode)) {
1223 return (SET_ERROR(ENOTDIR));
1227 * Check accessibility of directory.
1230 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1235 if (zsb->z_utf8 && u8_validate(nm, strlen(nm),
1236 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1238 return (SET_ERROR(EILSEQ));
1241 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1242 if ((error == 0) && (*ipp))
1243 zfs_inode_update(ITOZ(*ipp));
1248 EXPORT_SYMBOL(zfs_lookup);
1251 * Attempt to create a new entry in a directory. If the entry
1252 * already exists, truncate the file if permissible, else return
1253 * an error. Return the ip of the created or trunc'd file.
1255 * IN: dip - inode of directory to put new file entry in.
1256 * name - name of new file entry.
1257 * vap - attributes of new file.
1258 * excl - flag indicating exclusive or non-exclusive mode.
1259 * mode - mode to open file with.
1260 * cr - credentials of caller.
1261 * flag - large file flag [UNUSED].
1262 * vsecp - ACL to be set
1264 * OUT: ipp - inode of created or trunc'd entry.
1266 * RETURN: 0 on success, error code on failure.
1269 * dip - ctime|mtime updated if new entry created
1270 * ip - ctime|mtime always, atime if new
1275 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1276 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1278 znode_t *zp, *dzp = ITOZ(dip);
1279 zfs_sb_t *zsb = ITOZSB(dip);
1287 zfs_acl_ids_t acl_ids;
1288 boolean_t fuid_dirtied;
1289 boolean_t have_acl = B_FALSE;
1290 boolean_t waited = B_FALSE;
1293 * If we have an ephemeral id, ACL, or XVATTR then
1294 * make sure file system is at proper version
1300 if (zsb->z_use_fuids == B_FALSE &&
1301 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1302 return (SET_ERROR(EINVAL));
1309 if (zsb->z_utf8 && u8_validate(name, strlen(name),
1310 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1312 return (SET_ERROR(EILSEQ));
1315 if (vap->va_mask & ATTR_XVATTR) {
1316 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1317 crgetuid(cr), cr, vap->va_mode)) != 0) {
1325 if (*name == '\0') {
1327 * Null component name refers to the directory itself.
1334 /* possible igrab(zp) */
1337 if (flag & FIGNORECASE)
1340 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1344 zfs_acl_ids_free(&acl_ids);
1345 if (strcmp(name, "..") == 0)
1346 error = SET_ERROR(EISDIR);
1356 * Create a new file object and update the directory
1359 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1361 zfs_acl_ids_free(&acl_ids);
1366 * We only support the creation of regular files in
1367 * extended attribute directories.
1370 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1372 zfs_acl_ids_free(&acl_ids);
1373 error = SET_ERROR(EINVAL);
1377 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1378 cr, vsecp, &acl_ids)) != 0)
1382 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1383 zfs_acl_ids_free(&acl_ids);
1384 error = SET_ERROR(EDQUOT);
1388 tx = dmu_tx_create(os);
1390 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1391 ZFS_SA_BASE_ATTR_SIZE);
1393 fuid_dirtied = zsb->z_fuid_dirty;
1395 zfs_fuid_txhold(zsb, tx);
1396 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1397 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1398 if (!zsb->z_use_sa &&
1399 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1400 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1401 0, acl_ids.z_aclp->z_acl_bytes);
1403 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1405 zfs_dirent_unlock(dl);
1406 if (error == ERESTART) {
1412 zfs_acl_ids_free(&acl_ids);
1417 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1420 zfs_fuid_sync(zsb, tx);
1422 (void) zfs_link_create(dl, zp, tx, ZNEW);
1423 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1424 if (flag & FIGNORECASE)
1426 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1427 vsecp, acl_ids.z_fuidp, vap);
1428 zfs_acl_ids_free(&acl_ids);
1431 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1434 zfs_acl_ids_free(&acl_ids);
1438 * A directory entry already exists for this name.
1441 * Can't truncate an existing file if in exclusive mode.
1444 error = SET_ERROR(EEXIST);
1448 * Can't open a directory for writing.
1450 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1451 error = SET_ERROR(EISDIR);
1455 * Verify requested access to file.
1457 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1461 mutex_enter(&dzp->z_lock);
1463 mutex_exit(&dzp->z_lock);
1466 * Truncate regular files if requested.
1468 if (S_ISREG(ZTOI(zp)->i_mode) &&
1469 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1470 /* we can't hold any locks when calling zfs_freesp() */
1471 zfs_dirent_unlock(dl);
1473 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1479 zfs_dirent_unlock(dl);
1485 zfs_inode_update(dzp);
1486 zfs_inode_update(zp);
1490 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1491 zil_commit(zilog, 0);
1496 EXPORT_SYMBOL(zfs_create);
1499 * Remove an entry from a directory.
1501 * IN: dip - inode of directory to remove entry from.
1502 * name - name of entry to remove.
1503 * cr - credentials of caller.
1505 * RETURN: 0 if success
1506 * error code if failure
1510 * ip - ctime (if nlink > 0)
1513 uint64_t null_xattr = 0;
1517 zfs_remove(struct inode *dip, char *name, cred_t *cr, int flags)
1519 znode_t *zp, *dzp = ITOZ(dip);
1522 zfs_sb_t *zsb = ITOZSB(dip);
1524 uint64_t acl_obj, xattr_obj;
1525 uint64_t xattr_obj_unlinked = 0;
1529 boolean_t may_delete_now, delete_now = FALSE;
1530 boolean_t unlinked, toobig = FALSE;
1532 pathname_t *realnmp = NULL;
1536 boolean_t waited = B_FALSE;
1542 if (flags & FIGNORECASE) {
1552 * Attempt to lock directory; fail if entry doesn't exist.
1554 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1564 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1569 * Need to use rmdir for removing directories.
1571 if (S_ISDIR(ip->i_mode)) {
1572 error = SET_ERROR(EPERM);
1578 dnlc_remove(dvp, realnmp->pn_buf);
1580 dnlc_remove(dvp, name);
1581 #endif /* HAVE_DNLC */
1583 mutex_enter(&zp->z_lock);
1584 may_delete_now = atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped);
1585 mutex_exit(&zp->z_lock);
1588 * We may delete the znode now, or we may put it in the unlinked set;
1589 * it depends on whether we're the last link, and on whether there are
1590 * other holds on the inode. So we dmu_tx_hold() the right things to
1591 * allow for either case.
1594 tx = dmu_tx_create(zsb->z_os);
1595 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1596 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1597 zfs_sa_upgrade_txholds(tx, zp);
1598 zfs_sa_upgrade_txholds(tx, dzp);
1599 if (may_delete_now) {
1600 toobig = zp->z_size > zp->z_blksz * zfs_delete_blocks;
1601 /* if the file is too big, only hold_free a token amount */
1602 dmu_tx_hold_free(tx, zp->z_id, 0,
1603 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1606 /* are there any extended attributes? */
1607 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1608 &xattr_obj, sizeof (xattr_obj));
1609 if (error == 0 && xattr_obj) {
1610 error = zfs_zget(zsb, xattr_obj, &xzp);
1612 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1613 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1616 mutex_enter(&zp->z_lock);
1617 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1618 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1619 mutex_exit(&zp->z_lock);
1621 /* charge as an update -- would be nice not to charge at all */
1622 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
1625 * Mark this transaction as typically resulting in a net free of space
1627 dmu_tx_mark_netfree(tx);
1629 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1631 zfs_dirent_unlock(dl);
1635 if (error == ERESTART) {
1649 * Remove the directory entry.
1651 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1660 * Hold z_lock so that we can make sure that the ACL obj
1661 * hasn't changed. Could have been deleted due to
1664 mutex_enter(&zp->z_lock);
1665 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1666 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1667 delete_now = may_delete_now && !toobig &&
1668 atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped) &&
1669 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1674 if (xattr_obj_unlinked) {
1675 ASSERT3U(xzp->z_links, ==, 2);
1676 mutex_enter(&xzp->z_lock);
1677 xzp->z_unlinked = 1;
1679 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zsb),
1680 &xzp->z_links, sizeof (xzp->z_links), tx);
1681 ASSERT3U(error, ==, 0);
1682 mutex_exit(&xzp->z_lock);
1683 zfs_unlinked_add(xzp, tx);
1686 error = sa_remove(zp->z_sa_hdl,
1687 SA_ZPL_XATTR(zsb), tx);
1689 error = sa_update(zp->z_sa_hdl,
1690 SA_ZPL_XATTR(zsb), &null_xattr,
1691 sizeof (uint64_t), tx);
1695 * Add to the unlinked set because a new reference could be
1696 * taken concurrently resulting in a deferred destruction.
