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, uio->uio_loffset, uio->uio_resid, RL_READER);
489 * If we are reading past end-of-file we can skip
490 * to the end; but we might still need to set atime.
492 if (uio->uio_loffset >= zp->z_size) {
497 ASSERT(uio->uio_loffset < zp->z_size);
498 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
500 #ifdef HAVE_UIO_ZEROCOPY
501 if ((uio->uio_extflg == UIO_XUIO) &&
502 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
504 int blksz = zp->z_blksz;
505 uint64_t offset = uio->uio_loffset;
507 xuio = (xuio_t *)uio;
509 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
512 ASSERT(offset + n <= blksz);
515 (void) dmu_xuio_init(xuio, nblk);
517 if (vn_has_cached_data(ip)) {
519 * For simplicity, we always allocate a full buffer
520 * even if we only expect to read a portion of a block.
522 while (--nblk >= 0) {
523 (void) dmu_xuio_add(xuio,
524 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
529 #endif /* HAVE_UIO_ZEROCOPY */
532 nbytes = MIN(n, zfs_read_chunk_size -
533 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
535 if (zp->z_is_mapped && !(ioflag & O_DIRECT)) {
536 error = mappedread(ip, nbytes, uio);
538 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
543 /* convert checksum errors into IO errors */
545 error = SET_ERROR(EIO);
552 zfs_range_unlock(rl);
557 EXPORT_SYMBOL(zfs_read);
560 * Write the bytes to a file.
562 * IN: ip - inode of file to be written to.
563 * uio - structure supplying write location, range info,
565 * ioflag - FAPPEND flag set if in append mode.
566 * O_DIRECT flag; used to bypass page cache.
567 * cr - credentials of caller.
569 * OUT: uio - updated offset and range.
571 * RETURN: 0 if success
572 * error code if failure
575 * ip - ctime|mtime updated if byte count > 0
580 zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
582 znode_t *zp = ITOZ(ip);
583 rlim64_t limit = uio->uio_limit;
584 ssize_t start_resid = uio->uio_resid;
588 zfs_sb_t *zsb = ZTOZSB(zp);
593 int max_blksz = zsb->z_max_blksz;
596 const iovec_t *aiov = NULL;
599 const iovec_t *iovp = uio->uio_iov;
602 sa_bulk_attr_t bulk[4];
603 uint64_t mtime[2], ctime[2];
604 ASSERTV(int iovcnt = uio->uio_iovcnt);
607 * Fasttrack empty write
613 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
619 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
620 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
621 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb), NULL, &zp->z_size, 8);
622 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
626 * Callers might not be able to detect properly that we are read-only,
627 * so check it explicitly here.
629 if (zfs_is_readonly(zsb)) {
631 return (SET_ERROR(EROFS));
635 * If immutable or not appending then return EPERM
637 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
638 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
639 (uio->uio_loffset < zp->z_size))) {
641 return (SET_ERROR(EPERM));
647 * Validate file offset
649 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
652 return (SET_ERROR(EINVAL));
656 * Pre-fault the pages to ensure slow (eg NFS) pages
658 * Skip this if uio contains loaned arc_buf.
660 #ifdef HAVE_UIO_ZEROCOPY
661 if ((uio->uio_extflg == UIO_XUIO) &&
662 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
663 xuio = (xuio_t *)uio;
666 uio_prefaultpages(MIN(n, max_blksz), uio);
669 * If in append mode, set the io offset pointer to eof.
671 if (ioflag & FAPPEND) {
673 * Obtain an appending range lock to guarantee file append
674 * semantics. We reset the write offset once we have the lock.
676 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
678 if (rl->r_len == UINT64_MAX) {
680 * We overlocked the file because this write will cause
681 * the file block size to increase.
682 * Note that zp_size cannot change with this lock held.
686 uio->uio_loffset = woff;
689 * Note that if the file block size will change as a result of
690 * this write, then this range lock will lock the entire file
691 * so that we can re-write the block safely.
693 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
697 zfs_range_unlock(rl);
699 return (SET_ERROR(EFBIG));
702 if ((woff + n) > limit || woff > (limit - n))
705 /* Will this write extend the file length? */
706 write_eof = (woff + n > zp->z_size);
708 end_size = MAX(zp->z_size, woff + n);
711 * Write the file in reasonable size chunks. Each chunk is written
712 * in a separate transaction; this keeps the intent log records small
713 * and allows us to do more fine-grained space accounting.
717 woff = uio->uio_loffset;
718 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
719 zfs_owner_overquota(zsb, zp, B_TRUE)) {
721 dmu_return_arcbuf(abuf);
722 error = SET_ERROR(EDQUOT);
726 if (xuio && abuf == NULL) {
727 ASSERT(i_iov < iovcnt);
728 ASSERT3U(uio->uio_segflg, !=, UIO_BVEC);
730 abuf = dmu_xuio_arcbuf(xuio, i_iov);
731 dmu_xuio_clear(xuio, i_iov);
732 ASSERT((aiov->iov_base == abuf->b_data) ||
733 ((char *)aiov->iov_base - (char *)abuf->b_data +
734 aiov->iov_len == arc_buf_size(abuf)));
736 } else if (abuf == NULL && n >= max_blksz &&
737 woff >= zp->z_size &&
738 P2PHASE(woff, max_blksz) == 0 &&
739 zp->z_blksz == max_blksz) {
741 * This write covers a full block. "Borrow" a buffer
742 * from the dmu so that we can fill it before we enter
743 * a transaction. This avoids the possibility of
744 * holding up the transaction if the data copy hangs
745 * up on a pagefault (e.g., from an NFS server mapping).
749 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
751 ASSERT(abuf != NULL);
752 ASSERT(arc_buf_size(abuf) == max_blksz);
753 if ((error = uiocopy(abuf->b_data, max_blksz,
754 UIO_WRITE, uio, &cbytes))) {
755 dmu_return_arcbuf(abuf);
758 ASSERT(cbytes == max_blksz);
762 * Start a transaction.
764 tx = dmu_tx_create(zsb->z_os);
765 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
766 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
767 zfs_sa_upgrade_txholds(tx, zp);
768 error = dmu_tx_assign(tx, TXG_WAIT);
772 dmu_return_arcbuf(abuf);
777 * If zfs_range_lock() over-locked we grow the blocksize
778 * and then reduce the lock range. This will only happen
779 * on the first iteration since zfs_range_reduce() will
780 * shrink down r_len to the appropriate size.
782 if (rl->r_len == UINT64_MAX) {
785 if (zp->z_blksz > max_blksz) {
787 * File's blocksize is already larger than the
788 * "recordsize" property. Only let it grow to
789 * the next power of 2.
791 ASSERT(!ISP2(zp->z_blksz));
792 new_blksz = MIN(end_size,
793 1 << highbit64(zp->z_blksz));
795 new_blksz = MIN(end_size, max_blksz);
797 zfs_grow_blocksize(zp, new_blksz, tx);
798 zfs_range_reduce(rl, woff, n);
802 * XXX - should we really limit each write to z_max_blksz?
803 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
805 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
808 tx_bytes = uio->uio_resid;
809 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
811 tx_bytes -= uio->uio_resid;
814 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
816 * If this is not a full block write, but we are
817 * extending the file past EOF and this data starts
818 * block-aligned, use assign_arcbuf(). Otherwise,
819 * write via dmu_write().
821 if (tx_bytes < max_blksz && (!write_eof ||
822 aiov->iov_base != abuf->b_data)) {
824 dmu_write(zsb->z_os, zp->z_id, woff,
825 aiov->iov_len, aiov->iov_base, tx);
826 dmu_return_arcbuf(abuf);
827 xuio_stat_wbuf_copied();
829 ASSERT(xuio || tx_bytes == max_blksz);
830 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
833 ASSERT(tx_bytes <= uio->uio_resid);
834 uioskip(uio, tx_bytes);
837 if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT))
838 update_pages(ip, woff, tx_bytes, zsb->z_os, zp->z_id);
841 * If we made no progress, we're done. If we made even
842 * partial progress, update the znode and ZIL accordingly.
845 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
846 (void *)&zp->z_size, sizeof (uint64_t), tx);
853 * Clear Set-UID/Set-GID bits on successful write if not
854 * privileged and at least one of the excute bits is set.
856 * It would be nice to to this after all writes have
857 * been done, but that would still expose the ISUID/ISGID
858 * to another app after the partial write is committed.
860 * Note: we don't call zfs_fuid_map_id() here because
861 * user 0 is not an ephemeral uid.
863 mutex_enter(&zp->z_acl_lock);
864 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
865 (S_IXUSR >> 6))) != 0 &&
866 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
867 secpolicy_vnode_setid_retain(cr,
868 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
870 zp->z_mode &= ~(S_ISUID | S_ISGID);
871 newmode = zp->z_mode;
872 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zsb),
873 (void *)&newmode, sizeof (uint64_t), tx);
875 mutex_exit(&zp->z_acl_lock);
877 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
880 * Update the file size (zp_size) if it has changed;
881 * account for possible concurrent updates.
883 while ((end_size = zp->z_size) < uio->uio_loffset) {
884 (void) atomic_cas_64(&zp->z_size, end_size,
889 * If we are replaying and eof is non zero then force
890 * the file size to the specified eof. Note, there's no
891 * concurrency during replay.
893 if (zsb->z_replay && zsb->z_replay_eof != 0)
894 zp->z_size = zsb->z_replay_eof;
896 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
898 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag,
904 ASSERT(tx_bytes == nbytes);
908 uio_prefaultpages(MIN(n, max_blksz), uio);
911 zfs_inode_update(zp);
912 zfs_range_unlock(rl);
915 * If we're in replay mode, or we made no progress, return error.
916 * Otherwise, it's at least a partial write, so it's successful.
918 if (zsb->z_replay || uio->uio_resid == start_resid) {
923 if (ioflag & (FSYNC | FDSYNC) ||
924 zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
925 zil_commit(zilog, zp->z_id);
930 EXPORT_SYMBOL(zfs_write);
933 zfs_iput_async(struct inode *ip)
935 objset_t *os = ITOZSB(ip)->z_os;
937 ASSERT(atomic_read(&ip->i_count) > 0);
940 if (atomic_read(&ip->i_count) == 1)
941 taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os)),
942 (task_func_t *)iput, ip, TQ_SLEEP);
948 zfs_get_done(zgd_t *zgd, int error)
950 znode_t *zp = zgd->zgd_private;
953 dmu_buf_rele(zgd->zgd_db, zgd);
955 zfs_range_unlock(zgd->zgd_rl);
958 * Release the vnode asynchronously as we currently have the
959 * txg stopped from syncing.
