4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
25 * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
26 * Copyright 2017 Nexenta Systems, Inc.
29 /* Portions Copyright 2007 Jeremy Teo */
30 /* Portions Copyright 2010 Robert Milkowski */
33 #include <sys/types.h>
34 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/sysmacros.h>
38 #include <sys/resource.h>
40 #include <sys/vfs_opreg.h>
44 #include <sys/taskq.h>
46 #include <sys/vmsystm.h>
47 #include <sys/atomic.h>
49 #include <sys/pathname.h>
50 #include <sys/cmn_err.h>
51 #include <sys/errno.h>
52 #include <sys/unistd.h>
53 #include <sys/zfs_dir.h>
54 #include <sys/zfs_acl.h>
55 #include <sys/zfs_ioctl.h>
56 #include <sys/fs/zfs.h>
58 #include <sys/dmu_objset.h>
64 #include <sys/dirent.h>
65 #include <sys/policy.h>
66 #include <sys/sunddi.h>
69 #include "fs/fs_subr.h"
70 #include <sys/zfs_ctldir.h>
71 #include <sys/zfs_fuid.h>
72 #include <sys/zfs_sa.h>
73 #include <sys/zfs_vnops.h>
75 #include <sys/zfs_rlock.h>
76 #include <sys/extdirent.h>
77 #include <sys/kidmap.h>
85 * Each vnode op performs some logical unit of work. To do this, the ZPL must
86 * properly lock its in-core state, create a DMU transaction, do the work,
87 * record this work in the intent log (ZIL), commit the DMU transaction,
88 * and wait for the intent log to commit if it is a synchronous operation.
89 * Moreover, the vnode ops must work in both normal and log replay context.
90 * The ordering of events is important to avoid deadlocks and references
91 * to freed memory. The example below illustrates the following Big Rules:
93 * (1) A check must be made in each zfs thread for a mounted file system.
94 * This is done avoiding races using ZFS_ENTER(zfsvfs).
95 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
96 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
97 * can return EIO from the calling function.
99 * (2) iput() should always be the last thing except for zil_commit()
100 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
101 * First, if it's the last reference, the vnode/znode
102 * can be freed, so the zp may point to freed memory. Second, the last
103 * reference will call zfs_zinactive(), which may induce a lot of work --
104 * pushing cached pages (which acquires range locks) and syncing out
105 * cached atime changes. Third, zfs_zinactive() may require a new tx,
106 * which could deadlock the system if you were already holding one.
107 * If you must call iput() within a tx then use zfs_iput_async().
109 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
110 * as they can span dmu_tx_assign() calls.
112 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
113 * dmu_tx_assign(). This is critical because we don't want to block
114 * while holding locks.
116 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
117 * reduces lock contention and CPU usage when we must wait (note that if
118 * throughput is constrained by the storage, nearly every transaction
121 * Note, in particular, that if a lock is sometimes acquired before
122 * the tx assigns, and sometimes after (e.g. z_lock), then failing
123 * to use a non-blocking assign can deadlock the system. The scenario:
125 * Thread A has grabbed a lock before calling dmu_tx_assign().
126 * Thread B is in an already-assigned tx, and blocks for this lock.
127 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
128 * forever, because the previous txg can't quiesce until B's tx commits.
130 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
131 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
132 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
133 * to indicate that this operation has already called dmu_tx_wait().
134 * This will ensure that we don't retry forever, waiting a short bit
137 * (5) If the operation succeeded, generate the intent log entry for it
138 * before dropping locks. This ensures that the ordering of events
139 * in the intent log matches the order in which they actually occurred.
140 * During ZIL replay the zfs_log_* functions will update the sequence
141 * number to indicate the zil transaction has replayed.
143 * (6) At the end of each vnode op, the DMU tx must always commit,
144 * regardless of whether there were any errors.
146 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
147 * to ensure that synchronous semantics are provided when necessary.
149 * In general, this is how things should be ordered in each vnode op:
151 * ZFS_ENTER(zfsvfs); // exit if unmounted
153 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
154 * rw_enter(...); // grab any other locks you need
155 * tx = dmu_tx_create(...); // get DMU tx
156 * dmu_tx_hold_*(); // hold each object you might modify
157 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
159 * rw_exit(...); // drop locks
160 * zfs_dirent_unlock(dl); // unlock directory entry
161 * iput(...); // release held vnodes
162 * if (error == ERESTART) {
168 * dmu_tx_abort(tx); // abort DMU tx
169 * ZFS_EXIT(zfsvfs); // finished in zfs
170 * return (error); // really out of space
172 * error = do_real_work(); // do whatever this VOP does
174 * zfs_log_*(...); // on success, make ZIL entry
175 * dmu_tx_commit(tx); // commit DMU tx -- error or not
176 * rw_exit(...); // drop locks
177 * zfs_dirent_unlock(dl); // unlock directory entry
178 * iput(...); // release held vnodes
179 * zil_commit(zilog, foid); // synchronous when necessary
180 * ZFS_EXIT(zfsvfs); // finished in zfs
181 * return (error); // done, report error
185 * Virus scanning is unsupported. It would be possible to add a hook
186 * here to performance the required virus scan. This could be done
187 * entirely in the kernel or potentially as an update to invoke a
191 zfs_vscan(struct inode *ip, cred_t *cr, int async)
198 zfs_open(struct inode *ip, int mode, int flag, cred_t *cr)
200 znode_t *zp = ITOZ(ip);
201 zfsvfs_t *zfsvfs = ITOZSB(ip);
206 /* Honor ZFS_APPENDONLY file attribute */
207 if ((mode & FMODE_WRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
208 ((flag & O_APPEND) == 0)) {
210 return (SET_ERROR(EPERM));
213 /* Virus scan eligible files on open */
214 if (!zfs_has_ctldir(zp) && zfsvfs->z_vscan && S_ISREG(ip->i_mode) &&
215 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
216 if (zfs_vscan(ip, cr, 0) != 0) {
218 return (SET_ERROR(EACCES));
222 /* Keep a count of the synchronous opens in the znode */
224 atomic_inc_32(&zp->z_sync_cnt);
229 EXPORT_SYMBOL(zfs_open);
233 zfs_close(struct inode *ip, int flag, cred_t *cr)
235 znode_t *zp = ITOZ(ip);
236 zfsvfs_t *zfsvfs = ITOZSB(ip);
241 /* Decrement the synchronous opens in the znode */
243 atomic_dec_32(&zp->z_sync_cnt);
245 if (!zfs_has_ctldir(zp) && zfsvfs->z_vscan && S_ISREG(ip->i_mode) &&
246 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
247 VERIFY(zfs_vscan(ip, cr, 1) == 0);
252 EXPORT_SYMBOL(zfs_close);
254 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
256 * Lseek support for finding holes (cmd == SEEK_HOLE) and
257 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
260 zfs_holey_common(struct inode *ip, int cmd, loff_t *off)
262 znode_t *zp = ITOZ(ip);
263 uint64_t noff = (uint64_t)*off; /* new offset */
268 file_sz = zp->z_size;
269 if (noff >= file_sz) {
270 return (SET_ERROR(ENXIO));
273 if (cmd == SEEK_HOLE)
278 error = dmu_offset_next(ZTOZSB(zp)->z_os, zp->z_id, hole, &noff);
281 return (SET_ERROR(ENXIO));
284 * We could find a hole that begins after the logical end-of-file,
285 * because dmu_offset_next() only works on whole blocks. If the
286 * EOF falls mid-block, then indicate that the "virtual hole"
287 * at the end of the file begins at the logical EOF, rather than
288 * at the end of the last block.
290 if (noff > file_sz) {
302 zfs_holey(struct inode *ip, int cmd, loff_t *off)
304 znode_t *zp = ITOZ(ip);
305 zfsvfs_t *zfsvfs = ITOZSB(ip);
311 error = zfs_holey_common(ip, cmd, off);
316 EXPORT_SYMBOL(zfs_holey);
317 #endif /* SEEK_HOLE && SEEK_DATA */
321 * When a file is memory mapped, we must keep the IO data synchronized
322 * between the DMU cache and the memory mapped pages. What this means:
324 * On Write: If we find a memory mapped page, we write to *both*
325 * the page and the dmu buffer.
328 update_pages(struct inode *ip, int64_t start, int len,
329 objset_t *os, uint64_t oid)
331 struct address_space *mp = ip->i_mapping;
337 off = start & (PAGE_SIZE-1);
338 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
339 nbytes = MIN(PAGE_SIZE - off, len);
341 pp = find_lock_page(mp, start >> PAGE_SHIFT);
343 if (mapping_writably_mapped(mp))
344 flush_dcache_page(pp);
347 (void) dmu_read(os, oid, start+off, nbytes, pb+off,
351 if (mapping_writably_mapped(mp))
352 flush_dcache_page(pp);
354 mark_page_accessed(pp);
367 * When a file is memory mapped, we must keep the IO data synchronized
368 * between the DMU cache and the memory mapped pages. What this means:
370 * On Read: We "read" preferentially from memory mapped pages,
371 * else we default from the dmu buffer.
373 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
374 * the file is memory mapped.
377 mappedread(struct inode *ip, int nbytes, uio_t *uio)
379 struct address_space *mp = ip->i_mapping;
381 znode_t *zp = ITOZ(ip);
388 start = uio->uio_loffset;
389 off = start & (PAGE_SIZE-1);
390 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
391 bytes = MIN(PAGE_SIZE - off, len);
393 pp = find_lock_page(mp, start >> PAGE_SHIFT);
395 ASSERT(PageUptodate(pp));
398 error = uiomove(pb + off, bytes, UIO_READ, uio);
401 if (mapping_writably_mapped(mp))
402 flush_dcache_page(pp);
404 mark_page_accessed(pp);
408 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
421 unsigned long zfs_read_chunk_size = 1024 * 1024; /* Tunable */
422 unsigned long zfs_delete_blocks = DMU_MAX_DELETEBLKCNT;
425 * Read bytes from specified file into supplied buffer.
427 * IN: ip - inode of file to be read from.
428 * uio - structure supplying read location, range info,
430 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
431 * O_DIRECT flag; used to bypass page cache.
432 * cr - credentials of caller.
434 * OUT: uio - updated offset and range, buffer filled.
436 * RETURN: 0 on success, error code on failure.
439 * inode - atime updated if byte count > 0
443 zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
445 znode_t *zp = ITOZ(ip);
446 zfsvfs_t *zfsvfs = ITOZSB(ip);
450 #ifdef HAVE_UIO_ZEROCOPY
452 #endif /* HAVE_UIO_ZEROCOPY */
457 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
459 return (SET_ERROR(EACCES));
463 * Validate file offset
465 if (uio->uio_loffset < (offset_t)0) {
467 return (SET_ERROR(EINVAL));
471 * Fasttrack empty reads
473 if (uio->uio_resid == 0) {
479 * If we're in FRSYNC mode, sync out this znode before reading it.
481 if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
482 zil_commit(zfsvfs->z_log, zp->z_id);
485 * Lock the range against changes.
487 rl = zfs_range_lock(&zp->z_range_lock, uio->uio_loffset, uio->uio_resid,
491 * If we are reading past end-of-file we can skip
492 * to the end; but we might still need to set atime.
494 if (uio->uio_loffset >= zp->z_size) {
499 ASSERT(uio->uio_loffset < zp->z_size);
500 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
502 #ifdef HAVE_UIO_ZEROCOPY
503 if ((uio->uio_extflg == UIO_XUIO) &&
504 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
506 int blksz = zp->z_blksz;
507 uint64_t offset = uio->uio_loffset;
509 xuio = (xuio_t *)uio;
511 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
514 ASSERT(offset + n <= blksz);
517 (void) dmu_xuio_init(xuio, nblk);
519 if (vn_has_cached_data(ip)) {
521 * For simplicity, we always allocate a full buffer
522 * even if we only expect to read a portion of a block.
524 while (--nblk >= 0) {
525 (void) dmu_xuio_add(xuio,
526 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
531 #endif /* HAVE_UIO_ZEROCOPY */
534 nbytes = MIN(n, zfs_read_chunk_size -
535 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
537 if (zp->z_is_mapped && !(ioflag & O_DIRECT)) {
538 error = mappedread(ip, nbytes, uio);
540 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
545 /* convert checksum errors into IO errors */
547 error = SET_ERROR(EIO);
554 zfs_range_unlock(rl);
559 EXPORT_SYMBOL(zfs_read);
562 * Write the bytes to a file.
