4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25 /* Portions Copyright 2007 Jeremy Teo */
26 /* Portions Copyright 2010 Robert Milkowski */
29 #include <sys/types.h>
30 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
36 #include <sys/vfs_opreg.h>
40 #include <sys/taskq.h>
42 #include <sys/vmsystm.h>
43 #include <sys/atomic.h>
45 #include <sys/pathname.h>
46 #include <sys/cmn_err.h>
47 #include <sys/errno.h>
48 #include <sys/unistd.h>
49 #include <sys/zfs_dir.h>
50 #include <sys/zfs_acl.h>
51 #include <sys/zfs_ioctl.h>
52 #include <sys/fs/zfs.h>
54 #include <sys/dmu_objset.h>
60 #include <sys/dirent.h>
61 #include <sys/policy.h>
62 #include <sys/sunddi.h>
65 #include "fs/fs_subr.h"
66 #include <sys/zfs_fuid.h>
67 #include <sys/zfs_sa.h>
68 #include <sys/zfs_vnops.h>
70 #include <sys/zfs_rlock.h>
71 #include <sys/extdirent.h>
72 #include <sys/kidmap.h>
79 * Each vnode op performs some logical unit of work. To do this, the ZPL must
80 * properly lock its in-core state, create a DMU transaction, do the work,
81 * record this work in the intent log (ZIL), commit the DMU transaction,
82 * and wait for the intent log to commit if it is a synchronous operation.
83 * Moreover, the vnode ops must work in both normal and log replay context.
84 * The ordering of events is important to avoid deadlocks and references
85 * to freed memory. The example below illustrates the following Big Rules:
87 * (1) A check must be made in each zfs thread for a mounted file system.
88 * This is done avoiding races using ZFS_ENTER(zsb).
89 * A ZFS_EXIT(zsb) is needed before all returns. Any znodes
90 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
91 * can return EIO from the calling function.
93 * (2) iput() should always be the last thing except for zil_commit()
94 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
95 * First, if it's the last reference, the vnode/znode
96 * can be freed, so the zp may point to freed memory. Second, the last
97 * reference will call zfs_zinactive(), which may induce a lot of work --
98 * pushing cached pages (which acquires range locks) and syncing out
99 * cached atime changes. Third, zfs_zinactive() may require a new tx,
100 * which could deadlock the system if you were already holding one.
101 * If you must call iput() within a tx then use iput_ASYNC().
103 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
104 * as they can span dmu_tx_assign() calls.
106 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
107 * This is critical because we don't want to block while holding locks.
108 * Note, in particular, that if a lock is sometimes acquired before
109 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
110 * use a non-blocking assign can deadlock the system. The scenario:
112 * Thread A has grabbed a lock before calling dmu_tx_assign().
113 * Thread B is in an already-assigned tx, and blocks for this lock.
114 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
115 * forever, because the previous txg can't quiesce until B's tx commits.
117 * If dmu_tx_assign() returns ERESTART and zsb->z_assign is TXG_NOWAIT,
118 * then drop all locks, call dmu_tx_wait(), and try again.
120 * (5) If the operation succeeded, generate the intent log entry for it
121 * before dropping locks. This ensures that the ordering of events
122 * in the intent log matches the order in which they actually occurred.
123 * During ZIL replay the zfs_log_* functions will update the sequence
124 * number to indicate the zil transaction has replayed.
126 * (6) At the end of each vnode op, the DMU tx must always commit,
127 * regardless of whether there were any errors.
129 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
130 * to ensure that synchronous semantics are provided when necessary.
132 * In general, this is how things should be ordered in each vnode op:
134 * ZFS_ENTER(zsb); // exit if unmounted
136 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
137 * rw_enter(...); // grab any other locks you need
138 * tx = dmu_tx_create(...); // get DMU tx
139 * dmu_tx_hold_*(); // hold each object you might modify
140 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign
142 * rw_exit(...); // drop locks
143 * zfs_dirent_unlock(dl); // unlock directory entry
144 * iput(...); // release held vnodes
145 * if (error == ERESTART) {
150 * dmu_tx_abort(tx); // abort DMU tx
151 * ZFS_EXIT(zsb); // finished in zfs
152 * return (error); // really out of space
154 * error = do_real_work(); // do whatever this VOP does
156 * zfs_log_*(...); // on success, make ZIL entry
157 * dmu_tx_commit(tx); // commit DMU tx -- error or not
158 * rw_exit(...); // drop locks
159 * zfs_dirent_unlock(dl); // unlock directory entry
160 * iput(...); // release held vnodes
161 * zil_commit(zilog, foid); // synchronous when necessary
162 * ZFS_EXIT(zsb); // finished in zfs
163 * return (error); // done, report error
168 * When a file is memory mapped, we must keep the IO data synchronized
169 * between the DMU cache and the memory mapped pages. What this means:
171 * On Write: If we find a memory mapped page, we write to *both*
172 * the page and the dmu buffer.
175 update_pages(struct inode *ip, int64_t start, int len,
176 objset_t *os, uint64_t oid)
178 struct address_space *mp = ip->i_mapping;
184 off = start & (PAGE_CACHE_SIZE-1);
185 for (start &= PAGE_CACHE_MASK; len > 0; start += PAGE_CACHE_SIZE) {
186 nbytes = MIN(PAGE_CACHE_SIZE - off, len);
188 pp = find_lock_page(mp, start >> PAGE_CACHE_SHIFT);
190 if (mapping_writably_mapped(mp))
191 flush_dcache_page(pp);
194 (void) dmu_read(os, oid, start+off, nbytes, pb+off,
198 if (mapping_writably_mapped(mp))
199 flush_dcache_page(pp);
201 mark_page_accessed(pp);
205 page_cache_release(pp);
214 * When a file is memory mapped, we must keep the IO data synchronized
215 * between the DMU cache and the memory mapped pages. What this means:
217 * On Read: We "read" preferentially from memory mapped pages,
218 * else we default from the dmu buffer.
220 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
221 * the file is memory mapped.
224 mappedread(struct inode *ip, int nbytes, uio_t *uio)
226 struct address_space *mp = ip->i_mapping;
228 znode_t *zp = ITOZ(ip);
229 objset_t *os = ITOZSB(ip)->z_os;
236 start = uio->uio_loffset;
237 off = start & (PAGE_CACHE_SIZE-1);
238 for (start &= PAGE_CACHE_MASK; len > 0; start += PAGE_CACHE_SIZE) {
239 bytes = MIN(PAGE_CACHE_SIZE - off, len);
241 pp = find_lock_page(mp, start >> PAGE_CACHE_SHIFT);
243 ASSERT(PageUptodate(pp));
246 error = uiomove(pb + off, bytes, UIO_READ, uio);
249 if (mapping_writably_mapped(mp))
250 flush_dcache_page(pp);
252 mark_page_accessed(pp);
254 page_cache_release(pp);
256 error = dmu_read_uio(os, zp->z_id, uio, bytes);
268 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
271 * Read bytes from specified file into supplied buffer.
273 * IN: ip - inode of file to be read from.
274 * uio - structure supplying read location, range info,
276 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
277 * O_DIRECT flag; used to bypass page cache.
278 * cr - credentials of caller.
280 * OUT: uio - updated offset and range, buffer filled.
282 * RETURN: 0 if success
283 * error code if failure
286 * inode - atime updated if byte count > 0
290 zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
292 znode_t *zp = ITOZ(ip);
293 zfs_sb_t *zsb = ITOZSB(ip);
298 #ifdef HAVE_UIO_ZEROCOPY
300 #endif /* HAVE_UIO_ZEROCOPY */
306 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
312 * Validate file offset
314 if (uio->uio_loffset < (offset_t)0) {
320 * Fasttrack empty reads
322 if (uio->uio_resid == 0) {
327 #ifdef HAVE_MANDLOCKS
329 * Check for mandatory locks
331 if (MANDMODE(zp->z_mode)) {
332 if (error = chklock(ip, FREAD,
333 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
338 #endif /* HAVE_MANDLOCK */
341 * If we're in FRSYNC mode, sync out this znode before reading it.
343 if (ioflag & FRSYNC || zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
344 zil_commit(zsb->z_log, zp->z_id);
347 * Lock the range against changes.
349 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
352 * If we are reading past end-of-file we can skip
353 * to the end; but we might still need to set atime.
355 if (uio->uio_loffset >= zp->z_size) {
360 ASSERT(uio->uio_loffset < zp->z_size);
361 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
363 #ifdef HAVE_UIO_ZEROCOPY
364 if ((uio->uio_extflg == UIO_XUIO) &&
365 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
367 int blksz = zp->z_blksz;
368 uint64_t offset = uio->uio_loffset;
370 xuio = (xuio_t *)uio;
372 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
375 ASSERT(offset + n <= blksz);
378 (void) dmu_xuio_init(xuio, nblk);
380 if (vn_has_cached_data(ip)) {
382 * For simplicity, we always allocate a full buffer
383 * even if we only expect to read a portion of a block.
385 while (--nblk >= 0) {
386 (void) dmu_xuio_add(xuio,
387 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
392 #endif /* HAVE_UIO_ZEROCOPY */
395 nbytes = MIN(n, zfs_read_chunk_size -
396 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
398 if (zp->z_is_mapped && !(ioflag & O_DIRECT))
399 error = mappedread(ip, nbytes, uio);
401 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
404 /* convert checksum errors into IO errors */
413 zfs_range_unlock(rl);
415 ZFS_ACCESSTIME_STAMP(zsb, zp);
416 zfs_inode_update(zp);
420 EXPORT_SYMBOL(zfs_read);
423 * Write the bytes to a file.
425 * IN: ip - inode of file to be written to.
426 * uio - structure supplying write location, range info,
428 * ioflag - FAPPEND flag set if in append mode.
429 * O_DIRECT flag; used to bypass page cache.
430 * cr - credentials of caller.
432 * OUT: uio - updated offset and range.
434 * RETURN: 0 if success
435 * error code if failure
438 * ip - ctime|mtime updated if byte count > 0
443 zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
445 znode_t *zp = ITOZ(ip);
446 rlim64_t limit = uio->uio_limit;
447 ssize_t start_resid = uio->uio_resid;
451 zfs_sb_t *zsb = ZTOZSB(zp);
456 int max_blksz = zsb->z_max_blksz;
459 iovec_t *aiov = NULL;
462 iovec_t *iovp = uio->uio_iov;
465 sa_bulk_attr_t bulk[4];
466 uint64_t mtime[2], ctime[2];
467 ASSERTV(int iovcnt = uio->uio_iovcnt);
470 * Fasttrack empty write
476 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
482 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
483 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
484 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb), NULL, &zp->z_size, 8);
485 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
489 * If immutable or not appending then return EPERM
491 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
492 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
493 (uio->uio_loffset < zp->z_size))) {
501 * Validate file offset
503 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
509 #ifdef HAVE_MANDLOCKS
511 * Check for mandatory locks before calling zfs_range_lock()
512 * in order to prevent a deadlock with locks set via fcntl().
