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;
601 sa_bulk_attr_t bulk[4];
602 uint64_t mtime[2], ctime[2];
604 #ifdef HAVE_UIO_ZEROCOPY
606 const iovec_t *iovp = uio->uio_iov;
607 ASSERTV(int iovcnt = uio->uio_iovcnt);
611 * Fasttrack empty write
617 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
623 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
624 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
625 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb), NULL, &zp->z_size, 8);
626 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
630 * Callers might not be able to detect properly that we are read-only,
631 * so check it explicitly here.
633 if (zfs_is_readonly(zsb)) {
635 return (SET_ERROR(EROFS));
639 * If immutable or not appending then return EPERM
641 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
642 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
643 (uio->uio_loffset < zp->z_size))) {
645 return (SET_ERROR(EPERM));
651 * Validate file offset
653 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
656 return (SET_ERROR(EINVAL));
660 * Pre-fault the pages to ensure slow (eg NFS) pages
662 * Skip this if uio contains loaned arc_buf.
664 #ifdef HAVE_UIO_ZEROCOPY
665 if ((uio->uio_extflg == UIO_XUIO) &&
666 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
667 xuio = (xuio_t *)uio;
670 uio_prefaultpages(MIN(n, max_blksz), uio);
673 * If in append mode, set the io offset pointer to eof.
675 if (ioflag & FAPPEND) {
677 * Obtain an appending range lock to guarantee file append
678 * semantics. We reset the write offset once we have the lock.
680 rl = zfs_range_lock(&zp->z_range_lock, 0, n, RL_APPEND);
682 if (rl->r_len == UINT64_MAX) {
684 * We overlocked the file because this write will cause
685 * the file block size to increase.
686 * Note that zp_size cannot change with this lock held.
690 uio->uio_loffset = woff;
693 * Note that if the file block size will change as a result of
694 * this write, then this range lock will lock the entire file
695 * so that we can re-write the block safely.
697 rl = zfs_range_lock(&zp->z_range_lock, woff, n, RL_WRITER);
701 zfs_range_unlock(rl);
703 return (SET_ERROR(EFBIG));
706 if ((woff + n) > limit || woff > (limit - n))
709 /* Will this write extend the file length? */
710 write_eof = (woff + n > zp->z_size);
712 end_size = MAX(zp->z_size, woff + n);
715 * Write the file in reasonable size chunks. Each chunk is written
716 * in a separate transaction; this keeps the intent log records small
717 * and allows us to do more fine-grained space accounting.
721 woff = uio->uio_loffset;
722 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
723 zfs_owner_overquota(zsb, zp, B_TRUE)) {
725 dmu_return_arcbuf(abuf);
726 error = SET_ERROR(EDQUOT);
730 if (xuio && abuf == NULL) {
731 #ifdef HAVE_UIO_ZEROCOPY
732 ASSERT(i_iov < iovcnt);
733 ASSERT3U(uio->uio_segflg, !=, UIO_BVEC);
735 abuf = dmu_xuio_arcbuf(xuio, i_iov);
736 dmu_xuio_clear(xuio, i_iov);
737 ASSERT((aiov->iov_base == abuf->b_data) ||
738 ((char *)aiov->iov_base - (char *)abuf->b_data +
739 aiov->iov_len == arc_buf_size(abuf)));
742 } else if (abuf == NULL && n >= max_blksz &&
743 woff >= zp->z_size &&
744 P2PHASE(woff, max_blksz) == 0 &&
745 zp->z_blksz == max_blksz) {
747 * This write covers a full block. "Borrow" a buffer
748 * from the dmu so that we can fill it before we enter
749 * a transaction. This avoids the possibility of
750 * holding up the transaction if the data copy hangs
751 * up on a pagefault (e.g., from an NFS server mapping).
755 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
757 ASSERT(abuf != NULL);
758 ASSERT(arc_buf_size(abuf) == max_blksz);
759 if ((error = uiocopy(abuf->b_data, max_blksz,
760 UIO_WRITE, uio, &cbytes))) {
761 dmu_return_arcbuf(abuf);
764 ASSERT(cbytes == max_blksz);
768 * Start a transaction.
770 tx = dmu_tx_create(zsb->z_os);
771 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
772 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
773 zfs_sa_upgrade_txholds(tx, zp);
774 error = dmu_tx_assign(tx, TXG_WAIT);
778 dmu_return_arcbuf(abuf);
783 * If zfs_range_lock() over-locked we grow the blocksize
784 * and then reduce the lock range. This will only happen
785 * on the first iteration since zfs_range_reduce() will
786 * shrink down r_len to the appropriate size.
788 if (rl->r_len == UINT64_MAX) {
791 if (zp->z_blksz > max_blksz) {
793 * File's blocksize is already larger than the
794 * "recordsize" property. Only let it grow to
795 * the next power of 2.
797 ASSERT(!ISP2(zp->z_blksz));
798 new_blksz = MIN(end_size,
799 1 << highbit64(zp->z_blksz));
801 new_blksz = MIN(end_size, max_blksz);
803 zfs_grow_blocksize(zp, new_blksz, tx);
804 zfs_range_reduce(rl, woff, n);
808 * XXX - should we really limit each write to z_max_blksz?
809 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
811 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
814 tx_bytes = uio->uio_resid;
815 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
817 tx_bytes -= uio->uio_resid;
820 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
822 * If this is not a full block write, but we are
823 * extending the file past EOF and this data starts
824 * block-aligned, use assign_arcbuf(). Otherwise,
825 * write via dmu_write().
827 if (tx_bytes < max_blksz && (!write_eof ||
828 aiov->iov_base != abuf->b_data)) {
830 dmu_write(zsb->z_os, zp->z_id, woff,
831 aiov->iov_len, aiov->iov_base, tx);
832 dmu_return_arcbuf(abuf);
833 xuio_stat_wbuf_copied();
835 ASSERT(xuio || tx_bytes == max_blksz);
836 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
839 ASSERT(tx_bytes <= uio->uio_resid);
840 uioskip(uio, tx_bytes);
843 if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT))
844 update_pages(ip, woff, tx_bytes, zsb->z_os, zp->z_id);
847 * If we made no progress, we're done. If we made even
848 * partial progress, update the znode and ZIL accordingly.
851 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
852 (void *)&zp->z_size, sizeof (uint64_t), tx);
859 * Clear Set-UID/Set-GID bits on successful write if not
860 * privileged and at least one of the excute bits is set.
862 * It would be nice to to this after all writes have
863 * been done, but that would still expose the ISUID/ISGID
864 * to another app after the partial write is committed.
866 * Note: we don't call zfs_fuid_map_id() here because
867 * user 0 is not an ephemeral uid.
869 mutex_enter(&zp->z_acl_lock);
870 uid = KUID_TO_SUID(ip->i_uid);
871 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
872 (S_IXUSR >> 6))) != 0 &&
873 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
874 secpolicy_vnode_setid_retain(cr,
875 ((zp->z_mode & S_ISUID) != 0 && uid == 0)) != 0) {
877 zp->z_mode &= ~(S_ISUID | S_ISGID);
878 ip->i_mode = newmode = zp->z_mode;
879 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zsb),
880 (void *)&newmode, sizeof (uint64_t), tx);
882 mutex_exit(&zp->z_acl_lock);
884 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
887 * Update the file size (zp_size) if it has changed;
888 * account for possible concurrent updates.
890 while ((end_size = zp->z_size) < uio->uio_loffset) {
891 (void) atomic_cas_64(&zp->z_size, end_size,
896 * If we are replaying and eof is non zero then force
897 * the file size to the specified eof. Note, there's no
898 * concurrency during replay.
900 if (zsb->z_replay && zsb->z_replay_eof != 0)
901 zp->z_size = zsb->z_replay_eof;
903 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
905 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag,
911 ASSERT(tx_bytes == nbytes);
915 uio_prefaultpages(MIN(n, max_blksz), uio);
918 zfs_inode_update(zp);
919 zfs_range_unlock(rl);
922 * If we're in replay mode, or we made no progress, return error.
923 * Otherwise, it's at least a partial write, so it's successful.
925 if (zsb->z_replay || uio->uio_resid == start_resid) {
930 if (ioflag & (FSYNC | FDSYNC) ||
931 zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
932 zil_commit(zilog, zp->z_id);
937 EXPORT_SYMBOL(zfs_write);
940 * Drop a reference on the passed inode asynchronously. This ensures
941 * that the caller will never drop the last reference on an inode in
942 * the current context. Doing so while holding open a tx could result
943 * in a deadlock if iput_final() re-enters the filesystem code.
946 zfs_iput_async(struct inode *ip)
948 objset_t *os = ITOZSB(ip)->z_os;
950 ASSERT(atomic_read(&ip->i_count) > 0);
953 if (atomic_read(&ip->i_count) == 1)
954 VERIFY(taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os)),
955 (task_func_t *)iput, ip, TQ_SLEEP) != TASKQID_INVALID);
961 zfs_get_done(zgd_t *zgd, int error)
963 znode_t *zp = zgd->zgd_private;
966 dmu_buf_rele(zgd->zgd_db, zgd);
968 zfs_range_unlock(zgd->zgd_rl);
971 * Release the vnode asynchronously as we currently have the
972 * txg stopped from syncing.
974 zfs_iput_async(ZTOI(zp));
976 if (error == 0 && zgd->zgd_bp)
977 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
979 kmem_free(zgd, sizeof (zgd_t));
983 static int zil_fault_io = 0;
987 * Get data to generate a TX_WRITE intent log record.
990 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
993 objset_t *os = zsb->z_os;
995 uint64_t object = lr->lr_foid;
996 uint64_t offset = lr->lr_offset;
997 uint64_t size = lr->lr_length;
998 blkptr_t *bp = &lr->lr_blkptr;
1003 ASSERT(zio != NULL);
1007 * Nothing to do if the file has been removed
1009 if (zfs_zget(zsb, object, &zp) != 0)
1010 return (SET_ERROR(ENOENT));
1011 if (zp->z_unlinked) {
1013 * Release the vnode asynchronously as we currently have the
1014 * txg stopped from syncing.
1016 zfs_iput_async(ZTOI(zp));
1017 return (SET_ERROR(ENOENT));
1020 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1021 zgd->zgd_zilog = zsb->z_log;
1022 zgd->zgd_private = zp;
1025 * Write records come in two flavors: immediate and indirect.
1026 * For small writes it's cheaper to store the data with the
1027 * log record (immediate); for large writes it's cheaper to
1028 * sync the data and get a pointer to it (indirect) so that
1029 * we don't have to write the data twice.