961 zfs_iput_async(ZTOI(zp));
963 if (error == 0 && zgd->zgd_bp)
964 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
966 kmem_free(zgd, sizeof (zgd_t));
970 static int zil_fault_io = 0;
974 * Get data to generate a TX_WRITE intent log record.
977 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
980 objset_t *os = zsb->z_os;
982 uint64_t object = lr->lr_foid;
983 uint64_t offset = lr->lr_offset;
984 uint64_t size = lr->lr_length;
985 blkptr_t *bp = &lr->lr_blkptr;
994 * Nothing to do if the file has been removed
996 if (zfs_zget(zsb, object, &zp) != 0)
997 return (SET_ERROR(ENOENT));
998 if (zp->z_unlinked) {
1000 * Release the vnode asynchronously as we currently have the
1001 * txg stopped from syncing.
1003 zfs_iput_async(ZTOI(zp));
1004 return (SET_ERROR(ENOENT));
1007 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1008 zgd->zgd_zilog = zsb->z_log;
1009 zgd->zgd_private = zp;
1012 * Write records come in two flavors: immediate and indirect.
1013 * For small writes it's cheaper to store the data with the
1014 * log record (immediate); for large writes it's cheaper to
1015 * sync the data and get a pointer to it (indirect) so that
1016 * we don't have to write the data twice.
1018 if (buf != NULL) { /* immediate write */
1019 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1020 /* test for truncation needs to be done while range locked */
1021 if (offset >= zp->z_size) {
1022 error = SET_ERROR(ENOENT);
1024 error = dmu_read(os, object, offset, size, buf,
1025 DMU_READ_NO_PREFETCH);
1027 ASSERT(error == 0 || error == ENOENT);
1028 } else { /* indirect write */
1030 * Have to lock the whole block to ensure when it's
1031 * written out and it's checksum is being calculated
1032 * that no one can change the data. We need to re-check
1033 * blocksize after we get the lock in case it's changed!
1038 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1040 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1042 if (zp->z_blksz == size)
1045 zfs_range_unlock(zgd->zgd_rl);
1047 /* test for truncation needs to be done while range locked */
1048 if (lr->lr_offset >= zp->z_size)
1049 error = SET_ERROR(ENOENT);
1052 error = SET_ERROR(EIO);
1057 error = dmu_buf_hold(os, object, offset, zgd, &db,
1058 DMU_READ_NO_PREFETCH);
1061 blkptr_t *obp = dmu_buf_get_blkptr(db);
1063 ASSERT(BP_IS_HOLE(bp));
1070 ASSERT(db->db_offset == offset);
1071 ASSERT(db->db_size == size);
1073 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1075 ASSERT(error || lr->lr_length <= zp->z_blksz);
1078 * On success, we need to wait for the write I/O
1079 * initiated by dmu_sync() to complete before we can
1080 * release this dbuf. We will finish everything up
1081 * in the zfs_get_done() callback.
1086 if (error == EALREADY) {
1087 lr->lr_common.lrc_txtype = TX_WRITE2;
1093 zfs_get_done(zgd, error);
1100 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
1102 znode_t *zp = ITOZ(ip);
1103 zfs_sb_t *zsb = ITOZSB(ip);
1109 if (flag & V_ACE_MASK)
1110 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1112 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1117 EXPORT_SYMBOL(zfs_access);
1120 * Lookup an entry in a directory, or an extended attribute directory.
1121 * If it exists, return a held inode reference for it.
1123 * IN: dip - inode of directory to search.
1124 * nm - name of entry to lookup.
1125 * flags - LOOKUP_XATTR set if looking for an attribute.
1126 * cr - credentials of caller.
1127 * direntflags - directory lookup flags
1128 * realpnp - returned pathname.
1130 * OUT: ipp - inode of located entry, NULL if not found.
1132 * RETURN: 0 on success, error code on failure.
1139 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
1140 cred_t *cr, int *direntflags, pathname_t *realpnp)
1142 znode_t *zdp = ITOZ(dip);
1143 zfs_sb_t *zsb = ITOZSB(dip);
1147 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1149 if (!S_ISDIR(dip->i_mode)) {
1150 return (SET_ERROR(ENOTDIR));
1151 } else if (zdp->z_sa_hdl == NULL) {
1152 return (SET_ERROR(EIO));
1155 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1156 error = zfs_fastaccesschk_execute(zdp, cr);
1165 vnode_t *tvp = dnlc_lookup(dvp, nm);
1168 error = zfs_fastaccesschk_execute(zdp, cr);
1173 if (tvp == DNLC_NO_VNODE) {
1175 return (SET_ERROR(ENOENT));
1178 return (specvp_check(vpp, cr));
1181 #endif /* HAVE_DNLC */
1190 if (flags & LOOKUP_XATTR) {
1192 * We don't allow recursive attributes..
1193 * Maybe someday we will.
1195 if (zdp->z_pflags & ZFS_XATTR) {
1197 return (SET_ERROR(EINVAL));
1200 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1206 * Do we have permission to get into attribute directory?
1209 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1219 if (!S_ISDIR(dip->i_mode)) {
1221 return (SET_ERROR(ENOTDIR));
1225 * Check accessibility of directory.
1228 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1233 if (zsb->z_utf8 && u8_validate(nm, strlen(nm),
1234 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1236 return (SET_ERROR(EILSEQ));
1239 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1240 if ((error == 0) && (*ipp))
1241 zfs_inode_update(ITOZ(*ipp));
1246 EXPORT_SYMBOL(zfs_lookup);
1249 * Attempt to create a new entry in a directory. If the entry
1250 * already exists, truncate the file if permissible, else return
1251 * an error. Return the ip of the created or trunc'd file.
1253 * IN: dip - inode of directory to put new file entry in.
1254 * name - name of new file entry.
1255 * vap - attributes of new file.
1256 * excl - flag indicating exclusive or non-exclusive mode.
1257 * mode - mode to open file with.
1258 * cr - credentials of caller.
1259 * flag - large file flag [UNUSED].
1260 * vsecp - ACL to be set
1262 * OUT: ipp - inode of created or trunc'd entry.
1264 * RETURN: 0 on success, error code on failure.
1267 * dip - ctime|mtime updated if new entry created
1268 * ip - ctime|mtime always, atime if new
1273 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1274 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1276 znode_t *zp, *dzp = ITOZ(dip);
1277 zfs_sb_t *zsb = ITOZSB(dip);
1285 zfs_acl_ids_t acl_ids;
1286 boolean_t fuid_dirtied;
1287 boolean_t have_acl = B_FALSE;
1288 boolean_t waited = B_FALSE;
1291 * If we have an ephemeral id, ACL, or XVATTR then
1292 * make sure file system is at proper version
1298 if (zsb->z_use_fuids == B_FALSE &&
1299 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1300 return (SET_ERROR(EINVAL));
1307 if (zsb->z_utf8 && u8_validate(name, strlen(name),
1308 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1310 return (SET_ERROR(EILSEQ));
1313 if (vap->va_mask & ATTR_XVATTR) {
1314 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1315 crgetuid(cr), cr, vap->va_mode)) != 0) {
1323 if (*name == '\0') {
1325 * Null component name refers to the directory itself.
1332 /* possible igrab(zp) */
1335 if (flag & FIGNORECASE)
1338 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1342 zfs_acl_ids_free(&acl_ids);
1343 if (strcmp(name, "..") == 0)
1344 error = SET_ERROR(EISDIR);
1354 * Create a new file object and update the directory
1357 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1359 zfs_acl_ids_free(&acl_ids);
1364 * We only support the creation of regular files in
1365 * extended attribute directories.
1368 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1370 zfs_acl_ids_free(&acl_ids);
1371 error = SET_ERROR(EINVAL);
1375 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1376 cr, vsecp, &acl_ids)) != 0)
1380 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1381 zfs_acl_ids_free(&acl_ids);
1382 error = SET_ERROR(EDQUOT);
1386 tx = dmu_tx_create(os);
1388 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1389 ZFS_SA_BASE_ATTR_SIZE);
1391 fuid_dirtied = zsb->z_fuid_dirty;
1393 zfs_fuid_txhold(zsb, tx);
1394 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1395 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1396 if (!zsb->z_use_sa &&
1397 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1398 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1399 0, acl_ids.z_aclp->z_acl_bytes);
1401 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1403 zfs_dirent_unlock(dl);
1404 if (error == ERESTART) {
1410 zfs_acl_ids_free(&acl_ids);
1415 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1418 zfs_fuid_sync(zsb, tx);
1420 (void) zfs_link_create(dl, zp, tx, ZNEW);
1421 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1422 if (flag & FIGNORECASE)
1424 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1425 vsecp, acl_ids.z_fuidp, vap);
1426 zfs_acl_ids_free(&acl_ids);
1429 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1432 zfs_acl_ids_free(&acl_ids);
1436 * A directory entry already exists for this name.
1439 * Can't truncate an existing file if in exclusive mode.
1442 error = SET_ERROR(EEXIST);
1446 * Can't open a directory for writing.
1448 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1449 error = SET_ERROR(EISDIR);
1453 * Verify requested access to file.
1455 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1459 mutex_enter(&dzp->z_lock);
1461 mutex_exit(&dzp->z_lock);
1464 * Truncate regular files if requested.
1466 if (S_ISREG(ZTOI(zp)->i_mode) &&
1467 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1468 /* we can't hold any locks when calling zfs_freesp() */
1469 zfs_dirent_unlock(dl);
1471 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1477 zfs_dirent_unlock(dl);
1483 zfs_inode_update(dzp);
1484 zfs_inode_update(zp);
1488 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1489 zil_commit(zilog, 0);
1494 EXPORT_SYMBOL(zfs_create);
1497 * Remove an entry from a directory.
1499 * IN: dip - inode of directory to remove entry from.
1500 * name - name of entry to remove.
1501 * cr - credentials of caller.
1503 * RETURN: 0 if success
1504 * error code if failure
1508 * ip - ctime (if nlink > 0)
1511 uint64_t null_xattr = 0;
1515 zfs_remove(struct inode *dip, char *name, cred_t *cr)
1517 znode_t *zp, *dzp = ITOZ(dip);
1520 zfs_sb_t *zsb = ITOZSB(dip);
1522 uint64_t acl_obj, xattr_obj;
1523 uint64_t xattr_obj_unlinked = 0;
1527 boolean_t may_delete_now, delete_now = FALSE;
1528 boolean_t unlinked, toobig = FALSE;
1530 pathname_t *realnmp = NULL;
1531 #ifdef HAVE_PN_UTILS
1533 #endif /* HAVE_PN_UTILS */
1536 boolean_t waited = B_FALSE;
1542 #ifdef HAVE_PN_UTILS
1543 if (flags & FIGNORECASE) {
1548 #endif /* HAVE_PN_UTILS */
1554 * Attempt to lock directory; fail if entry doesn't exist.