514 if (MANDMODE((mode_t)zp->z_mode) &&
515 (error = chklock(ip, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
519 #endif /* HAVE_MANDLOCKS */
521 #ifdef HAVE_UIO_ZEROCOPY
523 * Pre-fault the pages to ensure slow (eg NFS) pages
525 * Skip this if uio contains loaned arc_buf.
527 if ((uio->uio_extflg == UIO_XUIO) &&
528 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
529 xuio = (xuio_t *)uio;
531 uio_prefaultpages(MIN(n, max_blksz), uio);
532 #endif /* HAVE_UIO_ZEROCOPY */
535 * If in append mode, set the io offset pointer to eof.
537 if (ioflag & FAPPEND) {
539 * Obtain an appending range lock to guarantee file append
540 * semantics. We reset the write offset once we have the lock.
542 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
544 if (rl->r_len == UINT64_MAX) {
546 * We overlocked the file because this write will cause
547 * the file block size to increase.
548 * Note that zp_size cannot change with this lock held.
552 uio->uio_loffset = woff;
555 * Note that if the file block size will change as a result of
556 * this write, then this range lock will lock the entire file
557 * so that we can re-write the block safely.
559 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
563 zfs_range_unlock(rl);
568 if ((woff + n) > limit || woff > (limit - n))
571 /* Will this write extend the file length? */
572 write_eof = (woff + n > zp->z_size);
574 end_size = MAX(zp->z_size, woff + n);
577 * Write the file in reasonable size chunks. Each chunk is written
578 * in a separate transaction; this keeps the intent log records small
579 * and allows us to do more fine-grained space accounting.
583 woff = uio->uio_loffset;
585 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
586 zfs_owner_overquota(zsb, zp, B_TRUE)) {
588 dmu_return_arcbuf(abuf);
593 if (xuio && abuf == NULL) {
594 ASSERT(i_iov < iovcnt);
596 abuf = dmu_xuio_arcbuf(xuio, i_iov);
597 dmu_xuio_clear(xuio, i_iov);
598 ASSERT((aiov->iov_base == abuf->b_data) ||
599 ((char *)aiov->iov_base - (char *)abuf->b_data +
600 aiov->iov_len == arc_buf_size(abuf)));
602 } else if (abuf == NULL && n >= max_blksz &&
603 woff >= zp->z_size &&
604 P2PHASE(woff, max_blksz) == 0 &&
605 zp->z_blksz == max_blksz) {
607 * This write covers a full block. "Borrow" a buffer
608 * from the dmu so that we can fill it before we enter
609 * a transaction. This avoids the possibility of
610 * holding up the transaction if the data copy hangs
611 * up on a pagefault (e.g., from an NFS server mapping).
615 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
617 ASSERT(abuf != NULL);
618 ASSERT(arc_buf_size(abuf) == max_blksz);
619 if ((error = uiocopy(abuf->b_data, max_blksz,
620 UIO_WRITE, uio, &cbytes))) {
621 dmu_return_arcbuf(abuf);
624 ASSERT(cbytes == max_blksz);
628 * Start a transaction.
630 tx = dmu_tx_create(zsb->z_os);
631 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
632 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
633 zfs_sa_upgrade_txholds(tx, zp);
634 error = dmu_tx_assign(tx, TXG_NOWAIT);
636 if (error == ERESTART) {
643 dmu_return_arcbuf(abuf);
648 * If zfs_range_lock() over-locked we grow the blocksize
649 * and then reduce the lock range. This will only happen
650 * on the first iteration since zfs_range_reduce() will
651 * shrink down r_len to the appropriate size.
653 if (rl->r_len == UINT64_MAX) {
656 if (zp->z_blksz > max_blksz) {
657 ASSERT(!ISP2(zp->z_blksz));
658 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
660 new_blksz = MIN(end_size, max_blksz);
662 zfs_grow_blocksize(zp, new_blksz, tx);
663 zfs_range_reduce(rl, woff, n);
667 * XXX - should we really limit each write to z_max_blksz?
668 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
670 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
673 tx_bytes = uio->uio_resid;
674 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
676 tx_bytes -= uio->uio_resid;
679 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
681 * If this is not a full block write, but we are
682 * extending the file past EOF and this data starts
683 * block-aligned, use assign_arcbuf(). Otherwise,
684 * write via dmu_write().
686 if (tx_bytes < max_blksz && (!write_eof ||
687 aiov->iov_base != abuf->b_data)) {
689 dmu_write(zsb->z_os, zp->z_id, woff,
690 aiov->iov_len, aiov->iov_base, tx);
691 dmu_return_arcbuf(abuf);
692 xuio_stat_wbuf_copied();
694 ASSERT(xuio || tx_bytes == max_blksz);
695 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
698 ASSERT(tx_bytes <= uio->uio_resid);
699 uioskip(uio, tx_bytes);
702 if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT))
703 update_pages(ip, woff, tx_bytes, zsb->z_os, zp->z_id);
706 * If we made no progress, we're done. If we made even
707 * partial progress, update the znode and ZIL accordingly.
710 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
711 (void *)&zp->z_size, sizeof (uint64_t), tx);
718 * Clear Set-UID/Set-GID bits on successful write if not
719 * privileged and at least one of the excute bits is set.
721 * It would be nice to to this after all writes have
722 * been done, but that would still expose the ISUID/ISGID
723 * to another app after the partial write is committed.
725 * Note: we don't call zfs_fuid_map_id() here because
726 * user 0 is not an ephemeral uid.
728 mutex_enter(&zp->z_acl_lock);
729 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
730 (S_IXUSR >> 6))) != 0 &&
731 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
732 secpolicy_vnode_setid_retain(cr,
733 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
735 zp->z_mode &= ~(S_ISUID | S_ISGID);
736 newmode = zp->z_mode;
737 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zsb),
738 (void *)&newmode, sizeof (uint64_t), tx);
740 mutex_exit(&zp->z_acl_lock);
742 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
746 * Update the file size (zp_size) if it has changed;
747 * account for possible concurrent updates.
749 while ((end_size = zp->z_size) < uio->uio_loffset) {
750 (void) atomic_cas_64(&zp->z_size, end_size,
755 * If we are replaying and eof is non zero then force
756 * the file size to the specified eof. Note, there's no
757 * concurrency during replay.
759 if (zsb->z_replay && zsb->z_replay_eof != 0)
760 zp->z_size = zsb->z_replay_eof;
762 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
764 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
769 ASSERT(tx_bytes == nbytes);
773 uio_prefaultpages(MIN(n, max_blksz), uio);
776 zfs_range_unlock(rl);
779 * If we're in replay mode, or we made no progress, return error.
780 * Otherwise, it's at least a partial write, so it's successful.
782 if (zsb->z_replay || uio->uio_resid == start_resid) {
787 if (ioflag & (FSYNC | FDSYNC) ||
788 zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
789 zil_commit(zilog, zp->z_id);
791 zfs_inode_update(zp);
795 EXPORT_SYMBOL(zfs_write);
798 iput_async(struct inode *ip, taskq_t *taskq)
800 ASSERT(atomic_read(&ip->i_count) > 0);
801 if (atomic_read(&ip->i_count) == 1)
802 taskq_dispatch(taskq, (task_func_t *)iput, ip, TQ_SLEEP);
808 zfs_get_done(zgd_t *zgd, int error)
810 znode_t *zp = zgd->zgd_private;
811 objset_t *os = ZTOZSB(zp)->z_os;
814 dmu_buf_rele(zgd->zgd_db, zgd);
816 zfs_range_unlock(zgd->zgd_rl);
819 * Release the vnode asynchronously as we currently have the
820 * txg stopped from syncing.
822 iput_async(ZTOI(zp), dsl_pool_iput_taskq(dmu_objset_pool(os)));
824 if (error == 0 && zgd->zgd_bp)
825 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
827 kmem_free(zgd, sizeof (zgd_t));
831 static int zil_fault_io = 0;
835 * Get data to generate a TX_WRITE intent log record.
838 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
841 objset_t *os = zsb->z_os;
843 uint64_t object = lr->lr_foid;
844 uint64_t offset = lr->lr_offset;
845 uint64_t size = lr->lr_length;
846 blkptr_t *bp = &lr->lr_blkptr;
855 * Nothing to do if the file has been removed
857 if (zfs_zget(zsb, object, &zp) != 0)
859 if (zp->z_unlinked) {
861 * Release the vnode asynchronously as we currently have the
862 * txg stopped from syncing.
864 iput_async(ZTOI(zp), dsl_pool_iput_taskq(dmu_objset_pool(os)));
868 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
869 zgd->zgd_zilog = zsb->z_log;
870 zgd->zgd_private = zp;
873 * Write records come in two flavors: immediate and indirect.
874 * For small writes it's cheaper to store the data with the
875 * log record (immediate); for large writes it's cheaper to
876 * sync the data and get a pointer to it (indirect) so that
877 * we don't have to write the data twice.
879 if (buf != NULL) { /* immediate write */
880 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
881 /* test for truncation needs to be done while range locked */
882 if (offset >= zp->z_size) {
885 error = dmu_read(os, object, offset, size, buf,
886 DMU_READ_NO_PREFETCH);
888 ASSERT(error == 0 || error == ENOENT);
889 } else { /* indirect write */
891 * Have to lock the whole block to ensure when it's
892 * written out and it's checksum is being calculated
893 * that no one can change the data. We need to re-check
894 * blocksize after we get the lock in case it's changed!
899 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
901 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
903 if (zp->z_blksz == size)
906 zfs_range_unlock(zgd->zgd_rl);
908 /* test for truncation needs to be done while range locked */
909 if (lr->lr_offset >= zp->z_size)
918 error = dmu_buf_hold(os, object, offset, zgd, &db,
919 DMU_READ_NO_PREFETCH);
925 ASSERT(db->db_offset == offset);
926 ASSERT(db->db_size == size);
928 error = dmu_sync(zio, lr->lr_common.lrc_txg,
930 ASSERT(error || lr->lr_length <= zp->z_blksz);
933 * On success, we need to wait for the write I/O
934 * initiated by dmu_sync() to complete before we can
935 * release this dbuf. We will finish everything up
936 * in the zfs_get_done() callback.