1031 if (buf != NULL) { /* immediate write */
1032 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset, size,
1034 /* test for truncation needs to be done while range locked */
1035 if (offset >= zp->z_size) {
1036 error = SET_ERROR(ENOENT);
1038 error = dmu_read(os, object, offset, size, buf,
1039 DMU_READ_NO_PREFETCH);
1041 ASSERT(error == 0 || error == ENOENT);
1042 } else { /* indirect write */
1044 * Have to lock the whole block to ensure when it's
1045 * written out and it's checksum is being calculated
1046 * that no one can change the data. We need to re-check
1047 * blocksize after we get the lock in case it's changed!
1052 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1054 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset,
1056 if (zp->z_blksz == size)
1059 zfs_range_unlock(zgd->zgd_rl);
1061 /* test for truncation needs to be done while range locked */
1062 if (lr->lr_offset >= zp->z_size)
1063 error = SET_ERROR(ENOENT);
1066 error = SET_ERROR(EIO);
1071 error = dmu_buf_hold(os, object, offset, zgd, &db,
1072 DMU_READ_NO_PREFETCH);
1075 blkptr_t *obp = dmu_buf_get_blkptr(db);
1077 ASSERT(BP_IS_HOLE(bp));
1084 ASSERT(db->db_offset == offset);
1085 ASSERT(db->db_size == size);
1087 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1089 ASSERT(error || lr->lr_length <= zp->z_blksz);
1092 * On success, we need to wait for the write I/O
1093 * initiated by dmu_sync() to complete before we can
1094 * release this dbuf. We will finish everything up
1095 * in the zfs_get_done() callback.
1100 if (error == EALREADY) {
1101 lr->lr_common.lrc_txtype = TX_WRITE2;
1107 zfs_get_done(zgd, error);
1114 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
1116 znode_t *zp = ITOZ(ip);
1117 zfs_sb_t *zsb = ITOZSB(ip);
1123 if (flag & V_ACE_MASK)
1124 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1126 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1131 EXPORT_SYMBOL(zfs_access);
1134 * Lookup an entry in a directory, or an extended attribute directory.
1135 * If it exists, return a held inode reference for it.
1137 * IN: dip - inode of directory to search.
1138 * nm - name of entry to lookup.
1139 * flags - LOOKUP_XATTR set if looking for an attribute.
1140 * cr - credentials of caller.
1141 * direntflags - directory lookup flags
1142 * realpnp - returned pathname.
1144 * OUT: ipp - inode of located entry, NULL if not found.
1146 * RETURN: 0 on success, error code on failure.
1153 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
1154 cred_t *cr, int *direntflags, pathname_t *realpnp)
1156 znode_t *zdp = ITOZ(dip);
1157 zfs_sb_t *zsb = ITOZSB(dip);
1161 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1163 if (!S_ISDIR(dip->i_mode)) {
1164 return (SET_ERROR(ENOTDIR));
1165 } else if (zdp->z_sa_hdl == NULL) {
1166 return (SET_ERROR(EIO));
1169 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1170 error = zfs_fastaccesschk_execute(zdp, cr);
1179 vnode_t *tvp = dnlc_lookup(dvp, nm);
1182 error = zfs_fastaccesschk_execute(zdp, cr);
1187 if (tvp == DNLC_NO_VNODE) {
1189 return (SET_ERROR(ENOENT));
1192 return (specvp_check(vpp, cr));
1195 #endif /* HAVE_DNLC */
1204 if (flags & LOOKUP_XATTR) {
1206 * We don't allow recursive attributes..
1207 * Maybe someday we will.
1209 if (zdp->z_pflags & ZFS_XATTR) {
1211 return (SET_ERROR(EINVAL));
1214 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1220 * Do we have permission to get into attribute directory?
1223 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1233 if (!S_ISDIR(dip->i_mode)) {
1235 return (SET_ERROR(ENOTDIR));
1239 * Check accessibility of directory.
1242 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1247 if (zsb->z_utf8 && u8_validate(nm, strlen(nm),
1248 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1250 return (SET_ERROR(EILSEQ));
1253 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1254 if ((error == 0) && (*ipp))
1255 zfs_inode_update(ITOZ(*ipp));
1260 EXPORT_SYMBOL(zfs_lookup);
1263 * Attempt to create a new entry in a directory. If the entry
1264 * already exists, truncate the file if permissible, else return
1265 * an error. Return the ip of the created or trunc'd file.
1267 * IN: dip - inode of directory to put new file entry in.
1268 * name - name of new file entry.
1269 * vap - attributes of new file.
1270 * excl - flag indicating exclusive or non-exclusive mode.
1271 * mode - mode to open file with.
1272 * cr - credentials of caller.
1273 * flag - large file flag [UNUSED].
1274 * vsecp - ACL to be set
1276 * OUT: ipp - inode of created or trunc'd entry.
1278 * RETURN: 0 on success, error code on failure.
1281 * dip - ctime|mtime updated if new entry created
1282 * ip - ctime|mtime always, atime if new
1287 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1288 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1290 znode_t *zp, *dzp = ITOZ(dip);
1291 zfs_sb_t *zsb = ITOZSB(dip);
1299 zfs_acl_ids_t acl_ids;
1300 boolean_t fuid_dirtied;
1301 boolean_t have_acl = B_FALSE;
1302 boolean_t waited = B_FALSE;
1305 * If we have an ephemeral id, ACL, or XVATTR then
1306 * make sure file system is at proper version
1312 if (zsb->z_use_fuids == B_FALSE &&
1313 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1314 return (SET_ERROR(EINVAL));
1321 if (zsb->z_utf8 && u8_validate(name, strlen(name),
1322 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1324 return (SET_ERROR(EILSEQ));
1327 if (vap->va_mask & ATTR_XVATTR) {
1328 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1329 crgetuid(cr), cr, vap->va_mode)) != 0) {
1337 if (*name == '\0') {
1339 * Null component name refers to the directory itself.
1346 /* possible igrab(zp) */
1349 if (flag & FIGNORECASE)
1352 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1356 zfs_acl_ids_free(&acl_ids);
1357 if (strcmp(name, "..") == 0)
1358 error = SET_ERROR(EISDIR);
1368 * Create a new file object and update the directory
1371 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1373 zfs_acl_ids_free(&acl_ids);
1378 * We only support the creation of regular files in
1379 * extended attribute directories.
1382 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1384 zfs_acl_ids_free(&acl_ids);
1385 error = SET_ERROR(EINVAL);
1389 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1390 cr, vsecp, &acl_ids)) != 0)
1394 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1395 zfs_acl_ids_free(&acl_ids);
1396 error = SET_ERROR(EDQUOT);
1400 tx = dmu_tx_create(os);
1402 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1403 ZFS_SA_BASE_ATTR_SIZE);
1405 fuid_dirtied = zsb->z_fuid_dirty;
1407 zfs_fuid_txhold(zsb, tx);
1408 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1409 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1410 if (!zsb->z_use_sa &&
1411 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1412 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1413 0, acl_ids.z_aclp->z_acl_bytes);
1415 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1417 zfs_dirent_unlock(dl);
1418 if (error == ERESTART) {
1424 zfs_acl_ids_free(&acl_ids);
1429 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1432 zfs_fuid_sync(zsb, tx);
1434 (void) zfs_link_create(dl, zp, tx, ZNEW);
1435 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1436 if (flag & FIGNORECASE)
1438 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1439 vsecp, acl_ids.z_fuidp, vap);
1440 zfs_acl_ids_free(&acl_ids);
1443 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1446 zfs_acl_ids_free(&acl_ids);
1450 * A directory entry already exists for this name.
1453 * Can't truncate an existing file if in exclusive mode.
1456 error = SET_ERROR(EEXIST);
1460 * Can't open a directory for writing.
1462 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1463 error = SET_ERROR(EISDIR);
1467 * Verify requested access to file.
1469 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1473 mutex_enter(&dzp->z_lock);
1475 mutex_exit(&dzp->z_lock);
1478 * Truncate regular files if requested.
1480 if (S_ISREG(ZTOI(zp)->i_mode) &&
1481 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1482 /* we can't hold any locks when calling zfs_freesp() */
1484 zfs_dirent_unlock(dl);
1487 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1493 zfs_dirent_unlock(dl);
1499 zfs_inode_update(dzp);
1500 zfs_inode_update(zp);
1504 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1505 zil_commit(zilog, 0);
1510 EXPORT_SYMBOL(zfs_create);
1514 zfs_tmpfile(struct inode *dip, vattr_t *vap, int excl,
1515 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1517 znode_t *zp = NULL, *dzp = ITOZ(dip);
1518 zfs_sb_t *zsb = ITOZSB(dip);
1524 zfs_acl_ids_t acl_ids;
1525 boolean_t fuid_dirtied;
1526 boolean_t have_acl = B_FALSE;
1527 boolean_t waited = B_FALSE;
1530 * If we have an ephemeral id, ACL, or XVATTR then
1531 * make sure file system is at proper version
1537 if (zsb->z_use_fuids == B_FALSE &&
1538 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1539 return (SET_ERROR(EINVAL));
1545 if (vap->va_mask & ATTR_XVATTR) {
1546 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1547 crgetuid(cr), cr, vap->va_mode)) != 0) {
1557 * Create a new file object and update the directory
1560 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1562 zfs_acl_ids_free(&acl_ids);
1566 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1567 cr, vsecp, &acl_ids)) != 0)
1571 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1572 zfs_acl_ids_free(&acl_ids);
1573 error = SET_ERROR(EDQUOT);
1577 tx = dmu_tx_create(os);
1579 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1580 ZFS_SA_BASE_ATTR_SIZE);
1581 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
1583 fuid_dirtied = zsb->z_fuid_dirty;
1585 zfs_fuid_txhold(zsb, tx);
1586 if (!zsb->z_use_sa &&
1587 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1588 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1589 0, acl_ids.z_aclp->z_acl_bytes);
1591 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1593 if (error == ERESTART) {
1599 zfs_acl_ids_free(&acl_ids);
1604 zfs_mknode(dzp, vap, tx, cr, IS_TMPFILE, &zp, &acl_ids);
1607 zfs_fuid_sync(zsb, tx);
1609 /* Add to unlinked set */
1611 zfs_unlinked_add(zp, tx);
1612 zfs_acl_ids_free(&acl_ids);
1620 zfs_inode_update(dzp);
1621 zfs_inode_update(zp);
1630 * Remove an entry from a directory.
1632 * IN: dip - inode of directory to remove entry from.
1633 * name - name of entry to remove.
1634 * cr - credentials of caller.
1636 * RETURN: 0 if success
1637 * error code if failure
1641 * ip - ctime (if nlink > 0)
1644 uint64_t null_xattr = 0;
1648 zfs_remove(struct inode *dip, char *name, cred_t *cr, int flags)
1650 znode_t *zp, *dzp = ITOZ(dip);
1653 zfs_sb_t *zsb = ITOZSB(dip);
1655 uint64_t acl_obj, xattr_obj;
1656 uint64_t xattr_obj_unlinked = 0;
1661 boolean_t may_delete_now, delete_now = FALSE;
1662 boolean_t unlinked, toobig = FALSE;
1664 pathname_t *realnmp = NULL;
1668 boolean_t waited = B_FALSE;
1674 if (flags & FIGNORECASE) {
1684 * Attempt to lock directory; fail if entry doesn't exist.
