4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
26 /* Portions Copyright 2007 Jeremy Teo */
27 /* Portions Copyright 2010 Robert Milkowski */
30 #include <sys/types.h>
31 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
37 #include <sys/vfs_opreg.h>
41 #include <sys/taskq.h>
43 #include <sys/vmsystm.h>
44 #include <sys/atomic.h>
46 #include <sys/pathname.h>
47 #include <sys/cmn_err.h>
48 #include <sys/errno.h>
49 #include <sys/unistd.h>
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_acl.h>
52 #include <sys/zfs_ioctl.h>
53 #include <sys/fs/zfs.h>
55 #include <sys/dmu_objset.h>
61 #include <sys/dirent.h>
62 #include <sys/policy.h>
63 #include <sys/sunddi.h>
66 #include "fs/fs_subr.h"
67 #include <sys/zfs_ctldir.h>
68 #include <sys/zfs_fuid.h>
69 #include <sys/zfs_sa.h>
70 #include <sys/zfs_vnops.h>
72 #include <sys/zfs_rlock.h>
73 #include <sys/extdirent.h>
74 #include <sys/kidmap.h>
82 * Each vnode op performs some logical unit of work. To do this, the ZPL must
83 * properly lock its in-core state, create a DMU transaction, do the work,
84 * record this work in the intent log (ZIL), commit the DMU transaction,
85 * and wait for the intent log to commit if it is a synchronous operation.
86 * Moreover, the vnode ops must work in both normal and log replay context.
87 * The ordering of events is important to avoid deadlocks and references
88 * to freed memory. The example below illustrates the following Big Rules:
90 * (1) A check must be made in each zfs thread for a mounted file system.
91 * This is done avoiding races using ZFS_ENTER(zsb).
92 * A ZFS_EXIT(zsb) is needed before all returns. Any znodes
93 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
94 * can return EIO from the calling function.
96 * (2) iput() should always be the last thing except for zil_commit()
97 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
98 * First, if it's the last reference, the vnode/znode
99 * can be freed, so the zp may point to freed memory. Second, the last
100 * reference will call zfs_zinactive(), which may induce a lot of work --
101 * pushing cached pages (which acquires range locks) and syncing out
102 * cached atime changes. Third, zfs_zinactive() may require a new tx,
103 * which could deadlock the system if you were already holding one.
104 * If you must call iput() within a tx then use zfs_iput_async().
106 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
107 * as they can span dmu_tx_assign() calls.
109 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
110 * dmu_tx_assign(). This is critical because we don't want to block
111 * while holding locks.
113 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
114 * reduces lock contention and CPU usage when we must wait (note that if
115 * throughput is constrained by the storage, nearly every transaction
118 * Note, in particular, that if a lock is sometimes acquired before
119 * the tx assigns, and sometimes after (e.g. z_lock), then failing
120 * to use a non-blocking assign can deadlock the system. The scenario:
122 * Thread A has grabbed a lock before calling dmu_tx_assign().
123 * Thread B is in an already-assigned tx, and blocks for this lock.
124 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
125 * forever, because the previous txg can't quiesce until B's tx commits.
127 * If dmu_tx_assign() returns ERESTART and zsb->z_assign is TXG_NOWAIT,
128 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
129 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
130 * to indicate that this operation has already called dmu_tx_wait().
131 * This will ensure that we don't retry forever, waiting a short bit
134 * (5) If the operation succeeded, generate the intent log entry for it
135 * before dropping locks. This ensures that the ordering of events
136 * in the intent log matches the order in which they actually occurred.
137 * During ZIL replay the zfs_log_* functions will update the sequence
138 * number to indicate the zil transaction has replayed.
140 * (6) At the end of each vnode op, the DMU tx must always commit,
141 * regardless of whether there were any errors.
143 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
144 * to ensure that synchronous semantics are provided when necessary.