941 if (error == EALREADY) {
942 lr->lr_common.lrc_txtype = TX_WRITE2;
948 zfs_get_done(zgd, error);
955 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
957 znode_t *zp = ITOZ(ip);
958 zfs_sb_t *zsb = ITOZSB(ip);
964 if (flag & V_ACE_MASK)
965 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
967 error = zfs_zaccess_rwx(zp, mode, flag, cr);
972 EXPORT_SYMBOL(zfs_access);
975 * Lookup an entry in a directory, or an extended attribute directory.
976 * If it exists, return a held inode reference for it.
978 * IN: dip - inode of directory to search.
979 * nm - name of entry to lookup.
980 * flags - LOOKUP_XATTR set if looking for an attribute.
981 * cr - credentials of caller.
982 * direntflags - directory lookup flags
983 * realpnp - returned pathname.
985 * OUT: ipp - inode of located entry, NULL if not found.
987 * RETURN: 0 if success
988 * error code if failure
995 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
996 cred_t *cr, int *direntflags, pathname_t *realpnp)
998 znode_t *zdp = ITOZ(dip);
999 zfs_sb_t *zsb = ITOZSB(dip);
1003 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1005 if (!S_ISDIR(dip->i_mode)) {
1007 } else if (zdp->z_sa_hdl == NULL) {
1011 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1012 error = zfs_fastaccesschk_execute(zdp, cr);
1021 vnode_t *tvp = dnlc_lookup(dvp, nm);
1024 error = zfs_fastaccesschk_execute(zdp, cr);
1029 if (tvp == DNLC_NO_VNODE) {
1034 return (specvp_check(vpp, cr));
1037 #endif /* HAVE_DNLC */
1046 if (flags & LOOKUP_XATTR) {
1048 * If the xattr property is off, refuse the lookup request.
1050 if (!(zsb->z_flags & ZSB_XATTR_USER)) {
1056 * We don't allow recursive attributes..
1057 * Maybe someday we will.
1059 if (zdp->z_pflags & ZFS_XATTR) {
1064 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1070 * Do we have permission to get into attribute directory?
1073 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1083 if (!S_ISDIR(dip->i_mode)) {
1089 * Check accessibility of directory.
1092 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1097 if (zsb->z_utf8 && u8_validate(nm, strlen(nm),
1098 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1103 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1104 if ((error == 0) && (*ipp))
1105 zfs_inode_update(ITOZ(*ipp));
1110 EXPORT_SYMBOL(zfs_lookup);
1113 * Attempt to create a new entry in a directory. If the entry
1114 * already exists, truncate the file if permissible, else return
1115 * an error. Return the ip of the created or trunc'd file.
1117 * IN: dip - inode of directory to put new file entry in.
1118 * name - name of new file entry.
1119 * vap - attributes of new file.
1120 * excl - flag indicating exclusive or non-exclusive mode.
1121 * mode - mode to open file with.
1122 * cr - credentials of caller.
1123 * flag - large file flag [UNUSED].
1124 * vsecp - ACL to be set
1126 * OUT: ipp - inode of created or trunc'd entry.
1128 * RETURN: 0 if success
1129 * error code if failure
1132 * dip - ctime|mtime updated if new entry created
1133 * ip - ctime|mtime always, atime if new
1138 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1139 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1141 znode_t *zp, *dzp = ITOZ(dip);
1142 zfs_sb_t *zsb = ITOZSB(dip);
1150 zfs_acl_ids_t acl_ids;
1151 boolean_t fuid_dirtied;
1152 boolean_t have_acl = B_FALSE;
1155 * If we have an ephemeral id, ACL, or XVATTR then
1156 * make sure file system is at proper version
1162 if (zsb->z_use_fuids == B_FALSE &&
1163 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1171 if (zsb->z_utf8 && u8_validate(name, strlen(name),
1172 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1178 if (vap->va_mask & AT_XVATTR) {
1179 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1180 crgetuid(cr), cr, vap->va_mode)) != 0) {
1185 #endif /* HAVE_XVATTR */
1189 if (*name == '\0') {
1191 * Null component name refers to the directory itself.
1198 /* possible igrab(zp) */
1201 if (flag & FIGNORECASE)
1204 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1208 zfs_acl_ids_free(&acl_ids);
1209 if (strcmp(name, "..") == 0)
1220 * Create a new file object and update the directory
1223 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1225 zfs_acl_ids_free(&acl_ids);
1230 * We only support the creation of regular files in
1231 * extended attribute directories.
1234 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1236 zfs_acl_ids_free(&acl_ids);
1241 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1242 cr, vsecp, &acl_ids)) != 0)
1246 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1247 zfs_acl_ids_free(&acl_ids);
1252 tx = dmu_tx_create(os);
1254 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1255 ZFS_SA_BASE_ATTR_SIZE);
1257 fuid_dirtied = zsb->z_fuid_dirty;
1259 zfs_fuid_txhold(zsb, tx);
1260 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1261 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1262 if (!zsb->z_use_sa &&
1263 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1264 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1265 0, acl_ids.z_aclp->z_acl_bytes);
1267 error = dmu_tx_assign(tx, TXG_NOWAIT);
1269 zfs_dirent_unlock(dl);
1270 if (error == ERESTART) {
1275 zfs_acl_ids_free(&acl_ids);
1280 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1283 zfs_fuid_sync(zsb, tx);
1285 (void) zfs_link_create(dl, zp, tx, ZNEW);
1286 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1287 if (flag & FIGNORECASE)
1289 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1290 vsecp, acl_ids.z_fuidp, vap);
1291 zfs_acl_ids_free(&acl_ids);
1294 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1297 zfs_acl_ids_free(&acl_ids);
1301 * A directory entry already exists for this name.
1304 * Can't truncate an existing file if in exclusive mode.
1311 * Can't open a directory for writing.
1313 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1318 * Verify requested access to file.
1320 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1324 mutex_enter(&dzp->z_lock);
1326 mutex_exit(&dzp->z_lock);
1329 * Truncate regular files if requested.
1331 if (S_ISREG(ZTOI(zp)->i_mode) &&
1332 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1333 /* we can't hold any locks when calling zfs_freesp() */
1334 zfs_dirent_unlock(dl);
1336 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1342 zfs_dirent_unlock(dl);
1348 zfs_inode_update(dzp);
1349 zfs_inode_update(zp);
1353 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1354 zil_commit(zilog, 0);
1359 EXPORT_SYMBOL(zfs_create);
1362 * Remove an entry from a directory.
1364 * IN: dip - inode of directory to remove entry from.
1365 * name - name of entry to remove.
1366 * cr - credentials of caller.
1368 * RETURN: 0 if success
1369 * error code if failure
1373 * ip - ctime (if nlink > 0)
1376 uint64_t null_xattr = 0;
1380 zfs_remove(struct inode *dip, char *name, cred_t *cr)
1382 znode_t *zp, *dzp = ITOZ(dip);
1385 zfs_sb_t *zsb = ITOZSB(dip);
1388 uint64_t xattr_obj_unlinked = 0;
1394 pathname_t *realnmp = NULL;
1395 #ifdef HAVE_PN_UTILS
1397 #endif /* HAVE_PN_UTILS */
1405 #ifdef HAVE_PN_UTILS
1406 if (flags & FIGNORECASE) {
1411 #endif /* HAVE_PN_UTILS */
1417 * Attempt to lock directory; fail if entry doesn't exist.
1419 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1421 #ifdef HAVE_PN_UTILS
1424 #endif /* HAVE_PN_UTILS */
1431 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1436 * Need to use rmdir for removing directories.
1438 if (S_ISDIR(ip->i_mode)) {
1445 dnlc_remove(dvp, realnmp->pn_buf);
1447 dnlc_remove(dvp, name);
1448 #endif /* HAVE_DNLC */
1451 * We never delete the znode and always place it in the unlinked
1452 * set. The dentry cache will always hold the last reference and
1453 * is responsible for safely freeing the znode.
1456 tx = dmu_tx_create(zsb->z_os);
1457 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1458 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1459 zfs_sa_upgrade_txholds(tx, zp);
1460 zfs_sa_upgrade_txholds(tx, dzp);
1462 /* are there any extended attributes? */
1463 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1464 &xattr_obj, sizeof (xattr_obj));
1465 if (error == 0 && xattr_obj) {
1466 error = zfs_zget(zsb, xattr_obj, &xzp);
1467 ASSERT3U(error, ==, 0);
1468 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1469 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1472 /* charge as an update -- would be nice not to charge at all */
1473 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
1475 error = dmu_tx_assign(tx, TXG_NOWAIT);
1477 zfs_dirent_unlock(dl);
1481 if (error == ERESTART) {
1486 #ifdef HAVE_PN_UTILS
1489 #endif /* HAVE_PN_UTILS */
1496 * Remove the directory entry.
1498 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1507 * Hold z_lock so that we can make sure that the ACL obj
1508 * hasn't changed. Could have been deleted due to
1511 mutex_enter(&zp->z_lock);
1512 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1513 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1514 mutex_exit(&zp->z_lock);
1515 zfs_unlinked_add(zp, tx);
1519 #ifdef HAVE_PN_UTILS
1520 if (flags & FIGNORECASE)
1522 #endif /* HAVE_PN_UTILS */
1523 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1527 #ifdef HAVE_PN_UTILS
1530 #endif /* HAVE_PN_UTILS */
1532 zfs_dirent_unlock(dl);
1533 zfs_inode_update(dzp);
1534 zfs_inode_update(zp);
1536 zfs_inode_update(xzp);
1542 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1543 zil_commit(zilog, 0);
1548 EXPORT_SYMBOL(zfs_remove);
1551 * Create a new directory and insert it into dip using the name
1552 * provided. Return a pointer to the inserted directory.
1554 * IN: dip - inode of directory to add subdir to.
1555 * dirname - name of new directory.
1556 * vap - attributes of new directory.
1557 * cr - credentials of caller.
1558 * vsecp - ACL to be set
1560 * OUT: ipp - inode of created directory.
1562 * RETURN: 0 if success
1563 * error code if failure
1566 * dip - ctime|mtime updated
1567 * ipp - ctime|mtime|atime updated
1571 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1572 cred_t *cr, int flags, vsecattr_t *vsecp)
1574 znode_t *zp, *dzp = ITOZ(dip);
1575 zfs_sb_t *zsb = ITOZSB(dip);
1583 gid_t gid = crgetgid(cr);
1584 zfs_acl_ids_t acl_ids;
1585 boolean_t fuid_dirtied;
1587 ASSERT(S_ISDIR(vap->va_mode));
1590 * If we have an ephemeral id, ACL, or XVATTR then
1591 * make sure file system is at proper version
1595 if (zsb->z_use_fuids == B_FALSE &&
1596 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1603 if (dzp->z_pflags & ZFS_XATTR) {
1608 if (zsb->z_utf8 && u8_validate(dirname,
1609 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1613 if (flags & FIGNORECASE)
1617 if (vap->va_mask & AT_XVATTR) {
1618 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1619 crgetuid(cr), cr, vap->va_mode)) != 0) {
1624 #endif /* HAVE_XVATTR */
1626 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1627 vsecp, &acl_ids)) != 0) {
1632 * First make sure the new directory doesn't exist.