1698 zfs_unlinked_add(zp, tx);
1699 mutex_exit(&zp->z_lock);
1700 zfs_inode_update(zp);
1702 } else if (unlinked) {
1703 mutex_exit(&zp->z_lock);
1704 zfs_unlinked_add(zp, tx);
1708 if (flags & FIGNORECASE)
1710 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1717 zfs_dirent_unlock(dl);
1718 zfs_inode_update(dzp);
1721 zfs_inode_update(zp);
1726 zfs_inode_update(xzp);
1727 zfs_iput_async(ZTOI(xzp));
1730 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1731 zil_commit(zilog, 0);
1736 EXPORT_SYMBOL(zfs_remove);
1739 * Create a new directory and insert it into dip using the name
1740 * provided. Return a pointer to the inserted directory.
1742 * IN: dip - inode of directory to add subdir to.
1743 * dirname - name of new directory.
1744 * vap - attributes of new directory.
1745 * cr - credentials of caller.
1746 * vsecp - ACL to be set
1748 * OUT: ipp - inode of created directory.
1750 * RETURN: 0 if success
1751 * error code if failure
1754 * dip - ctime|mtime updated
1755 * ipp - ctime|mtime|atime updated
1759 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1760 cred_t *cr, int flags, vsecattr_t *vsecp)
1762 znode_t *zp, *dzp = ITOZ(dip);
1763 zfs_sb_t *zsb = ITOZSB(dip);
1771 gid_t gid = crgetgid(cr);
1772 zfs_acl_ids_t acl_ids;
1773 boolean_t fuid_dirtied;
1774 boolean_t waited = B_FALSE;
1776 ASSERT(S_ISDIR(vap->va_mode));
1779 * If we have an ephemeral id, ACL, or XVATTR then
1780 * make sure file system is at proper version
1784 if (zsb->z_use_fuids == B_FALSE &&
1785 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1786 return (SET_ERROR(EINVAL));
1792 if (dzp->z_pflags & ZFS_XATTR) {
1794 return (SET_ERROR(EINVAL));
1797 if (zsb->z_utf8 && u8_validate(dirname,
1798 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1800 return (SET_ERROR(EILSEQ));
1802 if (flags & FIGNORECASE)
1805 if (vap->va_mask & ATTR_XVATTR) {
1806 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1807 crgetuid(cr), cr, vap->va_mode)) != 0) {
1813 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1814 vsecp, &acl_ids)) != 0) {
1819 * First make sure the new directory doesn't exist.
1821 * Existence is checked first to make sure we don't return
1822 * EACCES instead of EEXIST which can cause some applications
1828 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1830 zfs_acl_ids_free(&acl_ids);
1835 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
1836 zfs_acl_ids_free(&acl_ids);
1837 zfs_dirent_unlock(dl);
1842 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1843 zfs_acl_ids_free(&acl_ids);
1844 zfs_dirent_unlock(dl);
1846 return (SET_ERROR(EDQUOT));
1850 * Add a new entry to the directory.
1852 tx = dmu_tx_create(zsb->z_os);
1853 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1854 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1855 fuid_dirtied = zsb->z_fuid_dirty;
1857 zfs_fuid_txhold(zsb, tx);
1858 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1859 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1860 acl_ids.z_aclp->z_acl_bytes);
1863 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1864 ZFS_SA_BASE_ATTR_SIZE);
1866 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1868 zfs_dirent_unlock(dl);
1869 if (error == ERESTART) {
1875 zfs_acl_ids_free(&acl_ids);
1884 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1887 zfs_fuid_sync(zsb, tx);
1890 * Now put new name in parent dir.
1892 (void) zfs_link_create(dl, zp, tx, ZNEW);
1896 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1897 if (flags & FIGNORECASE)
1899 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1900 acl_ids.z_fuidp, vap);
1902 zfs_acl_ids_free(&acl_ids);
1906 zfs_dirent_unlock(dl);
1908 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1909 zil_commit(zilog, 0);
1911 zfs_inode_update(dzp);
1912 zfs_inode_update(zp);
1916 EXPORT_SYMBOL(zfs_mkdir);
1919 * Remove a directory subdir entry. If the current working
1920 * directory is the same as the subdir to be removed, the
1923 * IN: dip - inode of directory to remove from.
1924 * name - name of directory to be removed.
1925 * cwd - inode of current working directory.
1926 * cr - credentials of caller.
1927 * flags - case flags
1929 * RETURN: 0 on success, error code on failure.
1932 * dip - ctime|mtime updated
1936 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
1939 znode_t *dzp = ITOZ(dip);
1942 zfs_sb_t *zsb = ITOZSB(dip);
1948 boolean_t waited = B_FALSE;
1954 if (flags & FIGNORECASE)
1960 * Attempt to lock directory; fail if entry doesn't exist.
1962 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1970 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1974 if (!S_ISDIR(ip->i_mode)) {
1975 error = SET_ERROR(ENOTDIR);
1980 error = SET_ERROR(EINVAL);
1985 * Grab a lock on the directory to make sure that noone is
1986 * trying to add (or lookup) entries while we are removing it.
1988 rw_enter(&zp->z_name_lock, RW_WRITER);
1991 * Grab a lock on the parent pointer to make sure we play well
1992 * with the treewalk and directory rename code.
1994 rw_enter(&zp->z_parent_lock, RW_WRITER);
1996 tx = dmu_tx_create(zsb->z_os);
1997 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1998 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1999 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
2000 zfs_sa_upgrade_txholds(tx, zp);
2001 zfs_sa_upgrade_txholds(tx, dzp);
2002 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2004 rw_exit(&zp->z_parent_lock);
2005 rw_exit(&zp->z_name_lock);
2006 zfs_dirent_unlock(dl);
2008 if (error == ERESTART) {
2019 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2022 uint64_t txtype = TX_RMDIR;
2023 if (flags & FIGNORECASE)
2025 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2030 rw_exit(&zp->z_parent_lock);
2031 rw_exit(&zp->z_name_lock);
2033 zfs_dirent_unlock(dl);
2035 zfs_inode_update(dzp);
2036 zfs_inode_update(zp);
2039 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
2040 zil_commit(zilog, 0);
2045 EXPORT_SYMBOL(zfs_rmdir);
2048 * Read as many directory entries as will fit into the provided
2049 * dirent buffer from the given directory cursor position.
2051 * IN: ip - inode of directory to read.
2052 * dirent - buffer for directory entries.
2054 * OUT: dirent - filler buffer of directory entries.
2056 * RETURN: 0 if success
2057 * error code if failure
2060 * ip - atime updated
2062 * Note that the low 4 bits of the cookie returned by zap is always zero.
2063 * This allows us to use the low range for "special" directory entries:
2064 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2065 * we use the offset 2 for the '.zfs' directory.
2069 zfs_readdir(struct inode *ip, struct dir_context *ctx, cred_t *cr)
2071 znode_t *zp = ITOZ(ip);
2072 zfs_sb_t *zsb = ITOZSB(ip);
2075 zap_attribute_t zap;
2081 uint64_t offset; /* must be unsigned; checks for < 1 */
2086 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zsb),
2087 &parent, sizeof (parent))) != 0)
2091 * Quit if directory has been removed (posix)
2099 prefetch = zp->z_zn_prefetch;
2102 * Initialize the iterator cursor.
2106 * Start iteration from the beginning of the directory.
2108 zap_cursor_init(&zc, os, zp->z_id);
2111 * The offset is a serialized cursor.
2113 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2117 * Transform to file-system independent format
2122 * Special case `.', `..', and `.zfs'.
2125 (void) strcpy(zap.za_name, ".");
2126 zap.za_normalization_conflict = 0;
2129 } else if (offset == 1) {
2130 (void) strcpy(zap.za_name, "..");
2131 zap.za_normalization_conflict = 0;
2134 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2135 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2136 zap.za_normalization_conflict = 0;
2137 objnum = ZFSCTL_INO_ROOT;
2143 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2144 if (error == ENOENT)
2151 * Allow multiple entries provided the first entry is
2152 * the object id. Non-zpl consumers may safely make
2153 * use of the additional space.
2155 * XXX: This should be a feature flag for compatibility
2157 if (zap.za_integer_length != 8 ||
2158 zap.za_num_integers == 0) {
2159 cmn_err(CE_WARN, "zap_readdir: bad directory "
2160 "entry, obj = %lld, offset = %lld, "
2161 "length = %d, num = %lld\n",
2162 (u_longlong_t)zp->z_id,
2163 (u_longlong_t)offset,
2164 zap.za_integer_length,
2165 (u_longlong_t)zap.za_num_integers);
2166 error = SET_ERROR(ENXIO);
2170 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2171 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2174 done = !dir_emit(ctx, zap.za_name, strlen(zap.za_name),
2179 /* Prefetch znode */
2181 dmu_prefetch(os, objnum, 0, 0, 0,
2182 ZIO_PRIORITY_SYNC_READ);
2186 * Move to the next entry, fill in the previous offset.