1556 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1558 #ifdef HAVE_PN_UTILS
1561 #endif /* HAVE_PN_UTILS */
1568 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1573 * Need to use rmdir for removing directories.
1575 if (S_ISDIR(ip->i_mode)) {
1576 error = SET_ERROR(EPERM);
1582 dnlc_remove(dvp, realnmp->pn_buf);
1584 dnlc_remove(dvp, name);
1585 #endif /* HAVE_DNLC */
1587 mutex_enter(&zp->z_lock);
1588 may_delete_now = atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped);
1589 mutex_exit(&zp->z_lock);
1592 * We may delete the znode now, or we may put it in the unlinked set;
1593 * it depends on whether we're the last link, and on whether there are
1594 * other holds on the inode. So we dmu_tx_hold() the right things to
1595 * allow for either case.
1598 tx = dmu_tx_create(zsb->z_os);
1599 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1600 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1601 zfs_sa_upgrade_txholds(tx, zp);
1602 zfs_sa_upgrade_txholds(tx, dzp);
1603 if (may_delete_now) {
1604 toobig = zp->z_size > zp->z_blksz * zfs_delete_blocks;
1605 /* if the file is too big, only hold_free a token amount */
1606 dmu_tx_hold_free(tx, zp->z_id, 0,
1607 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1610 /* are there any extended attributes? */
1611 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1612 &xattr_obj, sizeof (xattr_obj));
1613 if (error == 0 && xattr_obj) {
1614 error = zfs_zget(zsb, xattr_obj, &xzp);
1616 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1617 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1620 mutex_enter(&zp->z_lock);
1621 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1622 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1623 mutex_exit(&zp->z_lock);
1625 /* charge as an update -- would be nice not to charge at all */
1626 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
1629 * Mark this transaction as typically resulting in a net free of space
1631 dmu_tx_mark_netfree(tx);
1633 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1635 zfs_dirent_unlock(dl);
1639 if (error == ERESTART) {
1645 #ifdef HAVE_PN_UTILS
1648 #endif /* HAVE_PN_UTILS */
1655 * Remove the directory entry.
1657 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1666 * Hold z_lock so that we can make sure that the ACL obj
1667 * hasn't changed. Could have been deleted due to
1670 mutex_enter(&zp->z_lock);
1671 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1672 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1673 delete_now = may_delete_now && !toobig &&
1674 atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped) &&
1675 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1680 if (xattr_obj_unlinked) {
1681 ASSERT3U(xzp->z_links, ==, 2);
1682 mutex_enter(&xzp->z_lock);
1683 xzp->z_unlinked = 1;
1685 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zsb),
1686 &xzp->z_links, sizeof (xzp->z_links), tx);
1687 ASSERT3U(error, ==, 0);
1688 mutex_exit(&xzp->z_lock);
1689 zfs_unlinked_add(xzp, tx);
1692 error = sa_remove(zp->z_sa_hdl,
1693 SA_ZPL_XATTR(zsb), tx);
1695 error = sa_update(zp->z_sa_hdl,
1696 SA_ZPL_XATTR(zsb), &null_xattr,
1697 sizeof (uint64_t), tx);
1701 * Add to the unlinked set because a new reference could be
1702 * taken concurrently resulting in a deferred destruction.
1704 zfs_unlinked_add(zp, tx);
1705 mutex_exit(&zp->z_lock);
1706 zfs_inode_update(zp);
1708 } else if (unlinked) {
1709 mutex_exit(&zp->z_lock);
1710 zfs_unlinked_add(zp, tx);
1714 #ifdef HAVE_PN_UTILS
1715 if (flags & FIGNORECASE)
1717 #endif /* HAVE_PN_UTILS */
1718 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1722 #ifdef HAVE_PN_UTILS
1725 #endif /* HAVE_PN_UTILS */
1727 zfs_dirent_unlock(dl);
1728 zfs_inode_update(dzp);
1731 zfs_inode_update(zp);
1736 zfs_inode_update(xzp);
1737 zfs_iput_async(ZTOI(xzp));
1740 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1741 zil_commit(zilog, 0);
1746 EXPORT_SYMBOL(zfs_remove);
1749 * Create a new directory and insert it into dip using the name
1750 * provided. Return a pointer to the inserted directory.
1752 * IN: dip - inode of directory to add subdir to.
1753 * dirname - name of new directory.
1754 * vap - attributes of new directory.
1755 * cr - credentials of caller.
1756 * vsecp - ACL to be set
1758 * OUT: ipp - inode of created directory.
1760 * RETURN: 0 if success
1761 * error code if failure
1764 * dip - ctime|mtime updated
1765 * ipp - ctime|mtime|atime updated
1769 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1770 cred_t *cr, int flags, vsecattr_t *vsecp)
1772 znode_t *zp, *dzp = ITOZ(dip);
1773 zfs_sb_t *zsb = ITOZSB(dip);
1781 gid_t gid = crgetgid(cr);
1782 zfs_acl_ids_t acl_ids;
1783 boolean_t fuid_dirtied;
1784 boolean_t waited = B_FALSE;
1786 ASSERT(S_ISDIR(vap->va_mode));
1789 * If we have an ephemeral id, ACL, or XVATTR then
1790 * make sure file system is at proper version
1794 if (zsb->z_use_fuids == B_FALSE &&
1795 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1796 return (SET_ERROR(EINVAL));
1802 if (dzp->z_pflags & ZFS_XATTR) {
1804 return (SET_ERROR(EINVAL));
1807 if (zsb->z_utf8 && u8_validate(dirname,
1808 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1810 return (SET_ERROR(EILSEQ));
1812 if (flags & FIGNORECASE)
1815 if (vap->va_mask & ATTR_XVATTR) {
1816 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1817 crgetuid(cr), cr, vap->va_mode)) != 0) {
1823 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1824 vsecp, &acl_ids)) != 0) {
1829 * First make sure the new directory doesn't exist.
1831 * Existence is checked first to make sure we don't return
1832 * EACCES instead of EEXIST which can cause some applications
1838 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1840 zfs_acl_ids_free(&acl_ids);
1845 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
1846 zfs_acl_ids_free(&acl_ids);
1847 zfs_dirent_unlock(dl);
1852 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1853 zfs_acl_ids_free(&acl_ids);
1854 zfs_dirent_unlock(dl);
1856 return (SET_ERROR(EDQUOT));
1860 * Add a new entry to the directory.
1862 tx = dmu_tx_create(zsb->z_os);
1863 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1864 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1865 fuid_dirtied = zsb->z_fuid_dirty;
1867 zfs_fuid_txhold(zsb, tx);
1868 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1869 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1870 acl_ids.z_aclp->z_acl_bytes);
1873 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1874 ZFS_SA_BASE_ATTR_SIZE);
1876 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1878 zfs_dirent_unlock(dl);
1879 if (error == ERESTART) {
1885 zfs_acl_ids_free(&acl_ids);
1894 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1897 zfs_fuid_sync(zsb, tx);
1900 * Now put new name in parent dir.
1902 (void) zfs_link_create(dl, zp, tx, ZNEW);
1906 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1907 if (flags & FIGNORECASE)
1909 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1910 acl_ids.z_fuidp, vap);
1912 zfs_acl_ids_free(&acl_ids);
1916 zfs_dirent_unlock(dl);
1918 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1919 zil_commit(zilog, 0);
1921 zfs_inode_update(dzp);
1922 zfs_inode_update(zp);
1926 EXPORT_SYMBOL(zfs_mkdir);
1929 * Remove a directory subdir entry. If the current working
1930 * directory is the same as the subdir to be removed, the
1933 * IN: dip - inode of directory to remove from.
1934 * name - name of directory to be removed.
1935 * cwd - inode of current working directory.
1936 * cr - credentials of caller.
1937 * flags - case flags
1939 * RETURN: 0 on success, error code on failure.
1942 * dip - ctime|mtime updated
1946 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
1949 znode_t *dzp = ITOZ(dip);
1952 zfs_sb_t *zsb = ITOZSB(dip);
1958 boolean_t waited = B_FALSE;
1964 if (flags & FIGNORECASE)
1970 * Attempt to lock directory; fail if entry doesn't exist.
1972 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1980 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1984 if (!S_ISDIR(ip->i_mode)) {
1985 error = SET_ERROR(ENOTDIR);
1990 error = SET_ERROR(EINVAL);
1995 * Grab a lock on the directory to make sure that noone is
1996 * trying to add (or lookup) entries while we are removing it.
1998 rw_enter(&zp->z_name_lock, RW_WRITER);
2001 * Grab a lock on the parent pointer to make sure we play well
2002 * with the treewalk and directory rename code.
2004 rw_enter(&zp->z_parent_lock, RW_WRITER);
2006 tx = dmu_tx_create(zsb->z_os);
2007 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2008 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2009 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
2010 zfs_sa_upgrade_txholds(tx, zp);
2011 zfs_sa_upgrade_txholds(tx, dzp);
2012 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2014 rw_exit(&zp->z_parent_lock);
2015 rw_exit(&zp->z_name_lock);
2016 zfs_dirent_unlock(dl);
2018 if (error == ERESTART) {
2029 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2032 uint64_t txtype = TX_RMDIR;
2033 if (flags & FIGNORECASE)
2035 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2040 rw_exit(&zp->z_parent_lock);
2041 rw_exit(&zp->z_name_lock);
2043 zfs_dirent_unlock(dl);
2045 zfs_inode_update(dzp);
2046 zfs_inode_update(zp);
2049 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
2050 zil_commit(zilog, 0);
2055 EXPORT_SYMBOL(zfs_rmdir);
2058 * Read as many directory entries as will fit into the provided
2059 * dirent buffer from the given directory cursor position.
2061 * IN: ip - inode of directory to read.
2062 * dirent - buffer for directory entries.
2064 * OUT: dirent - filler buffer of directory entries.
2066 * RETURN: 0 if success
2067 * error code if failure
2070 * ip - atime updated
2072 * Note that the low 4 bits of the cookie returned by zap is always zero.
2073 * This allows us to use the low range for "special" directory entries:
2074 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2075 * we use the offset 2 for the '.zfs' directory.
2079 zfs_readdir(struct inode *ip, struct dir_context *ctx, cred_t *cr)
2081 znode_t *zp = ITOZ(ip);
2082 zfs_sb_t *zsb = ITOZSB(ip);
2085 zap_attribute_t zap;
2091 uint64_t offset; /* must be unsigned; checks for < 1 */
2096 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zsb),
2097 &parent, sizeof (parent))) != 0)
2101 * Quit if directory has been removed (posix)
2109 prefetch = zp->z_zn_prefetch;
2112 * Initialize the iterator cursor.