1686 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1696 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1701 * Need to use rmdir for removing directories.
1703 if (S_ISDIR(ip->i_mode)) {
1704 error = SET_ERROR(EPERM);
1710 dnlc_remove(dvp, realnmp->pn_buf);
1712 dnlc_remove(dvp, name);
1713 #endif /* HAVE_DNLC */
1715 mutex_enter(&zp->z_lock);
1716 may_delete_now = atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped);
1717 mutex_exit(&zp->z_lock);
1720 * We may delete the znode now, or we may put it in the unlinked set;
1721 * it depends on whether we're the last link, and on whether there are
1722 * other holds on the inode. So we dmu_tx_hold() the right things to
1723 * allow for either case.
1726 tx = dmu_tx_create(zsb->z_os);
1727 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1728 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1729 zfs_sa_upgrade_txholds(tx, zp);
1730 zfs_sa_upgrade_txholds(tx, dzp);
1731 if (may_delete_now) {
1732 toobig = zp->z_size > zp->z_blksz * zfs_delete_blocks;
1733 /* if the file is too big, only hold_free a token amount */
1734 dmu_tx_hold_free(tx, zp->z_id, 0,
1735 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1738 /* are there any extended attributes? */
1739 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1740 &xattr_obj, sizeof (xattr_obj));
1741 if (error == 0 && xattr_obj) {
1742 error = zfs_zget(zsb, xattr_obj, &xzp);
1744 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1745 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1748 mutex_enter(&zp->z_lock);
1749 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1750 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1751 mutex_exit(&zp->z_lock);
1753 /* charge as an update -- would be nice not to charge at all */
1754 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
1757 * Mark this transaction as typically resulting in a net free of space
1759 dmu_tx_mark_netfree(tx);
1761 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1763 zfs_dirent_unlock(dl);
1767 if (error == ERESTART) {
1781 * Remove the directory entry.
1783 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1792 * Hold z_lock so that we can make sure that the ACL obj
1793 * hasn't changed. Could have been deleted due to
1796 mutex_enter(&zp->z_lock);
1797 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1798 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1799 delete_now = may_delete_now && !toobig &&
1800 atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped) &&
1801 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1806 if (xattr_obj_unlinked) {
1807 ASSERT3U(ZTOI(xzp)->i_nlink, ==, 2);
1808 mutex_enter(&xzp->z_lock);
1809 xzp->z_unlinked = 1;
1810 clear_nlink(ZTOI(xzp));
1812 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zsb),
1813 &links, sizeof (links), tx);
1814 ASSERT3U(error, ==, 0);
1815 mutex_exit(&xzp->z_lock);
1816 zfs_unlinked_add(xzp, tx);
1819 error = sa_remove(zp->z_sa_hdl,
1820 SA_ZPL_XATTR(zsb), tx);
1822 error = sa_update(zp->z_sa_hdl,
1823 SA_ZPL_XATTR(zsb), &null_xattr,
1824 sizeof (uint64_t), tx);
1828 * Add to the unlinked set because a new reference could be
1829 * taken concurrently resulting in a deferred destruction.
1831 zfs_unlinked_add(zp, tx);
1832 mutex_exit(&zp->z_lock);
1833 zfs_inode_update(zp);
1835 } else if (unlinked) {
1836 mutex_exit(&zp->z_lock);
1837 zfs_unlinked_add(zp, tx);
1841 if (flags & FIGNORECASE)
1843 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1850 zfs_dirent_unlock(dl);
1851 zfs_inode_update(dzp);
1854 zfs_inode_update(zp);
1859 zfs_inode_update(xzp);
1860 zfs_iput_async(ZTOI(xzp));
1863 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1864 zil_commit(zilog, 0);
1869 EXPORT_SYMBOL(zfs_remove);
1872 * Create a new directory and insert it into dip using the name
1873 * provided. Return a pointer to the inserted directory.
1875 * IN: dip - inode of directory to add subdir to.
1876 * dirname - name of new directory.
1877 * vap - attributes of new directory.
1878 * cr - credentials of caller.
1879 * vsecp - ACL to be set
1881 * OUT: ipp - inode of created directory.
1883 * RETURN: 0 if success
1884 * error code if failure
1887 * dip - ctime|mtime updated
1888 * ipp - ctime|mtime|atime updated
1892 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1893 cred_t *cr, int flags, vsecattr_t *vsecp)
1895 znode_t *zp, *dzp = ITOZ(dip);
1896 zfs_sb_t *zsb = ITOZSB(dip);
1904 gid_t gid = crgetgid(cr);
1905 zfs_acl_ids_t acl_ids;
1906 boolean_t fuid_dirtied;
1907 boolean_t waited = B_FALSE;
1909 ASSERT(S_ISDIR(vap->va_mode));
1912 * If we have an ephemeral id, ACL, or XVATTR then
1913 * make sure file system is at proper version
1917 if (zsb->z_use_fuids == B_FALSE &&
1918 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1919 return (SET_ERROR(EINVAL));
1925 if (dzp->z_pflags & ZFS_XATTR) {
1927 return (SET_ERROR(EINVAL));
1930 if (zsb->z_utf8 && u8_validate(dirname,
1931 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1933 return (SET_ERROR(EILSEQ));
1935 if (flags & FIGNORECASE)
1938 if (vap->va_mask & ATTR_XVATTR) {
1939 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1940 crgetuid(cr), cr, vap->va_mode)) != 0) {
1946 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1947 vsecp, &acl_ids)) != 0) {
1952 * First make sure the new directory doesn't exist.
1954 * Existence is checked first to make sure we don't return
1955 * EACCES instead of EEXIST which can cause some applications
1961 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1963 zfs_acl_ids_free(&acl_ids);
1968 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
1969 zfs_acl_ids_free(&acl_ids);
1970 zfs_dirent_unlock(dl);
1975 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1976 zfs_acl_ids_free(&acl_ids);
1977 zfs_dirent_unlock(dl);
1979 return (SET_ERROR(EDQUOT));
1983 * Add a new entry to the directory.
1985 tx = dmu_tx_create(zsb->z_os);
1986 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1987 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1988 fuid_dirtied = zsb->z_fuid_dirty;
1990 zfs_fuid_txhold(zsb, tx);
1991 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1992 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1993 acl_ids.z_aclp->z_acl_bytes);
1996 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1997 ZFS_SA_BASE_ATTR_SIZE);
1999 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2001 zfs_dirent_unlock(dl);
2002 if (error == ERESTART) {
2008 zfs_acl_ids_free(&acl_ids);
2017 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2020 zfs_fuid_sync(zsb, tx);
2023 * Now put new name in parent dir.
2025 (void) zfs_link_create(dl, zp, tx, ZNEW);
2029 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2030 if (flags & FIGNORECASE)
2032 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2033 acl_ids.z_fuidp, vap);
2035 zfs_acl_ids_free(&acl_ids);
2039 zfs_dirent_unlock(dl);
2041 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
2042 zil_commit(zilog, 0);
2044 zfs_inode_update(dzp);
2045 zfs_inode_update(zp);
2049 EXPORT_SYMBOL(zfs_mkdir);
2052 * Remove a directory subdir entry. If the current working
2053 * directory is the same as the subdir to be removed, the
2056 * IN: dip - inode of directory to remove from.
2057 * name - name of directory to be removed.
2058 * cwd - inode of current working directory.
2059 * cr - credentials of caller.
2060 * flags - case flags
2062 * RETURN: 0 on success, error code on failure.
2065 * dip - ctime|mtime updated
2069 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
2072 znode_t *dzp = ITOZ(dip);
2075 zfs_sb_t *zsb = ITOZSB(dip);
2081 boolean_t waited = B_FALSE;
2087 if (flags & FIGNORECASE)
2093 * Attempt to lock directory; fail if entry doesn't exist.
2095 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2103 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
2107 if (!S_ISDIR(ip->i_mode)) {
2108 error = SET_ERROR(ENOTDIR);
2113 error = SET_ERROR(EINVAL);
2118 * Grab a lock on the directory to make sure that noone is
2119 * trying to add (or lookup) entries while we are removing it.
2121 rw_enter(&zp->z_name_lock, RW_WRITER);
2124 * Grab a lock on the parent pointer to make sure we play well
2125 * with the treewalk and directory rename code.
2127 rw_enter(&zp->z_parent_lock, RW_WRITER);
2129 tx = dmu_tx_create(zsb->z_os);
2130 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2131 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2132 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
2133 zfs_sa_upgrade_txholds(tx, zp);
2134 zfs_sa_upgrade_txholds(tx, dzp);
2135 dmu_tx_mark_netfree(tx);
2136 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2138 rw_exit(&zp->z_parent_lock);
2139 rw_exit(&zp->z_name_lock);
2140 zfs_dirent_unlock(dl);
2142 if (error == ERESTART) {
2153 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2156 uint64_t txtype = TX_RMDIR;
2157 if (flags & FIGNORECASE)
2159 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2164 rw_exit(&zp->z_parent_lock);
2165 rw_exit(&zp->z_name_lock);
2167 zfs_dirent_unlock(dl);
2169 zfs_inode_update(dzp);
2170 zfs_inode_update(zp);
2173 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
2174 zil_commit(zilog, 0);
2179 EXPORT_SYMBOL(zfs_rmdir);
2182 * Read as many directory entries as will fit into the provided
2183 * dirent buffer from the given directory cursor position.
2185 * IN: ip - inode of directory to read.
2186 * dirent - buffer for directory entries.
2188 * OUT: dirent - filler buffer of directory entries.
2190 * RETURN: 0 if success
2191 * error code if failure
2194 * ip - atime updated
2196 * Note that the low 4 bits of the cookie returned by zap is always zero.
2197 * This allows us to use the low range for "special" directory entries:
2198 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2199 * we use the offset 2 for the '.zfs' directory.
2203 zfs_readdir(struct inode *ip, struct dir_context *ctx, cred_t *cr)
2205 znode_t *zp = ITOZ(ip);
2206 zfs_sb_t *zsb = ITOZSB(ip);
2209 zap_attribute_t zap;
2215 uint64_t offset; /* must be unsigned; checks for < 1 */
2220 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zsb),
2221 &parent, sizeof (parent))) != 0)
2225 * Quit if directory has been removed (posix)
2233 prefetch = zp->z_zn_prefetch;
2236 * Initialize the iterator cursor.
2240 * Start iteration from the beginning of the directory.
2242 zap_cursor_init(&zc, os, zp->z_id);
2245 * The offset is a serialized cursor.
2247 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2251 * Transform to file-system independent format
2256 * Special case `.', `..', and `.zfs'.