2188 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2189 zap_cursor_advance(&zc);
2190 offset = zap_cursor_serialize(&zc);
2196 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2199 zap_cursor_fini(&zc);
2200 if (error == ENOENT)
2207 EXPORT_SYMBOL(zfs_readdir);
2209 ulong_t zfs_fsync_sync_cnt = 4;
2212 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2214 znode_t *zp = ITOZ(ip);
2215 zfs_sb_t *zsb = ITOZSB(ip);
2217 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2219 if (zsb->z_os->os_sync != ZFS_SYNC_DISABLED) {
2222 zil_commit(zsb->z_log, zp->z_id);
2225 tsd_set(zfs_fsyncer_key, NULL);
2229 EXPORT_SYMBOL(zfs_fsync);
2233 * Get the requested file attributes and place them in the provided
2236 * IN: ip - inode of file.
2237 * vap - va_mask identifies requested attributes.
2238 * If ATTR_XVATTR set, then optional attrs are requested
2239 * flags - ATTR_NOACLCHECK (CIFS server context)
2240 * cr - credentials of caller.
2242 * OUT: vap - attribute values.
2244 * RETURN: 0 (always succeeds)
2248 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2250 znode_t *zp = ITOZ(ip);
2251 zfs_sb_t *zsb = ITOZSB(ip);
2254 uint64_t atime[2], mtime[2], ctime[2];
2255 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2256 xoptattr_t *xoap = NULL;
2257 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2258 sa_bulk_attr_t bulk[3];
2264 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2266 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zsb), NULL, &atime, 16);
2267 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
2268 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
2270 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2276 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2277 * Also, if we are the owner don't bother, since owner should
2278 * always be allowed to read basic attributes of file.
2280 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2281 (vap->va_uid != crgetuid(cr))) {
2282 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2290 * Return all attributes. It's cheaper to provide the answer
2291 * than to determine whether we were asked the question.
2294 mutex_enter(&zp->z_lock);
2295 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2296 vap->va_mode = zp->z_mode;
2297 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2298 vap->va_nodeid = zp->z_id;
2299 if ((zp->z_id == zsb->z_root) && zfs_show_ctldir(zp))
2300 links = zp->z_links + 1;
2302 links = zp->z_links;
2303 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2304 vap->va_size = i_size_read(ip);
2305 vap->va_rdev = ip->i_rdev;
2306 vap->va_seq = ip->i_generation;
2309 * Add in any requested optional attributes and the create time.
2310 * Also set the corresponding bits in the returned attribute bitmap.
2312 if ((xoap = xva_getxoptattr(xvap)) != NULL && zsb->z_use_fuids) {
2313 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2315 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2316 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2319 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2320 xoap->xoa_readonly =
2321 ((zp->z_pflags & ZFS_READONLY) != 0);
2322 XVA_SET_RTN(xvap, XAT_READONLY);
2325 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2327 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2328 XVA_SET_RTN(xvap, XAT_SYSTEM);
2331 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2333 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2334 XVA_SET_RTN(xvap, XAT_HIDDEN);
2337 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2338 xoap->xoa_nounlink =
2339 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2340 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2343 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2344 xoap->xoa_immutable =
2345 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2346 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2349 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2350 xoap->xoa_appendonly =
2351 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2352 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2355 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2357 ((zp->z_pflags & ZFS_NODUMP) != 0);
2358 XVA_SET_RTN(xvap, XAT_NODUMP);
2361 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2363 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2364 XVA_SET_RTN(xvap, XAT_OPAQUE);
2367 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2368 xoap->xoa_av_quarantined =
2369 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2370 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2373 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2374 xoap->xoa_av_modified =
2375 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2376 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2379 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2380 S_ISREG(ip->i_mode)) {
2381 zfs_sa_get_scanstamp(zp, xvap);
2384 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2387 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zsb),
2388 times, sizeof (times));
2389 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2390 XVA_SET_RTN(xvap, XAT_CREATETIME);
2393 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2394 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2395 XVA_SET_RTN(xvap, XAT_REPARSE);
2397 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2398 xoap->xoa_generation = ip->i_generation;
2399 XVA_SET_RTN(xvap, XAT_GEN);
2402 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2404 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2405 XVA_SET_RTN(xvap, XAT_OFFLINE);
2408 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2410 ((zp->z_pflags & ZFS_SPARSE) != 0);
2411 XVA_SET_RTN(xvap, XAT_SPARSE);
2415 ZFS_TIME_DECODE(&vap->va_atime, atime);
2416 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2417 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2419 mutex_exit(&zp->z_lock);
2421 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2423 if (zp->z_blksz == 0) {
2425 * Block size hasn't been set; suggest maximal I/O transfers.
2427 vap->va_blksize = zsb->z_max_blksz;
2433 EXPORT_SYMBOL(zfs_getattr);
2436 * Get the basic file attributes and place them in the provided kstat
2437 * structure. The inode is assumed to be the authoritative source
2438 * for most of the attributes. However, the znode currently has the
2439 * authoritative atime, blksize, and block count.
2441 * IN: ip - inode of file.
2443 * OUT: sp - kstat values.
2445 * RETURN: 0 (always succeeds)
2449 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2451 znode_t *zp = ITOZ(ip);
2452 zfs_sb_t *zsb = ITOZSB(ip);
2454 u_longlong_t nblocks;
2459 mutex_enter(&zp->z_lock);
2461 generic_fillattr(ip, sp);
2463 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2464 sp->blksize = blksize;
2465 sp->blocks = nblocks;
2467 if (unlikely(zp->z_blksz == 0)) {
2469 * Block size hasn't been set; suggest maximal I/O transfers.
2471 sp->blksize = zsb->z_max_blksz;
2474 mutex_exit(&zp->z_lock);
2477 * Required to prevent NFS client from detecting different inode
2478 * numbers of snapshot root dentry before and after snapshot mount.
2480 if (zsb->z_issnap) {
2481 if (ip->i_sb->s_root->d_inode == ip)
2482 sp->ino = ZFSCTL_INO_SNAPDIRS -
2483 dmu_objset_id(zsb->z_os);
2490 EXPORT_SYMBOL(zfs_getattr_fast);
2493 * Set the file attributes to the values contained in the
2496 * IN: ip - inode of file to be modified.
2497 * vap - new attribute values.
2498 * If ATTR_XVATTR set, then optional attrs are being set
2499 * flags - ATTR_UTIME set if non-default time values provided.
2500 * - ATTR_NOACLCHECK (CIFS context only).
2501 * cr - credentials of caller.
2503 * RETURN: 0 if success
2504 * error code if failure
2507 * ip - ctime updated, mtime updated if size changed.
2511 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2513 znode_t *zp = ITOZ(ip);
2514 zfs_sb_t *zsb = ITOZSB(ip);
2518 xvattr_t *tmpxvattr;
2519 uint_t mask = vap->va_mask;
2520 uint_t saved_mask = 0;
2523 uint64_t new_uid, new_gid;
2525 uint64_t mtime[2], ctime[2], atime[2];
2527 int need_policy = FALSE;
2529 zfs_fuid_info_t *fuidp = NULL;
2530 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2533 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2534 boolean_t fuid_dirtied = B_FALSE;
2535 sa_bulk_attr_t *bulk, *xattr_bulk;
2536 int count = 0, xattr_count = 0;
2547 * Make sure that if we have ephemeral uid/gid or xvattr specified
2548 * that file system is at proper version level
2551 if (zsb->z_use_fuids == B_FALSE &&
2552 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2553 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2554 (mask & ATTR_XVATTR))) {
2556 return (SET_ERROR(EINVAL));
2559 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2561 return (SET_ERROR(EISDIR));
2564 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2566 return (SET_ERROR(EINVAL));
2570 * If this is an xvattr_t, then get a pointer to the structure of
2571 * optional attributes. If this is NULL, then we have a vattr_t.
2573 xoap = xva_getxoptattr(xvap);
2575 tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP);
2576 xva_init(tmpxvattr);
2578 bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2579 xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2582 * Immutable files can only alter immutable bit and atime
2584 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2585 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2586 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2591 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2597 * Verify timestamps doesn't overflow 32 bits.
2598 * ZFS can handle large timestamps, but 32bit syscalls can't
2599 * handle times greater than 2039. This check should be removed
2600 * once large timestamps are fully supported.