564 * IN: ip - inode of file to be written to.
565 * uio - structure supplying write location, range info,
567 * ioflag - FAPPEND flag set if in append mode.
568 * O_DIRECT flag; used to bypass page cache.
569 * cr - credentials of caller.
571 * OUT: uio - updated offset and range.
573 * RETURN: 0 if success
574 * error code if failure
577 * ip - ctime|mtime updated if byte count > 0
582 zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
584 znode_t *zp = ITOZ(ip);
585 rlim64_t limit = uio->uio_limit;
586 ssize_t start_resid = uio->uio_resid;
590 zfsvfs_t *zfsvfs = ZTOZSB(zp);
595 int max_blksz = zfsvfs->z_max_blksz;
598 const iovec_t *aiov = NULL;
602 sa_bulk_attr_t bulk[4];
603 uint64_t mtime[2], ctime[2];
605 #ifdef HAVE_UIO_ZEROCOPY
607 const iovec_t *iovp = uio->uio_iov;
608 ASSERTV(int iovcnt = uio->uio_iovcnt);
612 * Fasttrack empty write
618 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
624 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
625 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
626 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
628 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
632 * Callers might not be able to detect properly that we are read-only,
633 * so check it explicitly here.
635 if (zfs_is_readonly(zfsvfs)) {
637 return (SET_ERROR(EROFS));
641 * If immutable or not appending then return EPERM
643 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
644 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
645 (uio->uio_loffset < zp->z_size))) {
647 return (SET_ERROR(EPERM));
650 zilog = zfsvfs->z_log;
653 * Validate file offset
655 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
658 return (SET_ERROR(EINVAL));
662 * Pre-fault the pages to ensure slow (eg NFS) pages
664 * Skip this if uio contains loaned arc_buf.
666 #ifdef HAVE_UIO_ZEROCOPY
667 if ((uio->uio_extflg == UIO_XUIO) &&
668 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
669 xuio = (xuio_t *)uio;
672 uio_prefaultpages(MIN(n, max_blksz), uio);
675 * If in append mode, set the io offset pointer to eof.
677 if (ioflag & FAPPEND) {
679 * Obtain an appending range lock to guarantee file append
680 * semantics. We reset the write offset once we have the lock.
682 rl = zfs_range_lock(&zp->z_range_lock, 0, n, RL_APPEND);
684 if (rl->r_len == UINT64_MAX) {
686 * We overlocked the file because this write will cause
687 * the file block size to increase.
688 * Note that zp_size cannot change with this lock held.
692 uio->uio_loffset = woff;
695 * Note that if the file block size will change as a result of
696 * this write, then this range lock will lock the entire file
697 * so that we can re-write the block safely.
699 rl = zfs_range_lock(&zp->z_range_lock, woff, n, RL_WRITER);
703 zfs_range_unlock(rl);
705 return (SET_ERROR(EFBIG));
708 if ((woff + n) > limit || woff > (limit - n))
711 /* Will this write extend the file length? */
712 write_eof = (woff + n > zp->z_size);
714 end_size = MAX(zp->z_size, woff + n);
717 * Write the file in reasonable size chunks. Each chunk is written
718 * in a separate transaction; this keeps the intent log records small
719 * and allows us to do more fine-grained space accounting.
723 woff = uio->uio_loffset;
724 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
725 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
727 dmu_return_arcbuf(abuf);
728 error = SET_ERROR(EDQUOT);
732 if (xuio && abuf == NULL) {
733 #ifdef HAVE_UIO_ZEROCOPY
734 ASSERT(i_iov < iovcnt);
735 ASSERT3U(uio->uio_segflg, !=, UIO_BVEC);
737 abuf = dmu_xuio_arcbuf(xuio, i_iov);
738 dmu_xuio_clear(xuio, i_iov);
739 ASSERT((aiov->iov_base == abuf->b_data) ||
740 ((char *)aiov->iov_base - (char *)abuf->b_data +
741 aiov->iov_len == arc_buf_size(abuf)));
744 } else if (abuf == NULL && n >= max_blksz &&
745 woff >= zp->z_size &&
746 P2PHASE(woff, max_blksz) == 0 &&
747 zp->z_blksz == max_blksz) {
749 * This write covers a full block. "Borrow" a buffer
750 * from the dmu so that we can fill it before we enter
751 * a transaction. This avoids the possibility of
752 * holding up the transaction if the data copy hangs
753 * up on a pagefault (e.g., from an NFS server mapping).
757 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
759 ASSERT(abuf != NULL);
760 ASSERT(arc_buf_size(abuf) == max_blksz);
761 if ((error = uiocopy(abuf->b_data, max_blksz,
762 UIO_WRITE, uio, &cbytes))) {
763 dmu_return_arcbuf(abuf);
766 ASSERT(cbytes == max_blksz);
770 * Start a transaction.
772 tx = dmu_tx_create(zfsvfs->z_os);
773 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
774 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
775 zfs_sa_upgrade_txholds(tx, zp);
776 error = dmu_tx_assign(tx, TXG_WAIT);
780 dmu_return_arcbuf(abuf);
785 * If zfs_range_lock() over-locked we grow the blocksize
786 * and then reduce the lock range. This will only happen
787 * on the first iteration since zfs_range_reduce() will
788 * shrink down r_len to the appropriate size.
790 if (rl->r_len == UINT64_MAX) {
793 if (zp->z_blksz > max_blksz) {
795 * File's blocksize is already larger than the
796 * "recordsize" property. Only let it grow to
797 * the next power of 2.
799 ASSERT(!ISP2(zp->z_blksz));
800 new_blksz = MIN(end_size,
801 1 << highbit64(zp->z_blksz));
803 new_blksz = MIN(end_size, max_blksz);
805 zfs_grow_blocksize(zp, new_blksz, tx);
806 zfs_range_reduce(rl, woff, n);
810 * XXX - should we really limit each write to z_max_blksz?
811 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
813 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
816 tx_bytes = uio->uio_resid;
817 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
819 tx_bytes -= uio->uio_resid;
822 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
824 * If this is not a full block write, but we are
825 * extending the file past EOF and this data starts
826 * block-aligned, use assign_arcbuf(). Otherwise,
827 * write via dmu_write().
829 if (tx_bytes < max_blksz && (!write_eof ||
830 aiov->iov_base != abuf->b_data)) {
832 dmu_write(zfsvfs->z_os, zp->z_id, woff,
833 // cppcheck-suppress nullPointer
834 aiov->iov_len, aiov->iov_base, tx);
835 dmu_return_arcbuf(abuf);
836 xuio_stat_wbuf_copied();
838 ASSERT(xuio || tx_bytes == max_blksz);
839 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
842 ASSERT(tx_bytes <= uio->uio_resid);
843 uioskip(uio, tx_bytes);
845 if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT)) {
846 update_pages(ip, woff,
847 tx_bytes, zfsvfs->z_os, zp->z_id);
851 * If we made no progress, we're done. If we made even
852 * partial progress, update the znode and ZIL accordingly.
855 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
856 (void *)&zp->z_size, sizeof (uint64_t), tx);
863 * Clear Set-UID/Set-GID bits on successful write if not
864 * privileged and at least one of the execute bits is set.
866 * It would be nice to to this after all writes have
867 * been done, but that would still expose the ISUID/ISGID
868 * to another app after the partial write is committed.
870 * Note: we don't call zfs_fuid_map_id() here because
871 * user 0 is not an ephemeral uid.
873 mutex_enter(&zp->z_acl_lock);
874 uid = KUID_TO_SUID(ip->i_uid);
875 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
876 (S_IXUSR >> 6))) != 0 &&
877 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
878 secpolicy_vnode_setid_retain(cr,
879 ((zp->z_mode & S_ISUID) != 0 && uid == 0)) != 0) {
881 zp->z_mode &= ~(S_ISUID | S_ISGID);
882 ip->i_mode = newmode = zp->z_mode;
883 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
884 (void *)&newmode, sizeof (uint64_t), tx);
886 mutex_exit(&zp->z_acl_lock);
888 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
891 * Update the file size (zp_size) if it has changed;
892 * account for possible concurrent updates.
894 while ((end_size = zp->z_size) < uio->uio_loffset) {
895 (void) atomic_cas_64(&zp->z_size, end_size,
900 * If we are replaying and eof is non zero then force
901 * the file size to the specified eof. Note, there's no
902 * concurrency during replay.
904 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
905 zp->z_size = zfsvfs->z_replay_eof;
907 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
909 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag,
915 ASSERT(tx_bytes == nbytes);
919 uio_prefaultpages(MIN(n, max_blksz), uio);
922 zfs_inode_update(zp);
923 zfs_range_unlock(rl);
926 * If we're in replay mode, or we made no progress, return error.
927 * Otherwise, it's at least a partial write, so it's successful.
929 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
934 if (ioflag & (FSYNC | FDSYNC) ||
935 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
936 zil_commit(zilog, zp->z_id);
941 EXPORT_SYMBOL(zfs_write);
944 * Drop a reference on the passed inode asynchronously. This ensures
945 * that the caller will never drop the last reference on an inode in
946 * the current context. Doing so while holding open a tx could result
947 * in a deadlock if iput_final() re-enters the filesystem code.
950 zfs_iput_async(struct inode *ip)
952 objset_t *os = ITOZSB(ip)->z_os;
954 ASSERT(atomic_read(&ip->i_count) > 0);
957 if (atomic_read(&ip->i_count) == 1)
958 VERIFY(taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os)),
959 (task_func_t *)iput, ip, TQ_SLEEP) != TASKQID_INVALID);
965 zfs_get_done(zgd_t *zgd, int error)
967 znode_t *zp = zgd->zgd_private;
970 dmu_buf_rele(zgd->zgd_db, zgd);
972 zfs_range_unlock(zgd->zgd_rl);
975 * Release the vnode asynchronously as we currently have the
976 * txg stopped from syncing.
978 zfs_iput_async(ZTOI(zp));
980 if (error == 0 && zgd->zgd_bp)
981 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
983 kmem_free(zgd, sizeof (zgd_t));
987 static int zil_fault_io = 0;
991 * Get data to generate a TX_WRITE intent log record.
994 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
996 zfsvfs_t *zfsvfs = arg;
997 objset_t *os = zfsvfs->z_os;
999 uint64_t object = lr->lr_foid;
1000 uint64_t offset = lr->lr_offset;
1001 uint64_t size = lr->lr_length;
1002 blkptr_t *bp = &lr->lr_blkptr;
1007 ASSERT(zio != NULL);
1011 * Nothing to do if the file has been removed
1013 if (zfs_zget(zfsvfs, object, &zp) != 0)
1014 return (SET_ERROR(ENOENT));
1015 if (zp->z_unlinked) {
1017 * Release the vnode asynchronously as we currently have the
1018 * txg stopped from syncing.
1020 zfs_iput_async(ZTOI(zp));
1021 return (SET_ERROR(ENOENT));
1024 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1025 zgd->zgd_zilog = zfsvfs->z_log;
1026 zgd->zgd_private = zp;
1029 * Write records come in two flavors: immediate and indirect.
1030 * For small writes it's cheaper to store the data with the
1031 * log record (immediate); for large writes it's cheaper to
1032 * sync the data and get a pointer to it (indirect) so that
1033 * we don't have to write the data twice.
1035 if (buf != NULL) { /* immediate write */
1036 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset, size,
1038 /* test for truncation needs to be done while range locked */
1039 if (offset >= zp->z_size) {
1040 error = SET_ERROR(ENOENT);
1042 error = dmu_read(os, object, offset, size, buf,
1043 DMU_READ_NO_PREFETCH);
1045 ASSERT(error == 0 || error == ENOENT);
1046 } else { /* indirect write */
1048 * Have to lock the whole block to ensure when it's
1049 * written out and it's checksum is being calculated
1050 * that no one can change the data. We need to re-check
1051 * blocksize after we get the lock in case it's changed!