2259 (void) strcpy(zap.za_name, ".");
2260 zap.za_normalization_conflict = 0;
2263 } else if (offset == 1) {
2264 (void) strcpy(zap.za_name, "..");
2265 zap.za_normalization_conflict = 0;
2268 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2269 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2270 zap.za_normalization_conflict = 0;
2271 objnum = ZFSCTL_INO_ROOT;
2277 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2278 if (error == ENOENT)
2285 * Allow multiple entries provided the first entry is
2286 * the object id. Non-zpl consumers may safely make
2287 * use of the additional space.
2289 * XXX: This should be a feature flag for compatibility
2291 if (zap.za_integer_length != 8 ||
2292 zap.za_num_integers == 0) {
2293 cmn_err(CE_WARN, "zap_readdir: bad directory "
2294 "entry, obj = %lld, offset = %lld, "
2295 "length = %d, num = %lld\n",
2296 (u_longlong_t)zp->z_id,
2297 (u_longlong_t)offset,
2298 zap.za_integer_length,
2299 (u_longlong_t)zap.za_num_integers);
2300 error = SET_ERROR(ENXIO);
2304 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2305 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2308 done = !dir_emit(ctx, zap.za_name, strlen(zap.za_name),
2313 /* Prefetch znode */
2315 dmu_prefetch(os, objnum, 0, 0, 0,
2316 ZIO_PRIORITY_SYNC_READ);
2320 * Move to the next entry, fill in the previous offset.
2322 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2323 zap_cursor_advance(&zc);
2324 offset = zap_cursor_serialize(&zc);
2330 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2333 zap_cursor_fini(&zc);
2334 if (error == ENOENT)
2341 EXPORT_SYMBOL(zfs_readdir);
2343 ulong_t zfs_fsync_sync_cnt = 4;
2346 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2348 znode_t *zp = ITOZ(ip);
2349 zfs_sb_t *zsb = ITOZSB(ip);
2351 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2353 if (zsb->z_os->os_sync != ZFS_SYNC_DISABLED) {
2356 zil_commit(zsb->z_log, zp->z_id);
2359 tsd_set(zfs_fsyncer_key, NULL);
2363 EXPORT_SYMBOL(zfs_fsync);
2367 * Get the requested file attributes and place them in the provided
2370 * IN: ip - inode of file.
2371 * vap - va_mask identifies requested attributes.
2372 * If ATTR_XVATTR set, then optional attrs are requested
2373 * flags - ATTR_NOACLCHECK (CIFS server context)
2374 * cr - credentials of caller.
2376 * OUT: vap - attribute values.
2378 * RETURN: 0 (always succeeds)
2382 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2384 znode_t *zp = ITOZ(ip);
2385 zfs_sb_t *zsb = ITOZSB(ip);
2388 uint64_t atime[2], mtime[2], ctime[2];
2389 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2390 xoptattr_t *xoap = NULL;
2391 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2392 sa_bulk_attr_t bulk[3];
2398 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2400 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zsb), NULL, &atime, 16);
2401 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
2402 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
2404 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2410 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2411 * Also, if we are the owner don't bother, since owner should
2412 * always be allowed to read basic attributes of file.
2414 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2415 (vap->va_uid != crgetuid(cr))) {
2416 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2424 * Return all attributes. It's cheaper to provide the answer
2425 * than to determine whether we were asked the question.
2428 mutex_enter(&zp->z_lock);
2429 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2430 vap->va_mode = zp->z_mode;
2431 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2432 vap->va_nodeid = zp->z_id;
2433 if ((zp->z_id == zsb->z_root) && zfs_show_ctldir(zp))
2434 links = ZTOI(zp)->i_nlink + 1;
2436 links = ZTOI(zp)->i_nlink;
2437 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2438 vap->va_size = i_size_read(ip);
2439 vap->va_rdev = ip->i_rdev;
2440 vap->va_seq = ip->i_generation;
2443 * Add in any requested optional attributes and the create time.
2444 * Also set the corresponding bits in the returned attribute bitmap.
2446 if ((xoap = xva_getxoptattr(xvap)) != NULL && zsb->z_use_fuids) {
2447 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2449 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2450 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2453 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2454 xoap->xoa_readonly =
2455 ((zp->z_pflags & ZFS_READONLY) != 0);
2456 XVA_SET_RTN(xvap, XAT_READONLY);
2459 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2461 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2462 XVA_SET_RTN(xvap, XAT_SYSTEM);
2465 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2467 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2468 XVA_SET_RTN(xvap, XAT_HIDDEN);
2471 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2472 xoap->xoa_nounlink =
2473 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2474 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2477 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2478 xoap->xoa_immutable =
2479 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2480 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2483 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2484 xoap->xoa_appendonly =
2485 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2486 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2489 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2491 ((zp->z_pflags & ZFS_NODUMP) != 0);
2492 XVA_SET_RTN(xvap, XAT_NODUMP);
2495 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2497 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2498 XVA_SET_RTN(xvap, XAT_OPAQUE);
2501 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2502 xoap->xoa_av_quarantined =
2503 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2504 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2507 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2508 xoap->xoa_av_modified =
2509 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2510 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2513 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2514 S_ISREG(ip->i_mode)) {
2515 zfs_sa_get_scanstamp(zp, xvap);
2518 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2521 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zsb),
2522 times, sizeof (times));
2523 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2524 XVA_SET_RTN(xvap, XAT_CREATETIME);
2527 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2528 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2529 XVA_SET_RTN(xvap, XAT_REPARSE);
2531 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2532 xoap->xoa_generation = ip->i_generation;
2533 XVA_SET_RTN(xvap, XAT_GEN);
2536 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2538 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2539 XVA_SET_RTN(xvap, XAT_OFFLINE);
2542 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2544 ((zp->z_pflags & ZFS_SPARSE) != 0);
2545 XVA_SET_RTN(xvap, XAT_SPARSE);
2549 ZFS_TIME_DECODE(&vap->va_atime, atime);
2550 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2551 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2553 mutex_exit(&zp->z_lock);
2555 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2557 if (zp->z_blksz == 0) {
2559 * Block size hasn't been set; suggest maximal I/O transfers.
2561 vap->va_blksize = zsb->z_max_blksz;
2567 EXPORT_SYMBOL(zfs_getattr);
2570 * Get the basic file attributes and place them in the provided kstat
2571 * structure. The inode is assumed to be the authoritative source
2572 * for most of the attributes. However, the znode currently has the
2573 * authoritative atime, blksize, and block count.
2575 * IN: ip - inode of file.
2577 * OUT: sp - kstat values.
2579 * RETURN: 0 (always succeeds)
2583 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2585 znode_t *zp = ITOZ(ip);
2586 zfs_sb_t *zsb = ITOZSB(ip);
2588 u_longlong_t nblocks;
2593 mutex_enter(&zp->z_lock);
2595 generic_fillattr(ip, sp);
2597 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2598 sp->blksize = blksize;
2599 sp->blocks = nblocks;
2601 if (unlikely(zp->z_blksz == 0)) {
2603 * Block size hasn't been set; suggest maximal I/O transfers.
2605 sp->blksize = zsb->z_max_blksz;
2608 mutex_exit(&zp->z_lock);
2611 * Required to prevent NFS client from detecting different inode
2612 * numbers of snapshot root dentry before and after snapshot mount.
2614 if (zsb->z_issnap) {
2615 if (ip->i_sb->s_root->d_inode == ip)
2616 sp->ino = ZFSCTL_INO_SNAPDIRS -
2617 dmu_objset_id(zsb->z_os);
2624 EXPORT_SYMBOL(zfs_getattr_fast);
2627 * Set the file attributes to the values contained in the
2630 * IN: ip - inode of file to be modified.
2631 * vap - new attribute values.
2632 * If ATTR_XVATTR set, then optional attrs are being set
2633 * flags - ATTR_UTIME set if non-default time values provided.
2634 * - ATTR_NOACLCHECK (CIFS context only).
2635 * cr - credentials of caller.
2637 * RETURN: 0 if success
2638 * error code if failure
2641 * ip - ctime updated, mtime updated if size changed.
2645 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2647 znode_t *zp = ITOZ(ip);
2648 zfs_sb_t *zsb = ITOZSB(ip);
2652 xvattr_t *tmpxvattr;
2653 uint_t mask = vap->va_mask;
2654 uint_t saved_mask = 0;
2657 uint64_t new_kuid = 0, new_kgid = 0, new_uid, new_gid;
2659 uint64_t mtime[2], ctime[2], atime[2];
2661 int need_policy = FALSE;
2663 zfs_fuid_info_t *fuidp = NULL;
2664 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2667 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2668 boolean_t fuid_dirtied = B_FALSE;
2669 sa_bulk_attr_t *bulk, *xattr_bulk;
2670 int count = 0, xattr_count = 0;
2681 * Make sure that if we have ephemeral uid/gid or xvattr specified
2682 * that file system is at proper version level
2685 if (zsb->z_use_fuids == B_FALSE &&
2686 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2687 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2688 (mask & ATTR_XVATTR))) {
2690 return (SET_ERROR(EINVAL));
2693 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2695 return (SET_ERROR(EISDIR));
2698 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2700 return (SET_ERROR(EINVAL));
2704 * If this is an xvattr_t, then get a pointer to the structure of
2705 * optional attributes. If this is NULL, then we have a vattr_t.
2707 xoap = xva_getxoptattr(xvap);
2709 tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP);
2710 xva_init(tmpxvattr);
2712 bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2713 xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2716 * Immutable files can only alter immutable bit and atime
2718 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2719 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2720 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2725 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2731 * Verify timestamps doesn't overflow 32 bits.
2732 * ZFS can handle large timestamps, but 32bit syscalls can't
2733 * handle times greater than 2039. This check should be removed
2734 * once large timestamps are fully supported.
2736 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2737 if (((mask & ATTR_ATIME) &&
2738 TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2739 ((mask & ATTR_MTIME) &&
2740 TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2750 /* Can this be moved to before the top label? */
2751 if (zfs_is_readonly(zsb)) {
2757 * First validate permissions
2760 if (mask & ATTR_SIZE) {
2761 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2766 * XXX - Note, we are not providing any open
2767 * mode flags here (like FNDELAY), so we may
2768 * block if there are locks present... this
2769 * should be addressed in openat().
2771 /* XXX - would it be OK to generate a log record here? */
2772 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2777 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2778 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2779 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2780 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2781 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2782 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2783 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2784 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2785 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2789 if (mask & (ATTR_UID|ATTR_GID)) {
2790 int idmask = (mask & (ATTR_UID|ATTR_GID));
2795 * NOTE: even if a new mode is being set,
2796 * we may clear S_ISUID/S_ISGID bits.