2602 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2603 if (((mask & ATTR_ATIME) &&
2604 TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2605 ((mask & ATTR_MTIME) &&
2606 TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2616 /* Can this be moved to before the top label? */
2617 if (zfs_is_readonly(zsb)) {
2623 * First validate permissions
2626 if (mask & ATTR_SIZE) {
2627 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2632 * XXX - Note, we are not providing any open
2633 * mode flags here (like FNDELAY), so we may
2634 * block if there are locks present... this
2635 * should be addressed in openat().
2637 /* XXX - would it be OK to generate a log record here? */
2638 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2643 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2644 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2645 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2646 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2647 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2648 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2649 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2650 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2651 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2655 if (mask & (ATTR_UID|ATTR_GID)) {
2656 int idmask = (mask & (ATTR_UID|ATTR_GID));
2661 * NOTE: even if a new mode is being set,
2662 * we may clear S_ISUID/S_ISGID bits.
2665 if (!(mask & ATTR_MODE))
2666 vap->va_mode = zp->z_mode;
2669 * Take ownership or chgrp to group we are a member of
2672 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
2673 take_group = (mask & ATTR_GID) &&
2674 zfs_groupmember(zsb, vap->va_gid, cr);
2677 * If both ATTR_UID and ATTR_GID are set then take_owner and
2678 * take_group must both be set in order to allow taking
2681 * Otherwise, send the check through secpolicy_vnode_setattr()
2685 if (((idmask == (ATTR_UID|ATTR_GID)) &&
2686 take_owner && take_group) ||
2687 ((idmask == ATTR_UID) && take_owner) ||
2688 ((idmask == ATTR_GID) && take_group)) {
2689 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2690 skipaclchk, cr) == 0) {
2692 * Remove setuid/setgid for non-privileged users
2694 (void) secpolicy_setid_clear(vap, cr);
2695 trim_mask = (mask & (ATTR_UID|ATTR_GID));
2704 mutex_enter(&zp->z_lock);
2705 oldva.va_mode = zp->z_mode;
2706 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2707 if (mask & ATTR_XVATTR) {
2709 * Update xvattr mask to include only those attributes
2710 * that are actually changing.
2712 * the bits will be restored prior to actually setting
2713 * the attributes so the caller thinks they were set.
2715 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2716 if (xoap->xoa_appendonly !=
2717 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2720 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2721 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
2725 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2726 if (xoap->xoa_nounlink !=
2727 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2730 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2731 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
2735 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2736 if (xoap->xoa_immutable !=
2737 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2740 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2741 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
2745 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2746 if (xoap->xoa_nodump !=
2747 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2750 XVA_CLR_REQ(xvap, XAT_NODUMP);
2751 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
2755 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2756 if (xoap->xoa_av_modified !=
2757 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2760 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2761 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
2765 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2766 if ((!S_ISREG(ip->i_mode) &&
2767 xoap->xoa_av_quarantined) ||
2768 xoap->xoa_av_quarantined !=
2769 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2772 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2773 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
2777 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2778 mutex_exit(&zp->z_lock);
2783 if (need_policy == FALSE &&
2784 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2785 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2790 mutex_exit(&zp->z_lock);
2792 if (mask & ATTR_MODE) {
2793 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2794 err = secpolicy_setid_setsticky_clear(ip, vap,
2799 trim_mask |= ATTR_MODE;
2807 * If trim_mask is set then take ownership
2808 * has been granted or write_acl is present and user
2809 * has the ability to modify mode. In that case remove
2810 * UID|GID and or MODE from mask so that
2811 * secpolicy_vnode_setattr() doesn't revoke it.
2815 saved_mask = vap->va_mask;
2816 vap->va_mask &= ~trim_mask;
2818 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
2819 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2824 vap->va_mask |= saved_mask;
2828 * secpolicy_vnode_setattr, or take ownership may have
2831 mask = vap->va_mask;
2833 if ((mask & (ATTR_UID | ATTR_GID))) {
2834 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
2835 &xattr_obj, sizeof (xattr_obj));
2837 if (err == 0 && xattr_obj) {
2838 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
2842 if (mask & ATTR_UID) {
2843 new_uid = zfs_fuid_create(zsb,
2844 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2845 if (new_uid != zp->z_uid &&
2846 zfs_fuid_overquota(zsb, B_FALSE, new_uid)) {
2854 if (mask & ATTR_GID) {
2855 new_gid = zfs_fuid_create(zsb, (uint64_t)vap->va_gid,
2856 cr, ZFS_GROUP, &fuidp);
2857 if (new_gid != zp->z_gid &&
2858 zfs_fuid_overquota(zsb, B_TRUE, new_gid)) {
2866 tx = dmu_tx_create(zsb->z_os);
2868 if (mask & ATTR_MODE) {
2869 uint64_t pmode = zp->z_mode;
2871 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2873 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
2875 mutex_enter(&zp->z_lock);
2876 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2878 * Are we upgrading ACL from old V0 format
2881 if (zsb->z_version >= ZPL_VERSION_FUID &&
2882 zfs_znode_acl_version(zp) ==
2883 ZFS_ACL_VERSION_INITIAL) {
2884 dmu_tx_hold_free(tx, acl_obj, 0,
2886 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2887 0, aclp->z_acl_bytes);
2889 dmu_tx_hold_write(tx, acl_obj, 0,
2892 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2893 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2894 0, aclp->z_acl_bytes);
2896 mutex_exit(&zp->z_lock);
2897 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2899 if ((mask & ATTR_XVATTR) &&
2900 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2901 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2903 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2907 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
2910 fuid_dirtied = zsb->z_fuid_dirty;
2912 zfs_fuid_txhold(zsb, tx);
2914 zfs_sa_upgrade_txholds(tx, zp);
2916 err = dmu_tx_assign(tx, TXG_WAIT);
2922 * Set each attribute requested.
2923 * We group settings according to the locks they need to acquire.
2925 * Note: you cannot set ctime directly, although it will be
2926 * updated as a side-effect of calling this function.
2930 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2931 mutex_enter(&zp->z_acl_lock);
2932 mutex_enter(&zp->z_lock);
2934 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
2935 &zp->z_pflags, sizeof (zp->z_pflags));
2938 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2939 mutex_enter(&attrzp->z_acl_lock);
2940 mutex_enter(&attrzp->z_lock);
2941 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2942 SA_ZPL_FLAGS(zsb), NULL, &attrzp->z_pflags,
2943 sizeof (attrzp->z_pflags));
2946 if (mask & (ATTR_UID|ATTR_GID)) {
2948 if (mask & ATTR_UID) {
2949 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zsb), NULL,
2950 &new_uid, sizeof (new_uid));
2951 zp->z_uid = new_uid;
2953 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2954 SA_ZPL_UID(zsb), NULL, &new_uid,
2956 attrzp->z_uid = new_uid;
2960 if (mask & ATTR_GID) {
2961 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zsb),
2962 NULL, &new_gid, sizeof (new_gid));
2963 zp->z_gid = new_gid;
2965 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2966 SA_ZPL_GID(zsb), NULL, &new_gid,
2968 attrzp->z_gid = new_gid;
2971 if (!(mask & ATTR_MODE)) {
2972 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb),
2973 NULL, &new_mode, sizeof (new_mode));
2974 new_mode = zp->z_mode;
2976 err = zfs_acl_chown_setattr(zp);
2979 err = zfs_acl_chown_setattr(attrzp);
2984 if (mask & ATTR_MODE) {
2985 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb), NULL,
2986 &new_mode, sizeof (new_mode));
2987 zp->z_mode = new_mode;
2988 ASSERT3P(aclp, !=, NULL);
2989 err = zfs_aclset_common(zp, aclp, cr, tx);
2991 if (zp->z_acl_cached)
2992 zfs_acl_free(zp->z_acl_cached);
2993 zp->z_acl_cached = aclp;
2998 if ((mask & ATTR_ATIME) || zp->z_atime_dirty) {
2999 zp->z_atime_dirty = 0;
3000 ZFS_TIME_ENCODE(&ip->i_atime, atime);
3001 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zsb), NULL,
3002 &atime, sizeof (atime));
3005 if (mask & ATTR_MTIME) {
3006 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3007 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL,
3008 mtime, sizeof (mtime));
3011 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3012 if (mask & ATTR_SIZE && !(mask & ATTR_MTIME)) {
3013 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb),
3014 NULL, mtime, sizeof (mtime));
3015 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
3016 &ctime, sizeof (ctime));
3017 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
3018 } else if (mask != 0) {
3019 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
3020 &ctime, sizeof (ctime));
3021 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime);
3023 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3024 SA_ZPL_CTIME(zsb), NULL,
3025 &ctime, sizeof (ctime));
3026 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3031 * Do this after setting timestamps to prevent timestamp
3032 * update from toggling bit
3035 if (xoap && (mask & ATTR_XVATTR)) {
3038 * restore trimmed off masks
3039 * so that return masks can be set for caller.