1056 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1058 zgd->zgd_rl = zfs_range_lock(&zp->z_range_lock, offset,
1060 if (zp->z_blksz == size)
1063 zfs_range_unlock(zgd->zgd_rl);
1065 /* test for truncation needs to be done while range locked */
1066 if (lr->lr_offset >= zp->z_size)
1067 error = SET_ERROR(ENOENT);
1070 error = SET_ERROR(EIO);
1075 error = dmu_buf_hold(os, object, offset, zgd, &db,
1076 DMU_READ_NO_PREFETCH);
1079 blkptr_t *obp = dmu_buf_get_blkptr(db);
1081 ASSERT(BP_IS_HOLE(bp));
1088 ASSERT(db->db_offset == offset);
1089 ASSERT(db->db_size == size);
1091 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1093 ASSERT(error || lr->lr_length <= size);
1096 * On success, we need to wait for the write I/O
1097 * initiated by dmu_sync() to complete before we can
1098 * release this dbuf. We will finish everything up
1099 * in the zfs_get_done() callback.
1104 if (error == EALREADY) {
1105 lr->lr_common.lrc_txtype = TX_WRITE2;
1111 zfs_get_done(zgd, error);
1118 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
1120 znode_t *zp = ITOZ(ip);
1121 zfsvfs_t *zfsvfs = ITOZSB(ip);
1127 if (flag & V_ACE_MASK)
1128 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1130 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1135 EXPORT_SYMBOL(zfs_access);
1138 * Lookup an entry in a directory, or an extended attribute directory.
1139 * If it exists, return a held inode reference for it.
1141 * IN: dip - inode of directory to search.
1142 * nm - name of entry to lookup.
1143 * flags - LOOKUP_XATTR set if looking for an attribute.
1144 * cr - credentials of caller.
1145 * direntflags - directory lookup flags
1146 * realpnp - returned pathname.
1148 * OUT: ipp - inode of located entry, NULL if not found.
1150 * RETURN: 0 on success, error code on failure.
1157 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
1158 cred_t *cr, int *direntflags, pathname_t *realpnp)
1160 znode_t *zdp = ITOZ(dip);
1161 zfsvfs_t *zfsvfs = ITOZSB(dip);
1165 * Fast path lookup, however we must skip DNLC lookup
1166 * for case folding or normalizing lookups because the
1167 * DNLC code only stores the passed in name. This means
1168 * creating 'a' and removing 'A' on a case insensitive
1169 * file system would work, but DNLC still thinks 'a'
1170 * exists and won't let you create it again on the next
1171 * pass through fast path.
1173 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1175 if (!S_ISDIR(dip->i_mode)) {
1176 return (SET_ERROR(ENOTDIR));
1177 } else if (zdp->z_sa_hdl == NULL) {
1178 return (SET_ERROR(EIO));
1181 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1182 error = zfs_fastaccesschk_execute(zdp, cr);
1190 } else if (!zdp->z_zfsvfs->z_norm &&
1191 (zdp->z_zfsvfs->z_case == ZFS_CASE_SENSITIVE)) {
1193 vnode_t *tvp = dnlc_lookup(dvp, nm);
1196 error = zfs_fastaccesschk_execute(zdp, cr);
1201 if (tvp == DNLC_NO_VNODE) {
1203 return (SET_ERROR(ENOENT));
1206 return (specvp_check(vpp, cr));
1209 #endif /* HAVE_DNLC */
1218 if (flags & LOOKUP_XATTR) {
1220 * We don't allow recursive attributes..
1221 * Maybe someday we will.
1223 if (zdp->z_pflags & ZFS_XATTR) {
1225 return (SET_ERROR(EINVAL));
1228 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1234 * Do we have permission to get into attribute directory?
1237 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1247 if (!S_ISDIR(dip->i_mode)) {
1249 return (SET_ERROR(ENOTDIR));
1253 * Check accessibility of directory.
1256 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1261 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1262 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1264 return (SET_ERROR(EILSEQ));
1267 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1268 if ((error == 0) && (*ipp))
1269 zfs_inode_update(ITOZ(*ipp));
1274 EXPORT_SYMBOL(zfs_lookup);
1277 * Attempt to create a new entry in a directory. If the entry
1278 * already exists, truncate the file if permissible, else return
1279 * an error. Return the ip of the created or trunc'd file.
1281 * IN: dip - inode of directory to put new file entry in.
1282 * name - name of new file entry.
1283 * vap - attributes of new file.
1284 * excl - flag indicating exclusive or non-exclusive mode.
1285 * mode - mode to open file with.
1286 * cr - credentials of caller.
1287 * flag - large file flag [UNUSED].
1288 * vsecp - ACL to be set
1290 * OUT: ipp - inode of created or trunc'd entry.
1292 * RETURN: 0 on success, error code on failure.
1295 * dip - ctime|mtime updated if new entry created
1296 * ip - ctime|mtime always, atime if new
1301 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1302 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1304 znode_t *zp, *dzp = ITOZ(dip);
1305 zfsvfs_t *zfsvfs = ITOZSB(dip);
1313 zfs_acl_ids_t acl_ids;
1314 boolean_t fuid_dirtied;
1315 boolean_t have_acl = B_FALSE;
1316 boolean_t waited = B_FALSE;
1319 * If we have an ephemeral id, ACL, or XVATTR then
1320 * make sure file system is at proper version
1326 if (zfsvfs->z_use_fuids == B_FALSE &&
1327 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1328 return (SET_ERROR(EINVAL));
1331 return (SET_ERROR(EINVAL));
1336 zilog = zfsvfs->z_log;
1338 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1339 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1341 return (SET_ERROR(EILSEQ));
1344 if (vap->va_mask & ATTR_XVATTR) {
1345 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1346 crgetuid(cr), cr, vap->va_mode)) != 0) {
1354 if (*name == '\0') {
1356 * Null component name refers to the directory itself.
1363 /* possible igrab(zp) */
1366 if (flag & FIGNORECASE)
1369 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1373 zfs_acl_ids_free(&acl_ids);
1374 if (strcmp(name, "..") == 0)
1375 error = SET_ERROR(EISDIR);
1385 * Create a new file object and update the directory
1388 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1390 zfs_acl_ids_free(&acl_ids);
1395 * We only support the creation of regular files in
1396 * extended attribute directories.
1399 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1401 zfs_acl_ids_free(&acl_ids);
1402 error = SET_ERROR(EINVAL);
1406 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1407 cr, vsecp, &acl_ids)) != 0)
1411 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1412 zfs_acl_ids_free(&acl_ids);
1413 error = SET_ERROR(EDQUOT);
1417 tx = dmu_tx_create(os);
1419 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1420 ZFS_SA_BASE_ATTR_SIZE);
1422 fuid_dirtied = zfsvfs->z_fuid_dirty;
1424 zfs_fuid_txhold(zfsvfs, tx);
1425 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1426 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1427 if (!zfsvfs->z_use_sa &&
1428 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1429 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1430 0, acl_ids.z_aclp->z_acl_bytes);
1432 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1434 zfs_dirent_unlock(dl);
1435 if (error == ERESTART) {
1441 zfs_acl_ids_free(&acl_ids);
1446 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1449 zfs_fuid_sync(zfsvfs, tx);
1451 (void) zfs_link_create(dl, zp, tx, ZNEW);
1452 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1453 if (flag & FIGNORECASE)
1455 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1456 vsecp, acl_ids.z_fuidp, vap);
1457 zfs_acl_ids_free(&acl_ids);
1460 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1463 zfs_acl_ids_free(&acl_ids);
1467 * A directory entry already exists for this name.
1470 * Can't truncate an existing file if in exclusive mode.
1473 error = SET_ERROR(EEXIST);
1477 * Can't open a directory for writing.
1479 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1480 error = SET_ERROR(EISDIR);
1484 * Verify requested access to file.
1486 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1490 mutex_enter(&dzp->z_lock);
1492 mutex_exit(&dzp->z_lock);
1495 * Truncate regular files if requested.
1497 if (S_ISREG(ZTOI(zp)->i_mode) &&
1498 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1499 /* we can't hold any locks when calling zfs_freesp() */
1501 zfs_dirent_unlock(dl);
1504 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1510 zfs_dirent_unlock(dl);
1516 zfs_inode_update(dzp);
1517 zfs_inode_update(zp);
1521 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1522 zil_commit(zilog, 0);
1527 EXPORT_SYMBOL(zfs_create);
1531 zfs_tmpfile(struct inode *dip, vattr_t *vap, int excl,
1532 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1534 znode_t *zp = NULL, *dzp = ITOZ(dip);
1535 zfsvfs_t *zfsvfs = ITOZSB(dip);
1541 zfs_acl_ids_t acl_ids;
1542 boolean_t fuid_dirtied;
1543 boolean_t have_acl = B_FALSE;
1544 boolean_t waited = B_FALSE;
1547 * If we have an ephemeral id, ACL, or XVATTR then
1548 * make sure file system is at proper version
1554 if (zfsvfs->z_use_fuids == B_FALSE &&
1555 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1556 return (SET_ERROR(EINVAL));
1562 if (vap->va_mask & ATTR_XVATTR) {
1563 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1564 crgetuid(cr), cr, vap->va_mode)) != 0) {
1574 * Create a new file object and update the directory
1577 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1579 zfs_acl_ids_free(&acl_ids);
1583 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1584 cr, vsecp, &acl_ids)) != 0)
1588 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1589 zfs_acl_ids_free(&acl_ids);
1590 error = SET_ERROR(EDQUOT);
1594 tx = dmu_tx_create(os);
1596 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1597 ZFS_SA_BASE_ATTR_SIZE);
1598 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1600 fuid_dirtied = zfsvfs->z_fuid_dirty;
1602 zfs_fuid_txhold(zfsvfs, tx);
1603 if (!zfsvfs->z_use_sa &&
1604 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1605 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1606 0, acl_ids.z_aclp->z_acl_bytes);
1608 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1610 if (error == ERESTART) {
1616 zfs_acl_ids_free(&acl_ids);
1621 zfs_mknode(dzp, vap, tx, cr, IS_TMPFILE, &zp, &acl_ids);
1624 zfs_fuid_sync(zfsvfs, tx);
1626 /* Add to unlinked set */
1628 zfs_unlinked_add(zp, tx);
1629 zfs_acl_ids_free(&acl_ids);
1637 zfs_inode_update(dzp);
1638 zfs_inode_update(zp);
1647 * Remove an entry from a directory.
1649 * IN: dip - inode of directory to remove entry from.
1650 * name - name of entry to remove.
1651 * cr - credentials of caller.
1653 * RETURN: 0 if success
1654 * error code if failure
1658 * ip - ctime (if nlink > 0)
1661 uint64_t null_xattr = 0;
1665 zfs_remove(struct inode *dip, char *name, cred_t *cr, int flags)
1667 znode_t *zp, *dzp = ITOZ(dip);
1670 zfsvfs_t *zfsvfs = ITOZSB(dip);
1672 uint64_t acl_obj, xattr_obj;
1673 uint64_t xattr_obj_unlinked = 0;
1678 boolean_t may_delete_now, delete_now = FALSE;
1679 boolean_t unlinked, toobig = FALSE;
1681 pathname_t *realnmp = NULL;
1685 boolean_t waited = B_FALSE;
1688 return (SET_ERROR(EINVAL));
1692 zilog = zfsvfs->z_log;
1694 if (flags & FIGNORECASE) {
1704 * Attempt to lock directory; fail if entry doesn't exist.
1706 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1716 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1721 * Need to use rmdir for removing directories.
1723 if (S_ISDIR(ip->i_mode)) {
1724 error = SET_ERROR(EPERM);
1730 dnlc_remove(dvp, realnmp->pn_buf);
1732 dnlc_remove(dvp, name);
1733 #endif /* HAVE_DNLC */
1735 mutex_enter(&zp->z_lock);
1736 may_delete_now = atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped);
1737 mutex_exit(&zp->z_lock);
1740 * We may delete the znode now, or we may put it in the unlinked set;
1741 * it depends on whether we're the last link, and on whether there are
1742 * other holds on the inode. So we dmu_tx_hold() the right things to
1743 * allow for either case.
1746 tx = dmu_tx_create(zfsvfs->z_os);
1747 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1748 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1749 zfs_sa_upgrade_txholds(tx, zp);
1750 zfs_sa_upgrade_txholds(tx, dzp);
1751 if (may_delete_now) {
1752 toobig = zp->z_size > zp->z_blksz * zfs_delete_blocks;
1753 /* if the file is too big, only hold_free a token amount */
1754 dmu_tx_hold_free(tx, zp->z_id, 0,
1755 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1758 /* are there any extended attributes? */
1759 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1760 &xattr_obj, sizeof (xattr_obj));
1761 if (error == 0 && xattr_obj) {
1762 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1764 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1765 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1768 mutex_enter(&zp->z_lock);
1769 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1770 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1771 mutex_exit(&zp->z_lock);
1773 /* charge as an update -- would be nice not to charge at all */
1774 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1777 * Mark this transaction as typically resulting in a net free of space
1779 dmu_tx_mark_netfree(tx);
1781 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1783 zfs_dirent_unlock(dl);
1784 if (error == ERESTART) {
1804 * Remove the directory entry.