2799 if (!(mask & ATTR_MODE))
2800 vap->va_mode = zp->z_mode;
2803 * Take ownership or chgrp to group we are a member of
2806 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
2807 take_group = (mask & ATTR_GID) &&
2808 zfs_groupmember(zsb, vap->va_gid, cr);
2811 * If both ATTR_UID and ATTR_GID are set then take_owner and
2812 * take_group must both be set in order to allow taking
2815 * Otherwise, send the check through secpolicy_vnode_setattr()
2819 if (((idmask == (ATTR_UID|ATTR_GID)) &&
2820 take_owner && take_group) ||
2821 ((idmask == ATTR_UID) && take_owner) ||
2822 ((idmask == ATTR_GID) && take_group)) {
2823 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2824 skipaclchk, cr) == 0) {
2826 * Remove setuid/setgid for non-privileged users
2828 (void) secpolicy_setid_clear(vap, cr);
2829 trim_mask = (mask & (ATTR_UID|ATTR_GID));
2838 mutex_enter(&zp->z_lock);
2839 oldva.va_mode = zp->z_mode;
2840 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2841 if (mask & ATTR_XVATTR) {
2843 * Update xvattr mask to include only those attributes
2844 * that are actually changing.
2846 * the bits will be restored prior to actually setting
2847 * the attributes so the caller thinks they were set.
2849 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2850 if (xoap->xoa_appendonly !=
2851 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2854 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2855 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
2859 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2860 if (xoap->xoa_nounlink !=
2861 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2864 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2865 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
2869 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2870 if (xoap->xoa_immutable !=
2871 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2874 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2875 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
2879 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2880 if (xoap->xoa_nodump !=
2881 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2884 XVA_CLR_REQ(xvap, XAT_NODUMP);
2885 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
2889 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2890 if (xoap->xoa_av_modified !=
2891 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2894 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2895 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
2899 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2900 if ((!S_ISREG(ip->i_mode) &&
2901 xoap->xoa_av_quarantined) ||
2902 xoap->xoa_av_quarantined !=
2903 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2906 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2907 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
2911 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2912 mutex_exit(&zp->z_lock);
2917 if (need_policy == FALSE &&
2918 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2919 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2924 mutex_exit(&zp->z_lock);
2926 if (mask & ATTR_MODE) {
2927 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2928 err = secpolicy_setid_setsticky_clear(ip, vap,
2933 trim_mask |= ATTR_MODE;
2941 * If trim_mask is set then take ownership
2942 * has been granted or write_acl is present and user
2943 * has the ability to modify mode. In that case remove
2944 * UID|GID and or MODE from mask so that
2945 * secpolicy_vnode_setattr() doesn't revoke it.
2949 saved_mask = vap->va_mask;
2950 vap->va_mask &= ~trim_mask;
2952 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
2953 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2958 vap->va_mask |= saved_mask;
2962 * secpolicy_vnode_setattr, or take ownership may have
2965 mask = vap->va_mask;
2967 if ((mask & (ATTR_UID | ATTR_GID))) {
2968 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
2969 &xattr_obj, sizeof (xattr_obj));
2971 if (err == 0 && xattr_obj) {
2972 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
2976 if (mask & ATTR_UID) {
2977 new_kuid = zfs_fuid_create(zsb,
2978 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2979 if (new_kuid != KUID_TO_SUID(ZTOI(zp)->i_uid) &&
2980 zfs_fuid_overquota(zsb, B_FALSE, new_kuid)) {
2988 if (mask & ATTR_GID) {
2989 new_kgid = zfs_fuid_create(zsb, (uint64_t)vap->va_gid,
2990 cr, ZFS_GROUP, &fuidp);
2991 if (new_kgid != KGID_TO_SGID(ZTOI(zp)->i_gid) &&
2992 zfs_fuid_overquota(zsb, B_TRUE, new_kgid)) {
3000 tx = dmu_tx_create(zsb->z_os);
3002 if (mask & ATTR_MODE) {
3003 uint64_t pmode = zp->z_mode;
3005 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3007 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
3009 mutex_enter(&zp->z_lock);
3010 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3012 * Are we upgrading ACL from old V0 format
3015 if (zsb->z_version >= ZPL_VERSION_FUID &&
3016 zfs_znode_acl_version(zp) ==
3017 ZFS_ACL_VERSION_INITIAL) {
3018 dmu_tx_hold_free(tx, acl_obj, 0,
3020 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3021 0, aclp->z_acl_bytes);
3023 dmu_tx_hold_write(tx, acl_obj, 0,
3026 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3027 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3028 0, aclp->z_acl_bytes);
3030 mutex_exit(&zp->z_lock);
3031 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3033 if ((mask & ATTR_XVATTR) &&
3034 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3035 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3037 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3041 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3044 fuid_dirtied = zsb->z_fuid_dirty;
3046 zfs_fuid_txhold(zsb, tx);
3048 zfs_sa_upgrade_txholds(tx, zp);
3050 err = dmu_tx_assign(tx, TXG_WAIT);
3056 * Set each attribute requested.
3057 * We group settings according to the locks they need to acquire.
3059 * Note: you cannot set ctime directly, although it will be
3060 * updated as a side-effect of calling this function.
3064 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3065 mutex_enter(&zp->z_acl_lock);
3066 mutex_enter(&zp->z_lock);
3068 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
3069 &zp->z_pflags, sizeof (zp->z_pflags));
3072 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3073 mutex_enter(&attrzp->z_acl_lock);
3074 mutex_enter(&attrzp->z_lock);
3075 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3076 SA_ZPL_FLAGS(zsb), NULL, &attrzp->z_pflags,
3077 sizeof (attrzp->z_pflags));
3080 if (mask & (ATTR_UID|ATTR_GID)) {
3082 if (mask & ATTR_UID) {
3083 ZTOI(zp)->i_uid = SUID_TO_KUID(new_kuid);
3084 new_uid = zfs_uid_read(ZTOI(zp));
3085 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zsb), NULL,
3086 &new_uid, sizeof (new_uid));
3088 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3089 SA_ZPL_UID(zsb), NULL, &new_uid,
3091 ZTOI(attrzp)->i_uid = SUID_TO_KUID(new_uid);
3095 if (mask & ATTR_GID) {
3096 ZTOI(zp)->i_gid = SGID_TO_KGID(new_kgid);
3097 new_gid = zfs_gid_read(ZTOI(zp));
3098 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zsb),
3099 NULL, &new_gid, sizeof (new_gid));
3101 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3102 SA_ZPL_GID(zsb), NULL, &new_gid,
3104 ZTOI(attrzp)->i_gid = SGID_TO_KGID(new_kgid);
3107 if (!(mask & ATTR_MODE)) {
3108 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb),
3109 NULL, &new_mode, sizeof (new_mode));
3110 new_mode = zp->z_mode;
3112 err = zfs_acl_chown_setattr(zp);
3115 err = zfs_acl_chown_setattr(attrzp);
3120 if (mask & ATTR_MODE) {
3121 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb), NULL,
3122 &new_mode, sizeof (new_mode));
3123 zp->z_mode = ZTOI(zp)->i_mode = new_mode;
3124 ASSERT3P(aclp, !=, NULL);
3125 err = zfs_aclset_common(zp, aclp, cr, tx);
3127 if (zp->z_acl_cached)
3128 zfs_acl_free(zp->z_acl_cached);
3129 zp->z_acl_cached = aclp;
3133 if ((mask & ATTR_ATIME) || zp->z_atime_dirty) {
3134 zp->z_atime_dirty = 0;
3135 ZFS_TIME_ENCODE(&ip->i_atime, atime);
3136 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zsb), NULL,
3137 &atime, sizeof (atime));
3140 if (mask & ATTR_MTIME) {
3141 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3142 ZTOI(zp)->i_mtime = timespec_trunc(vap->va_mtime,
3143 ZTOI(zp)->i_sb->s_time_gran);
3145 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL,
3146 mtime, sizeof (mtime));
3149 if (mask & ATTR_CTIME) {
3150 ZFS_TIME_ENCODE(&vap->va_ctime, ctime);
3151 ZTOI(zp)->i_ctime = timespec_trunc(vap->va_ctime,
3152 ZTOI(zp)->i_sb->s_time_gran);
3153 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
3154 ctime, sizeof (ctime));
3157 if (attrzp && mask) {
3158 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3159 SA_ZPL_CTIME(zsb), NULL, &ctime,
3164 * Do this after setting timestamps to prevent timestamp
3165 * update from toggling bit
3168 if (xoap && (mask & ATTR_XVATTR)) {
3171 * restore trimmed off masks
3172 * so that return masks can be set for caller.
3175 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
3176 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3178 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
3179 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3181 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
3182 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3184 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
3185 XVA_SET_REQ(xvap, XAT_NODUMP);
3187 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
3188 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3190 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
3191 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3194 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3195 ASSERT(S_ISREG(ip->i_mode));
3197 zfs_xvattr_set(zp, xvap, tx);
3201 zfs_fuid_sync(zsb, tx);
3204 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3206 mutex_exit(&zp->z_lock);
3207 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3208 mutex_exit(&zp->z_acl_lock);
3211 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3212 mutex_exit(&attrzp->z_acl_lock);
3213 mutex_exit(&attrzp->z_lock);
3216 if (err == 0 && attrzp) {
3217 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3228 zfs_fuid_info_free(fuidp);
3234 if (err == ERESTART)
3237 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3239 zfs_inode_update(zp);
3243 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3244 zil_commit(zilog, 0);
3247 kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * 7);
3248 kmem_free(bulk, sizeof (sa_bulk_attr_t) * 7);
3249 kmem_free(tmpxvattr, sizeof (xvattr_t));
3253 EXPORT_SYMBOL(zfs_setattr);
3255 typedef struct zfs_zlock {
3256 krwlock_t *zl_rwlock; /* lock we acquired */
3257 znode_t *zl_znode; /* znode we held */
3258 struct zfs_zlock *zl_next; /* next in list */
3262 * Drop locks and release vnodes that were held by zfs_rename_lock().
3265 zfs_rename_unlock(zfs_zlock_t **zlpp)
3269 while ((zl = *zlpp) != NULL) {
3270 if (zl->zl_znode != NULL)
3271 iput(ZTOI(zl->zl_znode));
3272 rw_exit(zl->zl_rwlock);
3273 *zlpp = zl->zl_next;
3274 kmem_free(zl, sizeof (*zl));
3279 * Search back through the directory tree, using the ".." entries.
3280 * Lock each directory in the chain to prevent concurrent renames.
3281 * Fail any attempt to move a directory into one of its own descendants.
3282 * XXX - z_parent_lock can overlap with map or grow locks
3285 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3289 uint64_t rootid = ZTOZSB(zp)->z_root;
3290 uint64_t oidp = zp->z_id;
3291 krwlock_t *rwlp = &szp->z_parent_lock;
3292 krw_t rw = RW_WRITER;
3295 * First pass write-locks szp and compares to zp->z_id.
3296 * Later passes read-lock zp and compare to zp->z_parent.
3299 if (!rw_tryenter(rwlp, rw)) {
3301 * Another thread is renaming in this path.