3042 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
3043 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3045 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
3046 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3048 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
3049 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3051 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
3052 XVA_SET_REQ(xvap, XAT_NODUMP);
3054 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
3055 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3057 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
3058 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3061 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3062 ASSERT(S_ISREG(ip->i_mode));
3064 zfs_xvattr_set(zp, xvap, tx);
3068 zfs_fuid_sync(zsb, tx);
3071 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3073 mutex_exit(&zp->z_lock);
3074 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3075 mutex_exit(&zp->z_acl_lock);
3078 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3079 mutex_exit(&attrzp->z_acl_lock);
3080 mutex_exit(&attrzp->z_lock);
3083 if (err == 0 && attrzp) {
3084 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3095 zfs_fuid_info_free(fuidp);
3101 if (err == ERESTART)
3104 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3106 zfs_inode_update(zp);
3110 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3111 zil_commit(zilog, 0);
3114 kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * 7);
3115 kmem_free(bulk, sizeof (sa_bulk_attr_t) * 7);
3116 kmem_free(tmpxvattr, sizeof (xvattr_t));
3120 EXPORT_SYMBOL(zfs_setattr);
3122 typedef struct zfs_zlock {
3123 krwlock_t *zl_rwlock; /* lock we acquired */
3124 znode_t *zl_znode; /* znode we held */
3125 struct zfs_zlock *zl_next; /* next in list */
3129 * Drop locks and release vnodes that were held by zfs_rename_lock().
3132 zfs_rename_unlock(zfs_zlock_t **zlpp)
3136 while ((zl = *zlpp) != NULL) {
3137 if (zl->zl_znode != NULL)
3138 iput(ZTOI(zl->zl_znode));
3139 rw_exit(zl->zl_rwlock);
3140 *zlpp = zl->zl_next;
3141 kmem_free(zl, sizeof (*zl));
3146 * Search back through the directory tree, using the ".." entries.
3147 * Lock each directory in the chain to prevent concurrent renames.
3148 * Fail any attempt to move a directory into one of its own descendants.
3149 * XXX - z_parent_lock can overlap with map or grow locks
3152 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3156 uint64_t rootid = ZTOZSB(zp)->z_root;
3157 uint64_t oidp = zp->z_id;
3158 krwlock_t *rwlp = &szp->z_parent_lock;
3159 krw_t rw = RW_WRITER;
3162 * First pass write-locks szp and compares to zp->z_id.
3163 * Later passes read-lock zp and compare to zp->z_parent.
3166 if (!rw_tryenter(rwlp, rw)) {
3168 * Another thread is renaming in this path.
3169 * Note that if we are a WRITER, we don't have any
3170 * parent_locks held yet.
3172 if (rw == RW_READER && zp->z_id > szp->z_id) {
3174 * Drop our locks and restart
3176 zfs_rename_unlock(&zl);
3180 rwlp = &szp->z_parent_lock;
3185 * Wait for other thread to drop its locks
3191 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3192 zl->zl_rwlock = rwlp;
3193 zl->zl_znode = NULL;
3194 zl->zl_next = *zlpp;
3197 if (oidp == szp->z_id) /* We're a descendant of szp */
3198 return (SET_ERROR(EINVAL));
3200 if (oidp == rootid) /* We've hit the top */
3203 if (rw == RW_READER) { /* i.e. not the first pass */
3204 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3209 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3210 &oidp, sizeof (oidp));
3211 rwlp = &zp->z_parent_lock;
3214 } while (zp->z_id != sdzp->z_id);
3220 * Move an entry from the provided source directory to the target
3221 * directory. Change the entry name as indicated.
3223 * IN: sdip - Source directory containing the "old entry".
3224 * snm - Old entry name.
3225 * tdip - Target directory to contain the "new entry".
3226 * tnm - New entry name.
3227 * cr - credentials of caller.
3228 * flags - case flags
3230 * RETURN: 0 on success, error code on failure.
3233 * sdip,tdip - ctime|mtime updated
3237 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3238 cred_t *cr, int flags)
3240 znode_t *tdzp, *szp, *tzp;
3241 znode_t *sdzp = ITOZ(sdip);
3242 zfs_sb_t *zsb = ITOZSB(sdip);
3244 zfs_dirlock_t *sdl, *tdl;
3247 int cmp, serr, terr;
3250 boolean_t waited = B_FALSE;
3253 ZFS_VERIFY_ZP(sdzp);
3257 ZFS_VERIFY_ZP(tdzp);
3260 * We check i_sb because snapshots and the ctldir must have different
3263 if (tdip->i_sb != sdip->i_sb || zfsctl_is_node(tdip)) {
3265 return (SET_ERROR(EXDEV));
3268 if (zsb->z_utf8 && u8_validate(tnm,
3269 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3271 return (SET_ERROR(EILSEQ));
3274 if (flags & FIGNORECASE)
3283 * This is to prevent the creation of links into attribute space
3284 * by renaming a linked file into/outof an attribute directory.
3285 * See the comment in zfs_link() for why this is considered bad.
3287 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3289 return (SET_ERROR(EINVAL));
3293 * Lock source and target directory entries. To prevent deadlock,
3294 * a lock ordering must be defined. We lock the directory with
3295 * the smallest object id first, or if it's a tie, the one with
3296 * the lexically first name.
3298 if (sdzp->z_id < tdzp->z_id) {
3300 } else if (sdzp->z_id > tdzp->z_id) {
3304 * First compare the two name arguments without
3305 * considering any case folding.
3307 int nofold = (zsb->z_norm & ~U8_TEXTPREP_TOUPPER);
3309 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3310 ASSERT(error == 0 || !zsb->z_utf8);
3313 * POSIX: "If the old argument and the new argument
3314 * both refer to links to the same existing file,
3315 * the rename() function shall return successfully
3316 * and perform no other action."
3322 * If the file system is case-folding, then we may
3323 * have some more checking to do. A case-folding file
3324 * system is either supporting mixed case sensitivity
3325 * access or is completely case-insensitive. Note
3326 * that the file system is always case preserving.
3328 * In mixed sensitivity mode case sensitive behavior
3329 * is the default. FIGNORECASE must be used to
3330 * explicitly request case insensitive behavior.
3332 * If the source and target names provided differ only
3333 * by case (e.g., a request to rename 'tim' to 'Tim'),
3334 * we will treat this as a special case in the
3335 * case-insensitive mode: as long as the source name
3336 * is an exact match, we will allow this to proceed as
3337 * a name-change request.
3339 if ((zsb->z_case == ZFS_CASE_INSENSITIVE ||
3340 (zsb->z_case == ZFS_CASE_MIXED &&
3341 flags & FIGNORECASE)) &&
3342 u8_strcmp(snm, tnm, 0, zsb->z_norm, U8_UNICODE_LATEST,
3345 * case preserving rename request, require exact
3354 * If the source and destination directories are the same, we should
3355 * grab the z_name_lock of that directory only once.
3359 rw_enter(&sdzp->z_name_lock, RW_READER);
3363 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3364 ZEXISTS | zflg, NULL, NULL);
3365 terr = zfs_dirent_lock(&tdl,
3366 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3368 terr = zfs_dirent_lock(&tdl,
3369 tdzp, tnm, &tzp, zflg, NULL, NULL);
3370 serr = zfs_dirent_lock(&sdl,
3371 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3377 * Source entry invalid or not there.
3380 zfs_dirent_unlock(tdl);
3386 rw_exit(&sdzp->z_name_lock);
3388 if (strcmp(snm, "..") == 0)
3394 zfs_dirent_unlock(sdl);
3398 rw_exit(&sdzp->z_name_lock);
3400 if (strcmp(tnm, "..") == 0)
3407 * Must have write access at the source to remove the old entry
3408 * and write access at the target to create the new entry.
3409 * Note that if target and source are the same, this can be
3410 * done in a single check.
3413 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3416 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3418 * Check to make sure rename is valid.
3419 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3421 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3426 * Does target exist?
3430 * Source and target must be the same type.
3432 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3433 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3434 error = SET_ERROR(ENOTDIR);
3438 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3439 error = SET_ERROR(EISDIR);
3444 * POSIX dictates that when the source and target
3445 * entries refer to the same file object, rename
3446 * must do nothing and exit without error.