1806 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1815 * Hold z_lock so that we can make sure that the ACL obj
1816 * hasn't changed. Could have been deleted due to
1819 mutex_enter(&zp->z_lock);
1820 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1821 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1822 delete_now = may_delete_now && !toobig &&
1823 atomic_read(&ip->i_count) == 1 && !(zp->z_is_mapped) &&
1824 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1829 if (xattr_obj_unlinked) {
1830 ASSERT3U(ZTOI(xzp)->i_nlink, ==, 2);
1831 mutex_enter(&xzp->z_lock);
1832 xzp->z_unlinked = 1;
1833 clear_nlink(ZTOI(xzp));
1835 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1836 &links, sizeof (links), tx);
1837 ASSERT3U(error, ==, 0);
1838 mutex_exit(&xzp->z_lock);
1839 zfs_unlinked_add(xzp, tx);
1842 error = sa_remove(zp->z_sa_hdl,
1843 SA_ZPL_XATTR(zfsvfs), tx);
1845 error = sa_update(zp->z_sa_hdl,
1846 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1847 sizeof (uint64_t), tx);
1851 * Add to the unlinked set because a new reference could be
1852 * taken concurrently resulting in a deferred destruction.
1854 zfs_unlinked_add(zp, tx);
1855 mutex_exit(&zp->z_lock);
1856 } else if (unlinked) {
1857 mutex_exit(&zp->z_lock);
1858 zfs_unlinked_add(zp, tx);
1862 if (flags & FIGNORECASE)
1864 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1871 zfs_dirent_unlock(dl);
1872 zfs_inode_update(dzp);
1873 zfs_inode_update(zp);
1881 zfs_inode_update(xzp);
1882 zfs_iput_async(ZTOI(xzp));
1885 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1886 zil_commit(zilog, 0);
1891 EXPORT_SYMBOL(zfs_remove);
1894 * Create a new directory and insert it into dip using the name
1895 * provided. Return a pointer to the inserted directory.
1897 * IN: dip - inode of directory to add subdir to.
1898 * dirname - name of new directory.
1899 * vap - attributes of new directory.
1900 * cr - credentials of caller.
1901 * vsecp - ACL to be set
1903 * OUT: ipp - inode of created directory.
1905 * RETURN: 0 if success
1906 * error code if failure
1909 * dip - ctime|mtime updated
1910 * ipp - ctime|mtime|atime updated
1914 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1915 cred_t *cr, int flags, vsecattr_t *vsecp)
1917 znode_t *zp, *dzp = ITOZ(dip);
1918 zfsvfs_t *zfsvfs = ITOZSB(dip);
1926 gid_t gid = crgetgid(cr);
1927 zfs_acl_ids_t acl_ids;
1928 boolean_t fuid_dirtied;
1929 boolean_t waited = B_FALSE;
1931 ASSERT(S_ISDIR(vap->va_mode));
1934 * If we have an ephemeral id, ACL, or XVATTR then
1935 * make sure file system is at proper version
1939 if (zfsvfs->z_use_fuids == B_FALSE &&
1940 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1941 return (SET_ERROR(EINVAL));
1943 if (dirname == NULL)
1944 return (SET_ERROR(EINVAL));
1948 zilog = zfsvfs->z_log;
1950 if (dzp->z_pflags & ZFS_XATTR) {
1952 return (SET_ERROR(EINVAL));
1955 if (zfsvfs->z_utf8 && u8_validate(dirname,
1956 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1958 return (SET_ERROR(EILSEQ));
1960 if (flags & FIGNORECASE)
1963 if (vap->va_mask & ATTR_XVATTR) {
1964 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1965 crgetuid(cr), cr, vap->va_mode)) != 0) {
1971 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1972 vsecp, &acl_ids)) != 0) {
1977 * First make sure the new directory doesn't exist.
1979 * Existence is checked first to make sure we don't return
1980 * EACCES instead of EEXIST which can cause some applications
1986 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1988 zfs_acl_ids_free(&acl_ids);
1993 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
1994 zfs_acl_ids_free(&acl_ids);
1995 zfs_dirent_unlock(dl);
2000 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2001 zfs_acl_ids_free(&acl_ids);
2002 zfs_dirent_unlock(dl);
2004 return (SET_ERROR(EDQUOT));
2008 * Add a new entry to the directory.
2010 tx = dmu_tx_create(zfsvfs->z_os);
2011 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2012 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2013 fuid_dirtied = zfsvfs->z_fuid_dirty;
2015 zfs_fuid_txhold(zfsvfs, tx);
2016 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2017 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2018 acl_ids.z_aclp->z_acl_bytes);
2021 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2022 ZFS_SA_BASE_ATTR_SIZE);
2024 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2026 zfs_dirent_unlock(dl);
2027 if (error == ERESTART) {
2033 zfs_acl_ids_free(&acl_ids);
2042 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2045 zfs_fuid_sync(zfsvfs, tx);
2048 * Now put new name in parent dir.
2050 (void) zfs_link_create(dl, zp, tx, ZNEW);
2054 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2055 if (flags & FIGNORECASE)
2057 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2058 acl_ids.z_fuidp, vap);
2060 zfs_acl_ids_free(&acl_ids);
2064 zfs_dirent_unlock(dl);
2066 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2067 zil_commit(zilog, 0);
2069 zfs_inode_update(dzp);
2070 zfs_inode_update(zp);
2074 EXPORT_SYMBOL(zfs_mkdir);
2077 * Remove a directory subdir entry. If the current working
2078 * directory is the same as the subdir to be removed, the
2081 * IN: dip - inode of directory to remove from.
2082 * name - name of directory to be removed.
2083 * cwd - inode of current working directory.
2084 * cr - credentials of caller.
2085 * flags - case flags
2087 * RETURN: 0 on success, error code on failure.
2090 * dip - ctime|mtime updated
2094 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
2097 znode_t *dzp = ITOZ(dip);
2100 zfsvfs_t *zfsvfs = ITOZSB(dip);
2106 boolean_t waited = B_FALSE;
2109 return (SET_ERROR(EINVAL));
2113 zilog = zfsvfs->z_log;
2115 if (flags & FIGNORECASE)
2121 * Attempt to lock directory; fail if entry doesn't exist.
2123 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2131 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
2135 if (!S_ISDIR(ip->i_mode)) {
2136 error = SET_ERROR(ENOTDIR);
2141 error = SET_ERROR(EINVAL);
2146 * Grab a lock on the directory to make sure that no one is
2147 * trying to add (or lookup) entries while we are removing it.
2149 rw_enter(&zp->z_name_lock, RW_WRITER);
2152 * Grab a lock on the parent pointer to make sure we play well
2153 * with the treewalk and directory rename code.
2155 rw_enter(&zp->z_parent_lock, RW_WRITER);
2157 tx = dmu_tx_create(zfsvfs->z_os);
2158 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2159 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2160 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2161 zfs_sa_upgrade_txholds(tx, zp);
2162 zfs_sa_upgrade_txholds(tx, dzp);
2163 dmu_tx_mark_netfree(tx);
2164 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2166 rw_exit(&zp->z_parent_lock);
2167 rw_exit(&zp->z_name_lock);
2168 zfs_dirent_unlock(dl);
2169 if (error == ERESTART) {
2182 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2185 uint64_t txtype = TX_RMDIR;
2186 if (flags & FIGNORECASE)
2188 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2193 rw_exit(&zp->z_parent_lock);
2194 rw_exit(&zp->z_name_lock);
2196 zfs_dirent_unlock(dl);
2198 zfs_inode_update(dzp);
2199 zfs_inode_update(zp);
2202 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2203 zil_commit(zilog, 0);
2208 EXPORT_SYMBOL(zfs_rmdir);
2211 * Read as many directory entries as will fit into the provided
2212 * dirent buffer from the given directory cursor position.
2214 * IN: ip - inode of directory to read.
2215 * dirent - buffer for directory entries.
2217 * OUT: dirent - filler buffer of directory entries.
2219 * RETURN: 0 if success
2220 * error code if failure
2223 * ip - atime updated
2225 * Note that the low 4 bits of the cookie returned by zap is always zero.
2226 * This allows us to use the low range for "special" directory entries:
2227 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2228 * we use the offset 2 for the '.zfs' directory.
2232 zfs_readdir(struct inode *ip, struct dir_context *ctx, cred_t *cr)
2234 znode_t *zp = ITOZ(ip);
2235 zfsvfs_t *zfsvfs = ITOZSB(ip);
2238 zap_attribute_t zap;
2244 uint64_t offset; /* must be unsigned; checks for < 1 */
2249 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2250 &parent, sizeof (parent))) != 0)
2254 * Quit if directory has been removed (posix)
2262 prefetch = zp->z_zn_prefetch;
2265 * Initialize the iterator cursor.
2269 * Start iteration from the beginning of the directory.
2271 zap_cursor_init(&zc, os, zp->z_id);
2274 * The offset is a serialized cursor.
2276 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2280 * Transform to file-system independent format
2285 * Special case `.', `..', and `.zfs'.
2288 (void) strcpy(zap.za_name, ".");
2289 zap.za_normalization_conflict = 0;
2292 } else if (offset == 1) {
2293 (void) strcpy(zap.za_name, "..");
2294 zap.za_normalization_conflict = 0;
2297 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2298 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2299 zap.za_normalization_conflict = 0;
2300 objnum = ZFSCTL_INO_ROOT;
2306 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2307 if (error == ENOENT)
2314 * Allow multiple entries provided the first entry is
2315 * the object id. Non-zpl consumers may safely make
2316 * use of the additional space.
2318 * XXX: This should be a feature flag for compatibility
2320 if (zap.za_integer_length != 8 ||
2321 zap.za_num_integers == 0) {
2322 cmn_err(CE_WARN, "zap_readdir: bad directory "
2323 "entry, obj = %lld, offset = %lld, "
2324 "length = %d, num = %lld\n",
2325 (u_longlong_t)zp->z_id,
2326 (u_longlong_t)offset,
2327 zap.za_integer_length,
2328 (u_longlong_t)zap.za_num_integers);
2329 error = SET_ERROR(ENXIO);
2333 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2334 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2337 done = !dir_emit(ctx, zap.za_name, strlen(zap.za_name),
2342 /* Prefetch znode */
2344 dmu_prefetch(os, objnum, 0, 0, 0,
2345 ZIO_PRIORITY_SYNC_READ);
2349 * Move to the next entry, fill in the previous offset.
2351 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2352 zap_cursor_advance(&zc);
2353 offset = zap_cursor_serialize(&zc);
2359 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2362 zap_cursor_fini(&zc);
2363 if (error == ENOENT)
2370 EXPORT_SYMBOL(zfs_readdir);
2372 ulong_t zfs_fsync_sync_cnt = 4;
2375 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2377 znode_t *zp = ITOZ(ip);
2378 zfsvfs_t *zfsvfs = ITOZSB(ip);
2380 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2382 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2385 zil_commit(zfsvfs->z_log, zp->z_id);
2388 tsd_set(zfs_fsyncer_key, NULL);
2392 EXPORT_SYMBOL(zfs_fsync);
2396 * Get the requested file attributes and place them in the provided
2399 * IN: ip - inode of file.
2400 * vap - va_mask identifies requested attributes.
2401 * If ATTR_XVATTR set, then optional attrs are requested
2402 * flags - ATTR_NOACLCHECK (CIFS server context)
2403 * cr - credentials of caller.
2405 * OUT: vap - attribute values.
2407 * RETURN: 0 (always succeeds)
2411 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2413 znode_t *zp = ITOZ(ip);
2414 zfsvfs_t *zfsvfs = ITOZSB(ip);
2417 uint64_t atime[2], mtime[2], ctime[2];
2418 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2419 xoptattr_t *xoap = NULL;
2420 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2421 sa_bulk_attr_t bulk[3];
2427 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2429 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
2430 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2431 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2433 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2439 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2440 * Also, if we are the owner don't bother, since owner should
2441 * always be allowed to read basic attributes of file.