1634 * Existence is checked first to make sure we don't return
1635 * EACCES instead of EEXIST which can cause some applications
1641 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1643 zfs_acl_ids_free(&acl_ids);
1648 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
1649 zfs_acl_ids_free(&acl_ids);
1650 zfs_dirent_unlock(dl);
1655 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1656 zfs_acl_ids_free(&acl_ids);
1657 zfs_dirent_unlock(dl);
1663 * Add a new entry to the directory.
1665 tx = dmu_tx_create(zsb->z_os);
1666 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1667 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1668 fuid_dirtied = zsb->z_fuid_dirty;
1670 zfs_fuid_txhold(zsb, tx);
1671 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1672 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1673 acl_ids.z_aclp->z_acl_bytes);
1676 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1677 ZFS_SA_BASE_ATTR_SIZE);
1679 error = dmu_tx_assign(tx, TXG_NOWAIT);
1681 zfs_dirent_unlock(dl);
1682 if (error == ERESTART) {
1687 zfs_acl_ids_free(&acl_ids);
1696 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1699 zfs_fuid_sync(zsb, tx);
1702 * Now put new name in parent dir.
1704 (void) zfs_link_create(dl, zp, tx, ZNEW);
1708 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1709 if (flags & FIGNORECASE)
1711 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1712 acl_ids.z_fuidp, vap);
1714 zfs_acl_ids_free(&acl_ids);
1718 zfs_dirent_unlock(dl);
1720 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1721 zil_commit(zilog, 0);
1723 zfs_inode_update(dzp);
1724 zfs_inode_update(zp);
1728 EXPORT_SYMBOL(zfs_mkdir);
1731 * Remove a directory subdir entry. If the current working
1732 * directory is the same as the subdir to be removed, the
1735 * IN: dip - inode of directory to remove from.
1736 * name - name of directory to be removed.
1737 * cwd - inode of current working directory.
1738 * cr - credentials of caller.
1739 * flags - case flags
1741 * RETURN: 0 if success
1742 * error code if failure
1745 * dip - ctime|mtime updated
1749 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
1752 znode_t *dzp = ITOZ(dip);
1755 zfs_sb_t *zsb = ITOZSB(dip);
1766 if (flags & FIGNORECASE)
1772 * Attempt to lock directory; fail if entry doesn't exist.
1774 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1782 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1786 if (!S_ISDIR(ip->i_mode)) {
1797 * Grab a lock on the directory to make sure that noone is
1798 * trying to add (or lookup) entries while we are removing it.
1800 rw_enter(&zp->z_name_lock, RW_WRITER);
1803 * Grab a lock on the parent pointer to make sure we play well
1804 * with the treewalk and directory rename code.
1806 rw_enter(&zp->z_parent_lock, RW_WRITER);
1808 tx = dmu_tx_create(zsb->z_os);
1809 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1810 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1811 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
1812 zfs_sa_upgrade_txholds(tx, zp);
1813 zfs_sa_upgrade_txholds(tx, dzp);
1814 error = dmu_tx_assign(tx, TXG_NOWAIT);
1816 rw_exit(&zp->z_parent_lock);
1817 rw_exit(&zp->z_name_lock);
1818 zfs_dirent_unlock(dl);
1820 if (error == ERESTART) {
1830 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
1833 uint64_t txtype = TX_RMDIR;
1834 if (flags & FIGNORECASE)
1836 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
1841 rw_exit(&zp->z_parent_lock);
1842 rw_exit(&zp->z_name_lock);
1844 zfs_dirent_unlock(dl);
1848 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1849 zil_commit(zilog, 0);
1851 zfs_inode_update(dzp);
1852 zfs_inode_update(zp);
1856 EXPORT_SYMBOL(zfs_rmdir);
1859 * Read as many directory entries as will fit into the provided
1860 * dirent buffer from the given directory cursor position.
1862 * IN: ip - inode of directory to read.
1863 * dirent - buffer for directory entries.
1865 * OUT: dirent - filler buffer of directory entries.
1867 * RETURN: 0 if success
1868 * error code if failure
1871 * ip - atime updated
1873 * Note that the low 4 bits of the cookie returned by zap is always zero.
1874 * This allows us to use the low range for "special" directory entries:
1875 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
1876 * we use the offset 2 for the '.zfs' directory.
1880 zfs_readdir(struct inode *ip, void *dirent, filldir_t filldir,
1881 loff_t *pos, cred_t *cr)
1883 znode_t *zp = ITOZ(ip);
1884 zfs_sb_t *zsb = ITOZSB(ip);
1887 zap_attribute_t zap;
1897 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zsb),
1898 &parent, sizeof (parent))) != 0)
1902 * Quit if directory has been removed (posix)
1909 prefetch = zp->z_zn_prefetch;
1912 * Initialize the iterator cursor.
1916 * Start iteration from the beginning of the directory.
1918 zap_cursor_init(&zc, os, zp->z_id);
1921 * The offset is a serialized cursor.
1923 zap_cursor_init_serialized(&zc, os, zp->z_id, *pos);
1927 * Transform to file-system independent format
1934 * Special case `.', `..', and `.zfs'.
1937 (void) strcpy(zap.za_name, ".");
1938 zap.za_normalization_conflict = 0;
1940 } else if (*pos == 1) {
1941 (void) strcpy(zap.za_name, "..");
1942 zap.za_normalization_conflict = 0;
1944 } else if (*pos == 2 && zfs_show_ctldir(zp)) {
1945 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1946 zap.za_normalization_conflict = 0;
1947 objnum = ZFSCTL_INO_ROOT;
1952 if ((error = zap_cursor_retrieve(&zc, &zap))) {
1953 if (error == ENOENT)
1959 if (zap.za_integer_length != 8 ||
1960 zap.za_num_integers != 1) {
1961 cmn_err(CE_WARN, "zap_readdir: bad directory "
1962 "entry, obj = %lld, offset = %lld\n",
1963 (u_longlong_t)zp->z_id,
1964 (u_longlong_t)*pos);
1969 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
1971 done = filldir(dirent, zap.za_name, strlen(zap.za_name),
1972 zap_cursor_serialize(&zc), objnum, 0);
1977 /* Prefetch znode */
1979 dmu_prefetch(os, objnum, 0, 0);
1983 zap_cursor_advance(&zc);
1984 *pos = zap_cursor_serialize(&zc);
1989 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
1992 zap_cursor_fini(&zc);
1993 if (error == ENOENT)
1996 ZFS_ACCESSTIME_STAMP(zsb, zp);
1997 zfs_inode_update(zp);
2004 EXPORT_SYMBOL(zfs_readdir);
2006 ulong_t zfs_fsync_sync_cnt = 4;
2009 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2011 znode_t *zp = ITOZ(ip);
2012 zfs_sb_t *zsb = ITOZSB(ip);
2014 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2016 if (zsb->z_os->os_sync != ZFS_SYNC_DISABLED) {
2019 zil_commit(zsb->z_log, zp->z_id);
2024 EXPORT_SYMBOL(zfs_fsync);
2028 * Get the requested file attributes and place them in the provided
2031 * IN: ip - inode of file.
2032 * stat - kstat structure to fill in.
2033 * flags - ATTR_NOACLCHECK (CIFS server context)
2034 * cr - credentials of caller.
2036 * OUT: stat - filled in kstat values.
2040 zfs_getattr(struct inode *ip, struct kstat *stat, int flags, cred_t *cr)
2042 znode_t *zp = ITOZ(ip);
2043 zfs_sb_t *zsb = ITOZSB(ip);
2046 uint64_t mtime[2], ctime[2];
2048 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2049 sa_bulk_attr_t bulk[2];
2055 zfs_fuid_map_ids(zp, cr, &stat->uid, &stat->gid);
2057 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
2058 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
2060 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2066 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2067 * Also, if we are the owner don't bother, since owner should
2068 * always be allowed to read basic attributes of file.
2070 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2071 (stat->uid != crgetuid(cr))) {
2072 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2080 * Return all attributes. It's cheaper to provide the answer
2081 * than to determine whether we were asked the question.
2084 mutex_enter(&zp->z_lock);
2085 stat->ino = ip->i_ino;
2086 stat->mode = zp->z_mode;
2087 stat->uid = zp->z_uid;
2088 stat->gid = zp->z_gid;
2089 if ((zp->z_id == zsb->z_root) && zfs_show_ctldir(zp))
2090 links = zp->z_links + 1;
2092 links = zp->z_links;
2093 stat->nlink = MIN(links, ZFS_LINK_MAX);
2094 stat->size = i_size_read(ip);
2095 stat->rdev = ip->i_rdev;
2096 stat->dev = ip->i_rdev;
2098 ZFS_TIME_DECODE(&stat->atime, zp->z_atime);
2099 ZFS_TIME_DECODE(&stat->mtime, mtime);
2100 ZFS_TIME_DECODE(&stat->ctime, ctime);
2102 mutex_exit(&zp->z_lock);
2104 sa_object_size(zp->z_sa_hdl, &blksz, &stat->blocks);
2105 stat->blksize = (1 << ip->i_blkbits);
2107 if (zp->z_blksz == 0) {
2109 * Block size hasn't been set; suggest maximal I/O transfers.
2111 stat->blksize = zsb->z_max_blksz;
2117 EXPORT_SYMBOL(zfs_getattr);
2120 * Set the file attributes to the values contained in the
2123 * IN: ip - inode of file to be modified.
2124 * vap - new attribute values.
2125 * If AT_XVATTR set, then optional attrs are being set
2126 * flags - ATTR_UTIME set if non-default time values provided.
2127 * - ATTR_NOACLCHECK (CIFS context only).
2128 * cr - credentials of caller.
2130 * RETURN: 0 if success
2131 * error code if failure
2134 * ip - ctime updated, mtime updated if size changed.