3448 if (szp->z_id == tzp->z_id) {
3454 tx = dmu_tx_create(zsb->z_os);
3455 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3456 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3457 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3458 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3460 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3461 zfs_sa_upgrade_txholds(tx, tdzp);
3464 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3465 zfs_sa_upgrade_txholds(tx, tzp);
3468 zfs_sa_upgrade_txholds(tx, szp);
3469 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
3470 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3473 zfs_rename_unlock(&zl);
3474 zfs_dirent_unlock(sdl);
3475 zfs_dirent_unlock(tdl);
3478 rw_exit(&sdzp->z_name_lock);
3483 if (error == ERESTART) {
3494 if (tzp) /* Attempt to remove the existing target */
3495 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3498 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3500 szp->z_pflags |= ZFS_AV_MODIFIED;
3502 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zsb),
3503 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3506 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3508 zfs_log_rename(zilog, tx, TX_RENAME |
3509 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3510 sdl->dl_name, tdzp, tdl->dl_name, szp);
3513 * At this point, we have successfully created
3514 * the target name, but have failed to remove
3515 * the source name. Since the create was done
3516 * with the ZRENAMING flag, there are
3517 * complications; for one, the link count is
3518 * wrong. The easiest way to deal with this
3519 * is to remove the newly created target, and
3520 * return the original error. This must
3521 * succeed; fortunately, it is very unlikely to
3522 * fail, since we just created it.
3524 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3525 ZRENAMING, NULL), ==, 0);
3533 zfs_rename_unlock(&zl);
3535 zfs_dirent_unlock(sdl);
3536 zfs_dirent_unlock(tdl);
3538 zfs_inode_update(sdzp);
3540 rw_exit(&sdzp->z_name_lock);
3543 zfs_inode_update(tdzp);
3545 zfs_inode_update(szp);
3548 zfs_inode_update(tzp);
3552 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3553 zil_commit(zilog, 0);
3558 EXPORT_SYMBOL(zfs_rename);
3561 * Insert the indicated symbolic reference entry into the directory.
3563 * IN: dip - Directory to contain new symbolic link.
3564 * link - Name for new symlink entry.
3565 * vap - Attributes of new entry.
3566 * target - Target path of new symlink.
3568 * cr - credentials of caller.
3569 * flags - case flags
3571 * RETURN: 0 on success, error code on failure.
3574 * dip - ctime|mtime updated
3578 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
3579 struct inode **ipp, cred_t *cr, int flags)
3581 znode_t *zp, *dzp = ITOZ(dip);
3584 zfs_sb_t *zsb = ITOZSB(dip);
3586 uint64_t len = strlen(link);
3589 zfs_acl_ids_t acl_ids;
3590 boolean_t fuid_dirtied;
3591 uint64_t txtype = TX_SYMLINK;
3592 boolean_t waited = B_FALSE;
3594 ASSERT(S_ISLNK(vap->va_mode));
3600 if (zsb->z_utf8 && u8_validate(name, strlen(name),
3601 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3603 return (SET_ERROR(EILSEQ));
3605 if (flags & FIGNORECASE)
3608 if (len > MAXPATHLEN) {
3610 return (SET_ERROR(ENAMETOOLONG));
3613 if ((error = zfs_acl_ids_create(dzp, 0,
3614 vap, cr, NULL, &acl_ids)) != 0) {
3622 * Attempt to lock directory; fail if entry already exists.
3624 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3626 zfs_acl_ids_free(&acl_ids);
3631 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3632 zfs_acl_ids_free(&acl_ids);
3633 zfs_dirent_unlock(dl);
3638 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
3639 zfs_acl_ids_free(&acl_ids);
3640 zfs_dirent_unlock(dl);
3642 return (SET_ERROR(EDQUOT));
3644 tx = dmu_tx_create(zsb->z_os);
3645 fuid_dirtied = zsb->z_fuid_dirty;
3646 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3647 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3648 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3649 ZFS_SA_BASE_ATTR_SIZE + len);
3650 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3651 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3652 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3653 acl_ids.z_aclp->z_acl_bytes);
3656 zfs_fuid_txhold(zsb, tx);
3657 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3659 zfs_dirent_unlock(dl);
3660 if (error == ERESTART) {
3666 zfs_acl_ids_free(&acl_ids);
3673 * Create a new object for the symlink.
3674 * for version 4 ZPL datsets the symlink will be an SA attribute
3676 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3679 zfs_fuid_sync(zsb, tx);
3681 mutex_enter(&zp->z_lock);
3683 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zsb),
3686 zfs_sa_symlink(zp, link, len, tx);
3687 mutex_exit(&zp->z_lock);
3690 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
3691 &zp->z_size, sizeof (zp->z_size), tx);
3693 * Insert the new object into the directory.
3695 (void) zfs_link_create(dl, zp, tx, ZNEW);
3697 if (flags & FIGNORECASE)
3699 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3701 zfs_inode_update(dzp);
3702 zfs_inode_update(zp);
3704 zfs_acl_ids_free(&acl_ids);
3708 zfs_dirent_unlock(dl);
3712 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3713 zil_commit(zilog, 0);
3718 EXPORT_SYMBOL(zfs_symlink);
3721 * Return, in the buffer contained in the provided uio structure,
3722 * the symbolic path referred to by ip.
3724 * IN: ip - inode of symbolic link
3725 * uio - structure to contain the link path.
3726 * cr - credentials of caller.
3728 * RETURN: 0 if success
3729 * error code if failure
3732 * ip - atime updated
3736 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
3738 znode_t *zp = ITOZ(ip);
3739 zfs_sb_t *zsb = ITOZSB(ip);
3745 mutex_enter(&zp->z_lock);
3747 error = sa_lookup_uio(zp->z_sa_hdl,
3748 SA_ZPL_SYMLINK(zsb), uio);
3750 error = zfs_sa_readlink(zp, uio);
3751 mutex_exit(&zp->z_lock);
3756 EXPORT_SYMBOL(zfs_readlink);
3759 * Insert a new entry into directory tdip referencing sip.
3761 * IN: tdip - Directory to contain new entry.
3762 * sip - inode of new entry.
3763 * name - name of new entry.
3764 * cr - credentials of caller.
3766 * RETURN: 0 if success
3767 * error code if failure
3770 * tdip - ctime|mtime updated
3771 * sip - ctime updated
3775 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr,
3778 znode_t *dzp = ITOZ(tdip);
3780 zfs_sb_t *zsb = ITOZSB(tdip);
3788 boolean_t waited = B_FALSE;
3790 ASSERT(S_ISDIR(tdip->i_mode));
3797 * POSIX dictates that we return EPERM here.
3798 * Better choices include ENOTSUP or EISDIR.
3800 if (S_ISDIR(sip->i_mode)) {
3802 return (SET_ERROR(EPERM));
3809 * We check i_sb because snapshots and the ctldir must have different
3812 if (sip->i_sb != tdip->i_sb || zfsctl_is_node(sip)) {
3814 return (SET_ERROR(EXDEV));
3817 /* Prevent links to .zfs/shares files */
3819 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zsb),
3820 &parent, sizeof (uint64_t))) != 0) {
3824 if (parent == zsb->z_shares_dir) {
3826 return (SET_ERROR(EPERM));
3829 if (zsb->z_utf8 && u8_validate(name,
3830 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3832 return (SET_ERROR(EILSEQ));
3834 if (flags & FIGNORECASE)
3838 * We do not support links between attributes and non-attributes
3839 * because of the potential security risk of creating links
3840 * into "normal" file space in order to circumvent restrictions
3841 * imposed in attribute space.
3843 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3845 return (SET_ERROR(EINVAL));
3848 owner = zfs_fuid_map_id(zsb, szp->z_uid, cr, ZFS_OWNER);
3849 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3851 return (SET_ERROR(EPERM));
3854 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3861 * Attempt to lock directory; fail if entry already exists.
3863 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3869 tx = dmu_tx_create(zsb->z_os);
3870 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3871 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3872 zfs_sa_upgrade_txholds(tx, szp);
3873 zfs_sa_upgrade_txholds(tx, dzp);
3874 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3876 zfs_dirent_unlock(dl);
3877 if (error == ERESTART) {
3888 error = zfs_link_create(dl, szp, tx, 0);
3891 uint64_t txtype = TX_LINK;
3892 if (flags & FIGNORECASE)
3894 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3899 zfs_dirent_unlock(dl);
3901 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3902 zil_commit(zilog, 0);
3904 zfs_inode_update(dzp);
3905 zfs_inode_update(szp);
3909 EXPORT_SYMBOL(zfs_link);
3912 zfs_putpage_commit_cb(void *arg)
3914 struct page *pp = arg;
3917 end_page_writeback(pp);
3921 * Push a page out to disk, once the page is on stable storage the
3922 * registered commit callback will be run as notification of completion.