146 * In general, this is how things should be ordered in each vnode op:
148 * ZFS_ENTER(zsb); // exit if unmounted
150 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
151 * rw_enter(...); // grab any other locks you need
152 * tx = dmu_tx_create(...); // get DMU tx
153 * dmu_tx_hold_*(); // hold each object you might modify
154 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
156 * rw_exit(...); // drop locks
157 * zfs_dirent_unlock(dl); // unlock directory entry
158 * iput(...); // release held vnodes
159 * if (error == ERESTART) {
165 * dmu_tx_abort(tx); // abort DMU tx
166 * ZFS_EXIT(zsb); // finished in zfs
167 * return (error); // really out of space
169 * error = do_real_work(); // do whatever this VOP does
171 * zfs_log_*(...); // on success, make ZIL entry
172 * dmu_tx_commit(tx); // commit DMU tx -- error or not
173 * rw_exit(...); // drop locks
174 * zfs_dirent_unlock(dl); // unlock directory entry
175 * iput(...); // release held vnodes
176 * zil_commit(zilog, foid); // synchronous when necessary
177 * ZFS_EXIT(zsb); // finished in zfs
178 * return (error); // done, report error
182 * Virus scanning is unsupported. It would be possible to add a hook
183 * here to performance the required virus scan. This could be done
184 * entirely in the kernel or potentially as an update to invoke a
188 zfs_vscan(struct inode *ip, cred_t *cr, int async)
195 zfs_open(struct inode *ip, int mode, int flag, cred_t *cr)
197 znode_t *zp = ITOZ(ip);
198 zfs_sb_t *zsb = ITOZSB(ip);
203 /* Honor ZFS_APPENDONLY file attribute */
204 if ((mode & FMODE_WRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
205 ((flag & O_APPEND) == 0)) {
207 return (SET_ERROR(EPERM));
210 /* Virus scan eligible files on open */
211 if (!zfs_has_ctldir(zp) && zsb->z_vscan && S_ISREG(ip->i_mode) &&
212 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
213 if (zfs_vscan(ip, cr, 0) != 0) {
215 return (SET_ERROR(EACCES));
219 /* Keep a count of the synchronous opens in the znode */
221 atomic_inc_32(&zp->z_sync_cnt);
226 EXPORT_SYMBOL(zfs_open);
230 zfs_close(struct inode *ip, int flag, cred_t *cr)
232 znode_t *zp = ITOZ(ip);
233 zfs_sb_t *zsb = ITOZSB(ip);
238 /* Decrement the synchronous opens in the znode */
240 atomic_dec_32(&zp->z_sync_cnt);
242 if (!zfs_has_ctldir(zp) && zsb->z_vscan && S_ISREG(ip->i_mode) &&
243 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
244 VERIFY(zfs_vscan(ip, cr, 1) == 0);
249 EXPORT_SYMBOL(zfs_close);
251 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
253 * Lseek support for finding holes (cmd == SEEK_HOLE) and
254 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
257 zfs_holey_common(struct inode *ip, int cmd, loff_t *off)
259 znode_t *zp = ITOZ(ip);
260 uint64_t noff = (uint64_t)*off; /* new offset */
265 file_sz = zp->z_size;
266 if (noff >= file_sz) {
267 return (SET_ERROR(ENXIO));
270 if (cmd == SEEK_HOLE)
275 error = dmu_offset_next(ZTOZSB(zp)->z_os, zp->z_id, hole, &noff);
278 if ((error == ESRCH) || (noff > file_sz)) {
280 * Handle the virtual hole at the end of file.
286 return (SET_ERROR(ENXIO));
296 zfs_holey(struct inode *ip, int cmd, loff_t *off)
298 znode_t *zp = ITOZ(ip);
299 zfs_sb_t *zsb = ITOZSB(ip);
305 error = zfs_holey_common(ip, cmd, off);
310 EXPORT_SYMBOL(zfs_holey);
311 #endif /* SEEK_HOLE && SEEK_DATA */
315 * When a file is memory mapped, we must keep the IO data synchronized
316 * between the DMU cache and the memory mapped pages. What this means:
318 * On Write: If we find a memory mapped page, we write to *both*
319 * the page and the dmu buffer.
322 update_pages(struct inode *ip, int64_t start, int len,
323 objset_t *os, uint64_t oid)
325 struct address_space *mp = ip->i_mapping;
331 off = start & (PAGE_CACHE_SIZE-1);
332 for (start &= PAGE_CACHE_MASK; len > 0; start += PAGE_CACHE_SIZE) {
333 nbytes = MIN(PAGE_CACHE_SIZE - off, len);
335 pp = find_lock_page(mp, start >> PAGE_CACHE_SHIFT);
337 if (mapping_writably_mapped(mp))
338 flush_dcache_page(pp);
341 (void) dmu_read(os, oid, start+off, nbytes, pb+off,
345 if (mapping_writably_mapped(mp))
346 flush_dcache_page(pp);
348 mark_page_accessed(pp);
352 page_cache_release(pp);
361 * When a file is memory mapped, we must keep the IO data synchronized
362 * between the DMU cache and the memory mapped pages. What this means:
364 * On Read: We "read" preferentially from memory mapped pages,
365 * else we default from the dmu buffer.
367 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
368 * the file is memory mapped.