2116 * Start iteration from the beginning of the directory.
2118 zap_cursor_init(&zc, os, zp->z_id);
2121 * The offset is a serialized cursor.
2123 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2127 * Transform to file-system independent format
2132 * Special case `.', `..', and `.zfs'.
2135 (void) strcpy(zap.za_name, ".");
2136 zap.za_normalization_conflict = 0;
2139 } else if (offset == 1) {
2140 (void) strcpy(zap.za_name, "..");
2141 zap.za_normalization_conflict = 0;
2144 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2145 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2146 zap.za_normalization_conflict = 0;
2147 objnum = ZFSCTL_INO_ROOT;
2153 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2154 if (error == ENOENT)
2161 * Allow multiple entries provided the first entry is
2162 * the object id. Non-zpl consumers may safely make
2163 * use of the additional space.
2165 * XXX: This should be a feature flag for compatibility
2167 if (zap.za_integer_length != 8 ||
2168 zap.za_num_integers == 0) {
2169 cmn_err(CE_WARN, "zap_readdir: bad directory "
2170 "entry, obj = %lld, offset = %lld, "
2171 "length = %d, num = %lld\n",
2172 (u_longlong_t)zp->z_id,
2173 (u_longlong_t)offset,
2174 zap.za_integer_length,
2175 (u_longlong_t)zap.za_num_integers);
2176 error = SET_ERROR(ENXIO);
2180 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2181 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2184 done = !dir_emit(ctx, zap.za_name, strlen(zap.za_name),
2189 /* Prefetch znode */
2191 dmu_prefetch(os, objnum, 0, 0, 0,
2192 ZIO_PRIORITY_SYNC_READ);
2196 * Move to the next entry, fill in the previous offset.
2198 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2199 zap_cursor_advance(&zc);
2200 offset = zap_cursor_serialize(&zc);
2206 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2209 zap_cursor_fini(&zc);
2210 if (error == ENOENT)
2217 EXPORT_SYMBOL(zfs_readdir);
2219 ulong_t zfs_fsync_sync_cnt = 4;
2222 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2224 znode_t *zp = ITOZ(ip);
2225 zfs_sb_t *zsb = ITOZSB(ip);
2227 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2229 if (zsb->z_os->os_sync != ZFS_SYNC_DISABLED) {
2232 zil_commit(zsb->z_log, zp->z_id);
2235 tsd_set(zfs_fsyncer_key, NULL);
2239 EXPORT_SYMBOL(zfs_fsync);
2243 * Get the requested file attributes and place them in the provided
2246 * IN: ip - inode of file.
2247 * vap - va_mask identifies requested attributes.
2248 * If ATTR_XVATTR set, then optional attrs are requested
2249 * flags - ATTR_NOACLCHECK (CIFS server context)
2250 * cr - credentials of caller.
2252 * OUT: vap - attribute values.
2254 * RETURN: 0 (always succeeds)
2258 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2260 znode_t *zp = ITOZ(ip);
2261 zfs_sb_t *zsb = ITOZSB(ip);
2264 uint64_t atime[2], mtime[2], ctime[2];
2265 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2266 xoptattr_t *xoap = NULL;
2267 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2268 sa_bulk_attr_t bulk[3];
2274 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2276 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zsb), NULL, &atime, 16);
2277 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
2278 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
2280 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2286 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2287 * Also, if we are the owner don't bother, since owner should
2288 * always be allowed to read basic attributes of file.
2290 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2291 (vap->va_uid != crgetuid(cr))) {
2292 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2300 * Return all attributes. It's cheaper to provide the answer
2301 * than to determine whether we were asked the question.
2304 mutex_enter(&zp->z_lock);
2305 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2306 vap->va_mode = zp->z_mode;
2307 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2308 vap->va_nodeid = zp->z_id;
2309 if ((zp->z_id == zsb->z_root) && zfs_show_ctldir(zp))
2310 links = zp->z_links + 1;
2312 links = zp->z_links;
2313 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2314 vap->va_size = i_size_read(ip);
2315 vap->va_rdev = ip->i_rdev;
2316 vap->va_seq = ip->i_generation;
2319 * Add in any requested optional attributes and the create time.
2320 * Also set the corresponding bits in the returned attribute bitmap.
2322 if ((xoap = xva_getxoptattr(xvap)) != NULL && zsb->z_use_fuids) {
2323 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2325 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2326 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2329 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2330 xoap->xoa_readonly =
2331 ((zp->z_pflags & ZFS_READONLY) != 0);
2332 XVA_SET_RTN(xvap, XAT_READONLY);
2335 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2337 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2338 XVA_SET_RTN(xvap, XAT_SYSTEM);
2341 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2343 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2344 XVA_SET_RTN(xvap, XAT_HIDDEN);
2347 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2348 xoap->xoa_nounlink =
2349 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2350 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2353 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2354 xoap->xoa_immutable =
2355 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2356 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2359 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2360 xoap->xoa_appendonly =
2361 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2362 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2365 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2367 ((zp->z_pflags & ZFS_NODUMP) != 0);
2368 XVA_SET_RTN(xvap, XAT_NODUMP);
2371 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2373 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2374 XVA_SET_RTN(xvap, XAT_OPAQUE);
2377 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2378 xoap->xoa_av_quarantined =
2379 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2380 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2383 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2384 xoap->xoa_av_modified =
2385 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2386 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2389 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2390 S_ISREG(ip->i_mode)) {
2391 zfs_sa_get_scanstamp(zp, xvap);
2394 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2397 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zsb),
2398 times, sizeof (times));
2399 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2400 XVA_SET_RTN(xvap, XAT_CREATETIME);
2403 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2404 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2405 XVA_SET_RTN(xvap, XAT_REPARSE);
2407 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2408 xoap->xoa_generation = zp->z_gen;
2409 XVA_SET_RTN(xvap, XAT_GEN);
2412 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2414 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2415 XVA_SET_RTN(xvap, XAT_OFFLINE);
2418 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2420 ((zp->z_pflags & ZFS_SPARSE) != 0);
2421 XVA_SET_RTN(xvap, XAT_SPARSE);
2425 ZFS_TIME_DECODE(&vap->va_atime, atime);
2426 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2427 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2429 mutex_exit(&zp->z_lock);
2431 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2433 if (zp->z_blksz == 0) {
2435 * Block size hasn't been set; suggest maximal I/O transfers.
2437 vap->va_blksize = zsb->z_max_blksz;
2443 EXPORT_SYMBOL(zfs_getattr);
2446 * Get the basic file attributes and place them in the provided kstat
2447 * structure. The inode is assumed to be the authoritative source
2448 * for most of the attributes. However, the znode currently has the
2449 * authoritative atime, blksize, and block count.
2451 * IN: ip - inode of file.
2453 * OUT: sp - kstat values.
2455 * RETURN: 0 (always succeeds)
2459 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2461 znode_t *zp = ITOZ(ip);
2462 zfs_sb_t *zsb = ITOZSB(ip);
2464 u_longlong_t nblocks;
2469 mutex_enter(&zp->z_lock);
2471 generic_fillattr(ip, sp);
2473 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2474 sp->blksize = blksize;
2475 sp->blocks = nblocks;
2477 if (unlikely(zp->z_blksz == 0)) {
2479 * Block size hasn't been set; suggest maximal I/O transfers.
2481 sp->blksize = zsb->z_max_blksz;
2484 mutex_exit(&zp->z_lock);
2487 * Required to prevent NFS client from detecting different inode
2488 * numbers of snapshot root dentry before and after snapshot mount.
2490 if (zsb->z_issnap) {
2491 if (ip->i_sb->s_root->d_inode == ip)
2492 sp->ino = ZFSCTL_INO_SNAPDIRS -
2493 dmu_objset_id(zsb->z_os);
2500 EXPORT_SYMBOL(zfs_getattr_fast);
2503 * Set the file attributes to the values contained in the
2506 * IN: ip - inode of file to be modified.
2507 * vap - new attribute values.
2508 * If ATTR_XVATTR set, then optional attrs are being set
2509 * flags - ATTR_UTIME set if non-default time values provided.
2510 * - ATTR_NOACLCHECK (CIFS context only).
2511 * cr - credentials of caller.
2513 * RETURN: 0 if success
2514 * error code if failure
2517 * ip - ctime updated, mtime updated if size changed.
2521 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2523 znode_t *zp = ITOZ(ip);
2524 zfs_sb_t *zsb = ITOZSB(ip);
2528 xvattr_t *tmpxvattr;
2529 uint_t mask = vap->va_mask;
2530 uint_t saved_mask = 0;
2533 uint64_t new_uid, new_gid;
2535 uint64_t mtime[2], ctime[2], atime[2];
2537 int need_policy = FALSE;
2539 zfs_fuid_info_t *fuidp = NULL;
2540 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2543 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2544 boolean_t fuid_dirtied = B_FALSE;
2545 sa_bulk_attr_t *bulk, *xattr_bulk;
2546 int count = 0, xattr_count = 0;
2557 * Make sure that if we have ephemeral uid/gid or xvattr specified
2558 * that file system is at proper version level
2561 if (zsb->z_use_fuids == B_FALSE &&
2562 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2563 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2564 (mask & ATTR_XVATTR))) {
2566 return (SET_ERROR(EINVAL));
2569 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2571 return (SET_ERROR(EISDIR));
2574 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2576 return (SET_ERROR(EINVAL));
2580 * If this is an xvattr_t, then get a pointer to the structure of
2581 * optional attributes. If this is NULL, then we have a vattr_t.
2583 xoap = xva_getxoptattr(xvap);
2585 tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP);
2586 xva_init(tmpxvattr);
2588 bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2589 xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2592 * Immutable files can only alter immutable bit and atime
2594 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2595 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2596 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2601 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2607 * Verify timestamps doesn't overflow 32 bits.
2608 * ZFS can handle large timestamps, but 32bit syscalls can't
2609 * handle times greater than 2039. This check should be removed
2610 * once large timestamps are fully supported.
2612 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2613 if (((mask & ATTR_ATIME) &&
2614 TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2615 ((mask & ATTR_MTIME) &&
2616 TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2626 /* Can this be moved to before the top label? */
2627 if (zfs_is_readonly(zsb)) {
2633 * First validate permissions
2636 if (mask & ATTR_SIZE) {
2637 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2642 * XXX - Note, we are not providing any open
2643 * mode flags here (like FNDELAY), so we may
2644 * block if there are locks present... this
2645 * should be addressed in openat().