2443 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2444 (vap->va_uid != crgetuid(cr))) {
2445 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2453 * Return all attributes. It's cheaper to provide the answer
2454 * than to determine whether we were asked the question.
2457 mutex_enter(&zp->z_lock);
2458 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2459 vap->va_mode = zp->z_mode;
2460 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2461 vap->va_nodeid = zp->z_id;
2462 if ((zp->z_id == zfsvfs->z_root) && zfs_show_ctldir(zp))
2463 links = ZTOI(zp)->i_nlink + 1;
2465 links = ZTOI(zp)->i_nlink;
2466 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2467 vap->va_size = i_size_read(ip);
2468 vap->va_rdev = ip->i_rdev;
2469 vap->va_seq = ip->i_generation;
2472 * Add in any requested optional attributes and the create time.
2473 * Also set the corresponding bits in the returned attribute bitmap.
2475 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2476 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2478 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2479 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2482 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2483 xoap->xoa_readonly =
2484 ((zp->z_pflags & ZFS_READONLY) != 0);
2485 XVA_SET_RTN(xvap, XAT_READONLY);
2488 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2490 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2491 XVA_SET_RTN(xvap, XAT_SYSTEM);
2494 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2496 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2497 XVA_SET_RTN(xvap, XAT_HIDDEN);
2500 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2501 xoap->xoa_nounlink =
2502 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2503 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2506 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2507 xoap->xoa_immutable =
2508 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2509 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2512 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2513 xoap->xoa_appendonly =
2514 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2515 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2518 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2520 ((zp->z_pflags & ZFS_NODUMP) != 0);
2521 XVA_SET_RTN(xvap, XAT_NODUMP);
2524 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2526 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2527 XVA_SET_RTN(xvap, XAT_OPAQUE);
2530 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2531 xoap->xoa_av_quarantined =
2532 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2533 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2536 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2537 xoap->xoa_av_modified =
2538 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2539 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2542 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2543 S_ISREG(ip->i_mode)) {
2544 zfs_sa_get_scanstamp(zp, xvap);
2547 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2550 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2551 times, sizeof (times));
2552 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2553 XVA_SET_RTN(xvap, XAT_CREATETIME);
2556 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2557 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2558 XVA_SET_RTN(xvap, XAT_REPARSE);
2560 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2561 xoap->xoa_generation = ip->i_generation;
2562 XVA_SET_RTN(xvap, XAT_GEN);
2565 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2567 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2568 XVA_SET_RTN(xvap, XAT_OFFLINE);
2571 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2573 ((zp->z_pflags & ZFS_SPARSE) != 0);
2574 XVA_SET_RTN(xvap, XAT_SPARSE);
2578 ZFS_TIME_DECODE(&vap->va_atime, atime);
2579 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2580 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2582 mutex_exit(&zp->z_lock);
2584 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2586 if (zp->z_blksz == 0) {
2588 * Block size hasn't been set; suggest maximal I/O transfers.
2590 vap->va_blksize = zfsvfs->z_max_blksz;
2596 EXPORT_SYMBOL(zfs_getattr);
2599 * Get the basic file attributes and place them in the provided kstat
2600 * structure. The inode is assumed to be the authoritative source
2601 * for most of the attributes. However, the znode currently has the
2602 * authoritative atime, blksize, and block count.
2604 * IN: ip - inode of file.
2606 * OUT: sp - kstat values.
2608 * RETURN: 0 (always succeeds)
2612 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2614 znode_t *zp = ITOZ(ip);
2615 zfsvfs_t *zfsvfs = ITOZSB(ip);
2617 u_longlong_t nblocks;
2622 mutex_enter(&zp->z_lock);
2624 generic_fillattr(ip, sp);
2626 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2627 sp->blksize = blksize;
2628 sp->blocks = nblocks;
2630 if (unlikely(zp->z_blksz == 0)) {
2632 * Block size hasn't been set; suggest maximal I/O transfers.
2634 sp->blksize = zfsvfs->z_max_blksz;
2637 mutex_exit(&zp->z_lock);
2640 * Required to prevent NFS client from detecting different inode
2641 * numbers of snapshot root dentry before and after snapshot mount.
2643 if (zfsvfs->z_issnap) {
2644 if (ip->i_sb->s_root->d_inode == ip)
2645 sp->ino = ZFSCTL_INO_SNAPDIRS -
2646 dmu_objset_id(zfsvfs->z_os);
2653 EXPORT_SYMBOL(zfs_getattr_fast);
2656 * Set the file attributes to the values contained in the
2659 * IN: ip - inode of file to be modified.
2660 * vap - new attribute values.
2661 * If ATTR_XVATTR set, then optional attrs are being set
2662 * flags - ATTR_UTIME set if non-default time values provided.
2663 * - ATTR_NOACLCHECK (CIFS context only).
2664 * cr - credentials of caller.
2666 * RETURN: 0 if success
2667 * error code if failure
2670 * ip - ctime updated, mtime updated if size changed.
2674 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2676 znode_t *zp = ITOZ(ip);
2677 zfsvfs_t *zfsvfs = ITOZSB(ip);
2681 xvattr_t *tmpxvattr;
2682 uint_t mask = vap->va_mask;
2683 uint_t saved_mask = 0;
2686 uint64_t new_kuid = 0, new_kgid = 0, new_uid, new_gid;
2688 uint64_t mtime[2], ctime[2], atime[2];
2690 int need_policy = FALSE;
2692 zfs_fuid_info_t *fuidp = NULL;
2693 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2696 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2697 boolean_t fuid_dirtied = B_FALSE;
2698 sa_bulk_attr_t *bulk, *xattr_bulk;
2699 int count = 0, xattr_count = 0;
2707 zilog = zfsvfs->z_log;
2710 * Make sure that if we have ephemeral uid/gid or xvattr specified
2711 * that file system is at proper version level
2714 if (zfsvfs->z_use_fuids == B_FALSE &&
2715 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2716 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2717 (mask & ATTR_XVATTR))) {
2719 return (SET_ERROR(EINVAL));
2722 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2724 return (SET_ERROR(EISDIR));
2727 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2729 return (SET_ERROR(EINVAL));
2733 * If this is an xvattr_t, then get a pointer to the structure of
2734 * optional attributes. If this is NULL, then we have a vattr_t.
2736 xoap = xva_getxoptattr(xvap);
2738 tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP);
2739 xva_init(tmpxvattr);
2741 bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2742 xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2745 * Immutable files can only alter immutable bit and atime
2747 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2748 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2749 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2754 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2760 * Verify timestamps doesn't overflow 32 bits.
2761 * ZFS can handle large timestamps, but 32bit syscalls can't
2762 * handle times greater than 2039. This check should be removed
2763 * once large timestamps are fully supported.
2765 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2766 if (((mask & ATTR_ATIME) &&
2767 TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2768 ((mask & ATTR_MTIME) &&
2769 TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2779 /* Can this be moved to before the top label? */
2780 if (zfs_is_readonly(zfsvfs)) {
2786 * First validate permissions
2789 if (mask & ATTR_SIZE) {
2790 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2795 * XXX - Note, we are not providing any open
2796 * mode flags here (like FNDELAY), so we may
2797 * block if there are locks present... this
2798 * should be addressed in openat().
2800 /* XXX - would it be OK to generate a log record here? */
2801 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2806 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2807 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2808 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2809 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2810 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2811 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2812 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2813 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2814 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2818 if (mask & (ATTR_UID|ATTR_GID)) {
2819 int idmask = (mask & (ATTR_UID|ATTR_GID));
2824 * NOTE: even if a new mode is being set,
2825 * we may clear S_ISUID/S_ISGID bits.
2828 if (!(mask & ATTR_MODE))
2829 vap->va_mode = zp->z_mode;
2832 * Take ownership or chgrp to group we are a member of
2835 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
2836 take_group = (mask & ATTR_GID) &&
2837 zfs_groupmember(zfsvfs, vap->va_gid, cr);
2840 * If both ATTR_UID and ATTR_GID are set then take_owner and
2841 * take_group must both be set in order to allow taking
2844 * Otherwise, send the check through secpolicy_vnode_setattr()
2848 if (((idmask == (ATTR_UID|ATTR_GID)) &&
2849 take_owner && take_group) ||
2850 ((idmask == ATTR_UID) && take_owner) ||
2851 ((idmask == ATTR_GID) && take_group)) {
2852 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2853 skipaclchk, cr) == 0) {
2855 * Remove setuid/setgid for non-privileged users
2857 (void) secpolicy_setid_clear(vap, cr);
2858 trim_mask = (mask & (ATTR_UID|ATTR_GID));
2867 mutex_enter(&zp->z_lock);
2868 oldva.va_mode = zp->z_mode;
2869 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2870 if (mask & ATTR_XVATTR) {
2872 * Update xvattr mask to include only those attributes
2873 * that are actually changing.
2875 * the bits will be restored prior to actually setting
2876 * the attributes so the caller thinks they were set.
2878 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2879 if (xoap->xoa_appendonly !=
2880 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2883 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2884 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
2888 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2889 if (xoap->xoa_nounlink !=
2890 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2893 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2894 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
2898 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2899 if (xoap->xoa_immutable !=
2900 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2903 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2904 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
2908 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2909 if (xoap->xoa_nodump !=
2910 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2913 XVA_CLR_REQ(xvap, XAT_NODUMP);
2914 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
2918 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2919 if (xoap->xoa_av_modified !=
2920 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2923 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2924 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
2928 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2929 if ((!S_ISREG(ip->i_mode) &&
2930 xoap->xoa_av_quarantined) ||
2931 xoap->xoa_av_quarantined !=
2932 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2935 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2936 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
2940 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2941 mutex_exit(&zp->z_lock);
2946 if (need_policy == FALSE &&
2947 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2948 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2953 mutex_exit(&zp->z_lock);
2955 if (mask & ATTR_MODE) {
2956 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2957 err = secpolicy_setid_setsticky_clear(ip, vap,
2962 trim_mask |= ATTR_MODE;
2970 * If trim_mask is set then take ownership
2971 * has been granted or write_acl is present and user
2972 * has the ability to modify mode. In that case remove
2973 * UID|GID and or MODE from mask so that
2974 * secpolicy_vnode_setattr() doesn't revoke it.
2978 saved_mask = vap->va_mask;
2979 vap->va_mask &= ~trim_mask;
2981 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
2982 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2987 vap->va_mask |= saved_mask;
2991 * secpolicy_vnode_setattr, or take ownership may have
2994 mask = vap->va_mask;
2996 if ((mask & (ATTR_UID | ATTR_GID))) {
2997 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2998 &xattr_obj, sizeof (xattr_obj));
3000 if (err == 0 && xattr_obj) {
3001 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
3005 if (mask & ATTR_UID) {
3006 new_kuid = zfs_fuid_create(zfsvfs,
3007 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3008 if (new_kuid != KUID_TO_SUID(ZTOI(zp)->i_uid) &&
3009 zfs_fuid_overquota(zfsvfs, B_FALSE, new_kuid)) {
3017 if (mask & ATTR_GID) {
3018 new_kgid = zfs_fuid_create(zfsvfs,
3019 (uint64_t)vap->va_gid, cr, ZFS_GROUP, &fuidp);
3020 if (new_kgid != KGID_TO_SGID(ZTOI(zp)->i_gid) &&
3021 zfs_fuid_overquota(zfsvfs, B_TRUE, new_kgid)) {
3029 tx = dmu_tx_create(zfsvfs->z_os);
3031 if (mask & ATTR_MODE) {
3032 uint64_t pmode = zp->z_mode;
3034 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3036 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
3038 mutex_enter(&zp->z_lock);
3039 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3041 * Are we upgrading ACL from old V0 format
3044 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3045 zfs_znode_acl_version(zp) ==
3046 ZFS_ACL_VERSION_INITIAL) {
3047 dmu_tx_hold_free(tx, acl_obj, 0,
3049 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3050 0, aclp->z_acl_bytes);
3052 dmu_tx_hold_write(tx, acl_obj, 0,
3055 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3056 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3057 0, aclp->z_acl_bytes);
3059 mutex_exit(&zp->z_lock);
3060 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3062 if ((mask & ATTR_XVATTR) &&
3063 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3064 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3066 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3070 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3073 fuid_dirtied = zfsvfs->z_fuid_dirty;
3075 zfs_fuid_txhold(zfsvfs, tx);
3077 zfs_sa_upgrade_txholds(tx, zp);
3079 err = dmu_tx_assign(tx, TXG_WAIT);
3085 * Set each attribute requested.