3302 * Note that if we are a WRITER, we don't have any
3303 * parent_locks held yet.
3305 if (rw == RW_READER && zp->z_id > szp->z_id) {
3307 * Drop our locks and restart
3309 zfs_rename_unlock(&zl);
3313 rwlp = &szp->z_parent_lock;
3318 * Wait for other thread to drop its locks
3324 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3325 zl->zl_rwlock = rwlp;
3326 zl->zl_znode = NULL;
3327 zl->zl_next = *zlpp;
3330 if (oidp == szp->z_id) /* We're a descendant of szp */
3331 return (SET_ERROR(EINVAL));
3333 if (oidp == rootid) /* We've hit the top */
3336 if (rw == RW_READER) { /* i.e. not the first pass */
3337 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3342 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3343 &oidp, sizeof (oidp));
3344 rwlp = &zp->z_parent_lock;
3347 } while (zp->z_id != sdzp->z_id);
3353 * Move an entry from the provided source directory to the target
3354 * directory. Change the entry name as indicated.
3356 * IN: sdip - Source directory containing the "old entry".
3357 * snm - Old entry name.
3358 * tdip - Target directory to contain the "new entry".
3359 * tnm - New entry name.
3360 * cr - credentials of caller.
3361 * flags - case flags
3363 * RETURN: 0 on success, error code on failure.
3366 * sdip,tdip - ctime|mtime updated
3370 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3371 cred_t *cr, int flags)
3373 znode_t *tdzp, *szp, *tzp;
3374 znode_t *sdzp = ITOZ(sdip);
3375 zfs_sb_t *zsb = ITOZSB(sdip);
3377 zfs_dirlock_t *sdl, *tdl;
3380 int cmp, serr, terr;
3383 boolean_t waited = B_FALSE;
3386 ZFS_VERIFY_ZP(sdzp);
3390 ZFS_VERIFY_ZP(tdzp);
3393 * We check i_sb because snapshots and the ctldir must have different
3396 if (tdip->i_sb != sdip->i_sb || zfsctl_is_node(tdip)) {
3398 return (SET_ERROR(EXDEV));
3401 if (zsb->z_utf8 && u8_validate(tnm,
3402 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3404 return (SET_ERROR(EILSEQ));
3407 if (flags & FIGNORECASE)
3416 * This is to prevent the creation of links into attribute space
3417 * by renaming a linked file into/outof an attribute directory.
3418 * See the comment in zfs_link() for why this is considered bad.
3420 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3422 return (SET_ERROR(EINVAL));
3426 * Lock source and target directory entries. To prevent deadlock,
3427 * a lock ordering must be defined. We lock the directory with
3428 * the smallest object id first, or if it's a tie, the one with
3429 * the lexically first name.
3431 if (sdzp->z_id < tdzp->z_id) {
3433 } else if (sdzp->z_id > tdzp->z_id) {
3437 * First compare the two name arguments without
3438 * considering any case folding.
3440 int nofold = (zsb->z_norm & ~U8_TEXTPREP_TOUPPER);
3442 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3443 ASSERT(error == 0 || !zsb->z_utf8);
3446 * POSIX: "If the old argument and the new argument
3447 * both refer to links to the same existing file,
3448 * the rename() function shall return successfully
3449 * and perform no other action."
3455 * If the file system is case-folding, then we may
3456 * have some more checking to do. A case-folding file
3457 * system is either supporting mixed case sensitivity
3458 * access or is completely case-insensitive. Note
3459 * that the file system is always case preserving.
3461 * In mixed sensitivity mode case sensitive behavior
3462 * is the default. FIGNORECASE must be used to
3463 * explicitly request case insensitive behavior.
3465 * If the source and target names provided differ only
3466 * by case (e.g., a request to rename 'tim' to 'Tim'),
3467 * we will treat this as a special case in the
3468 * case-insensitive mode: as long as the source name
3469 * is an exact match, we will allow this to proceed as
3470 * a name-change request.
3472 if ((zsb->z_case == ZFS_CASE_INSENSITIVE ||
3473 (zsb->z_case == ZFS_CASE_MIXED &&
3474 flags & FIGNORECASE)) &&
3475 u8_strcmp(snm, tnm, 0, zsb->z_norm, U8_UNICODE_LATEST,
3478 * case preserving rename request, require exact
3487 * If the source and destination directories are the same, we should
3488 * grab the z_name_lock of that directory only once.
3492 rw_enter(&sdzp->z_name_lock, RW_READER);
3496 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3497 ZEXISTS | zflg, NULL, NULL);
3498 terr = zfs_dirent_lock(&tdl,
3499 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3501 terr = zfs_dirent_lock(&tdl,
3502 tdzp, tnm, &tzp, zflg, NULL, NULL);
3503 serr = zfs_dirent_lock(&sdl,
3504 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3510 * Source entry invalid or not there.
3513 zfs_dirent_unlock(tdl);
3519 rw_exit(&sdzp->z_name_lock);
3521 if (strcmp(snm, "..") == 0)
3527 zfs_dirent_unlock(sdl);
3531 rw_exit(&sdzp->z_name_lock);
3533 if (strcmp(tnm, "..") == 0)
3540 * Must have write access at the source to remove the old entry
3541 * and write access at the target to create the new entry.
3542 * Note that if target and source are the same, this can be
3543 * done in a single check.
3546 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3549 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3551 * Check to make sure rename is valid.
3552 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3554 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3559 * Does target exist?
3563 * Source and target must be the same type.
3565 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3566 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3567 error = SET_ERROR(ENOTDIR);
3571 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3572 error = SET_ERROR(EISDIR);
3577 * POSIX dictates that when the source and target
3578 * entries refer to the same file object, rename
3579 * must do nothing and exit without error.
3581 if (szp->z_id == tzp->z_id) {
3587 tx = dmu_tx_create(zsb->z_os);
3588 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3589 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3590 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3591 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3593 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3594 zfs_sa_upgrade_txholds(tx, tdzp);
3597 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3598 zfs_sa_upgrade_txholds(tx, tzp);
3601 zfs_sa_upgrade_txholds(tx, szp);
3602 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
3603 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3606 zfs_rename_unlock(&zl);
3607 zfs_dirent_unlock(sdl);
3608 zfs_dirent_unlock(tdl);
3611 rw_exit(&sdzp->z_name_lock);
3616 if (error == ERESTART) {
3627 if (tzp) /* Attempt to remove the existing target */
3628 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3631 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3633 szp->z_pflags |= ZFS_AV_MODIFIED;
3635 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zsb),
3636 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3639 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3641 zfs_log_rename(zilog, tx, TX_RENAME |
3642 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3643 sdl->dl_name, tdzp, tdl->dl_name, szp);
3646 * At this point, we have successfully created
3647 * the target name, but have failed to remove
3648 * the source name. Since the create was done
3649 * with the ZRENAMING flag, there are
3650 * complications; for one, the link count is
3651 * wrong. The easiest way to deal with this
3652 * is to remove the newly created target, and
3653 * return the original error. This must
3654 * succeed; fortunately, it is very unlikely to
3655 * fail, since we just created it.
3657 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3658 ZRENAMING, NULL), ==, 0);
3666 zfs_rename_unlock(&zl);
3668 zfs_dirent_unlock(sdl);
3669 zfs_dirent_unlock(tdl);
3671 zfs_inode_update(sdzp);
3673 rw_exit(&sdzp->z_name_lock);
3676 zfs_inode_update(tdzp);
3678 zfs_inode_update(szp);
3681 zfs_inode_update(tzp);
3685 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3686 zil_commit(zilog, 0);
3691 EXPORT_SYMBOL(zfs_rename);
3694 * Insert the indicated symbolic reference entry into the directory.
3696 * IN: dip - Directory to contain new symbolic link.
3697 * link - Name for new symlink entry.
3698 * vap - Attributes of new entry.
3699 * target - Target path of new symlink.
3701 * cr - credentials of caller.
3702 * flags - case flags
3704 * RETURN: 0 on success, error code on failure.
3707 * dip - ctime|mtime updated
3711 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
3712 struct inode **ipp, cred_t *cr, int flags)
3714 znode_t *zp, *dzp = ITOZ(dip);
3717 zfs_sb_t *zsb = ITOZSB(dip);
3719 uint64_t len = strlen(link);
3722 zfs_acl_ids_t acl_ids;
3723 boolean_t fuid_dirtied;
3724 uint64_t txtype = TX_SYMLINK;
3725 boolean_t waited = B_FALSE;
3727 ASSERT(S_ISLNK(vap->va_mode));
3733 if (zsb->z_utf8 && u8_validate(name, strlen(name),
3734 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3736 return (SET_ERROR(EILSEQ));
3738 if (flags & FIGNORECASE)
3741 if (len > MAXPATHLEN) {
3743 return (SET_ERROR(ENAMETOOLONG));
3746 if ((error = zfs_acl_ids_create(dzp, 0,
3747 vap, cr, NULL, &acl_ids)) != 0) {
3755 * Attempt to lock directory; fail if entry already exists.
3757 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3759 zfs_acl_ids_free(&acl_ids);
3764 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3765 zfs_acl_ids_free(&acl_ids);
3766 zfs_dirent_unlock(dl);
3771 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
3772 zfs_acl_ids_free(&acl_ids);
3773 zfs_dirent_unlock(dl);
3775 return (SET_ERROR(EDQUOT));
3777 tx = dmu_tx_create(zsb->z_os);
3778 fuid_dirtied = zsb->z_fuid_dirty;
3779 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3780 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3781 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3782 ZFS_SA_BASE_ATTR_SIZE + len);
3783 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3784 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3785 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3786 acl_ids.z_aclp->z_acl_bytes);
3789 zfs_fuid_txhold(zsb, tx);
3790 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3792 zfs_dirent_unlock(dl);
3793 if (error == ERESTART) {
3799 zfs_acl_ids_free(&acl_ids);
3806 * Create a new object for the symlink.
3807 * for version 4 ZPL datsets the symlink will be an SA attribute
3809 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3812 zfs_fuid_sync(zsb, tx);
3814 mutex_enter(&zp->z_lock);
3816 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zsb),
3819 zfs_sa_symlink(zp, link, len, tx);
3820 mutex_exit(&zp->z_lock);
3823 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
3824 &zp->z_size, sizeof (zp->z_size), tx);
3826 * Insert the new object into the directory.
3828 (void) zfs_link_create(dl, zp, tx, ZNEW);
3830 if (flags & FIGNORECASE)
3832 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3834 zfs_inode_update(dzp);
3835 zfs_inode_update(zp);
3837 zfs_acl_ids_free(&acl_ids);
3841 zfs_dirent_unlock(dl);
3845 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3846 zil_commit(zilog, 0);
3851 EXPORT_SYMBOL(zfs_symlink);
3854 * Return, in the buffer contained in the provided uio structure,
3855 * the symbolic path referred to by ip.