2138 zfs_setattr(struct inode *ip, struct iattr *attr, int flags, cred_t *cr)
2140 znode_t *zp = ITOZ(ip);
2141 zfs_sb_t *zsb = ITOZSB(ip);
2145 uint_t mask = attr->ia_valid;
2149 uint64_t new_uid, new_gid;
2151 uint64_t mtime[2], ctime[2];
2153 int need_policy = FALSE;
2155 zfs_fuid_info_t *fuidp = NULL;
2156 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2157 zfs_acl_t *aclp = NULL;
2158 boolean_t fuid_dirtied = B_FALSE;
2159 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2160 int count = 0, xattr_count = 0;
2171 * Make sure that if we have ephemeral uid/gid or xvattr specified
2172 * that file system is at proper version level
2174 if (zsb->z_use_fuids == B_FALSE &&
2175 (((mask & ATTR_UID) && IS_EPHEMERAL(attr->ia_uid)) ||
2176 ((mask & ATTR_GID) && IS_EPHEMERAL(attr->ia_gid)))) {
2181 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2186 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2191 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2200 /* Can this be moved to before the top label? */
2201 if (zsb->z_vfs->mnt_flags & MNT_READONLY) {
2207 * First validate permissions
2210 if (mask & ATTR_SIZE) {
2211 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2217 * XXX - Note, we are not providing any open
2218 * mode flags here (like FNDELAY), so we may
2219 * block if there are locks present... this
2220 * should be addressed in openat().
2222 /* XXX - would it be OK to generate a log record here? */
2223 err = zfs_freesp(zp, attr->ia_size, 0, 0, FALSE);
2229 /* Careful negative Linux return code here */
2230 err = -vmtruncate(ip, attr->ia_size);
2237 if (mask & (ATTR_UID|ATTR_GID)) {
2238 int idmask = (mask & (ATTR_UID|ATTR_GID));
2243 * NOTE: even if a new mode is being set,
2244 * we may clear S_ISUID/S_ISGID bits.
2247 if (!(mask & ATTR_MODE))
2248 attr->ia_mode = zp->z_mode;
2251 * Take ownership or chgrp to group we are a member of
2254 take_owner = (mask & ATTR_UID) &&
2255 (attr->ia_uid == crgetuid(cr));
2256 take_group = (mask & ATTR_GID) &&
2257 zfs_groupmember(zsb, attr->ia_gid, cr);
2260 * If both AT_UID and AT_GID are set then take_owner and
2261 * take_group must both be set in order to allow taking
2264 * Otherwise, send the check through secpolicy_vnode_setattr()
2268 if (((idmask == (ATTR_UID|ATTR_GID)) &&
2269 take_owner && take_group) ||
2270 ((idmask == ATTR_UID) && take_owner) ||
2271 ((idmask == ATTR_GID) && take_group)) {
2272 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2273 skipaclchk, cr) == 0) {
2275 * Remove setuid/setgid for non-privileged users
2277 (void) secpolicy_setid_clear(attr, cr);
2278 trim_mask = (mask & (ATTR_UID|ATTR_GID));
2287 mutex_enter(&zp->z_lock);
2288 oldva.va_mode = zp->z_mode;
2289 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2291 mutex_exit(&zp->z_lock);
2293 if (mask & ATTR_MODE) {
2294 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2295 err = secpolicy_setid_setsticky_clear(ip, attr,
2301 trim_mask |= ATTR_MODE;
2309 * If trim_mask is set then take ownership
2310 * has been granted or write_acl is present and user
2311 * has the ability to modify mode. In that case remove
2312 * UID|GID and or MODE from mask so that
2313 * secpolicy_vnode_setattr() doesn't revoke it.
2317 saved_mask = attr->ia_valid;
2318 attr->ia_valid &= ~trim_mask;
2320 err = secpolicy_vnode_setattr(cr, ip, attr, &oldva, flags,
2321 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2328 attr->ia_valid |= saved_mask;
2332 * secpolicy_vnode_setattr, or take ownership may have
2335 mask = attr->ia_valid;
2337 if ((mask & (ATTR_UID | ATTR_GID))) {
2338 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
2339 &xattr_obj, sizeof (xattr_obj));
2341 if (err == 0 && xattr_obj) {
2342 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
2346 if (mask & ATTR_UID) {
2347 new_uid = zfs_fuid_create(zsb,
2348 (uint64_t)attr->ia_uid, cr, ZFS_OWNER, &fuidp);
2349 if (new_uid != zp->z_uid &&
2350 zfs_fuid_overquota(zsb, B_FALSE, new_uid)) {
2358 if (mask & ATTR_GID) {
2359 new_gid = zfs_fuid_create(zsb, (uint64_t)attr->ia_gid,
2360 cr, ZFS_GROUP, &fuidp);
2361 if (new_gid != zp->z_gid &&
2362 zfs_fuid_overquota(zsb, B_TRUE, new_gid)) {
2370 tx = dmu_tx_create(zsb->z_os);
2372 if (mask & ATTR_MODE) {
2373 uint64_t pmode = zp->z_mode;
2375 new_mode = (pmode & S_IFMT) | (attr->ia_mode & ~S_IFMT);
2377 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
2379 mutex_enter(&zp->z_lock);
2380 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2382 * Are we upgrading ACL from old V0 format
2385 if (zsb->z_version >= ZPL_VERSION_FUID &&
2386 zfs_znode_acl_version(zp) ==
2387 ZFS_ACL_VERSION_INITIAL) {
2388 dmu_tx_hold_free(tx, acl_obj, 0,
2390 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2391 0, aclp->z_acl_bytes);
2393 dmu_tx_hold_write(tx, acl_obj, 0,
2396 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2397 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2398 0, aclp->z_acl_bytes);
2400 mutex_exit(&zp->z_lock);
2401 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2403 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2407 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
2410 fuid_dirtied = zsb->z_fuid_dirty;
2412 zfs_fuid_txhold(zsb, tx);
2414 zfs_sa_upgrade_txholds(tx, zp);
2416 err = dmu_tx_assign(tx, TXG_NOWAIT);
2418 if (err == ERESTART)
2425 * Set each attribute requested.
2426 * We group settings according to the locks they need to acquire.
2428 * Note: you cannot set ctime directly, although it will be
2429 * updated as a side-effect of calling this function.
2433 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2434 mutex_enter(&zp->z_acl_lock);
2435 mutex_enter(&zp->z_lock);
2437 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
2438 &zp->z_pflags, sizeof (zp->z_pflags));
2441 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2442 mutex_enter(&attrzp->z_acl_lock);
2443 mutex_enter(&attrzp->z_lock);
2444 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2445 SA_ZPL_FLAGS(zsb), NULL, &attrzp->z_pflags,
2446 sizeof (attrzp->z_pflags));
2449 if (mask & (ATTR_UID|ATTR_GID)) {
2451 if (mask & ATTR_UID) {
2452 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zsb), NULL,
2453 &new_uid, sizeof (new_uid));
2454 zp->z_uid = new_uid;
2456 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2457 SA_ZPL_UID(zsb), NULL, &new_uid,
2459 attrzp->z_uid = new_uid;
2463 if (mask & ATTR_GID) {
2464 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zsb),
2465 NULL, &new_gid, sizeof (new_gid));
2466 zp->z_gid = new_gid;
2468 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2469 SA_ZPL_GID(zsb), NULL, &new_gid,
2471 attrzp->z_gid = new_gid;
2474 if (!(mask & ATTR_MODE)) {
2475 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb),
2476 NULL, &new_mode, sizeof (new_mode));
2477 new_mode = zp->z_mode;
2479 err = zfs_acl_chown_setattr(zp);
2482 err = zfs_acl_chown_setattr(attrzp);
2487 if (mask & ATTR_MODE) {
2488 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb), NULL,
2489 &new_mode, sizeof (new_mode));
2490 zp->z_mode = new_mode;
2491 ASSERT3P(aclp, !=, NULL);
2492 err = zfs_aclset_common(zp, aclp, cr, tx);
2493 ASSERT3U(err, ==, 0);
2494 if (zp->z_acl_cached)
2495 zfs_acl_free(zp->z_acl_cached);
2496 zp->z_acl_cached = aclp;
2501 if (mask & ATTR_ATIME) {
2502 ZFS_TIME_ENCODE(&attr->ia_atime, zp->z_atime);
2503 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zsb), NULL,
2504 &zp->z_atime, sizeof (zp->z_atime));
2507 if (mask & ATTR_MTIME) {
2508 ZFS_TIME_ENCODE(&attr->ia_mtime, mtime);
2509 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL,
2510 mtime, sizeof (mtime));
2513 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2514 if (mask & ATTR_SIZE && !(mask & ATTR_MTIME)) {
2515 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb),
2516 NULL, mtime, sizeof (mtime));
2517 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
2518 &ctime, sizeof (ctime));
2519 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
2521 } else if (mask != 0) {
2522 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
2523 &ctime, sizeof (ctime));
2524 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
2527 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2528 SA_ZPL_CTIME(zsb), NULL,
2529 &ctime, sizeof (ctime));
2530 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
2531 mtime, ctime, B_TRUE);
2535 * Do this after setting timestamps to prevent timestamp
2536 * update from toggling bit
2540 zfs_fuid_sync(zsb, tx);
2543 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, attr, mask, fuidp);
2545 mutex_exit(&zp->z_lock);
2546 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2547 mutex_exit(&zp->z_acl_lock);
2550 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2551 mutex_exit(&attrzp->z_acl_lock);
2552 mutex_exit(&attrzp->z_lock);
2555 if (err == 0 && attrzp) {
2556 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
2567 zfs_fuid_info_free(fuidp);
2573 if (err == ERESTART)
2576 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
2578 zfs_inode_update(zp);
2582 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
2583 zil_commit(zilog, 0);
2588 EXPORT_SYMBOL(zfs_setattr);
2590 typedef struct zfs_zlock {
2591 krwlock_t *zl_rwlock; /* lock we acquired */
2592 znode_t *zl_znode; /* znode we held */
2593 struct zfs_zlock *zl_next; /* next in list */
2597 * Drop locks and release vnodes that were held by zfs_rename_lock().
2600 zfs_rename_unlock(zfs_zlock_t **zlpp)
2604 while ((zl = *zlpp) != NULL) {
2605 if (zl->zl_znode != NULL)
2606 iput(ZTOI(zl->zl_znode));
2607 rw_exit(zl->zl_rwlock);
2608 *zlpp = zl->zl_next;
2609 kmem_free(zl, sizeof (*zl));
2614 * Search back through the directory tree, using the ".." entries.
2615 * Lock each directory in the chain to prevent concurrent renames.
2616 * Fail any attempt to move a directory into one of its own descendants.
2617 * XXX - z_parent_lock can overlap with map or grow locks
2620 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2624 uint64_t rootid = ZTOZSB(zp)->z_root;
2625 uint64_t oidp = zp->z_id;
2626 krwlock_t *rwlp = &szp->z_parent_lock;
2627 krw_t rw = RW_WRITER;
2630 * First pass write-locks szp and compares to zp->z_id.