3086 * We group settings according to the locks they need to acquire.
3088 * Note: you cannot set ctime directly, although it will be
3089 * updated as a side-effect of calling this function.
3093 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3094 mutex_enter(&zp->z_acl_lock);
3095 mutex_enter(&zp->z_lock);
3097 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3098 &zp->z_pflags, sizeof (zp->z_pflags));
3101 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3102 mutex_enter(&attrzp->z_acl_lock);
3103 mutex_enter(&attrzp->z_lock);
3104 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3105 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3106 sizeof (attrzp->z_pflags));
3109 if (mask & (ATTR_UID|ATTR_GID)) {
3111 if (mask & ATTR_UID) {
3112 ZTOI(zp)->i_uid = SUID_TO_KUID(new_kuid);
3113 new_uid = zfs_uid_read(ZTOI(zp));
3114 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3115 &new_uid, sizeof (new_uid));
3117 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3118 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3120 ZTOI(attrzp)->i_uid = SUID_TO_KUID(new_uid);
3124 if (mask & ATTR_GID) {
3125 ZTOI(zp)->i_gid = SGID_TO_KGID(new_kgid);
3126 new_gid = zfs_gid_read(ZTOI(zp));
3127 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3128 NULL, &new_gid, sizeof (new_gid));
3130 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3131 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3133 ZTOI(attrzp)->i_gid = SGID_TO_KGID(new_kgid);
3136 if (!(mask & ATTR_MODE)) {
3137 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3138 NULL, &new_mode, sizeof (new_mode));
3139 new_mode = zp->z_mode;
3141 err = zfs_acl_chown_setattr(zp);
3144 err = zfs_acl_chown_setattr(attrzp);
3149 if (mask & ATTR_MODE) {
3150 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3151 &new_mode, sizeof (new_mode));
3152 zp->z_mode = ZTOI(zp)->i_mode = new_mode;
3153 ASSERT3P(aclp, !=, NULL);
3154 err = zfs_aclset_common(zp, aclp, cr, tx);
3156 if (zp->z_acl_cached)
3157 zfs_acl_free(zp->z_acl_cached);
3158 zp->z_acl_cached = aclp;
3162 if ((mask & ATTR_ATIME) || zp->z_atime_dirty) {
3163 zp->z_atime_dirty = 0;
3164 ZFS_TIME_ENCODE(&ip->i_atime, atime);
3165 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3166 &atime, sizeof (atime));
3169 if (mask & ATTR_MTIME) {
3170 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3171 ZTOI(zp)->i_mtime = timespec_trunc(vap->va_mtime,
3172 ZTOI(zp)->i_sb->s_time_gran);
3174 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3175 mtime, sizeof (mtime));
3178 if (mask & ATTR_CTIME) {
3179 ZFS_TIME_ENCODE(&vap->va_ctime, ctime);
3180 ZTOI(zp)->i_ctime = timespec_trunc(vap->va_ctime,
3181 ZTOI(zp)->i_sb->s_time_gran);
3182 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3183 ctime, sizeof (ctime));
3186 if (attrzp && mask) {
3187 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3188 SA_ZPL_CTIME(zfsvfs), NULL, &ctime,
3193 * Do this after setting timestamps to prevent timestamp
3194 * update from toggling bit
3197 if (xoap && (mask & ATTR_XVATTR)) {
3200 * restore trimmed off masks
3201 * so that return masks can be set for caller.
3204 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
3205 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3207 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
3208 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3210 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
3211 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3213 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
3214 XVA_SET_REQ(xvap, XAT_NODUMP);
3216 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
3217 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3219 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
3220 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3223 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3224 ASSERT(S_ISREG(ip->i_mode));
3226 zfs_xvattr_set(zp, xvap, tx);
3230 zfs_fuid_sync(zfsvfs, tx);
3233 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3235 mutex_exit(&zp->z_lock);
3236 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3237 mutex_exit(&zp->z_acl_lock);
3240 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3241 mutex_exit(&attrzp->z_acl_lock);
3242 mutex_exit(&attrzp->z_lock);
3245 if (err == 0 && attrzp) {
3246 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3255 zfs_fuid_info_free(fuidp);
3263 if (err == ERESTART)
3266 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3270 zfs_inode_update(zp);
3274 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3275 zil_commit(zilog, 0);
3278 kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * 7);
3279 kmem_free(bulk, sizeof (sa_bulk_attr_t) * 7);
3280 kmem_free(tmpxvattr, sizeof (xvattr_t));
3284 EXPORT_SYMBOL(zfs_setattr);
3286 typedef struct zfs_zlock {
3287 krwlock_t *zl_rwlock; /* lock we acquired */
3288 znode_t *zl_znode; /* znode we held */
3289 struct zfs_zlock *zl_next; /* next in list */
3293 * Drop locks and release vnodes that were held by zfs_rename_lock().
3296 zfs_rename_unlock(zfs_zlock_t **zlpp)
3300 while ((zl = *zlpp) != NULL) {
3301 if (zl->zl_znode != NULL)
3302 zfs_iput_async(ZTOI(zl->zl_znode));
3303 rw_exit(zl->zl_rwlock);
3304 *zlpp = zl->zl_next;
3305 kmem_free(zl, sizeof (*zl));
3310 * Search back through the directory tree, using the ".." entries.
3311 * Lock each directory in the chain to prevent concurrent renames.
3312 * Fail any attempt to move a directory into one of its own descendants.
3313 * XXX - z_parent_lock can overlap with map or grow locks
3316 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3320 uint64_t rootid = ZTOZSB(zp)->z_root;
3321 uint64_t oidp = zp->z_id;
3322 krwlock_t *rwlp = &szp->z_parent_lock;
3323 krw_t rw = RW_WRITER;
3326 * First pass write-locks szp and compares to zp->z_id.
3327 * Later passes read-lock zp and compare to zp->z_parent.
3330 if (!rw_tryenter(rwlp, rw)) {
3332 * Another thread is renaming in this path.
3333 * Note that if we are a WRITER, we don't have any
3334 * parent_locks held yet.
3336 if (rw == RW_READER && zp->z_id > szp->z_id) {
3338 * Drop our locks and restart
3340 zfs_rename_unlock(&zl);
3344 rwlp = &szp->z_parent_lock;
3349 * Wait for other thread to drop its locks
3355 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3356 zl->zl_rwlock = rwlp;
3357 zl->zl_znode = NULL;
3358 zl->zl_next = *zlpp;
3361 if (oidp == szp->z_id) /* We're a descendant of szp */
3362 return (SET_ERROR(EINVAL));
3364 if (oidp == rootid) /* We've hit the top */
3367 if (rw == RW_READER) { /* i.e. not the first pass */
3368 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3373 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3374 &oidp, sizeof (oidp));
3375 rwlp = &zp->z_parent_lock;
3378 } while (zp->z_id != sdzp->z_id);
3384 * Move an entry from the provided source directory to the target
3385 * directory. Change the entry name as indicated.
3387 * IN: sdip - Source directory containing the "old entry".
3388 * snm - Old entry name.
3389 * tdip - Target directory to contain the "new entry".
3390 * tnm - New entry name.
3391 * cr - credentials of caller.
3392 * flags - case flags
3394 * RETURN: 0 on success, error code on failure.
3397 * sdip,tdip - ctime|mtime updated
3401 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3402 cred_t *cr, int flags)
3404 znode_t *tdzp, *szp, *tzp;
3405 znode_t *sdzp = ITOZ(sdip);
3406 zfsvfs_t *zfsvfs = ITOZSB(sdip);
3408 zfs_dirlock_t *sdl, *tdl;
3411 int cmp, serr, terr;
3414 boolean_t waited = B_FALSE;
3416 if (snm == NULL || tnm == NULL)
3417 return (SET_ERROR(EINVAL));
3420 ZFS_VERIFY_ZP(sdzp);
3421 zilog = zfsvfs->z_log;
3424 ZFS_VERIFY_ZP(tdzp);
3427 * We check i_sb because snapshots and the ctldir must have different
3430 if (tdip->i_sb != sdip->i_sb || zfsctl_is_node(tdip)) {
3432 return (SET_ERROR(EXDEV));
3435 if (zfsvfs->z_utf8 && u8_validate(tnm,
3436 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3438 return (SET_ERROR(EILSEQ));
3441 if (flags & FIGNORECASE)
3450 * This is to prevent the creation of links into attribute space
3451 * by renaming a linked file into/outof an attribute directory.
3452 * See the comment in zfs_link() for why this is considered bad.
3454 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3456 return (SET_ERROR(EINVAL));
3460 * Lock source and target directory entries. To prevent deadlock,
3461 * a lock ordering must be defined. We lock the directory with
3462 * the smallest object id first, or if it's a tie, the one with
3463 * the lexically first name.
3465 if (sdzp->z_id < tdzp->z_id) {
3467 } else if (sdzp->z_id > tdzp->z_id) {
3471 * First compare the two name arguments without
3472 * considering any case folding.
3474 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3476 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3477 ASSERT(error == 0 || !zfsvfs->z_utf8);
3480 * POSIX: "If the old argument and the new argument
3481 * both refer to links to the same existing file,
3482 * the rename() function shall return successfully
3483 * and perform no other action."
3489 * If the file system is case-folding, then we may
3490 * have some more checking to do. A case-folding file
3491 * system is either supporting mixed case sensitivity
3492 * access or is completely case-insensitive. Note
3493 * that the file system is always case preserving.
3495 * In mixed sensitivity mode case sensitive behavior
3496 * is the default. FIGNORECASE must be used to
3497 * explicitly request case insensitive behavior.
3499 * If the source and target names provided differ only
3500 * by case (e.g., a request to rename 'tim' to 'Tim'),
3501 * we will treat this as a special case in the
3502 * case-insensitive mode: as long as the source name
3503 * is an exact match, we will allow this to proceed as
3504 * a name-change request.
3506 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3507 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3508 flags & FIGNORECASE)) &&
3509 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3512 * case preserving rename request, require exact
3521 * If the source and destination directories are the same, we should
3522 * grab the z_name_lock of that directory only once.
3526 rw_enter(&sdzp->z_name_lock, RW_READER);
3530 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3531 ZEXISTS | zflg, NULL, NULL);
3532 terr = zfs_dirent_lock(&tdl,
3533 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3535 terr = zfs_dirent_lock(&tdl,
3536 tdzp, tnm, &tzp, zflg, NULL, NULL);
3537 serr = zfs_dirent_lock(&sdl,
3538 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3544 * Source entry invalid or not there.
3547 zfs_dirent_unlock(tdl);
3553 rw_exit(&sdzp->z_name_lock);
3555 if (strcmp(snm, "..") == 0)
3561 zfs_dirent_unlock(sdl);
3565 rw_exit(&sdzp->z_name_lock);
3567 if (strcmp(tnm, "..") == 0)
3574 * Must have write access at the source to remove the old entry
3575 * and write access at the target to create the new entry.
3576 * Note that if target and source are the same, this can be
3577 * done in a single check.
3580 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3583 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3585 * Check to make sure rename is valid.
3586 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3588 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3593 * Does target exist?
3597 * Source and target must be the same type.
3599 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3600 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3601 error = SET_ERROR(ENOTDIR);
3605 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3606 error = SET_ERROR(EISDIR);
3611 * POSIX dictates that when the source and target
3612 * entries refer to the same file object, rename
3613 * must do nothing and exit without error.
3615 if (szp->z_id == tzp->z_id) {
3621 tx = dmu_tx_create(zfsvfs->z_os);
3622 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3623 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3624 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3625 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3627 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3628 zfs_sa_upgrade_txholds(tx, tdzp);
3631 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3632 zfs_sa_upgrade_txholds(tx, tzp);
3635 zfs_sa_upgrade_txholds(tx, szp);
3636 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3637 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3640 zfs_rename_unlock(&zl);
3641 zfs_dirent_unlock(sdl);
3642 zfs_dirent_unlock(tdl);
3645 rw_exit(&sdzp->z_name_lock);
3647 if (error == ERESTART) {
3664 if (tzp) /* Attempt to remove the existing target */
3665 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3668 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3670 szp->z_pflags |= ZFS_AV_MODIFIED;
3672 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3673 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3676 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3678 zfs_log_rename(zilog, tx, TX_RENAME |
3679 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3680 sdl->dl_name, tdzp, tdl->dl_name, szp);
3683 * At this point, we have successfully created
3684 * the target name, but have failed to remove
3685 * the source name. Since the create was done
3686 * with the ZRENAMING flag, there are
3687 * complications; for one, the link count is
3688 * wrong. The easiest way to deal with this
3689 * is to remove the newly created target, and
3690 * return the original error. This must
3691 * succeed; fortunately, it is very unlikely to
3692 * fail, since we just created it.