3924 * IN: ip - page mapped for inode.
3925 * pp - page to push (page is locked)
3926 * wbc - writeback control data
3928 * RETURN: 0 if success
3929 * error code if failure
3932 * ip - ctime|mtime updated
3936 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
3938 znode_t *zp = ITOZ(ip);
3939 zfs_sb_t *zsb = ITOZSB(ip);
3947 uint64_t mtime[2], ctime[2];
3948 sa_bulk_attr_t bulk[3];
3950 struct address_space *mapping;
3955 ASSERT(PageLocked(pp));
3957 pgoff = page_offset(pp); /* Page byte-offset in file */
3958 offset = i_size_read(ip); /* File length in bytes */
3959 pglen = MIN(PAGE_SIZE, /* Page length in bytes */
3960 P2ROUNDUP(offset, PAGE_SIZE)-pgoff);
3962 /* Page is beyond end of file */
3963 if (pgoff >= offset) {
3969 /* Truncate page length to end of file */
3970 if (pgoff + pglen > offset)
3971 pglen = offset - pgoff;
3975 * FIXME: Allow mmap writes past its quota. The correct fix
3976 * is to register a page_mkwrite() handler to count the page
3977 * against its quota when it is about to be dirtied.
3979 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
3980 zfs_owner_overquota(zsb, zp, B_TRUE)) {
3986 * The ordering here is critical and must adhere to the following
3987 * rules in order to avoid deadlocking in either zfs_read() or
3988 * zfs_free_range() due to a lock inversion.
3990 * 1) The page must be unlocked prior to acquiring the range lock.
3991 * This is critical because zfs_read() calls find_lock_page()
3992 * which may block on the page lock while holding the range lock.
3994 * 2) Before setting or clearing write back on a page the range lock
3995 * must be held in order to prevent a lock inversion with the
3996 * zfs_free_range() function.
3998 * This presents a problem because upon entering this function the
3999 * page lock is already held. To safely acquire the range lock the
4000 * page lock must be dropped. This creates a window where another
4001 * process could truncate, invalidate, dirty, or write out the page.
4003 * Therefore, after successfully reacquiring the range and page locks
4004 * the current page state is checked. In the common case everything
4005 * will be as is expected and it can be written out. However, if
4006 * the page state has changed it must be handled accordingly.
4008 mapping = pp->mapping;
4009 redirty_page_for_writepage(wbc, pp);
4012 rl = zfs_range_lock(&zp->z_range_lock, pgoff, pglen, RL_WRITER);
4015 /* Page mapping changed or it was no longer dirty, we're done */
4016 if (unlikely((mapping != pp->mapping) || !PageDirty(pp))) {
4018 zfs_range_unlock(rl);
4023 /* Another process started write block if required */
4024 if (PageWriteback(pp)) {
4026 zfs_range_unlock(rl);
4028 if (wbc->sync_mode != WB_SYNC_NONE)
4029 wait_on_page_writeback(pp);
4035 /* Clear the dirty flag the required locks are held */
4036 if (!clear_page_dirty_for_io(pp)) {
4038 zfs_range_unlock(rl);
4044 * Counterpart for redirty_page_for_writepage() above. This page
4045 * was in fact not skipped and should not be counted as if it were.
4047 wbc->pages_skipped--;
4048 set_page_writeback(pp);
4051 tx = dmu_tx_create(zsb->z_os);
4052 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
4053 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4054 zfs_sa_upgrade_txholds(tx, zp);
4056 err = dmu_tx_assign(tx, TXG_NOWAIT);
4058 if (err == ERESTART)
4062 __set_page_dirty_nobuffers(pp);
4064 end_page_writeback(pp);
4065 zfs_range_unlock(rl);
4071 ASSERT3U(pglen, <=, PAGE_SIZE);
4072 dmu_write(zsb->z_os, zp->z_id, pgoff, pglen, va, tx);
4075 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
4076 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
4077 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zsb), NULL, &zp->z_pflags, 8);
4079 /* Preserve the mtime and ctime provided by the inode */
4080 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4081 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4082 zp->z_atime_dirty = 0;
4085 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4087 zfs_log_write(zsb->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0,
4088 zfs_putpage_commit_cb, pp);
4091 zfs_range_unlock(rl);
4093 if (wbc->sync_mode != WB_SYNC_NONE) {
4095 * Note that this is rarely called under writepages(), because
4096 * writepages() normally handles the entire commit for
4097 * performance reasons.
4099 if (zsb->z_log != NULL)
4100 zil_commit(zsb->z_log, zp->z_id);
4108 * Update the system attributes when the inode has been dirtied. For the
4109 * moment we only update the mode, atime, mtime, and ctime.
4112 zfs_dirty_inode(struct inode *ip, int flags)
4114 znode_t *zp = ITOZ(ip);
4115 zfs_sb_t *zsb = ITOZSB(ip);
4117 uint64_t mode, atime[2], mtime[2], ctime[2];
4118 sa_bulk_attr_t bulk[4];
4122 if (zfs_is_readonly(zsb) || dmu_objset_is_snapshot(zsb->z_os))
4130 * This is the lazytime semantic indroduced in Linux 4.0
4131 * This flag will only be called from update_time when lazytime is set.
4132 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4133 * Fortunately mtime and ctime are managed within ZFS itself, so we
4134 * only need to dirty atime.
4136 if (flags == I_DIRTY_TIME) {
4137 zp->z_atime_dirty = 1;
4142 tx = dmu_tx_create(zsb->z_os);
4144 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4145 zfs_sa_upgrade_txholds(tx, zp);
4147 error = dmu_tx_assign(tx, TXG_WAIT);
4153 mutex_enter(&zp->z_lock);
4154 zp->z_atime_dirty = 0;
4156 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zsb), NULL, &mode, 8);
4157 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zsb), NULL, &atime, 16);
4158 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
4159 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
4161 /* Preserve the mode, mtime and ctime provided by the inode */
4162 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4163 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4164 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4169 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4170 mutex_exit(&zp->z_lock);
4177 EXPORT_SYMBOL(zfs_dirty_inode);
4181 zfs_inactive(struct inode *ip)
4183 znode_t *zp = ITOZ(ip);
4184 zfs_sb_t *zsb = ITOZSB(ip);
4187 int need_unlock = 0;
4189 /* Only read lock if we haven't already write locked, e.g. rollback */
4190 if (!RW_WRITE_HELD(&zsb->z_teardown_inactive_lock)) {
4192 rw_enter(&zsb->z_teardown_inactive_lock, RW_READER);
4194 if (zp->z_sa_hdl == NULL) {
4196 rw_exit(&zsb->z_teardown_inactive_lock);
4200 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4201 dmu_tx_t *tx = dmu_tx_create(zsb->z_os);
4203 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4204 zfs_sa_upgrade_txholds(tx, zp);
4205 error = dmu_tx_assign(tx, TXG_WAIT);
4209 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4210 mutex_enter(&zp->z_lock);
4211 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zsb),
4212 (void *)&atime, sizeof (atime), tx);
4213 zp->z_atime_dirty = 0;
4214 mutex_exit(&zp->z_lock);
4221 rw_exit(&zsb->z_teardown_inactive_lock);
4223 EXPORT_SYMBOL(zfs_inactive);
4226 * Bounds-check the seek operation.
4228 * IN: ip - inode seeking within
4229 * ooff - old file offset
4230 * noffp - pointer to new file offset
4231 * ct - caller context
4233 * RETURN: 0 if success
4234 * EINVAL if new offset invalid
4238 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
4240 if (S_ISDIR(ip->i_mode))
4242 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4244 EXPORT_SYMBOL(zfs_seek);
4247 * Fill pages with data from the disk.
4250 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
4252 znode_t *zp = ITOZ(ip);
4253 zfs_sb_t *zsb = ITOZSB(ip);
4255 struct page *cur_pp;
4256 u_offset_t io_off, total;
4263 io_len = nr_pages << PAGE_SHIFT;
4264 i_size = i_size_read(ip);
4265 io_off = page_offset(pl[0]);
4267 if (io_off + io_len > i_size)
4268 io_len = i_size - io_off;
4271 * Iterate over list of pages and read each page individually.
4274 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4277 cur_pp = pl[page_idx++];
4279 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4283 /* convert checksum errors into IO errors */
4285 err = SET_ERROR(EIO);
4294 * Uses zfs_fillpage to read data from the file and fill the pages.
4296 * IN: ip - inode of file to get data from.
4297 * pl - list of pages to read
4298 * nr_pages - number of pages to read
4300 * RETURN: 0 on success, error code on failure.