2631 * Later passes read-lock zp and compare to zp->z_parent.
2634 if (!rw_tryenter(rwlp, rw)) {
2636 * Another thread is renaming in this path.
2637 * Note that if we are a WRITER, we don't have any
2638 * parent_locks held yet.
2640 if (rw == RW_READER && zp->z_id > szp->z_id) {
2642 * Drop our locks and restart
2644 zfs_rename_unlock(&zl);
2648 rwlp = &szp->z_parent_lock;
2653 * Wait for other thread to drop its locks
2659 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2660 zl->zl_rwlock = rwlp;
2661 zl->zl_znode = NULL;
2662 zl->zl_next = *zlpp;
2665 if (oidp == szp->z_id) /* We're a descendant of szp */
2668 if (oidp == rootid) /* We've hit the top */
2671 if (rw == RW_READER) { /* i.e. not the first pass */
2672 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
2677 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
2678 &oidp, sizeof (oidp));
2679 rwlp = &zp->z_parent_lock;
2682 } while (zp->z_id != sdzp->z_id);
2688 * Move an entry from the provided source directory to the target
2689 * directory. Change the entry name as indicated.
2691 * IN: sdip - Source directory containing the "old entry".
2692 * snm - Old entry name.
2693 * tdip - Target directory to contain the "new entry".
2694 * tnm - New entry name.
2695 * cr - credentials of caller.
2696 * flags - case flags
2698 * RETURN: 0 if success
2699 * error code if failure
2702 * sdip,tdip - ctime|mtime updated
2706 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
2707 cred_t *cr, int flags)
2709 znode_t *tdzp, *szp, *tzp;
2710 znode_t *sdzp = ITOZ(sdip);
2711 zfs_sb_t *zsb = ITOZSB(sdip);
2713 zfs_dirlock_t *sdl, *tdl;
2716 int cmp, serr, terr;
2721 ZFS_VERIFY_ZP(sdzp);
2724 if (tdip->i_sb != sdip->i_sb) {
2730 ZFS_VERIFY_ZP(tdzp);
2731 if (zsb->z_utf8 && u8_validate(tnm,
2732 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2737 if (flags & FIGNORECASE)
2746 * This is to prevent the creation of links into attribute space
2747 * by renaming a linked file into/outof an attribute directory.
2748 * See the comment in zfs_link() for why this is considered bad.
2750 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
2756 * Lock source and target directory entries. To prevent deadlock,
2757 * a lock ordering must be defined. We lock the directory with
2758 * the smallest object id first, or if it's a tie, the one with
2759 * the lexically first name.
2761 if (sdzp->z_id < tdzp->z_id) {
2763 } else if (sdzp->z_id > tdzp->z_id) {
2767 * First compare the two name arguments without
2768 * considering any case folding.
2770 int nofold = (zsb->z_norm & ~U8_TEXTPREP_TOUPPER);
2772 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
2773 ASSERT(error == 0 || !zsb->z_utf8);
2776 * POSIX: "If the old argument and the new argument
2777 * both refer to links to the same existing file,
2778 * the rename() function shall return successfully
2779 * and perform no other action."
2785 * If the file system is case-folding, then we may
2786 * have some more checking to do. A case-folding file
2787 * system is either supporting mixed case sensitivity
2788 * access or is completely case-insensitive. Note
2789 * that the file system is always case preserving.
2791 * In mixed sensitivity mode case sensitive behavior
2792 * is the default. FIGNORECASE must be used to
2793 * explicitly request case insensitive behavior.
2795 * If the source and target names provided differ only
2796 * by case (e.g., a request to rename 'tim' to 'Tim'),
2797 * we will treat this as a special case in the
2798 * case-insensitive mode: as long as the source name
2799 * is an exact match, we will allow this to proceed as
2800 * a name-change request.
2802 if ((zsb->z_case == ZFS_CASE_INSENSITIVE ||
2803 (zsb->z_case == ZFS_CASE_MIXED &&
2804 flags & FIGNORECASE)) &&
2805 u8_strcmp(snm, tnm, 0, zsb->z_norm, U8_UNICODE_LATEST,
2808 * case preserving rename request, require exact
2817 * If the source and destination directories are the same, we should
2818 * grab the z_name_lock of that directory only once.
2822 rw_enter(&sdzp->z_name_lock, RW_READER);
2826 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
2827 ZEXISTS | zflg, NULL, NULL);
2828 terr = zfs_dirent_lock(&tdl,
2829 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
2831 terr = zfs_dirent_lock(&tdl,
2832 tdzp, tnm, &tzp, zflg, NULL, NULL);
2833 serr = zfs_dirent_lock(&sdl,
2834 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
2840 * Source entry invalid or not there.
2843 zfs_dirent_unlock(tdl);
2849 rw_exit(&sdzp->z_name_lock);
2851 if (strcmp(snm, "..") == 0)
2857 zfs_dirent_unlock(sdl);
2861 rw_exit(&sdzp->z_name_lock);
2863 if (strcmp(tnm, "..") == 0)
2870 * Must have write access at the source to remove the old entry
2871 * and write access at the target to create the new entry.
2872 * Note that if target and source are the same, this can be
2873 * done in a single check.
2876 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
2879 if (S_ISDIR(ZTOI(szp)->i_mode)) {
2881 * Check to make sure rename is valid.
2882 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
2884 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
2889 * Does target exist?
2893 * Source and target must be the same type.
2895 if (S_ISDIR(ZTOI(szp)->i_mode)) {
2896 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
2901 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
2907 * POSIX dictates that when the source and target
2908 * entries refer to the same file object, rename
2909 * must do nothing and exit without error.
2911 if (szp->z_id == tzp->z_id) {
2917 tx = dmu_tx_create(zsb->z_os);
2918 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
2919 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
2920 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
2921 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
2923 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
2924 zfs_sa_upgrade_txholds(tx, tdzp);
2927 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
2928 zfs_sa_upgrade_txholds(tx, tzp);
2931 zfs_sa_upgrade_txholds(tx, szp);
2932 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
2933 error = dmu_tx_assign(tx, TXG_NOWAIT);
2936 zfs_rename_unlock(&zl);
2937 zfs_dirent_unlock(sdl);
2938 zfs_dirent_unlock(tdl);
2941 rw_exit(&sdzp->z_name_lock);
2946 if (error == ERESTART) {
2956 if (tzp) /* Attempt to remove the existing target */
2957 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
2960 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
2962 szp->z_pflags |= ZFS_AV_MODIFIED;
2964 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zsb),
2965 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
2966 ASSERT3U(error, ==, 0);
2968 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
2970 zfs_log_rename(zilog, tx, TX_RENAME |
2971 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
2972 sdl->dl_name, tdzp, tdl->dl_name, szp);
2975 * At this point, we have successfully created
2976 * the target name, but have failed to remove
2977 * the source name. Since the create was done
2978 * with the ZRENAMING flag, there are
2979 * complications; for one, the link count is
2980 * wrong. The easiest way to deal with this
2981 * is to remove the newly created target, and
2982 * return the original error. This must
2983 * succeed; fortunately, it is very unlikely to
2984 * fail, since we just created it.
2986 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
2987 ZRENAMING, NULL), ==, 0);
2995 zfs_rename_unlock(&zl);
2997 zfs_dirent_unlock(sdl);
2998 zfs_dirent_unlock(tdl);
3000 zfs_inode_update(sdzp);
3002 rw_exit(&sdzp->z_name_lock);
3005 zfs_inode_update(tdzp);
3007 zfs_inode_update(szp);
3010 zfs_inode_update(tzp);
3014 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3015 zil_commit(zilog, 0);
3020 EXPORT_SYMBOL(zfs_rename);
3023 * Insert the indicated symbolic reference entry into the directory.
3025 * IN: dip - Directory to contain new symbolic link.
3026 * link - Name for new symlink entry.
3027 * vap - Attributes of new entry.
3028 * target - Target path of new symlink.
3030 * cr - credentials of caller.
3031 * flags - case flags
3033 * RETURN: 0 if success
3034 * error code if failure
3037 * dip - ctime|mtime updated
3041 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
3042 struct inode **ipp, cred_t *cr, int flags)
3044 znode_t *zp, *dzp = ITOZ(dip);
3047 zfs_sb_t *zsb = ITOZSB(dip);
3049 uint64_t len = strlen(link);
3052 zfs_acl_ids_t acl_ids;
3053 boolean_t fuid_dirtied;
3054 uint64_t txtype = TX_SYMLINK;
3056 ASSERT(S_ISLNK(vap->va_mode));
3062 if (zsb->z_utf8 && u8_validate(name, strlen(name),
3063 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3067 if (flags & FIGNORECASE)
3070 if (len > MAXPATHLEN) {
3072 return (ENAMETOOLONG);
3075 if ((error = zfs_acl_ids_create(dzp, 0,
3076 vap, cr, NULL, &acl_ids)) != 0) {
3084 * Attempt to lock directory; fail if entry already exists.
3086 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3088 zfs_acl_ids_free(&acl_ids);
3093 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3094 zfs_acl_ids_free(&acl_ids);
3095 zfs_dirent_unlock(dl);
3100 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
3101 zfs_acl_ids_free(&acl_ids);
3102 zfs_dirent_unlock(dl);
3106 tx = dmu_tx_create(zsb->z_os);
3107 fuid_dirtied = zsb->z_fuid_dirty;
3108 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3109 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3110 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3111 ZFS_SA_BASE_ATTR_SIZE + len);
3112 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3113 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3114 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3115 acl_ids.z_aclp->z_acl_bytes);
3118 zfs_fuid_txhold(zsb, tx);
3119 error = dmu_tx_assign(tx, TXG_NOWAIT);
3121 zfs_dirent_unlock(dl);
3122 if (error == ERESTART) {
3127 zfs_acl_ids_free(&acl_ids);
3134 * Create a new object for the symlink.
3135 * for version 4 ZPL datsets the symlink will be an SA attribute
3137 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3140 zfs_fuid_sync(zsb, tx);
3142 mutex_enter(&zp->z_lock);
3144 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zsb),
3147 zfs_sa_symlink(zp, link, len, tx);
3148 mutex_exit(&zp->z_lock);
3151 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
3152 &zp->z_size, sizeof (zp->z_size), tx);
3154 * Insert the new object into the directory.