2647 /* XXX - would it be OK to generate a log record here? */
2648 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2653 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2654 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2655 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2656 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2657 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2658 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2659 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2660 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2661 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2665 if (mask & (ATTR_UID|ATTR_GID)) {
2666 int idmask = (mask & (ATTR_UID|ATTR_GID));
2671 * NOTE: even if a new mode is being set,
2672 * we may clear S_ISUID/S_ISGID bits.
2675 if (!(mask & ATTR_MODE))
2676 vap->va_mode = zp->z_mode;
2679 * Take ownership or chgrp to group we are a member of
2682 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
2683 take_group = (mask & ATTR_GID) &&
2684 zfs_groupmember(zsb, vap->va_gid, cr);
2687 * If both ATTR_UID and ATTR_GID are set then take_owner and
2688 * take_group must both be set in order to allow taking
2691 * Otherwise, send the check through secpolicy_vnode_setattr()
2695 if (((idmask == (ATTR_UID|ATTR_GID)) &&
2696 take_owner && take_group) ||
2697 ((idmask == ATTR_UID) && take_owner) ||
2698 ((idmask == ATTR_GID) && take_group)) {
2699 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2700 skipaclchk, cr) == 0) {
2702 * Remove setuid/setgid for non-privileged users
2704 (void) secpolicy_setid_clear(vap, cr);
2705 trim_mask = (mask & (ATTR_UID|ATTR_GID));
2714 mutex_enter(&zp->z_lock);
2715 oldva.va_mode = zp->z_mode;
2716 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2717 if (mask & ATTR_XVATTR) {
2719 * Update xvattr mask to include only those attributes
2720 * that are actually changing.
2722 * the bits will be restored prior to actually setting
2723 * the attributes so the caller thinks they were set.
2725 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2726 if (xoap->xoa_appendonly !=
2727 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2730 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2731 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
2735 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2736 if (xoap->xoa_nounlink !=
2737 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2740 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2741 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
2745 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2746 if (xoap->xoa_immutable !=
2747 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2750 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2751 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
2755 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2756 if (xoap->xoa_nodump !=
2757 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2760 XVA_CLR_REQ(xvap, XAT_NODUMP);
2761 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
2765 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2766 if (xoap->xoa_av_modified !=
2767 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2770 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2771 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
2775 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2776 if ((!S_ISREG(ip->i_mode) &&
2777 xoap->xoa_av_quarantined) ||
2778 xoap->xoa_av_quarantined !=
2779 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2782 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2783 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
2787 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2788 mutex_exit(&zp->z_lock);
2793 if (need_policy == FALSE &&
2794 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2795 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2800 mutex_exit(&zp->z_lock);
2802 if (mask & ATTR_MODE) {
2803 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2804 err = secpolicy_setid_setsticky_clear(ip, vap,
2809 trim_mask |= ATTR_MODE;
2817 * If trim_mask is set then take ownership
2818 * has been granted or write_acl is present and user
2819 * has the ability to modify mode. In that case remove
2820 * UID|GID and or MODE from mask so that
2821 * secpolicy_vnode_setattr() doesn't revoke it.
2825 saved_mask = vap->va_mask;
2826 vap->va_mask &= ~trim_mask;
2828 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
2829 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2834 vap->va_mask |= saved_mask;
2838 * secpolicy_vnode_setattr, or take ownership may have
2841 mask = vap->va_mask;
2843 if ((mask & (ATTR_UID | ATTR_GID))) {
2844 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
2845 &xattr_obj, sizeof (xattr_obj));
2847 if (err == 0 && xattr_obj) {
2848 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
2852 if (mask & ATTR_UID) {
2853 new_uid = zfs_fuid_create(zsb,
2854 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2855 if (new_uid != zp->z_uid &&
2856 zfs_fuid_overquota(zsb, B_FALSE, new_uid)) {
2864 if (mask & ATTR_GID) {
2865 new_gid = zfs_fuid_create(zsb, (uint64_t)vap->va_gid,
2866 cr, ZFS_GROUP, &fuidp);
2867 if (new_gid != zp->z_gid &&
2868 zfs_fuid_overquota(zsb, B_TRUE, new_gid)) {
2876 tx = dmu_tx_create(zsb->z_os);
2878 if (mask & ATTR_MODE) {
2879 uint64_t pmode = zp->z_mode;
2881 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2883 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
2885 mutex_enter(&zp->z_lock);
2886 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2888 * Are we upgrading ACL from old V0 format
2891 if (zsb->z_version >= ZPL_VERSION_FUID &&
2892 zfs_znode_acl_version(zp) ==
2893 ZFS_ACL_VERSION_INITIAL) {
2894 dmu_tx_hold_free(tx, acl_obj, 0,
2896 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2897 0, aclp->z_acl_bytes);
2899 dmu_tx_hold_write(tx, acl_obj, 0,
2902 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2903 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2904 0, aclp->z_acl_bytes);
2906 mutex_exit(&zp->z_lock);
2907 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2909 if ((mask & ATTR_XVATTR) &&
2910 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2911 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2913 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2917 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
2920 fuid_dirtied = zsb->z_fuid_dirty;
2922 zfs_fuid_txhold(zsb, tx);
2924 zfs_sa_upgrade_txholds(tx, zp);
2926 err = dmu_tx_assign(tx, TXG_WAIT);
2932 * Set each attribute requested.
2933 * We group settings according to the locks they need to acquire.
2935 * Note: you cannot set ctime directly, although it will be
2936 * updated as a side-effect of calling this function.
2940 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2941 mutex_enter(&zp->z_acl_lock);
2942 mutex_enter(&zp->z_lock);
2944 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
2945 &zp->z_pflags, sizeof (zp->z_pflags));
2948 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2949 mutex_enter(&attrzp->z_acl_lock);
2950 mutex_enter(&attrzp->z_lock);
2951 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2952 SA_ZPL_FLAGS(zsb), NULL, &attrzp->z_pflags,
2953 sizeof (attrzp->z_pflags));
2956 if (mask & (ATTR_UID|ATTR_GID)) {
2958 if (mask & ATTR_UID) {
2959 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zsb), NULL,
2960 &new_uid, sizeof (new_uid));
2961 zp->z_uid = new_uid;
2963 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2964 SA_ZPL_UID(zsb), NULL, &new_uid,
2966 attrzp->z_uid = new_uid;
2970 if (mask & ATTR_GID) {
2971 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zsb),
2972 NULL, &new_gid, sizeof (new_gid));
2973 zp->z_gid = new_gid;
2975 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2976 SA_ZPL_GID(zsb), NULL, &new_gid,
2978 attrzp->z_gid = new_gid;
2981 if (!(mask & ATTR_MODE)) {
2982 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb),
2983 NULL, &new_mode, sizeof (new_mode));
2984 new_mode = zp->z_mode;
2986 err = zfs_acl_chown_setattr(zp);
2989 err = zfs_acl_chown_setattr(attrzp);
2994 if (mask & ATTR_MODE) {
2995 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb), NULL,
2996 &new_mode, sizeof (new_mode));
2997 zp->z_mode = new_mode;
2998 ASSERT3P(aclp, !=, NULL);
2999 err = zfs_aclset_common(zp, aclp, cr, tx);
3001 if (zp->z_acl_cached)
3002 zfs_acl_free(zp->z_acl_cached);
3003 zp->z_acl_cached = aclp;
3008 if ((mask & ATTR_ATIME) || zp->z_atime_dirty) {
3009 zp->z_atime_dirty = 0;
3010 ZFS_TIME_ENCODE(&ip->i_atime, atime);
3011 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zsb), NULL,
3012 &atime, sizeof (atime));
3015 if (mask & ATTR_MTIME) {
3016 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3017 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL,
3018 mtime, sizeof (mtime));
3021 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3022 if (mask & ATTR_SIZE && !(mask & ATTR_MTIME)) {
3023 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb),
3024 NULL, mtime, sizeof (mtime));
3025 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
3026 &ctime, sizeof (ctime));
3027 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
3028 } else if (mask != 0) {
3029 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
3030 &ctime, sizeof (ctime));
3031 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime);
3033 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3034 SA_ZPL_CTIME(zsb), NULL,
3035 &ctime, sizeof (ctime));
3036 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3041 * Do this after setting timestamps to prevent timestamp
3042 * update from toggling bit
3045 if (xoap && (mask & ATTR_XVATTR)) {
3048 * restore trimmed off masks
3049 * so that return masks can be set for caller.
3052 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
3053 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3055 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
3056 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3058 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
3059 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3061 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
3062 XVA_SET_REQ(xvap, XAT_NODUMP);
3064 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
3065 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3067 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
3068 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3071 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3072 ASSERT(S_ISREG(ip->i_mode));
3074 zfs_xvattr_set(zp, xvap, tx);
3078 zfs_fuid_sync(zsb, tx);
3081 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3083 mutex_exit(&zp->z_lock);
3084 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3085 mutex_exit(&zp->z_acl_lock);
3088 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3089 mutex_exit(&attrzp->z_acl_lock);
3090 mutex_exit(&attrzp->z_lock);
3093 if (err == 0 && attrzp) {
3094 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3105 zfs_fuid_info_free(fuidp);
3111 if (err == ERESTART)
3114 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3116 zfs_inode_update(zp);
3120 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3121 zil_commit(zilog, 0);
3124 kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * 7);
3125 kmem_free(bulk, sizeof (sa_bulk_attr_t) * 7);
3126 kmem_free(tmpxvattr, sizeof (xvattr_t));
3130 EXPORT_SYMBOL(zfs_setattr);
3132 typedef struct zfs_zlock {
3133 krwlock_t *zl_rwlock; /* lock we acquired */
3134 znode_t *zl_znode; /* znode we held */
3135 struct zfs_zlock *zl_next; /* next in list */
3139 * Drop locks and release vnodes that were held by zfs_rename_lock().
3142 zfs_rename_unlock(zfs_zlock_t **zlpp)
3146 while ((zl = *zlpp) != NULL) {
3147 if (zl->zl_znode != NULL)
3148 iput(ZTOI(zl->zl_znode));
3149 rw_exit(zl->zl_rwlock);
3150 *zlpp = zl->zl_next;
3151 kmem_free(zl, sizeof (*zl));
3156 * Search back through the directory tree, using the ".." entries.
3157 * Lock each directory in the chain to prevent concurrent renames.
3158 * Fail any attempt to move a directory into one of its own descendants.
3159 * XXX - z_parent_lock can overlap with map or grow locks
3162 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3166 uint64_t rootid = ZTOZSB(zp)->z_root;
3167 uint64_t oidp = zp->z_id;
3168 krwlock_t *rwlp = &szp->z_parent_lock;
3169 krw_t rw = RW_WRITER;
3172 * First pass write-locks szp and compares to zp->z_id.