371 mappedread(struct inode *ip, int nbytes, uio_t *uio)
373 struct address_space *mp = ip->i_mapping;
375 znode_t *zp = ITOZ(ip);
376 objset_t *os = ITOZSB(ip)->z_os;
383 start = uio->uio_loffset;
384 off = start & (PAGE_CACHE_SIZE-1);
385 for (start &= PAGE_CACHE_MASK; len > 0; start += PAGE_CACHE_SIZE) {
386 bytes = MIN(PAGE_CACHE_SIZE - off, len);
388 pp = find_lock_page(mp, start >> PAGE_CACHE_SHIFT);
390 ASSERT(PageUptodate(pp));
393 error = uiomove(pb + off, bytes, UIO_READ, uio);
396 if (mapping_writably_mapped(mp))
397 flush_dcache_page(pp);
399 mark_page_accessed(pp);
401 page_cache_release(pp);
403 error = dmu_read_uio(os, zp->z_id, uio, bytes);
415 unsigned long zfs_read_chunk_size = 1024 * 1024; /* Tunable */
418 * Read bytes from specified file into supplied buffer.
420 * IN: ip - inode of file to be read from.
421 * uio - structure supplying read location, range info,
423 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
424 * O_DIRECT flag; used to bypass page cache.
425 * cr - credentials of caller.
427 * OUT: uio - updated offset and range, buffer filled.
429 * RETURN: 0 on success, error code on failure.
432 * inode - atime updated if byte count > 0
436 zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
438 znode_t *zp = ITOZ(ip);
439 zfs_sb_t *zsb = ITOZSB(ip);
444 #ifdef HAVE_UIO_ZEROCOPY
446 #endif /* HAVE_UIO_ZEROCOPY */
452 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
454 return (SET_ERROR(EACCES));
458 * Validate file offset
460 if (uio->uio_loffset < (offset_t)0) {
462 return (SET_ERROR(EINVAL));
466 * Fasttrack empty reads
468 if (uio->uio_resid == 0) {
474 * Check for mandatory locks
476 if (mandatory_lock(ip) &&
477 !lock_may_read(ip, uio->uio_loffset, uio->uio_resid)) {
479 return (SET_ERROR(EAGAIN));
483 * If we're in FRSYNC mode, sync out this znode before reading it.
485 if (ioflag & FRSYNC || zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
486 zil_commit(zsb->z_log, zp->z_id);
489 * Lock the range against changes.
491 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
494 * If we are reading past end-of-file we can skip
495 * to the end; but we might still need to set atime.
497 if (uio->uio_loffset >= zp->z_size) {
502 ASSERT(uio->uio_loffset < zp->z_size);
503 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
505 #ifdef HAVE_UIO_ZEROCOPY
506 if ((uio->uio_extflg == UIO_XUIO) &&
507 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
509 int blksz = zp->z_blksz;
510 uint64_t offset = uio->uio_loffset;
512 xuio = (xuio_t *)uio;
514 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
517 ASSERT(offset + n <= blksz);
520 (void) dmu_xuio_init(xuio, nblk);
522 if (vn_has_cached_data(ip)) {
524 * For simplicity, we always allocate a full buffer
525 * even if we only expect to read a portion of a block.
527 while (--nblk >= 0) {
528 (void) dmu_xuio_add(xuio,
529 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
534 #endif /* HAVE_UIO_ZEROCOPY */
537 nbytes = MIN(n, zfs_read_chunk_size -
538 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
540 if (zp->z_is_mapped && !(ioflag & O_DIRECT))
541 error = mappedread(ip, nbytes, uio);
543 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
546 /* convert checksum errors into IO errors */
548 error = SET_ERROR(EIO);
555 zfs_range_unlock(rl);
557 ZFS_ACCESSTIME_STAMP(zsb, zp);
561 EXPORT_SYMBOL(zfs_read);
564 * Write the bytes to a file.
566 * IN: ip - inode of file to be written to.
567 * uio - structure supplying write location, range info,
569 * ioflag - FAPPEND flag set if in append mode.
570 * O_DIRECT flag; used to bypass page cache.
571 * cr - credentials of caller.
573 * OUT: uio - updated offset and range.