3156 (void) zfs_link_create(dl, zp, tx, ZNEW);
3158 if (flags & FIGNORECASE)
3160 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3162 zfs_inode_update(dzp);
3163 zfs_inode_update(zp);
3165 zfs_acl_ids_free(&acl_ids);
3169 zfs_dirent_unlock(dl);
3173 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3174 zil_commit(zilog, 0);
3179 EXPORT_SYMBOL(zfs_symlink);
3182 * Return, in the buffer contained in the provided uio structure,
3183 * the symbolic path referred to by ip.
3185 * IN: ip - inode of symbolic link
3186 * uio - structure to contain the link path.
3187 * cr - credentials of caller.
3189 * RETURN: 0 if success
3190 * error code if failure
3193 * ip - atime updated
3197 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
3199 znode_t *zp = ITOZ(ip);
3200 zfs_sb_t *zsb = ITOZSB(ip);
3206 mutex_enter(&zp->z_lock);
3208 error = sa_lookup_uio(zp->z_sa_hdl,
3209 SA_ZPL_SYMLINK(zsb), uio);
3211 error = zfs_sa_readlink(zp, uio);
3212 mutex_exit(&zp->z_lock);
3214 ZFS_ACCESSTIME_STAMP(zsb, zp);
3215 zfs_inode_update(zp);
3219 EXPORT_SYMBOL(zfs_readlink);
3222 * Insert a new entry into directory tdip referencing sip.
3224 * IN: tdip - Directory to contain new entry.
3225 * sip - inode of new entry.
3226 * name - name of new entry.
3227 * cr - credentials of caller.
3229 * RETURN: 0 if success
3230 * error code if failure
3233 * tdip - ctime|mtime updated
3234 * sip - ctime updated
3238 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr)
3240 znode_t *dzp = ITOZ(tdip);
3242 zfs_sb_t *zsb = ITOZSB(tdip);
3251 ASSERT(S_ISDIR(tdip->i_mode));
3258 * POSIX dictates that we return EPERM here.
3259 * Better choices include ENOTSUP or EISDIR.
3261 if (S_ISDIR(sip->i_mode)) {
3266 if (sip->i_sb != tdip->i_sb) {
3274 /* Prevent links to .zfs/shares files */
3276 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zsb),
3277 &parent, sizeof (uint64_t))) != 0) {
3281 if (parent == zsb->z_shares_dir) {
3286 if (zsb->z_utf8 && u8_validate(name,
3287 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3291 #ifdef HAVE_PN_UTILS
3292 if (flags & FIGNORECASE)
3294 #endif /* HAVE_PN_UTILS */
3297 * We do not support links between attributes and non-attributes
3298 * because of the potential security risk of creating links
3299 * into "normal" file space in order to circumvent restrictions
3300 * imposed in attribute space.
3302 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3307 owner = zfs_fuid_map_id(zsb, szp->z_uid, cr, ZFS_OWNER);
3308 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3313 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3320 * Attempt to lock directory; fail if entry already exists.
3322 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3328 tx = dmu_tx_create(zsb->z_os);
3329 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3330 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3331 zfs_sa_upgrade_txholds(tx, szp);
3332 zfs_sa_upgrade_txholds(tx, dzp);
3333 error = dmu_tx_assign(tx, TXG_NOWAIT);
3335 zfs_dirent_unlock(dl);
3336 if (error == ERESTART) {
3346 error = zfs_link_create(dl, szp, tx, 0);
3349 uint64_t txtype = TX_LINK;
3350 #ifdef HAVE_PN_UTILS
3351 if (flags & FIGNORECASE)
3353 #endif /* HAVE_PN_UTILS */
3354 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3359 zfs_dirent_unlock(dl);
3361 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3362 zil_commit(zilog, 0);
3364 zfs_inode_update(dzp);
3365 zfs_inode_update(szp);
3369 EXPORT_SYMBOL(zfs_link);
3373 * zfs_null_putapage() is used when the file system has been force
3374 * unmounted. It just drops the pages.
3378 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3379 size_t *lenp, int flags, cred_t *cr)
3381 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
3386 * Push a page out to disk, klustering if possible.
3388 * IN: vp - file to push page to.
3389 * pp - page to push.
3390 * flags - additional flags.
3391 * cr - credentials of caller.
3393 * OUT: offp - start of range pushed.
3394 * lenp - len of range pushed.
3396 * RETURN: 0 if success
3397 * error code if failure
3399 * NOTE: callers must have locked the page to be pushed. On
3400 * exit, the page (and all other pages in the kluster) must be
3405 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3406 size_t *lenp, int flags, cred_t *cr)
3408 znode_t *zp = VTOZ(vp);
3409 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3411 u_offset_t off, koff;
3418 * If our blocksize is bigger than the page size, try to kluster
3419 * multiple pages so that we write a full block (thus avoiding
3420 * a read-modify-write).
3422 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
3423 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3424 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
3425 ASSERT(koff <= zp->z_size);
3426 if (koff + klen > zp->z_size)
3427 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
3428 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
3430 ASSERT3U(btop(len), ==, btopr(len));
3433 * Can't push pages past end-of-file.
3435 if (off >= zp->z_size) {
3436 /* ignore all pages */
3439 } else if (off + len > zp->z_size) {
3440 int npages = btopr(zp->z_size - off);
3443 page_list_break(&pp, &trunc, npages);
3444 /* ignore pages past end of file */
3446 pvn_write_done(trunc, flags);
3447 len = zp->z_size - off;
3450 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
3451 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
3456 tx = dmu_tx_create(zfsvfs->z_os);
3457 dmu_tx_hold_write(tx, zp->z_id, off, len);
3459 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3460 zfs_sa_upgrade_txholds(tx, zp);
3461 err = dmu_tx_assign(tx, TXG_NOWAIT);
3463 if (err == ERESTART) {
3472 if (zp->z_blksz <= PAGESIZE) {
3473 caddr_t va = zfs_map_page(pp, S_READ);
3474 ASSERT3U(len, <=, PAGESIZE);
3475 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
3476 zfs_unmap_page(pp, va);
3478 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
3482 uint64_t mtime[2], ctime[2];
3483 sa_bulk_attr_t bulk[3];
3486 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3488 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3490 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3492 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3494 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
3499 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
3509 * Copy the portion of the file indicated from pages into the file.
3510 * The pages are stored in a page list attached to the files vnode.
3512 * IN: vp - vnode of file to push page data to.
3513 * off - position in file to put data.
3514 * len - amount of data to write.
3515 * flags - flags to control the operation.
3516 * cr - credentials of caller.
3517 * ct - caller context.
3519 * RETURN: 0 if success
3520 * error code if failure
3523 * vp - ctime|mtime updated
3527 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr)
3529 znode_t *zp = VTOZ(vp);
3530 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3542 * Align this request to the file block size in case we kluster.
3543 * XXX - this can result in pretty aggresive locking, which can
3544 * impact simultanious read/write access. One option might be
3545 * to break up long requests (len == 0) into block-by-block
3546 * operations to get narrower locking.
3548 blksz = zp->z_blksz;
3550 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
3553 if (len > 0 && ISP2(blksz))
3554 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
3560 * Search the entire vp list for pages >= io_off.
3562 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
3563 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
3566 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
3568 if (off > zp->z_size) {
3569 /* past end of file */
3570 zfs_range_unlock(rl);
3575 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
3577 for (off = io_off; io_off < off + len; io_off += io_len) {
3578 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
3579 pp = page_lookup(vp, io_off,
3580 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
3582 pp = page_lookup_nowait(vp, io_off,
3583 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
3586 if (pp != NULL && pvn_getdirty(pp, flags)) {
3590 * Found a dirty page to push
3592 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
3600 zfs_range_unlock(rl);
3601 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3602 zil_commit(zfsvfs->z_log, zp->z_id);
3606 #endif /* HAVE_MMAP */
3610 zfs_inactive(struct inode *ip)
3612 znode_t *zp = ITOZ(ip);
3613 zfs_sb_t *zsb = ITOZSB(ip);
3616 #ifdef HAVE_SNAPSHOT
3617 /* Early return for snapshot inode? */
3618 #endif /* HAVE_SNAPSHOT */
3620 rw_enter(&zsb->z_teardown_inactive_lock, RW_READER);
3621 if (zp->z_sa_hdl == NULL) {
3622 rw_exit(&zsb->z_teardown_inactive_lock);
3626 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3627 dmu_tx_t *tx = dmu_tx_create(zsb->z_os);
3629 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3630 zfs_sa_upgrade_txholds(tx, zp);
3631 error = dmu_tx_assign(tx, TXG_WAIT);
3635 mutex_enter(&zp->z_lock);
3636 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zsb),
3637 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
3638 zp->z_atime_dirty = 0;
3639 mutex_exit(&zp->z_lock);
3645 rw_exit(&zsb->z_teardown_inactive_lock);
3647 EXPORT_SYMBOL(zfs_inactive);
3650 * Bounds-check the seek operation.
3652 * IN: ip - inode seeking within
3653 * ooff - old file offset
3654 * noffp - pointer to new file offset
3655 * ct - caller context
3657 * RETURN: 0 if success
3658 * EINVAL if new offset invalid
3662 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
3664 if (S_ISDIR(ip->i_mode))
3666 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3668 EXPORT_SYMBOL(zfs_seek);
3672 * Pre-filter the generic locking function to trap attempts to place
3673 * a mandatory lock on a memory mapped file.
3676 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
3677 flk_callback_t *flk_cbp, cred_t *cr)
3679 znode_t *zp = VTOZ(vp);
3680 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3686 * We are following the UFS semantics with respect to mapcnt
3687 * here: If we see that the file is mapped already, then we will
3688 * return an error, but we don't worry about races between this
3689 * function and zfs_map().
3691 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
3696 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
3700 * If we can't find a page in the cache, we will create a new page
3701 * and fill it with file data. For efficiency, we may try to fill
3702 * multiple pages at once (klustering) to fill up the supplied page
3703 * list. Note that the pages to be filled are held with an exclusive
3704 * lock to prevent access by other threads while they are being filled.
3707 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3708 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3710 znode_t *zp = VTOZ(vp);
3711 page_t *pp, *cur_pp;
3712 objset_t *os = zp->z_zfsvfs->z_os;
3713 u_offset_t io_off, total;
3717 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
3719 * We only have a single page, don't bother klustering
3723 pp = page_create_va(vp, io_off, io_len,
3724 PG_EXCL | PG_WAIT, seg, addr);
3727 * Try to find enough pages to fill the page list
3729 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3730 &io_len, off, plsz, 0);
3734 * The page already exists, nothing to do here.
3741 * Fill the pages in the kluster.