3173 * Later passes read-lock zp and compare to zp->z_parent.
3176 if (!rw_tryenter(rwlp, rw)) {
3178 * Another thread is renaming in this path.
3179 * Note that if we are a WRITER, we don't have any
3180 * parent_locks held yet.
3182 if (rw == RW_READER && zp->z_id > szp->z_id) {
3184 * Drop our locks and restart
3186 zfs_rename_unlock(&zl);
3190 rwlp = &szp->z_parent_lock;
3195 * Wait for other thread to drop its locks
3201 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3202 zl->zl_rwlock = rwlp;
3203 zl->zl_znode = NULL;
3204 zl->zl_next = *zlpp;
3207 if (oidp == szp->z_id) /* We're a descendant of szp */
3208 return (SET_ERROR(EINVAL));
3210 if (oidp == rootid) /* We've hit the top */
3213 if (rw == RW_READER) { /* i.e. not the first pass */
3214 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3219 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3220 &oidp, sizeof (oidp));
3221 rwlp = &zp->z_parent_lock;
3224 } while (zp->z_id != sdzp->z_id);
3230 * Move an entry from the provided source directory to the target
3231 * directory. Change the entry name as indicated.
3233 * IN: sdip - Source directory containing the "old entry".
3234 * snm - Old entry name.
3235 * tdip - Target directory to contain the "new entry".
3236 * tnm - New entry name.
3237 * cr - credentials of caller.
3238 * flags - case flags
3240 * RETURN: 0 on success, error code on failure.
3243 * sdip,tdip - ctime|mtime updated
3247 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3248 cred_t *cr, int flags)
3250 znode_t *tdzp, *szp, *tzp;
3251 znode_t *sdzp = ITOZ(sdip);
3252 zfs_sb_t *zsb = ITOZSB(sdip);
3254 zfs_dirlock_t *sdl, *tdl;
3257 int cmp, serr, terr;
3260 boolean_t waited = B_FALSE;
3263 ZFS_VERIFY_ZP(sdzp);
3267 ZFS_VERIFY_ZP(tdzp);
3270 * We check i_sb because snapshots and the ctldir must have different
3273 if (tdip->i_sb != sdip->i_sb || zfsctl_is_node(tdip)) {
3275 return (SET_ERROR(EXDEV));
3278 if (zsb->z_utf8 && u8_validate(tnm,
3279 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3281 return (SET_ERROR(EILSEQ));
3284 if (flags & FIGNORECASE)
3293 * This is to prevent the creation of links into attribute space
3294 * by renaming a linked file into/outof an attribute directory.
3295 * See the comment in zfs_link() for why this is considered bad.
3297 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3299 return (SET_ERROR(EINVAL));
3303 * Lock source and target directory entries. To prevent deadlock,
3304 * a lock ordering must be defined. We lock the directory with
3305 * the smallest object id first, or if it's a tie, the one with
3306 * the lexically first name.
3308 if (sdzp->z_id < tdzp->z_id) {
3310 } else if (sdzp->z_id > tdzp->z_id) {
3314 * First compare the two name arguments without
3315 * considering any case folding.
3317 int nofold = (zsb->z_norm & ~U8_TEXTPREP_TOUPPER);
3319 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3320 ASSERT(error == 0 || !zsb->z_utf8);
3323 * POSIX: "If the old argument and the new argument
3324 * both refer to links to the same existing file,
3325 * the rename() function shall return successfully
3326 * and perform no other action."
3332 * If the file system is case-folding, then we may
3333 * have some more checking to do. A case-folding file
3334 * system is either supporting mixed case sensitivity
3335 * access or is completely case-insensitive. Note
3336 * that the file system is always case preserving.
3338 * In mixed sensitivity mode case sensitive behavior
3339 * is the default. FIGNORECASE must be used to
3340 * explicitly request case insensitive behavior.
3342 * If the source and target names provided differ only
3343 * by case (e.g., a request to rename 'tim' to 'Tim'),
3344 * we will treat this as a special case in the
3345 * case-insensitive mode: as long as the source name
3346 * is an exact match, we will allow this to proceed as
3347 * a name-change request.
3349 if ((zsb->z_case == ZFS_CASE_INSENSITIVE ||
3350 (zsb->z_case == ZFS_CASE_MIXED &&
3351 flags & FIGNORECASE)) &&
3352 u8_strcmp(snm, tnm, 0, zsb->z_norm, U8_UNICODE_LATEST,
3355 * case preserving rename request, require exact
3364 * If the source and destination directories are the same, we should
3365 * grab the z_name_lock of that directory only once.
3369 rw_enter(&sdzp->z_name_lock, RW_READER);
3373 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3374 ZEXISTS | zflg, NULL, NULL);
3375 terr = zfs_dirent_lock(&tdl,
3376 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3378 terr = zfs_dirent_lock(&tdl,
3379 tdzp, tnm, &tzp, zflg, NULL, NULL);
3380 serr = zfs_dirent_lock(&sdl,
3381 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3387 * Source entry invalid or not there.
3390 zfs_dirent_unlock(tdl);
3396 rw_exit(&sdzp->z_name_lock);
3398 if (strcmp(snm, "..") == 0)
3404 zfs_dirent_unlock(sdl);
3408 rw_exit(&sdzp->z_name_lock);
3410 if (strcmp(tnm, "..") == 0)
3417 * Must have write access at the source to remove the old entry
3418 * and write access at the target to create the new entry.
3419 * Note that if target and source are the same, this can be
3420 * done in a single check.
3423 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3426 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3428 * Check to make sure rename is valid.
3429 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3431 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3436 * Does target exist?
3440 * Source and target must be the same type.
3442 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3443 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3444 error = SET_ERROR(ENOTDIR);
3448 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3449 error = SET_ERROR(EISDIR);
3454 * POSIX dictates that when the source and target
3455 * entries refer to the same file object, rename
3456 * must do nothing and exit without error.
3458 if (szp->z_id == tzp->z_id) {
3464 tx = dmu_tx_create(zsb->z_os);
3465 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3466 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3467 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3468 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3470 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3471 zfs_sa_upgrade_txholds(tx, tdzp);
3474 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3475 zfs_sa_upgrade_txholds(tx, tzp);
3478 zfs_sa_upgrade_txholds(tx, szp);
3479 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
3480 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3483 zfs_rename_unlock(&zl);
3484 zfs_dirent_unlock(sdl);
3485 zfs_dirent_unlock(tdl);
3488 rw_exit(&sdzp->z_name_lock);
3493 if (error == ERESTART) {
3504 if (tzp) /* Attempt to remove the existing target */
3505 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3508 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3510 szp->z_pflags |= ZFS_AV_MODIFIED;
3512 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zsb),
3513 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3516 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3518 zfs_log_rename(zilog, tx, TX_RENAME |
3519 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3520 sdl->dl_name, tdzp, tdl->dl_name, szp);
3523 * At this point, we have successfully created
3524 * the target name, but have failed to remove
3525 * the source name. Since the create was done
3526 * with the ZRENAMING flag, there are
3527 * complications; for one, the link count is
3528 * wrong. The easiest way to deal with this
3529 * is to remove the newly created target, and
3530 * return the original error. This must
3531 * succeed; fortunately, it is very unlikely to
3532 * fail, since we just created it.
3534 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3535 ZRENAMING, NULL), ==, 0);
3543 zfs_rename_unlock(&zl);
3545 zfs_dirent_unlock(sdl);
3546 zfs_dirent_unlock(tdl);
3548 zfs_inode_update(sdzp);
3550 rw_exit(&sdzp->z_name_lock);
3553 zfs_inode_update(tdzp);
3555 zfs_inode_update(szp);
3558 zfs_inode_update(tzp);
3562 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3563 zil_commit(zilog, 0);
3568 EXPORT_SYMBOL(zfs_rename);
3571 * Insert the indicated symbolic reference entry into the directory.
3573 * IN: dip - Directory to contain new symbolic link.
3574 * link - Name for new symlink entry.
3575 * vap - Attributes of new entry.
3576 * target - Target path of new symlink.
3578 * cr - credentials of caller.
3579 * flags - case flags
3581 * RETURN: 0 on success, error code on failure.
3584 * dip - ctime|mtime updated
3588 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
3589 struct inode **ipp, cred_t *cr, int flags)
3591 znode_t *zp, *dzp = ITOZ(dip);
3594 zfs_sb_t *zsb = ITOZSB(dip);
3596 uint64_t len = strlen(link);
3599 zfs_acl_ids_t acl_ids;
3600 boolean_t fuid_dirtied;
3601 uint64_t txtype = TX_SYMLINK;
3602 boolean_t waited = B_FALSE;
3604 ASSERT(S_ISLNK(vap->va_mode));
3610 if (zsb->z_utf8 && u8_validate(name, strlen(name),
3611 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3613 return (SET_ERROR(EILSEQ));
3615 if (flags & FIGNORECASE)
3618 if (len > MAXPATHLEN) {
3620 return (SET_ERROR(ENAMETOOLONG));
3623 if ((error = zfs_acl_ids_create(dzp, 0,
3624 vap, cr, NULL, &acl_ids)) != 0) {
3632 * Attempt to lock directory; fail if entry already exists.
3634 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3636 zfs_acl_ids_free(&acl_ids);
3641 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3642 zfs_acl_ids_free(&acl_ids);
3643 zfs_dirent_unlock(dl);
3648 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
3649 zfs_acl_ids_free(&acl_ids);
3650 zfs_dirent_unlock(dl);
3652 return (SET_ERROR(EDQUOT));
3654 tx = dmu_tx_create(zsb->z_os);
3655 fuid_dirtied = zsb->z_fuid_dirty;
3656 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3657 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3658 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3659 ZFS_SA_BASE_ATTR_SIZE + len);
3660 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3661 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3662 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3663 acl_ids.z_aclp->z_acl_bytes);
3666 zfs_fuid_txhold(zsb, tx);
3667 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3669 zfs_dirent_unlock(dl);
3670 if (error == ERESTART) {
3676 zfs_acl_ids_free(&acl_ids);
3683 * Create a new object for the symlink.
3684 * for version 4 ZPL datsets the symlink will be an SA attribute
3686 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3689 zfs_fuid_sync(zsb, tx);
3691 mutex_enter(&zp->z_lock);
3693 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zsb),
3696 zfs_sa_symlink(zp, link, len, tx);
3697 mutex_exit(&zp->z_lock);
3700 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
3701 &zp->z_size, sizeof (zp->z_size), tx);
3703 * Insert the new object into the directory.