3857 * IN: ip - inode of symbolic link
3858 * uio - structure to contain the link path.
3859 * cr - credentials of caller.
3861 * RETURN: 0 if success
3862 * error code if failure
3865 * ip - atime updated
3869 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
3871 znode_t *zp = ITOZ(ip);
3872 zfs_sb_t *zsb = ITOZSB(ip);
3878 mutex_enter(&zp->z_lock);
3880 error = sa_lookup_uio(zp->z_sa_hdl,
3881 SA_ZPL_SYMLINK(zsb), uio);
3883 error = zfs_sa_readlink(zp, uio);
3884 mutex_exit(&zp->z_lock);
3889 EXPORT_SYMBOL(zfs_readlink);
3892 * Insert a new entry into directory tdip referencing sip.
3894 * IN: tdip - Directory to contain new entry.
3895 * sip - inode of new entry.
3896 * name - name of new entry.
3897 * cr - credentials of caller.
3899 * RETURN: 0 if success
3900 * error code if failure
3903 * tdip - ctime|mtime updated
3904 * sip - ctime updated
3908 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr,
3911 znode_t *dzp = ITOZ(tdip);
3913 zfs_sb_t *zsb = ITOZSB(tdip);
3921 boolean_t waited = B_FALSE;
3922 boolean_t is_tmpfile = 0;
3925 is_tmpfile = (sip->i_nlink == 0 && (sip->i_state & I_LINKABLE));
3927 ASSERT(S_ISDIR(tdip->i_mode));
3934 * POSIX dictates that we return EPERM here.
3935 * Better choices include ENOTSUP or EISDIR.
3937 if (S_ISDIR(sip->i_mode)) {
3939 return (SET_ERROR(EPERM));
3946 * We check i_sb because snapshots and the ctldir must have different
3949 if (sip->i_sb != tdip->i_sb || zfsctl_is_node(sip)) {
3951 return (SET_ERROR(EXDEV));
3954 /* Prevent links to .zfs/shares files */
3956 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zsb),
3957 &parent, sizeof (uint64_t))) != 0) {
3961 if (parent == zsb->z_shares_dir) {
3963 return (SET_ERROR(EPERM));
3966 if (zsb->z_utf8 && u8_validate(name,
3967 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3969 return (SET_ERROR(EILSEQ));
3971 if (flags & FIGNORECASE)
3975 * We do not support links between attributes and non-attributes
3976 * because of the potential security risk of creating links
3977 * into "normal" file space in order to circumvent restrictions
3978 * imposed in attribute space.
3980 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3982 return (SET_ERROR(EINVAL));
3985 owner = zfs_fuid_map_id(zsb, KUID_TO_SUID(sip->i_uid), cr, ZFS_OWNER);
3986 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3988 return (SET_ERROR(EPERM));
3991 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3998 * Attempt to lock directory; fail if entry already exists.
4000 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4006 tx = dmu_tx_create(zsb->z_os);
4007 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4008 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4010 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
4012 zfs_sa_upgrade_txholds(tx, szp);
4013 zfs_sa_upgrade_txholds(tx, dzp);
4014 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4016 zfs_dirent_unlock(dl);
4017 if (error == ERESTART) {
4027 /* unmark z_unlinked so zfs_link_create will not reject */
4029 szp->z_unlinked = 0;
4030 error = zfs_link_create(dl, szp, tx, 0);
4033 uint64_t txtype = TX_LINK;
4035 * tmpfile is created to be in z_unlinkedobj, so remove it.
4036 * Also, we don't log in ZIL, be cause all previous file
4037 * operation on the tmpfile are ignored by ZIL. Instead we
4038 * always wait for txg to sync to make sure all previous
4039 * operation are sync safe.
4042 VERIFY(zap_remove_int(zsb->z_os, zsb->z_unlinkedobj,
4043 szp->z_id, tx) == 0);
4045 if (flags & FIGNORECASE)
4047 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4049 } else if (is_tmpfile) {
4050 /* restore z_unlinked since when linking failed */
4051 szp->z_unlinked = 1;
4053 txg = dmu_tx_get_txg(tx);
4056 zfs_dirent_unlock(dl);
4058 if (!is_tmpfile && zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
4059 zil_commit(zilog, 0);
4062 txg_wait_synced(dmu_objset_pool(zsb->z_os), txg);
4064 zfs_inode_update(dzp);
4065 zfs_inode_update(szp);
4069 EXPORT_SYMBOL(zfs_link);
4072 zfs_putpage_commit_cb(void *arg)
4074 struct page *pp = arg;
4077 end_page_writeback(pp);
4081 * Push a page out to disk, once the page is on stable storage the
4082 * registered commit callback will be run as notification of completion.
4084 * IN: ip - page mapped for inode.
4085 * pp - page to push (page is locked)
4086 * wbc - writeback control data
4088 * RETURN: 0 if success
4089 * error code if failure
4092 * ip - ctime|mtime updated
4096 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
4098 znode_t *zp = ITOZ(ip);
4099 zfs_sb_t *zsb = ITOZSB(ip);
4107 uint64_t mtime[2], ctime[2];
4108 sa_bulk_attr_t bulk[3];
4110 struct address_space *mapping;
4115 ASSERT(PageLocked(pp));
4117 pgoff = page_offset(pp); /* Page byte-offset in file */
4118 offset = i_size_read(ip); /* File length in bytes */
4119 pglen = MIN(PAGE_SIZE, /* Page length in bytes */
4120 P2ROUNDUP(offset, PAGE_SIZE)-pgoff);
4122 /* Page is beyond end of file */
4123 if (pgoff >= offset) {
4129 /* Truncate page length to end of file */
4130 if (pgoff + pglen > offset)
4131 pglen = offset - pgoff;
4135 * FIXME: Allow mmap writes past its quota. The correct fix
4136 * is to register a page_mkwrite() handler to count the page
4137 * against its quota when it is about to be dirtied.
4139 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
4140 zfs_owner_overquota(zsb, zp, B_TRUE)) {
4146 * The ordering here is critical and must adhere to the following
4147 * rules in order to avoid deadlocking in either zfs_read() or
4148 * zfs_free_range() due to a lock inversion.
4150 * 1) The page must be unlocked prior to acquiring the range lock.
4151 * This is critical because zfs_read() calls find_lock_page()
4152 * which may block on the page lock while holding the range lock.
4154 * 2) Before setting or clearing write back on a page the range lock
4155 * must be held in order to prevent a lock inversion with the
4156 * zfs_free_range() function.
4158 * This presents a problem because upon entering this function the
4159 * page lock is already held. To safely acquire the range lock the
4160 * page lock must be dropped. This creates a window where another
4161 * process could truncate, invalidate, dirty, or write out the page.
4163 * Therefore, after successfully reacquiring the range and page locks
4164 * the current page state is checked. In the common case everything
4165 * will be as is expected and it can be written out. However, if
4166 * the page state has changed it must be handled accordingly.
4168 mapping = pp->mapping;
4169 redirty_page_for_writepage(wbc, pp);
4172 rl = zfs_range_lock(&zp->z_range_lock, pgoff, pglen, RL_WRITER);
4175 /* Page mapping changed or it was no longer dirty, we're done */
4176 if (unlikely((mapping != pp->mapping) || !PageDirty(pp))) {
4178 zfs_range_unlock(rl);
4183 /* Another process started write block if required */
4184 if (PageWriteback(pp)) {
4186 zfs_range_unlock(rl);
4188 if (wbc->sync_mode != WB_SYNC_NONE)
4189 wait_on_page_writeback(pp);
4195 /* Clear the dirty flag the required locks are held */
4196 if (!clear_page_dirty_for_io(pp)) {
4198 zfs_range_unlock(rl);
4204 * Counterpart for redirty_page_for_writepage() above. This page
4205 * was in fact not skipped and should not be counted as if it were.
4207 wbc->pages_skipped--;
4208 set_page_writeback(pp);
4211 tx = dmu_tx_create(zsb->z_os);
4212 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
4213 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4214 zfs_sa_upgrade_txholds(tx, zp);
4216 err = dmu_tx_assign(tx, TXG_NOWAIT);
4218 if (err == ERESTART)
4222 __set_page_dirty_nobuffers(pp);
4224 end_page_writeback(pp);
4225 zfs_range_unlock(rl);
4231 ASSERT3U(pglen, <=, PAGE_SIZE);
4232 dmu_write(zsb->z_os, zp->z_id, pgoff, pglen, va, tx);
4235 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
4236 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
4237 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zsb), NULL, &zp->z_pflags, 8);
4239 /* Preserve the mtime and ctime provided by the inode */
4240 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4241 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4242 zp->z_atime_dirty = 0;
4245 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4247 zfs_log_write(zsb->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0,
4248 zfs_putpage_commit_cb, pp);
4251 zfs_range_unlock(rl);
4253 if (wbc->sync_mode != WB_SYNC_NONE) {
4255 * Note that this is rarely called under writepages(), because
4256 * writepages() normally handles the entire commit for
4257 * performance reasons.
4259 if (zsb->z_log != NULL)
4260 zil_commit(zsb->z_log, zp->z_id);
4268 * Update the system attributes when the inode has been dirtied. For the
4269 * moment we only update the mode, atime, mtime, and ctime.
4272 zfs_dirty_inode(struct inode *ip, int flags)
4274 znode_t *zp = ITOZ(ip);
4275 zfs_sb_t *zsb = ITOZSB(ip);
4277 uint64_t mode, atime[2], mtime[2], ctime[2];
4278 sa_bulk_attr_t bulk[4];
4282 if (zfs_is_readonly(zsb) || dmu_objset_is_snapshot(zsb->z_os))
4290 * This is the lazytime semantic indroduced in Linux 4.0
4291 * This flag will only be called from update_time when lazytime is set.
4292 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4293 * Fortunately mtime and ctime are managed within ZFS itself, so we
4294 * only need to dirty atime.