3694 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3695 ZRENAMING, NULL), ==, 0);
3703 zfs_rename_unlock(&zl);
3705 zfs_dirent_unlock(sdl);
3706 zfs_dirent_unlock(tdl);
3708 zfs_inode_update(sdzp);
3710 rw_exit(&sdzp->z_name_lock);
3713 zfs_inode_update(tdzp);
3715 zfs_inode_update(szp);
3718 zfs_inode_update(tzp);
3722 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3723 zil_commit(zilog, 0);
3728 EXPORT_SYMBOL(zfs_rename);
3731 * Insert the indicated symbolic reference entry into the directory.
3733 * IN: dip - Directory to contain new symbolic link.
3734 * link - Name for new symlink entry.
3735 * vap - Attributes of new entry.
3736 * target - Target path of new symlink.
3738 * cr - credentials of caller.
3739 * flags - case flags
3741 * RETURN: 0 on success, error code on failure.
3744 * dip - ctime|mtime updated
3748 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
3749 struct inode **ipp, cred_t *cr, int flags)
3751 znode_t *zp, *dzp = ITOZ(dip);
3754 zfsvfs_t *zfsvfs = ITOZSB(dip);
3756 uint64_t len = strlen(link);
3759 zfs_acl_ids_t acl_ids;
3760 boolean_t fuid_dirtied;
3761 uint64_t txtype = TX_SYMLINK;
3762 boolean_t waited = B_FALSE;
3764 ASSERT(S_ISLNK(vap->va_mode));
3767 return (SET_ERROR(EINVAL));
3771 zilog = zfsvfs->z_log;
3773 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3774 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3776 return (SET_ERROR(EILSEQ));
3778 if (flags & FIGNORECASE)
3781 if (len > MAXPATHLEN) {
3783 return (SET_ERROR(ENAMETOOLONG));
3786 if ((error = zfs_acl_ids_create(dzp, 0,
3787 vap, cr, NULL, &acl_ids)) != 0) {
3795 * Attempt to lock directory; fail if entry already exists.
3797 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3799 zfs_acl_ids_free(&acl_ids);
3804 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3805 zfs_acl_ids_free(&acl_ids);
3806 zfs_dirent_unlock(dl);
3811 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3812 zfs_acl_ids_free(&acl_ids);
3813 zfs_dirent_unlock(dl);
3815 return (SET_ERROR(EDQUOT));
3817 tx = dmu_tx_create(zfsvfs->z_os);
3818 fuid_dirtied = zfsvfs->z_fuid_dirty;
3819 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3820 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3821 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3822 ZFS_SA_BASE_ATTR_SIZE + len);
3823 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3824 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3825 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3826 acl_ids.z_aclp->z_acl_bytes);
3829 zfs_fuid_txhold(zfsvfs, tx);
3830 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3832 zfs_dirent_unlock(dl);
3833 if (error == ERESTART) {
3839 zfs_acl_ids_free(&acl_ids);
3846 * Create a new object for the symlink.
3847 * for version 4 ZPL datsets the symlink will be an SA attribute
3849 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3852 zfs_fuid_sync(zfsvfs, tx);
3854 mutex_enter(&zp->z_lock);
3856 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3859 zfs_sa_symlink(zp, link, len, tx);
3860 mutex_exit(&zp->z_lock);
3863 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3864 &zp->z_size, sizeof (zp->z_size), tx);
3866 * Insert the new object into the directory.
3868 (void) zfs_link_create(dl, zp, tx, ZNEW);
3870 if (flags & FIGNORECASE)
3872 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3874 zfs_inode_update(dzp);
3875 zfs_inode_update(zp);
3877 zfs_acl_ids_free(&acl_ids);
3881 zfs_dirent_unlock(dl);
3885 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3886 zil_commit(zilog, 0);
3891 EXPORT_SYMBOL(zfs_symlink);
3894 * Return, in the buffer contained in the provided uio structure,
3895 * the symbolic path referred to by ip.
3897 * IN: ip - inode of symbolic link
3898 * uio - structure to contain the link path.
3899 * cr - credentials of caller.
3901 * RETURN: 0 if success
3902 * error code if failure
3905 * ip - atime updated
3909 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
3911 znode_t *zp = ITOZ(ip);
3912 zfsvfs_t *zfsvfs = ITOZSB(ip);
3918 mutex_enter(&zp->z_lock);
3920 error = sa_lookup_uio(zp->z_sa_hdl,
3921 SA_ZPL_SYMLINK(zfsvfs), uio);
3923 error = zfs_sa_readlink(zp, uio);
3924 mutex_exit(&zp->z_lock);
3929 EXPORT_SYMBOL(zfs_readlink);
3932 * Insert a new entry into directory tdip referencing sip.
3934 * IN: tdip - Directory to contain new entry.
3935 * sip - inode of new entry.
3936 * name - name of new entry.
3937 * cr - credentials of caller.
3939 * RETURN: 0 if success
3940 * error code if failure
3943 * tdip - ctime|mtime updated
3944 * sip - ctime updated
3948 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr,
3951 znode_t *dzp = ITOZ(tdip);
3953 zfsvfs_t *zfsvfs = ITOZSB(tdip);
3961 boolean_t waited = B_FALSE;
3962 boolean_t is_tmpfile = 0;
3965 is_tmpfile = (sip->i_nlink == 0 && (sip->i_state & I_LINKABLE));
3967 ASSERT(S_ISDIR(tdip->i_mode));
3970 return (SET_ERROR(EINVAL));
3974 zilog = zfsvfs->z_log;
3977 * POSIX dictates that we return EPERM here.
3978 * Better choices include ENOTSUP or EISDIR.
3980 if (S_ISDIR(sip->i_mode)) {
3982 return (SET_ERROR(EPERM));
3989 * We check i_sb because snapshots and the ctldir must have different
3992 if (sip->i_sb != tdip->i_sb || zfsctl_is_node(sip)) {
3994 return (SET_ERROR(EXDEV));
3997 /* Prevent links to .zfs/shares files */
3999 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4000 &parent, sizeof (uint64_t))) != 0) {
4004 if (parent == zfsvfs->z_shares_dir) {
4006 return (SET_ERROR(EPERM));
4009 if (zfsvfs->z_utf8 && u8_validate(name,
4010 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4012 return (SET_ERROR(EILSEQ));
4014 if (flags & FIGNORECASE)
4018 * We do not support links between attributes and non-attributes
4019 * because of the potential security risk of creating links
4020 * into "normal" file space in order to circumvent restrictions
4021 * imposed in attribute space.
4023 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4025 return (SET_ERROR(EINVAL));
4028 owner = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(sip->i_uid),
4030 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4032 return (SET_ERROR(EPERM));
4035 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
4042 * Attempt to lock directory; fail if entry already exists.
4044 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4050 tx = dmu_tx_create(zfsvfs->z_os);
4051 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4052 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4054 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
4056 zfs_sa_upgrade_txholds(tx, szp);
4057 zfs_sa_upgrade_txholds(tx, dzp);
4058 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4060 zfs_dirent_unlock(dl);
4061 if (error == ERESTART) {
4071 /* unmark z_unlinked so zfs_link_create will not reject */
4073 szp->z_unlinked = 0;
4074 error = zfs_link_create(dl, szp, tx, 0);
4077 uint64_t txtype = TX_LINK;
4079 * tmpfile is created to be in z_unlinkedobj, so remove it.
4080 * Also, we don't log in ZIL, be cause all previous file
4081 * operation on the tmpfile are ignored by ZIL. Instead we
4082 * always wait for txg to sync to make sure all previous
4083 * operation are sync safe.
4086 VERIFY(zap_remove_int(zfsvfs->z_os,
4087 zfsvfs->z_unlinkedobj, szp->z_id, tx) == 0);
4089 if (flags & FIGNORECASE)
4091 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4093 } else if (is_tmpfile) {
4094 /* restore z_unlinked since when linking failed */
4095 szp->z_unlinked = 1;
4097 txg = dmu_tx_get_txg(tx);
4100 zfs_dirent_unlock(dl);
4102 if (!is_tmpfile && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4103 zil_commit(zilog, 0);
4106 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), txg);
4108 zfs_inode_update(dzp);
4109 zfs_inode_update(szp);
4113 EXPORT_SYMBOL(zfs_link);
4116 zfs_putpage_commit_cb(void *arg)
4118 struct page *pp = arg;
4121 end_page_writeback(pp);
4125 * Push a page out to disk, once the page is on stable storage the
4126 * registered commit callback will be run as notification of completion.
4128 * IN: ip - page mapped for inode.
4129 * pp - page to push (page is locked)
4130 * wbc - writeback control data
4132 * RETURN: 0 if success
4133 * error code if failure
4136 * ip - ctime|mtime updated
4140 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
4142 znode_t *zp = ITOZ(ip);
4143 zfsvfs_t *zfsvfs = ITOZSB(ip);
4151 uint64_t mtime[2], ctime[2];
4152 sa_bulk_attr_t bulk[3];
4154 struct address_space *mapping;
4159 ASSERT(PageLocked(pp));
4161 pgoff = page_offset(pp); /* Page byte-offset in file */
4162 offset = i_size_read(ip); /* File length in bytes */
4163 pglen = MIN(PAGE_SIZE, /* Page length in bytes */
4164 P2ROUNDUP(offset, PAGE_SIZE)-pgoff);
4166 /* Page is beyond end of file */
4167 if (pgoff >= offset) {
4173 /* Truncate page length to end of file */
4174 if (pgoff + pglen > offset)
4175 pglen = offset - pgoff;
4179 * FIXME: Allow mmap writes past its quota. The correct fix
4180 * is to register a page_mkwrite() handler to count the page
4181 * against its quota when it is about to be dirtied.
4183 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4184 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4190 * The ordering here is critical and must adhere to the following
4191 * rules in order to avoid deadlocking in either zfs_read() or
4192 * zfs_free_range() due to a lock inversion.
4194 * 1) The page must be unlocked prior to acquiring the range lock.
4195 * This is critical because zfs_read() calls find_lock_page()
4196 * which may block on the page lock while holding the range lock.
4198 * 2) Before setting or clearing write back on a page the range lock
4199 * must be held in order to prevent a lock inversion with the
4200 * zfs_free_range() function.
4202 * This presents a problem because upon entering this function the
4203 * page lock is already held. To safely acquire the range lock the
4204 * page lock must be dropped. This creates a window where another
4205 * process could truncate, invalidate, dirty, or write out the page.
4207 * Therefore, after successfully reacquiring the range and page locks
4208 * the current page state is checked. In the common case everything
4209 * will be as is expected and it can be written out. However, if
4210 * the page state has changed it must be handled accordingly.
4212 mapping = pp->mapping;
4213 redirty_page_for_writepage(wbc, pp);
4216 rl = zfs_range_lock(&zp->z_range_lock, pgoff, pglen, RL_WRITER);
4219 /* Page mapping changed or it was no longer dirty, we're done */
4220 if (unlikely((mapping != pp->mapping) || !PageDirty(pp))) {
4222 zfs_range_unlock(rl);
4227 /* Another process started write block if required */
4228 if (PageWriteback(pp)) {
4230 zfs_range_unlock(rl);
4232 if (wbc->sync_mode != WB_SYNC_NONE)
4233 wait_on_page_writeback(pp);
4239 /* Clear the dirty flag the required locks are held */
4240 if (!clear_page_dirty_for_io(pp)) {
4242 zfs_range_unlock(rl);
4248 * Counterpart for redirty_page_for_writepage() above. This page
4249 * was in fact not skipped and should not be counted as if it were.