3744 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3747 ASSERT3U(io_off, ==, cur_pp->p_offset);
3748 va = zfs_map_page(cur_pp, S_WRITE);
3749 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
3751 zfs_unmap_page(cur_pp, va);
3753 /* On error, toss the entire kluster */
3754 pvn_read_done(pp, B_ERROR);
3755 /* convert checksum errors into IO errors */
3760 cur_pp = cur_pp->p_next;
3764 * Fill in the page list array from the kluster starting
3765 * from the desired offset `off'.
3766 * NOTE: the page list will always be null terminated.
3768 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3769 ASSERT(pl == NULL || (*pl)->p_offset == off);
3775 * Return pointers to the pages for the file region [off, off + len]
3776 * in the pl array. If plsz is greater than len, this function may
3777 * also return page pointers from after the specified region
3778 * (i.e. the region [off, off + plsz]). These additional pages are
3779 * only returned if they are already in the cache, or were created as
3780 * part of a klustered read.
3782 * IN: vp - vnode of file to get data from.
3783 * off - position in file to get data from.
3784 * len - amount of data to retrieve.
3785 * plsz - length of provided page list.
3786 * seg - segment to obtain pages for.
3787 * addr - virtual address of fault.
3788 * rw - mode of created pages.
3789 * cr - credentials of caller.
3790 * ct - caller context.
3792 * OUT: protp - protection mode of created pages.
3793 * pl - list of pages created.
3795 * RETURN: 0 if success
3796 * error code if failure
3799 * vp - atime updated
3803 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3804 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3805 enum seg_rw rw, cred_t *cr)
3807 znode_t *zp = VTOZ(vp);
3808 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3812 /* we do our own caching, faultahead is unnecessary */
3815 else if (len > plsz)
3818 len = P2ROUNDUP(len, PAGESIZE);
3819 ASSERT(plsz >= len);
3828 * Loop through the requested range [off, off + len) looking
3829 * for pages. If we don't find a page, we will need to create
3830 * a new page and fill it with data from the file.
3833 if (*pl = page_lookup(vp, off, SE_SHARED))
3835 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
3838 ASSERT3U((*pl)->p_offset, ==, off);
3842 ASSERT3U(len, >=, PAGESIZE);
3845 ASSERT3U(plsz, >=, PAGESIZE);
3852 * Fill out the page array with any pages already in the cache.
3855 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
3862 * Release any pages we have previously locked.
3867 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3877 * Request a memory map for a section of a file. This code interacts
3878 * with common code and the VM system as follows:
3880 * common code calls mmap(), which ends up in smmap_common()
3882 * this calls VOP_MAP(), which takes you into (say) zfs
3884 * zfs_map() calls as_map(), passing segvn_create() as the callback
3886 * segvn_create() creates the new segment and calls VOP_ADDMAP()
3888 * zfs_addmap() updates z_mapcnt
3892 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
3893 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr)
3895 znode_t *zp = VTOZ(vp);
3896 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3897 segvn_crargs_t vn_a;
3903 if ((prot & PROT_WRITE) && (zp->z_pflags &
3904 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
3909 if ((prot & (PROT_READ | PROT_EXEC)) &&
3910 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
3915 if (vp->v_flag & VNOMAP) {
3920 if (off < 0 || len > MAXOFFSET_T - off) {
3925 if (vp->v_type != VREG) {
3931 * If file is locked, disallow mapping.
3933 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
3939 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
3947 vn_a.offset = (u_offset_t)off;
3948 vn_a.type = flags & MAP_TYPE;
3950 vn_a.maxprot = maxprot;
3953 vn_a.flags = flags & ~MAP_TYPE;
3955 vn_a.lgrp_mem_policy_flags = 0;
3957 error = as_map(as, *addrp, len, segvn_create, &vn_a);
3966 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
3967 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr)
3969 uint64_t pages = btopr(len);
3971 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
3976 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
3977 * more accurate mtime for the associated file. Since we don't have a way of
3978 * detecting when the data was actually modified, we have to resort to
3979 * heuristics. If an explicit msync() is done, then we mark the mtime when the
3980 * last page is pushed. The problem occurs when the msync() call is omitted,
3981 * which by far the most common case:
3989 * putpage() via fsflush
3991 * If we wait until fsflush to come along, we can have a modification time that
3992 * is some arbitrary point in the future. In order to prevent this in the
3993 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
3998 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
3999 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr)
4001 uint64_t pages = btopr(len);
4003 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4004 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4006 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4007 vn_has_cached_data(vp))
4008 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4012 #endif /* HAVE_MMAP */
4015 * convoff - converts the given data (start, whence) to the
4019 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4024 if ((lckdat->l_whence == 2) || (whence == 2)) {
4025 if ((error = zfs_getattr(ip, &stat, 0, CRED()) != 0))
4029 switch (lckdat->l_whence) {
4031 lckdat->l_start += offset;
4034 lckdat->l_start += stat.size;
4042 if (lckdat->l_start < 0)
4047 lckdat->l_start -= offset;
4050 lckdat->l_start -= stat.size;
4058 lckdat->l_whence = (short)whence;
4063 * Free or allocate space in a file. Currently, this function only
4064 * supports the `F_FREESP' command. However, this command is somewhat
4065 * misnamed, as its functionality includes the ability to allocate as
4066 * well as free space.
4068 * IN: ip - inode of file to free data in.
4069 * cmd - action to take (only F_FREESP supported).
4070 * bfp - section of file to free/alloc.
4071 * flag - current file open mode flags.
4072 * offset - current file offset.
4073 * cr - credentials of caller [UNUSED].
4075 * RETURN: 0 if success
4076 * error code if failure
4079 * ip - ctime|mtime updated
4083 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4084 offset_t offset, cred_t *cr)
4086 znode_t *zp = ITOZ(ip);
4087 zfs_sb_t *zsb = ITOZSB(ip);
4094 if (cmd != F_FREESP) {
4099 if ((error = convoff(ip, bfp, 0, offset))) {
4104 if (bfp->l_len < 0) {
4110 len = bfp->l_len; /* 0 means from off to end of file */
4112 error = zfs_freesp(zp, off, len, flag, TRUE);
4117 EXPORT_SYMBOL(zfs_space);
4121 zfs_fid(struct inode *ip, fid_t *fidp)
4123 znode_t *zp = ITOZ(ip);
4124 zfs_sb_t *zsb = ITOZSB(ip);
4127 uint64_t object = zp->z_id;
4134 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zsb),
4135 &gen64, sizeof (uint64_t))) != 0) {
4140 gen = (uint32_t)gen64;
4142 size = (zsb->z_parent != zsb) ? LONG_FID_LEN : SHORT_FID_LEN;
4143 if (fidp->fid_len < size) {
4144 fidp->fid_len = size;
4149 zfid = (zfid_short_t *)fidp;
4151 zfid->zf_len = size;
4153 for (i = 0; i < sizeof (zfid->zf_object); i++)
4154 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4156 /* Must have a non-zero generation number to distinguish from .zfs */
4159 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4160 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4162 if (size == LONG_FID_LEN) {
4163 uint64_t objsetid = dmu_objset_id(zsb->z_os);
4166 zlfid = (zfid_long_t *)fidp;
4168 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4169 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4171 /* XXX - this should be the generation number for the objset */
4172 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4173 zlfid->zf_setgen[i] = 0;
4179 EXPORT_SYMBOL(zfs_fid);
4183 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4185 znode_t *zp = ITOZ(ip);
4186 zfs_sb_t *zsb = ITOZSB(ip);
4188 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4192 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4197 EXPORT_SYMBOL(zfs_getsecattr);
4201 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4203 znode_t *zp = ITOZ(ip);
4204 zfs_sb_t *zsb = ITOZSB(ip);
4206 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4207 zilog_t *zilog = zsb->z_log;
4212 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4214 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
4215 zil_commit(zilog, 0);
4220 EXPORT_SYMBOL(zfs_setsecattr);
4222 #ifdef HAVE_UIO_ZEROCOPY
4224 * Tunable, both must be a power of 2.
4226 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4227 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4228 * an arcbuf for a partial block read
4230 int zcr_blksz_min = (1 << 10); /* 1K */
4231 int zcr_blksz_max = (1 << 17); /* 128K */
4235 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
4237 znode_t *zp = ITOZ(ip);
4238 zfs_sb_t *zsb = ITOZSB(ip);
4239 int max_blksz = zsb->z_max_blksz;
4240 uio_t *uio = &xuio->xu_uio;
4241 ssize_t size = uio->uio_resid;
4242 offset_t offset = uio->uio_loffset;
4247 int preamble, postamble;
4249 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4257 * Loan out an arc_buf for write if write size is bigger than
4258 * max_blksz, and the file's block size is also max_blksz.
4261 if (size < blksz || zp->z_blksz != blksz) {
4266 * Caller requests buffers for write before knowing where the
4267 * write offset might be (e.g. NFS TCP write).
4272 preamble = P2PHASE(offset, blksz);
4274 preamble = blksz - preamble;
4279 postamble = P2PHASE(size, blksz);
4282 fullblk = size / blksz;
4283 (void) dmu_xuio_init(xuio,
4284 (preamble != 0) + fullblk + (postamble != 0));
4287 * Have to fix iov base/len for partial buffers. They
4288 * currently represent full arc_buf's.
4291 /* data begins in the middle of the arc_buf */
4292 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4295 (void) dmu_xuio_add(xuio, abuf,
4296 blksz - preamble, preamble);
4299 for (i = 0; i < fullblk; i++) {
4300 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4303 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
4307 /* data ends in the middle of the arc_buf */
4308 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4311 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
4316 * Loan out an arc_buf for read if the read size is larger than
4317 * the current file block size. Block alignment is not
4318 * considered. Partial arc_buf will be loaned out for read.
4320 blksz = zp->z_blksz;
4321 if (blksz < zcr_blksz_min)
4322 blksz = zcr_blksz_min;
4323 if (blksz > zcr_blksz_max)
4324 blksz = zcr_blksz_max;
4325 /* avoid potential complexity of dealing with it */
4326 if (blksz > max_blksz) {
4331 maxsize = zp->z_size - uio->uio_loffset;
4345 uio->uio_extflg = UIO_XUIO;
4346 XUIO_XUZC_RW(xuio) = ioflag;
4353 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
4357 int ioflag = XUIO_XUZC_RW(xuio);
4359 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
4361 i = dmu_xuio_cnt(xuio);
4363 abuf = dmu_xuio_arcbuf(xuio, i);
4365 * if abuf == NULL, it must be a write buffer
4366 * that has been returned in zfs_write().
4369 dmu_return_arcbuf(abuf);
4370 ASSERT(abuf || ioflag == UIO_WRITE);
4373 dmu_xuio_fini(xuio);
4376 #endif /* HAVE_UIO_ZEROCOPY */