4303 * vp - atime updated
4307 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
4309 znode_t *zp = ITOZ(ip);
4310 zfs_sb_t *zsb = ITOZSB(ip);
4319 err = zfs_fillpage(ip, pl, nr_pages);
4324 EXPORT_SYMBOL(zfs_getpage);
4327 * Check ZFS specific permissions to memory map a section of a file.
4329 * IN: ip - inode of the file to mmap
4331 * addrp - start address in memory region
4332 * len - length of memory region
4333 * vm_flags- address flags
4335 * RETURN: 0 if success
4336 * error code if failure
4340 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4341 unsigned long vm_flags)
4343 znode_t *zp = ITOZ(ip);
4344 zfs_sb_t *zsb = ITOZSB(ip);
4349 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4350 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4352 return (SET_ERROR(EPERM));
4355 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4356 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4358 return (SET_ERROR(EACCES));
4361 if (off < 0 || len > MAXOFFSET_T - off) {
4363 return (SET_ERROR(ENXIO));
4369 EXPORT_SYMBOL(zfs_map);
4372 * convoff - converts the given data (start, whence) to the
4376 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4381 if ((lckdat->l_whence == 2) || (whence == 2)) {
4382 if ((error = zfs_getattr(ip, &vap, 0, CRED()) != 0))
4386 switch (lckdat->l_whence) {
4388 lckdat->l_start += offset;
4391 lckdat->l_start += vap.va_size;
4396 return (SET_ERROR(EINVAL));
4399 if (lckdat->l_start < 0)
4400 return (SET_ERROR(EINVAL));
4404 lckdat->l_start -= offset;
4407 lckdat->l_start -= vap.va_size;
4412 return (SET_ERROR(EINVAL));
4415 lckdat->l_whence = (short)whence;
4420 * Free or allocate space in a file. Currently, this function only
4421 * supports the `F_FREESP' command. However, this command is somewhat
4422 * misnamed, as its functionality includes the ability to allocate as
4423 * well as free space.
4425 * IN: ip - inode of file to free data in.
4426 * cmd - action to take (only F_FREESP supported).
4427 * bfp - section of file to free/alloc.
4428 * flag - current file open mode flags.
4429 * offset - current file offset.
4430 * cr - credentials of caller [UNUSED].
4432 * RETURN: 0 on success, error code on failure.
4435 * ip - ctime|mtime updated
4439 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4440 offset_t offset, cred_t *cr)
4442 znode_t *zp = ITOZ(ip);
4443 zfs_sb_t *zsb = ITOZSB(ip);
4450 if (cmd != F_FREESP) {
4452 return (SET_ERROR(EINVAL));
4456 * Callers might not be able to detect properly that we are read-only,
4457 * so check it explicitly here.
4459 if (zfs_is_readonly(zsb)) {
4461 return (SET_ERROR(EROFS));
4464 if ((error = convoff(ip, bfp, 0, offset))) {
4469 if (bfp->l_len < 0) {
4471 return (SET_ERROR(EINVAL));
4475 * Permissions aren't checked on Solaris because on this OS
4476 * zfs_space() can only be called with an opened file handle.
4477 * On Linux we can get here through truncate_range() which
4478 * operates directly on inodes, so we need to check access rights.
4480 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4486 len = bfp->l_len; /* 0 means from off to end of file */
4488 error = zfs_freesp(zp, off, len, flag, TRUE);
4493 EXPORT_SYMBOL(zfs_space);
4497 zfs_fid(struct inode *ip, fid_t *fidp)
4499 znode_t *zp = ITOZ(ip);
4500 zfs_sb_t *zsb = ITOZSB(ip);
4503 uint64_t object = zp->z_id;
4510 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zsb),
4511 &gen64, sizeof (uint64_t))) != 0) {
4516 gen = (uint32_t)gen64;
4518 size = (zsb->z_parent != zsb) ? LONG_FID_LEN : SHORT_FID_LEN;
4519 if (fidp->fid_len < size) {
4520 fidp->fid_len = size;
4522 return (SET_ERROR(ENOSPC));
4525 zfid = (zfid_short_t *)fidp;
4527 zfid->zf_len = size;
4529 for (i = 0; i < sizeof (zfid->zf_object); i++)
4530 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4532 /* Must have a non-zero generation number to distinguish from .zfs */
4535 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4536 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4538 if (size == LONG_FID_LEN) {
4539 uint64_t objsetid = dmu_objset_id(zsb->z_os);
4542 zlfid = (zfid_long_t *)fidp;
4544 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4545 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4547 /* XXX - this should be the generation number for the objset */
4548 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4549 zlfid->zf_setgen[i] = 0;
4555 EXPORT_SYMBOL(zfs_fid);
4559 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4561 znode_t *zp = ITOZ(ip);
4562 zfs_sb_t *zsb = ITOZSB(ip);
4564 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4568 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4573 EXPORT_SYMBOL(zfs_getsecattr);
4577 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4579 znode_t *zp = ITOZ(ip);
4580 zfs_sb_t *zsb = ITOZSB(ip);
4582 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4583 zilog_t *zilog = zsb->z_log;
4588 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4590 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
4591 zil_commit(zilog, 0);
4596 EXPORT_SYMBOL(zfs_setsecattr);
4598 #ifdef HAVE_UIO_ZEROCOPY
4600 * Tunable, both must be a power of 2.
4602 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4603 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4604 * an arcbuf for a partial block read
4606 int zcr_blksz_min = (1 << 10); /* 1K */
4607 int zcr_blksz_max = (1 << 17); /* 128K */
4611 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
4613 znode_t *zp = ITOZ(ip);
4614 zfs_sb_t *zsb = ITOZSB(ip);
4615 int max_blksz = zsb->z_max_blksz;
4616 uio_t *uio = &xuio->xu_uio;
4617 ssize_t size = uio->uio_resid;
4618 offset_t offset = uio->uio_loffset;
4623 int preamble, postamble;
4625 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4626 return (SET_ERROR(EINVAL));
4633 * Loan out an arc_buf for write if write size is bigger than
4634 * max_blksz, and the file's block size is also max_blksz.
4637 if (size < blksz || zp->z_blksz != blksz) {
4639 return (SET_ERROR(EINVAL));
4642 * Caller requests buffers for write before knowing where the
4643 * write offset might be (e.g. NFS TCP write).
4648 preamble = P2PHASE(offset, blksz);
4650 preamble = blksz - preamble;
4655 postamble = P2PHASE(size, blksz);
4658 fullblk = size / blksz;
4659 (void) dmu_xuio_init(xuio,
4660 (preamble != 0) + fullblk + (postamble != 0));
4663 * Have to fix iov base/len for partial buffers. They
4664 * currently represent full arc_buf's.
4667 /* data begins in the middle of the arc_buf */
4668 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4671 (void) dmu_xuio_add(xuio, abuf,
4672 blksz - preamble, preamble);
4675 for (i = 0; i < fullblk; i++) {
4676 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4679 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
4683 /* data ends in the middle of the arc_buf */
4684 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4687 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
4692 * Loan out an arc_buf for read if the read size is larger than
4693 * the current file block size. Block alignment is not
4694 * considered. Partial arc_buf will be loaned out for read.
4696 blksz = zp->z_blksz;
4697 if (blksz < zcr_blksz_min)
4698 blksz = zcr_blksz_min;
4699 if (blksz > zcr_blksz_max)
4700 blksz = zcr_blksz_max;
4701 /* avoid potential complexity of dealing with it */
4702 if (blksz > max_blksz) {
4704 return (SET_ERROR(EINVAL));
4707 maxsize = zp->z_size - uio->uio_loffset;
4713 return (SET_ERROR(EINVAL));
4718 return (SET_ERROR(EINVAL));
4721 uio->uio_extflg = UIO_XUIO;
4722 XUIO_XUZC_RW(xuio) = ioflag;
4729 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
4733 int ioflag = XUIO_XUZC_RW(xuio);
4735 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
4737 i = dmu_xuio_cnt(xuio);
4739 abuf = dmu_xuio_arcbuf(xuio, i);
4741 * if abuf == NULL, it must be a write buffer
4742 * that has been returned in zfs_write().
4745 dmu_return_arcbuf(abuf);
4746 ASSERT(abuf || ioflag == UIO_WRITE);
4749 dmu_xuio_fini(xuio);
4752 #endif /* HAVE_UIO_ZEROCOPY */
4754 #if defined(_KERNEL) && defined(HAVE_SPL)
4755 module_param(zfs_delete_blocks, ulong, 0644);
4756 MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async");
4757 module_param(zfs_read_chunk_size, long, 0644);
4758 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");