4251 wbc->pages_skipped--;
4252 set_page_writeback(pp);
4255 tx = dmu_tx_create(zfsvfs->z_os);
4256 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
4257 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4258 zfs_sa_upgrade_txholds(tx, zp);
4260 err = dmu_tx_assign(tx, TXG_NOWAIT);
4262 if (err == ERESTART)
4266 __set_page_dirty_nobuffers(pp);
4268 end_page_writeback(pp);
4269 zfs_range_unlock(rl);
4275 ASSERT3U(pglen, <=, PAGE_SIZE);
4276 dmu_write(zfsvfs->z_os, zp->z_id, pgoff, pglen, va, tx);
4279 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
4280 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
4281 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zfsvfs), NULL,
4284 /* Preserve the mtime and ctime provided by the inode */
4285 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4286 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4287 zp->z_atime_dirty = 0;
4290 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4292 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0,
4293 zfs_putpage_commit_cb, pp);
4296 zfs_range_unlock(rl);
4298 if (wbc->sync_mode != WB_SYNC_NONE) {
4300 * Note that this is rarely called under writepages(), because
4301 * writepages() normally handles the entire commit for
4302 * performance reasons.
4304 zil_commit(zfsvfs->z_log, zp->z_id);
4312 * Update the system attributes when the inode has been dirtied. For the
4313 * moment we only update the mode, atime, mtime, and ctime.
4316 zfs_dirty_inode(struct inode *ip, int flags)
4318 znode_t *zp = ITOZ(ip);
4319 zfsvfs_t *zfsvfs = ITOZSB(ip);
4321 uint64_t mode, atime[2], mtime[2], ctime[2];
4322 sa_bulk_attr_t bulk[4];
4326 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
4334 * This is the lazytime semantic indroduced in Linux 4.0
4335 * This flag will only be called from update_time when lazytime is set.
4336 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4337 * Fortunately mtime and ctime are managed within ZFS itself, so we
4338 * only need to dirty atime.
4340 if (flags == I_DIRTY_TIME) {
4341 zp->z_atime_dirty = 1;
4346 tx = dmu_tx_create(zfsvfs->z_os);
4348 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4349 zfs_sa_upgrade_txholds(tx, zp);
4351 error = dmu_tx_assign(tx, TXG_WAIT);
4357 mutex_enter(&zp->z_lock);
4358 zp->z_atime_dirty = 0;
4360 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
4361 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
4362 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
4363 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
4365 /* Preserve the mode, mtime and ctime provided by the inode */
4366 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4367 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4368 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4373 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4374 mutex_exit(&zp->z_lock);
4381 EXPORT_SYMBOL(zfs_dirty_inode);
4385 zfs_inactive(struct inode *ip)
4387 znode_t *zp = ITOZ(ip);
4388 zfsvfs_t *zfsvfs = ITOZSB(ip);
4391 int need_unlock = 0;
4393 /* Only read lock if we haven't already write locked, e.g. rollback */
4394 if (!RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)) {
4396 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4398 if (zp->z_sa_hdl == NULL) {
4400 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4404 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4405 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4407 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4408 zfs_sa_upgrade_txholds(tx, zp);
4409 error = dmu_tx_assign(tx, TXG_WAIT);
4413 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4414 mutex_enter(&zp->z_lock);
4415 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4416 (void *)&atime, sizeof (atime), tx);
4417 zp->z_atime_dirty = 0;
4418 mutex_exit(&zp->z_lock);
4425 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4427 EXPORT_SYMBOL(zfs_inactive);
4430 * Bounds-check the seek operation.
4432 * IN: ip - inode seeking within
4433 * ooff - old file offset
4434 * noffp - pointer to new file offset
4435 * ct - caller context
4437 * RETURN: 0 if success
4438 * EINVAL if new offset invalid
4442 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
4444 if (S_ISDIR(ip->i_mode))
4446 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4448 EXPORT_SYMBOL(zfs_seek);
4451 * Fill pages with data from the disk.
4454 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
4456 znode_t *zp = ITOZ(ip);
4457 zfsvfs_t *zfsvfs = ITOZSB(ip);
4459 struct page *cur_pp;
4460 u_offset_t io_off, total;
4467 io_len = nr_pages << PAGE_SHIFT;
4468 i_size = i_size_read(ip);
4469 io_off = page_offset(pl[0]);
4471 if (io_off + io_len > i_size)
4472 io_len = i_size - io_off;
4475 * Iterate over list of pages and read each page individually.
4478 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4481 cur_pp = pl[page_idx++];
4483 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4487 /* convert checksum errors into IO errors */
4489 err = SET_ERROR(EIO);
4498 * Uses zfs_fillpage to read data from the file and fill the pages.
4500 * IN: ip - inode of file to get data from.
4501 * pl - list of pages to read
4502 * nr_pages - number of pages to read
4504 * RETURN: 0 on success, error code on failure.
4507 * vp - atime updated
4511 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
4513 znode_t *zp = ITOZ(ip);
4514 zfsvfs_t *zfsvfs = ITOZSB(ip);
4523 err = zfs_fillpage(ip, pl, nr_pages);
4528 EXPORT_SYMBOL(zfs_getpage);
4531 * Check ZFS specific permissions to memory map a section of a file.
4533 * IN: ip - inode of the file to mmap
4535 * addrp - start address in memory region
4536 * len - length of memory region
4537 * vm_flags- address flags
4539 * RETURN: 0 if success
4540 * error code if failure
4544 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4545 unsigned long vm_flags)
4547 znode_t *zp = ITOZ(ip);
4548 zfsvfs_t *zfsvfs = ITOZSB(ip);
4553 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4554 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4556 return (SET_ERROR(EPERM));
4559 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4560 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4562 return (SET_ERROR(EACCES));
4565 if (off < 0 || len > MAXOFFSET_T - off) {
4567 return (SET_ERROR(ENXIO));
4573 EXPORT_SYMBOL(zfs_map);
4576 * convoff - converts the given data (start, whence) to the
4580 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4585 if ((lckdat->l_whence == 2) || (whence == 2)) {
4586 if ((error = zfs_getattr(ip, &vap, 0, CRED()) != 0))
4590 switch (lckdat->l_whence) {
4592 lckdat->l_start += offset;
4595 lckdat->l_start += vap.va_size;
4600 return (SET_ERROR(EINVAL));
4603 if (lckdat->l_start < 0)
4604 return (SET_ERROR(EINVAL));
4608 lckdat->l_start -= offset;
4611 lckdat->l_start -= vap.va_size;
4616 return (SET_ERROR(EINVAL));
4619 lckdat->l_whence = (short)whence;
4624 * Free or allocate space in a file. Currently, this function only
4625 * supports the `F_FREESP' command. However, this command is somewhat
4626 * misnamed, as its functionality includes the ability to allocate as
4627 * well as free space.
4629 * IN: ip - inode of file to free data in.
4630 * cmd - action to take (only F_FREESP supported).
4631 * bfp - section of file to free/alloc.
4632 * flag - current file open mode flags.
4633 * offset - current file offset.
4634 * cr - credentials of caller [UNUSED].
4636 * RETURN: 0 on success, error code on failure.
4639 * ip - ctime|mtime updated
4643 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4644 offset_t offset, cred_t *cr)
4646 znode_t *zp = ITOZ(ip);
4647 zfsvfs_t *zfsvfs = ITOZSB(ip);
4654 if (cmd != F_FREESP) {
4656 return (SET_ERROR(EINVAL));
4660 * Callers might not be able to detect properly that we are read-only,
4661 * so check it explicitly here.
4663 if (zfs_is_readonly(zfsvfs)) {
4665 return (SET_ERROR(EROFS));
4668 if ((error = convoff(ip, bfp, 0, offset))) {
4673 if (bfp->l_len < 0) {
4675 return (SET_ERROR(EINVAL));
4679 * Permissions aren't checked on Solaris because on this OS
4680 * zfs_space() can only be called with an opened file handle.
4681 * On Linux we can get here through truncate_range() which
4682 * operates directly on inodes, so we need to check access rights.
4684 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4690 len = bfp->l_len; /* 0 means from off to end of file */
4692 error = zfs_freesp(zp, off, len, flag, TRUE);
4697 EXPORT_SYMBOL(zfs_space);
4701 zfs_fid(struct inode *ip, fid_t *fidp)
4703 znode_t *zp = ITOZ(ip);
4704 zfsvfs_t *zfsvfs = ITOZSB(ip);
4707 uint64_t object = zp->z_id;
4714 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4715 &gen64, sizeof (uint64_t))) != 0) {
4720 gen = (uint32_t)gen64;
4722 size = SHORT_FID_LEN;
4724 zfid = (zfid_short_t *)fidp;
4726 zfid->zf_len = size;
4728 for (i = 0; i < sizeof (zfid->zf_object); i++)
4729 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4731 /* Must have a non-zero generation number to distinguish from .zfs */
4734 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4735 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4740 EXPORT_SYMBOL(zfs_fid);
4744 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4746 znode_t *zp = ITOZ(ip);
4747 zfsvfs_t *zfsvfs = ITOZSB(ip);
4749 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4753 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4758 EXPORT_SYMBOL(zfs_getsecattr);
4762 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4764 znode_t *zp = ITOZ(ip);
4765 zfsvfs_t *zfsvfs = ITOZSB(ip);
4767 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4768 zilog_t *zilog = zfsvfs->z_log;
4773 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4775 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4776 zil_commit(zilog, 0);
4781 EXPORT_SYMBOL(zfs_setsecattr);
4783 #ifdef HAVE_UIO_ZEROCOPY
4785 * Tunable, both must be a power of 2.
4787 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4788 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4789 * an arcbuf for a partial block read
4791 int zcr_blksz_min = (1 << 10); /* 1K */
4792 int zcr_blksz_max = (1 << 17); /* 128K */
4796 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
4798 znode_t *zp = ITOZ(ip);
4799 zfsvfs_t *zfsvfs = ITOZSB(ip);
4800 int max_blksz = zfsvfs->z_max_blksz;
4801 uio_t *uio = &xuio->xu_uio;
4802 ssize_t size = uio->uio_resid;
4803 offset_t offset = uio->uio_loffset;
4808 int preamble, postamble;
4810 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4811 return (SET_ERROR(EINVAL));
4818 * Loan out an arc_buf for write if write size is bigger than
4819 * max_blksz, and the file's block size is also max_blksz.
4822 if (size < blksz || zp->z_blksz != blksz) {
4824 return (SET_ERROR(EINVAL));
4827 * Caller requests buffers for write before knowing where the
4828 * write offset might be (e.g. NFS TCP write).
4833 preamble = P2PHASE(offset, blksz);
4835 preamble = blksz - preamble;
4840 postamble = P2PHASE(size, blksz);
4843 fullblk = size / blksz;
4844 (void) dmu_xuio_init(xuio,
4845 (preamble != 0) + fullblk + (postamble != 0));
4848 * Have to fix iov base/len for partial buffers. They
4849 * currently represent full arc_buf's.
4852 /* data begins in the middle of the arc_buf */
4853 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4856 (void) dmu_xuio_add(xuio, abuf,
4857 blksz - preamble, preamble);
4860 for (i = 0; i < fullblk; i++) {
4861 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4864 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
4868 /* data ends in the middle of the arc_buf */
4869 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4872 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
4877 * Loan out an arc_buf for read if the read size is larger than
4878 * the current file block size. Block alignment is not
4879 * considered. Partial arc_buf will be loaned out for read.
4881 blksz = zp->z_blksz;
4882 if (blksz < zcr_blksz_min)
4883 blksz = zcr_blksz_min;
4884 if (blksz > zcr_blksz_max)
4885 blksz = zcr_blksz_max;
4886 /* avoid potential complexity of dealing with it */
4887 if (blksz > max_blksz) {
4889 return (SET_ERROR(EINVAL));
4892 maxsize = zp->z_size - uio->uio_loffset;
4898 return (SET_ERROR(EINVAL));
4903 return (SET_ERROR(EINVAL));
4906 uio->uio_extflg = UIO_XUIO;
4907 XUIO_XUZC_RW(xuio) = ioflag;
4914 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
4918 int ioflag = XUIO_XUZC_RW(xuio);
4920 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
4922 i = dmu_xuio_cnt(xuio);
4924 abuf = dmu_xuio_arcbuf(xuio, i);
4926 * if abuf == NULL, it must be a write buffer
4927 * that has been returned in zfs_write().
4930 dmu_return_arcbuf(abuf);
4931 ASSERT(abuf || ioflag == UIO_WRITE);
4934 dmu_xuio_fini(xuio);
4937 #endif /* HAVE_UIO_ZEROCOPY */
4939 #if defined(_KERNEL) && defined(HAVE_SPL)
4941 module_param(zfs_delete_blocks, ulong, 0644);
4942 MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async");
4943 module_param(zfs_read_chunk_size, long, 0644);
4944 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");