4296 if (flags == I_DIRTY_TIME) {
4297 zp->z_atime_dirty = 1;
4302 tx = dmu_tx_create(zsb->z_os);
4304 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4305 zfs_sa_upgrade_txholds(tx, zp);
4307 error = dmu_tx_assign(tx, TXG_WAIT);
4313 mutex_enter(&zp->z_lock);
4314 zp->z_atime_dirty = 0;
4316 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zsb), NULL, &mode, 8);
4317 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zsb), NULL, &atime, 16);
4318 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
4319 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
4321 /* Preserve the mode, mtime and ctime provided by the inode */
4322 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4323 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4324 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4329 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4330 mutex_exit(&zp->z_lock);
4337 EXPORT_SYMBOL(zfs_dirty_inode);
4341 zfs_inactive(struct inode *ip)
4343 znode_t *zp = ITOZ(ip);
4344 zfs_sb_t *zsb = ITOZSB(ip);
4347 int need_unlock = 0;
4349 /* Only read lock if we haven't already write locked, e.g. rollback */
4350 if (!RW_WRITE_HELD(&zsb->z_teardown_inactive_lock)) {
4352 rw_enter(&zsb->z_teardown_inactive_lock, RW_READER);
4354 if (zp->z_sa_hdl == NULL) {
4356 rw_exit(&zsb->z_teardown_inactive_lock);
4360 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4361 dmu_tx_t *tx = dmu_tx_create(zsb->z_os);
4363 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4364 zfs_sa_upgrade_txholds(tx, zp);
4365 error = dmu_tx_assign(tx, TXG_WAIT);
4369 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4370 mutex_enter(&zp->z_lock);
4371 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zsb),
4372 (void *)&atime, sizeof (atime), tx);
4373 zp->z_atime_dirty = 0;
4374 mutex_exit(&zp->z_lock);
4381 rw_exit(&zsb->z_teardown_inactive_lock);
4383 EXPORT_SYMBOL(zfs_inactive);
4386 * Bounds-check the seek operation.
4388 * IN: ip - inode seeking within
4389 * ooff - old file offset
4390 * noffp - pointer to new file offset
4391 * ct - caller context
4393 * RETURN: 0 if success
4394 * EINVAL if new offset invalid
4398 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
4400 if (S_ISDIR(ip->i_mode))
4402 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4404 EXPORT_SYMBOL(zfs_seek);
4407 * Fill pages with data from the disk.
4410 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
4412 znode_t *zp = ITOZ(ip);
4413 zfs_sb_t *zsb = ITOZSB(ip);
4415 struct page *cur_pp;
4416 u_offset_t io_off, total;
4423 io_len = nr_pages << PAGE_SHIFT;
4424 i_size = i_size_read(ip);
4425 io_off = page_offset(pl[0]);
4427 if (io_off + io_len > i_size)
4428 io_len = i_size - io_off;
4431 * Iterate over list of pages and read each page individually.
4434 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4437 cur_pp = pl[page_idx++];
4439 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4443 /* convert checksum errors into IO errors */
4445 err = SET_ERROR(EIO);
4454 * Uses zfs_fillpage to read data from the file and fill the pages.
4456 * IN: ip - inode of file to get data from.
4457 * pl - list of pages to read
4458 * nr_pages - number of pages to read
4460 * RETURN: 0 on success, error code on failure.
4463 * vp - atime updated
4467 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
4469 znode_t *zp = ITOZ(ip);
4470 zfs_sb_t *zsb = ITOZSB(ip);
4479 err = zfs_fillpage(ip, pl, nr_pages);
4484 EXPORT_SYMBOL(zfs_getpage);
4487 * Check ZFS specific permissions to memory map a section of a file.
4489 * IN: ip - inode of the file to mmap
4491 * addrp - start address in memory region
4492 * len - length of memory region
4493 * vm_flags- address flags
4495 * RETURN: 0 if success
4496 * error code if failure
4500 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4501 unsigned long vm_flags)
4503 znode_t *zp = ITOZ(ip);
4504 zfs_sb_t *zsb = ITOZSB(ip);
4509 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4510 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4512 return (SET_ERROR(EPERM));
4515 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4516 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4518 return (SET_ERROR(EACCES));
4521 if (off < 0 || len > MAXOFFSET_T - off) {
4523 return (SET_ERROR(ENXIO));
4529 EXPORT_SYMBOL(zfs_map);
4532 * convoff - converts the given data (start, whence) to the
4536 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4541 if ((lckdat->l_whence == 2) || (whence == 2)) {
4542 if ((error = zfs_getattr(ip, &vap, 0, CRED()) != 0))
4546 switch (lckdat->l_whence) {
4548 lckdat->l_start += offset;
4551 lckdat->l_start += vap.va_size;
4556 return (SET_ERROR(EINVAL));
4559 if (lckdat->l_start < 0)
4560 return (SET_ERROR(EINVAL));
4564 lckdat->l_start -= offset;
4567 lckdat->l_start -= vap.va_size;
4572 return (SET_ERROR(EINVAL));
4575 lckdat->l_whence = (short)whence;
4580 * Free or allocate space in a file. Currently, this function only
4581 * supports the `F_FREESP' command. However, this command is somewhat
4582 * misnamed, as its functionality includes the ability to allocate as
4583 * well as free space.
4585 * IN: ip - inode of file to free data in.
4586 * cmd - action to take (only F_FREESP supported).
4587 * bfp - section of file to free/alloc.
4588 * flag - current file open mode flags.
4589 * offset - current file offset.
4590 * cr - credentials of caller [UNUSED].
4592 * RETURN: 0 on success, error code on failure.
4595 * ip - ctime|mtime updated
4599 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4600 offset_t offset, cred_t *cr)
4602 znode_t *zp = ITOZ(ip);
4603 zfs_sb_t *zsb = ITOZSB(ip);
4610 if (cmd != F_FREESP) {
4612 return (SET_ERROR(EINVAL));
4616 * Callers might not be able to detect properly that we are read-only,
4617 * so check it explicitly here.
4619 if (zfs_is_readonly(zsb)) {
4621 return (SET_ERROR(EROFS));
4624 if ((error = convoff(ip, bfp, 0, offset))) {
4629 if (bfp->l_len < 0) {
4631 return (SET_ERROR(EINVAL));
4635 * Permissions aren't checked on Solaris because on this OS
4636 * zfs_space() can only be called with an opened file handle.
4637 * On Linux we can get here through truncate_range() which
4638 * operates directly on inodes, so we need to check access rights.
4640 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4646 len = bfp->l_len; /* 0 means from off to end of file */
4648 error = zfs_freesp(zp, off, len, flag, TRUE);
4653 EXPORT_SYMBOL(zfs_space);
4657 zfs_fid(struct inode *ip, fid_t *fidp)
4659 znode_t *zp = ITOZ(ip);
4660 zfs_sb_t *zsb = ITOZSB(ip);
4663 uint64_t object = zp->z_id;
4670 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zsb),
4671 &gen64, sizeof (uint64_t))) != 0) {
4676 gen = (uint32_t)gen64;
4678 size = (zsb->z_parent != zsb) ? LONG_FID_LEN : SHORT_FID_LEN;
4679 if (fidp->fid_len < size) {
4680 fidp->fid_len = size;
4682 return (SET_ERROR(ENOSPC));
4685 zfid = (zfid_short_t *)fidp;
4687 zfid->zf_len = size;
4689 for (i = 0; i < sizeof (zfid->zf_object); i++)
4690 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4692 /* Must have a non-zero generation number to distinguish from .zfs */
4695 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4696 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4698 if (size == LONG_FID_LEN) {
4699 uint64_t objsetid = dmu_objset_id(zsb->z_os);
4702 zlfid = (zfid_long_t *)fidp;
4704 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4705 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4707 /* XXX - this should be the generation number for the objset */
4708 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4709 zlfid->zf_setgen[i] = 0;
4715 EXPORT_SYMBOL(zfs_fid);
4719 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4721 znode_t *zp = ITOZ(ip);
4722 zfs_sb_t *zsb = ITOZSB(ip);
4724 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4728 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4733 EXPORT_SYMBOL(zfs_getsecattr);
4737 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4739 znode_t *zp = ITOZ(ip);
4740 zfs_sb_t *zsb = ITOZSB(ip);
4742 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4743 zilog_t *zilog = zsb->z_log;
4748 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4750 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
4751 zil_commit(zilog, 0);
4756 EXPORT_SYMBOL(zfs_setsecattr);
4758 #ifdef HAVE_UIO_ZEROCOPY
4760 * Tunable, both must be a power of 2.
4762 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4763 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4764 * an arcbuf for a partial block read
4766 int zcr_blksz_min = (1 << 10); /* 1K */
4767 int zcr_blksz_max = (1 << 17); /* 128K */
4771 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
4773 znode_t *zp = ITOZ(ip);
4774 zfs_sb_t *zsb = ITOZSB(ip);
4775 int max_blksz = zsb->z_max_blksz;
4776 uio_t *uio = &xuio->xu_uio;
4777 ssize_t size = uio->uio_resid;
4778 offset_t offset = uio->uio_loffset;
4783 int preamble, postamble;
4785 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4786 return (SET_ERROR(EINVAL));
4793 * Loan out an arc_buf for write if write size is bigger than
4794 * max_blksz, and the file's block size is also max_blksz.
4797 if (size < blksz || zp->z_blksz != blksz) {
4799 return (SET_ERROR(EINVAL));
4802 * Caller requests buffers for write before knowing where the
4803 * write offset might be (e.g. NFS TCP write).
4808 preamble = P2PHASE(offset, blksz);
4810 preamble = blksz - preamble;
4815 postamble = P2PHASE(size, blksz);
4818 fullblk = size / blksz;
4819 (void) dmu_xuio_init(xuio,
4820 (preamble != 0) + fullblk + (postamble != 0));
4823 * Have to fix iov base/len for partial buffers. They
4824 * currently represent full arc_buf's.
4827 /* data begins in the middle of the arc_buf */
4828 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4831 (void) dmu_xuio_add(xuio, abuf,
4832 blksz - preamble, preamble);
4835 for (i = 0; i < fullblk; i++) {
4836 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4839 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
4843 /* data ends in the middle of the arc_buf */
4844 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4847 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
4852 * Loan out an arc_buf for read if the read size is larger than
4853 * the current file block size. Block alignment is not
4854 * considered. Partial arc_buf will be loaned out for read.
4856 blksz = zp->z_blksz;
4857 if (blksz < zcr_blksz_min)
4858 blksz = zcr_blksz_min;
4859 if (blksz > zcr_blksz_max)
4860 blksz = zcr_blksz_max;
4861 /* avoid potential complexity of dealing with it */
4862 if (blksz > max_blksz) {
4864 return (SET_ERROR(EINVAL));
4867 maxsize = zp->z_size - uio->uio_loffset;
4873 return (SET_ERROR(EINVAL));
4878 return (SET_ERROR(EINVAL));
4881 uio->uio_extflg = UIO_XUIO;
4882 XUIO_XUZC_RW(xuio) = ioflag;
4889 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
4893 int ioflag = XUIO_XUZC_RW(xuio);
4895 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
4897 i = dmu_xuio_cnt(xuio);
4899 abuf = dmu_xuio_arcbuf(xuio, i);
4901 * if abuf == NULL, it must be a write buffer
4902 * that has been returned in zfs_write().
4905 dmu_return_arcbuf(abuf);
4906 ASSERT(abuf || ioflag == UIO_WRITE);
4909 dmu_xuio_fini(xuio);
4912 #endif /* HAVE_UIO_ZEROCOPY */
4914 #if defined(_KERNEL) && defined(HAVE_SPL)
4915 module_param(zfs_delete_blocks, ulong, 0644);
4916 MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async");
4917 module_param(zfs_read_chunk_size, long, 0644);
4918 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");