3705 (void) zfs_link_create(dl, zp, tx, ZNEW);
3707 if (flags & FIGNORECASE)
3709 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3711 zfs_inode_update(dzp);
3712 zfs_inode_update(zp);
3714 zfs_acl_ids_free(&acl_ids);
3718 zfs_dirent_unlock(dl);
3722 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3723 zil_commit(zilog, 0);
3728 EXPORT_SYMBOL(zfs_symlink);
3731 * Return, in the buffer contained in the provided uio structure,
3732 * the symbolic path referred to by ip.
3734 * IN: ip - inode of symbolic link
3735 * uio - structure to contain the link path.
3736 * cr - credentials of caller.
3738 * RETURN: 0 if success
3739 * error code if failure
3742 * ip - atime updated
3746 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
3748 znode_t *zp = ITOZ(ip);
3749 zfs_sb_t *zsb = ITOZSB(ip);
3755 mutex_enter(&zp->z_lock);
3757 error = sa_lookup_uio(zp->z_sa_hdl,
3758 SA_ZPL_SYMLINK(zsb), uio);
3760 error = zfs_sa_readlink(zp, uio);
3761 mutex_exit(&zp->z_lock);
3766 EXPORT_SYMBOL(zfs_readlink);
3769 * Insert a new entry into directory tdip referencing sip.
3771 * IN: tdip - Directory to contain new entry.
3772 * sip - inode of new entry.
3773 * name - name of new entry.
3774 * cr - credentials of caller.
3776 * RETURN: 0 if success
3777 * error code if failure
3780 * tdip - ctime|mtime updated
3781 * sip - ctime updated
3785 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr)
3787 znode_t *dzp = ITOZ(tdip);
3789 zfs_sb_t *zsb = ITOZSB(tdip);
3797 boolean_t waited = B_FALSE;
3799 ASSERT(S_ISDIR(tdip->i_mode));
3806 * POSIX dictates that we return EPERM here.
3807 * Better choices include ENOTSUP or EISDIR.
3809 if (S_ISDIR(sip->i_mode)) {
3811 return (SET_ERROR(EPERM));
3818 * We check i_sb because snapshots and the ctldir must have different
3821 if (sip->i_sb != tdip->i_sb || zfsctl_is_node(sip)) {
3823 return (SET_ERROR(EXDEV));
3826 /* Prevent links to .zfs/shares files */
3828 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zsb),
3829 &parent, sizeof (uint64_t))) != 0) {
3833 if (parent == zsb->z_shares_dir) {
3835 return (SET_ERROR(EPERM));
3838 if (zsb->z_utf8 && u8_validate(name,
3839 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3841 return (SET_ERROR(EILSEQ));
3843 #ifdef HAVE_PN_UTILS
3844 if (flags & FIGNORECASE)
3846 #endif /* HAVE_PN_UTILS */
3849 * We do not support links between attributes and non-attributes
3850 * because of the potential security risk of creating links
3851 * into "normal" file space in order to circumvent restrictions
3852 * imposed in attribute space.
3854 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3856 return (SET_ERROR(EINVAL));
3859 owner = zfs_fuid_map_id(zsb, szp->z_uid, cr, ZFS_OWNER);
3860 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3862 return (SET_ERROR(EPERM));
3865 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3872 * Attempt to lock directory; fail if entry already exists.
3874 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3880 tx = dmu_tx_create(zsb->z_os);
3881 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3882 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3883 zfs_sa_upgrade_txholds(tx, szp);
3884 zfs_sa_upgrade_txholds(tx, dzp);
3885 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3887 zfs_dirent_unlock(dl);
3888 if (error == ERESTART) {
3899 error = zfs_link_create(dl, szp, tx, 0);
3902 uint64_t txtype = TX_LINK;
3903 #ifdef HAVE_PN_UTILS
3904 if (flags & FIGNORECASE)
3906 #endif /* HAVE_PN_UTILS */
3907 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3912 zfs_dirent_unlock(dl);
3914 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3915 zil_commit(zilog, 0);
3917 zfs_inode_update(dzp);
3918 zfs_inode_update(szp);
3922 EXPORT_SYMBOL(zfs_link);
3925 zfs_putpage_commit_cb(void *arg)
3927 struct page *pp = arg;
3930 end_page_writeback(pp);
3934 * Push a page out to disk, once the page is on stable storage the
3935 * registered commit callback will be run as notification of completion.
3937 * IN: ip - page mapped for inode.
3938 * pp - page to push (page is locked)
3939 * wbc - writeback control data
3941 * RETURN: 0 if success
3942 * error code if failure
3945 * ip - ctime|mtime updated
3949 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
3951 znode_t *zp = ITOZ(ip);
3952 zfs_sb_t *zsb = ITOZSB(ip);
3960 uint64_t mtime[2], ctime[2];
3961 sa_bulk_attr_t bulk[3];
3963 struct address_space *mapping;
3968 ASSERT(PageLocked(pp));
3970 pgoff = page_offset(pp); /* Page byte-offset in file */
3971 offset = i_size_read(ip); /* File length in bytes */
3972 pglen = MIN(PAGE_SIZE, /* Page length in bytes */
3973 P2ROUNDUP(offset, PAGE_SIZE)-pgoff);
3975 /* Page is beyond end of file */
3976 if (pgoff >= offset) {
3982 /* Truncate page length to end of file */
3983 if (pgoff + pglen > offset)
3984 pglen = offset - pgoff;
3988 * FIXME: Allow mmap writes past its quota. The correct fix
3989 * is to register a page_mkwrite() handler to count the page
3990 * against its quota when it is about to be dirtied.
3992 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
3993 zfs_owner_overquota(zsb, zp, B_TRUE)) {
3999 * The ordering here is critical and must adhere to the following
4000 * rules in order to avoid deadlocking in either zfs_read() or
4001 * zfs_free_range() due to a lock inversion.
4003 * 1) The page must be unlocked prior to acquiring the range lock.
4004 * This is critical because zfs_read() calls find_lock_page()
4005 * which may block on the page lock while holding the range lock.
4007 * 2) Before setting or clearing write back on a page the range lock
4008 * must be held in order to prevent a lock inversion with the
4009 * zfs_free_range() function.
4011 * This presents a problem because upon entering this function the
4012 * page lock is already held. To safely acquire the range lock the
4013 * page lock must be dropped. This creates a window where another
4014 * process could truncate, invalidate, dirty, or write out the page.
4016 * Therefore, after successfully reacquiring the range and page locks
4017 * the current page state is checked. In the common case everything
4018 * will be as is expected and it can be written out. However, if
4019 * the page state has changed it must be handled accordingly.
4021 mapping = pp->mapping;
4022 redirty_page_for_writepage(wbc, pp);
4025 rl = zfs_range_lock(zp, pgoff, pglen, RL_WRITER);
4028 /* Page mapping changed or it was no longer dirty, we're done */
4029 if (unlikely((mapping != pp->mapping) || !PageDirty(pp))) {
4031 zfs_range_unlock(rl);
4036 /* Another process started write block if required */
4037 if (PageWriteback(pp)) {
4039 zfs_range_unlock(rl);
4041 if (wbc->sync_mode != WB_SYNC_NONE)
4042 wait_on_page_writeback(pp);
4048 /* Clear the dirty flag the required locks are held */
4049 if (!clear_page_dirty_for_io(pp)) {
4051 zfs_range_unlock(rl);
4057 * Counterpart for redirty_page_for_writepage() above. This page
4058 * was in fact not skipped and should not be counted as if it were.
4060 wbc->pages_skipped--;
4061 set_page_writeback(pp);
4064 tx = dmu_tx_create(zsb->z_os);
4065 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
4066 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4067 zfs_sa_upgrade_txholds(tx, zp);
4069 err = dmu_tx_assign(tx, TXG_NOWAIT);
4071 if (err == ERESTART)
4075 __set_page_dirty_nobuffers(pp);
4077 end_page_writeback(pp);
4078 zfs_range_unlock(rl);
4084 ASSERT3U(pglen, <=, PAGE_SIZE);
4085 dmu_write(zsb->z_os, zp->z_id, pgoff, pglen, va, tx);
4088 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
4089 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
4090 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zsb), NULL, &zp->z_pflags, 8);
4092 /* Preserve the mtime and ctime provided by the inode */
4093 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4094 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4095 zp->z_atime_dirty = 0;
4098 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4100 zfs_log_write(zsb->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0,
4101 zfs_putpage_commit_cb, pp);
4104 zfs_range_unlock(rl);
4106 if (wbc->sync_mode != WB_SYNC_NONE) {
4108 * Note that this is rarely called under writepages(), because
4109 * writepages() normally handles the entire commit for
4110 * performance reasons.
4112 if (zsb->z_log != NULL)
4113 zil_commit(zsb->z_log, zp->z_id);
4121 * Update the system attributes when the inode has been dirtied. For the
4122 * moment we only update the mode, atime, mtime, and ctime.
4125 zfs_dirty_inode(struct inode *ip, int flags)
4127 znode_t *zp = ITOZ(ip);
4128 zfs_sb_t *zsb = ITOZSB(ip);
4130 uint64_t mode, atime[2], mtime[2], ctime[2];
4131 sa_bulk_attr_t bulk[4];
4135 if (zfs_is_readonly(zsb) || dmu_objset_is_snapshot(zsb->z_os))
4143 * This is the lazytime semantic indroduced in Linux 4.0
4144 * This flag will only be called from update_time when lazytime is set.
4145 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4146 * Fortunately mtime and ctime are managed within ZFS itself, so we
4147 * only need to dirty atime.
4149 if (flags == I_DIRTY_TIME) {
4150 zp->z_atime_dirty = 1;
4155 tx = dmu_tx_create(zsb->z_os);
4157 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4158 zfs_sa_upgrade_txholds(tx, zp);
4160 error = dmu_tx_assign(tx, TXG_WAIT);
4166 mutex_enter(&zp->z_lock);
4167 zp->z_atime_dirty = 0;
4169 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zsb), NULL, &mode, 8);
4170 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zsb), NULL, &atime, 16);
4171 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
4172 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
4174 /* Preserve the mode, mtime and ctime provided by the inode */
4175 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4176 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4177 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4182 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4183 mutex_exit(&zp->z_lock);
4190 EXPORT_SYMBOL(zfs_dirty_inode);
4194 zfs_inactive(struct inode *ip)
4196 znode_t *zp = ITOZ(ip);
4197 zfs_sb_t *zsb = ITOZSB(ip);
4200 int need_unlock = 0;
4202 /* Only read lock if we haven't already write locked, e.g. rollback */
4203 if (!RW_WRITE_HELD(&zsb->z_teardown_inactive_lock)) {
4205 rw_enter(&zsb->z_teardown_inactive_lock, RW_READER);
4207 if (zp->z_sa_hdl == NULL) {
4209 rw_exit(&zsb->z_teardown_inactive_lock);
4213 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4214 dmu_tx_t *tx = dmu_tx_create(zsb->z_os);
4216 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4217 zfs_sa_upgrade_txholds(tx, zp);
4218 error = dmu_tx_assign(tx, TXG_WAIT);
4222 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4223 mutex_enter(&zp->z_lock);
4224 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zsb),
4225 (void *)&atime, sizeof (atime), tx);
4226 zp->z_atime_dirty = 0;
4227 mutex_exit(&zp->z_lock);
4234 rw_exit(&zsb->z_teardown_inactive_lock);
4236 EXPORT_SYMBOL(zfs_inactive);
4239 * Bounds-check the seek operation.
4241 * IN: ip - inode seeking within
4242 * ooff - old file offset
4243 * noffp - pointer to new file offset
4244 * ct - caller context
4246 * RETURN: 0 if success
4247 * EINVAL if new offset invalid
4251 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
4253 if (S_ISDIR(ip->i_mode))
4255 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4257 EXPORT_SYMBOL(zfs_seek);
4260 * Fill pages with data from the disk.
4263 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
4265 znode_t *zp = ITOZ(ip);
4266 zfs_sb_t *zsb = ITOZSB(ip);
4268 struct page *cur_pp;
4269 u_offset_t io_off, total;
4276 io_len = nr_pages << PAGE_SHIFT;
4277 i_size = i_size_read(ip);
4278 io_off = page_offset(pl[0]);
4280 if (io_off + io_len > i_size)
4281 io_len = i_size - io_off;
4284 * Iterate over list of pages and read each page individually.
4288 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4292 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4296 /* convert checksum errors into IO errors */
4298 err = SET_ERROR(EIO);
4301 cur_pp = pl[++page_idx];
4308 * Uses zfs_fillpage to read data from the file and fill the pages.
4310 * IN: ip - inode of file to get data from.
4311 * pl - list of pages to read
4312 * nr_pages - number of pages to read
4314 * RETURN: 0 on success, error code on failure.
4317 * vp - atime updated
4321 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
4323 znode_t *zp = ITOZ(ip);
4324 zfs_sb_t *zsb = ITOZSB(ip);
4333 err = zfs_fillpage(ip, pl, nr_pages);
4338 EXPORT_SYMBOL(zfs_getpage);
4341 * Check ZFS specific permissions to memory map a section of a file.
4343 * IN: ip - inode of the file to mmap
4345 * addrp - start address in memory region
4346 * len - length of memory region
4347 * vm_flags- address flags
4349 * RETURN: 0 if success
4350 * error code if failure
4354 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4355 unsigned long vm_flags)
4357 znode_t *zp = ITOZ(ip);
4358 zfs_sb_t *zsb = ITOZSB(ip);
4363 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4364 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4366 return (SET_ERROR(EPERM));
4369 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4370 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4372 return (SET_ERROR(EACCES));
4375 if (off < 0 || len > MAXOFFSET_T - off) {
4377 return (SET_ERROR(ENXIO));
4383 EXPORT_SYMBOL(zfs_map);
4386 * convoff - converts the given data (start, whence) to the
4390 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4395 if ((lckdat->l_whence == 2) || (whence == 2)) {
4396 if ((error = zfs_getattr(ip, &vap, 0, CRED()) != 0))
4400 switch (lckdat->l_whence) {
4402 lckdat->l_start += offset;
4405 lckdat->l_start += vap.va_size;
4410 return (SET_ERROR(EINVAL));
4413 if (lckdat->l_start < 0)
4414 return (SET_ERROR(EINVAL));
4418 lckdat->l_start -= offset;
4421 lckdat->l_start -= vap.va_size;
4426 return (SET_ERROR(EINVAL));
4429 lckdat->l_whence = (short)whence;
4434 * Free or allocate space in a file. Currently, this function only
4435 * supports the `F_FREESP' command. However, this command is somewhat
4436 * misnamed, as its functionality includes the ability to allocate as
4437 * well as free space.
4439 * IN: ip - inode of file to free data in.
4440 * cmd - action to take (only F_FREESP supported).
4441 * bfp - section of file to free/alloc.
4442 * flag - current file open mode flags.
4443 * offset - current file offset.
4444 * cr - credentials of caller [UNUSED].
4446 * RETURN: 0 on success, error code on failure.
4449 * ip - ctime|mtime updated
4453 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4454 offset_t offset, cred_t *cr)
4456 znode_t *zp = ITOZ(ip);
4457 zfs_sb_t *zsb = ITOZSB(ip);
4464 if (cmd != F_FREESP) {
4466 return (SET_ERROR(EINVAL));
4470 * Callers might not be able to detect properly that we are read-only,
4471 * so check it explicitly here.
4473 if (zfs_is_readonly(zsb)) {
4475 return (SET_ERROR(EROFS));
4478 if ((error = convoff(ip, bfp, 0, offset))) {
4483 if (bfp->l_len < 0) {
4485 return (SET_ERROR(EINVAL));
4489 * Permissions aren't checked on Solaris because on this OS
4490 * zfs_space() can only be called with an opened file handle.
4491 * On Linux we can get here through truncate_range() which
4492 * operates directly on inodes, so we need to check access rights.
4494 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4500 len = bfp->l_len; /* 0 means from off to end of file */
4502 error = zfs_freesp(zp, off, len, flag, TRUE);
4507 EXPORT_SYMBOL(zfs_space);
4511 zfs_fid(struct inode *ip, fid_t *fidp)
4513 znode_t *zp = ITOZ(ip);
4514 zfs_sb_t *zsb = ITOZSB(ip);
4517 uint64_t object = zp->z_id;
4524 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zsb),
4525 &gen64, sizeof (uint64_t))) != 0) {
4530 gen = (uint32_t)gen64;
4532 size = (zsb->z_parent != zsb) ? LONG_FID_LEN : SHORT_FID_LEN;
4533 if (fidp->fid_len < size) {
4534 fidp->fid_len = size;
4536 return (SET_ERROR(ENOSPC));
4539 zfid = (zfid_short_t *)fidp;
4541 zfid->zf_len = size;
4543 for (i = 0; i < sizeof (zfid->zf_object); i++)
4544 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4546 /* Must have a non-zero generation number to distinguish from .zfs */
4549 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4550 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4552 if (size == LONG_FID_LEN) {
4553 uint64_t objsetid = dmu_objset_id(zsb->z_os);
4556 zlfid = (zfid_long_t *)fidp;
4558 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4559 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4561 /* XXX - this should be the generation number for the objset */
4562 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4563 zlfid->zf_setgen[i] = 0;
4569 EXPORT_SYMBOL(zfs_fid);
4573 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4575 znode_t *zp = ITOZ(ip);
4576 zfs_sb_t *zsb = ITOZSB(ip);
4578 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4582 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4587 EXPORT_SYMBOL(zfs_getsecattr);
4591 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4593 znode_t *zp = ITOZ(ip);
4594 zfs_sb_t *zsb = ITOZSB(ip);
4596 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4597 zilog_t *zilog = zsb->z_log;
4602 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4604 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
4605 zil_commit(zilog, 0);
4610 EXPORT_SYMBOL(zfs_setsecattr);
4612 #ifdef HAVE_UIO_ZEROCOPY
4614 * Tunable, both must be a power of 2.
4616 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4617 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4618 * an arcbuf for a partial block read
4620 int zcr_blksz_min = (1 << 10); /* 1K */
4621 int zcr_blksz_max = (1 << 17); /* 128K */
4625 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
4627 znode_t *zp = ITOZ(ip);
4628 zfs_sb_t *zsb = ITOZSB(ip);
4629 int max_blksz = zsb->z_max_blksz;
4630 uio_t *uio = &xuio->xu_uio;
4631 ssize_t size = uio->uio_resid;
4632 offset_t offset = uio->uio_loffset;
4637 int preamble, postamble;
4639 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4640 return (SET_ERROR(EINVAL));
4647 * Loan out an arc_buf for write if write size is bigger than
4648 * max_blksz, and the file's block size is also max_blksz.
4651 if (size < blksz || zp->z_blksz != blksz) {
4653 return (SET_ERROR(EINVAL));
4656 * Caller requests buffers for write before knowing where the
4657 * write offset might be (e.g. NFS TCP write).
4662 preamble = P2PHASE(offset, blksz);
4664 preamble = blksz - preamble;
4669 postamble = P2PHASE(size, blksz);
4672 fullblk = size / blksz;
4673 (void) dmu_xuio_init(xuio,
4674 (preamble != 0) + fullblk + (postamble != 0));
4677 * Have to fix iov base/len for partial buffers. They
4678 * currently represent full arc_buf's.
4681 /* data begins in the middle of the arc_buf */
4682 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4685 (void) dmu_xuio_add(xuio, abuf,
4686 blksz - preamble, preamble);
4689 for (i = 0; i < fullblk; i++) {
4690 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4693 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
4697 /* data ends in the middle of the arc_buf */
4698 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4701 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
4706 * Loan out an arc_buf for read if the read size is larger than
4707 * the current file block size. Block alignment is not
4708 * considered. Partial arc_buf will be loaned out for read.
4710 blksz = zp->z_blksz;
4711 if (blksz < zcr_blksz_min)
4712 blksz = zcr_blksz_min;
4713 if (blksz > zcr_blksz_max)
4714 blksz = zcr_blksz_max;
4715 /* avoid potential complexity of dealing with it */
4716 if (blksz > max_blksz) {
4718 return (SET_ERROR(EINVAL));
4721 maxsize = zp->z_size - uio->uio_loffset;
4727 return (SET_ERROR(EINVAL));
4732 return (SET_ERROR(EINVAL));
4735 uio->uio_extflg = UIO_XUIO;
4736 XUIO_XUZC_RW(xuio) = ioflag;
4743 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
4747 int ioflag = XUIO_XUZC_RW(xuio);
4749 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
4751 i = dmu_xuio_cnt(xuio);
4753 abuf = dmu_xuio_arcbuf(xuio, i);
4755 * if abuf == NULL, it must be a write buffer
4756 * that has been returned in zfs_write().
4759 dmu_return_arcbuf(abuf);
4760 ASSERT(abuf || ioflag == UIO_WRITE);
4763 dmu_xuio_fini(xuio);
4766 #endif /* HAVE_UIO_ZEROCOPY */
4768 #if defined(_KERNEL) && defined(HAVE_SPL)
4769 module_param(zfs_delete_blocks, ulong, 0644);
4770 MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async");
4771 module_param(zfs_read_chunk_size, long, 0644);
4772 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");