575 * RETURN: 0 if success
576 * error code if failure
579 * ip - ctime|mtime updated if byte count > 0
584 zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
586 znode_t *zp = ITOZ(ip);
587 rlim64_t limit = uio->uio_limit;
588 ssize_t start_resid = uio->uio_resid;
592 zfs_sb_t *zsb = ZTOZSB(zp);
597 int max_blksz = zsb->z_max_blksz;
600 iovec_t *aiov = NULL;
603 iovec_t *iovp = uio->uio_iov;
606 sa_bulk_attr_t bulk[4];
607 uint64_t mtime[2], ctime[2];
608 ASSERTV(int iovcnt = uio->uio_iovcnt);
611 * Fasttrack empty write
617 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
623 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
624 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
625 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb), NULL, &zp->z_size, 8);
626 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
630 * If immutable or not appending then return EPERM
632 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
633 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
634 (uio->uio_loffset < zp->z_size))) {
636 return (SET_ERROR(EPERM));
642 * Validate file offset
644 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
647 return (SET_ERROR(EINVAL));
651 * Check for mandatory locks before calling zfs_range_lock()
652 * in order to prevent a deadlock with locks set via fcntl().
654 if (mandatory_lock(ip) && !lock_may_write(ip, woff, n)) {
656 return (SET_ERROR(EAGAIN));
660 * Pre-fault the pages to ensure slow (eg NFS) pages
662 * Skip this if uio contains loaned arc_buf.
664 #ifdef HAVE_UIO_ZEROCOPY
665 if ((uio->uio_extflg == UIO_XUIO) &&
666 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
667 xuio = (xuio_t *)uio;
670 uio_prefaultpages(MIN(n, max_blksz), uio);
673 * If in append mode, set the io offset pointer to eof.
675 if (ioflag & FAPPEND) {
677 * Obtain an appending range lock to guarantee file append
678 * semantics. We reset the write offset once we have the lock.
680 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
682 if (rl->r_len == UINT64_MAX) {
684 * We overlocked the file because this write will cause
685 * the file block size to increase.
686 * Note that zp_size cannot change with this lock held.
690 uio->uio_loffset = woff;
693 * Note that if the file block size will change as a result of
694 * this write, then this range lock will lock the entire file
695 * so that we can re-write the block safely.
697 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
701 zfs_range_unlock(rl);
703 return (SET_ERROR(EFBIG));
706 if ((woff + n) > limit || woff > (limit - n))
709 /* Will this write extend the file length? */
710 write_eof = (woff + n > zp->z_size);
712 end_size = MAX(zp->z_size, woff + n);
715 * Write the file in reasonable size chunks. Each chunk is written
716 * in a separate transaction; this keeps the intent log records small
717 * and allows us to do more fine-grained space accounting.
721 woff = uio->uio_loffset;
722 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
723 zfs_owner_overquota(zsb, zp, B_TRUE)) {
725 dmu_return_arcbuf(abuf);
726 error = SET_ERROR(EDQUOT);
730 if (xuio && abuf == NULL) {
731 ASSERT(i_iov < iovcnt);
733 abuf = dmu_xuio_arcbuf(xuio, i_iov);
734 dmu_xuio_clear(xuio, i_iov);
735 ASSERT((aiov->iov_base == abuf->b_data) ||
736 ((char *)aiov->iov_base - (char *)abuf->b_data +
737 aiov->iov_len == arc_buf_size(abuf)));
739 } else if (abuf == NULL && n >= max_blksz &&
740 woff >= zp->z_size &&
741 P2PHASE(woff, max_blksz) == 0 &&
742 zp->z_blksz == max_blksz) {
744 * This write covers a full block. "Borrow" a buffer
745 * from the dmu so that we can fill it before we enter
746 * a transaction. This avoids the possibility of
747 * holding up the transaction if the data copy hangs
748 * up on a pagefault (e.g., from an NFS server mapping).
752 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
754 ASSERT(abuf != NULL);
755 ASSERT(arc_buf_size(abuf) == max_blksz);
756 if ((error = uiocopy(abuf->b_data, max_blksz,
757 UIO_WRITE, uio, &cbytes))) {
758 dmu_return_arcbuf(abuf);
761 ASSERT(cbytes == max_blksz);
765 * Start a transaction.
767 tx = dmu_tx_create(zsb->z_os);
768 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
769 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
770 zfs_sa_upgrade_txholds(tx, zp);
771 error = dmu_tx_assign(tx, TXG_WAIT);
775 dmu_return_arcbuf(abuf);
780 * If zfs_range_lock() over-locked we grow the blocksize
781 * and then reduce the lock range. This will only happen
782 * on the first iteration since zfs_range_reduce() will
783 * shrink down r_len to the appropriate size.
785 if (rl->r_len == UINT64_MAX) {
788 if (zp->z_blksz > max_blksz) {
789 ASSERT(!ISP2(zp->z_blksz));
790 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
792 new_blksz = MIN(end_size, max_blksz);
794 zfs_grow_blocksize(zp, new_blksz, tx);
795 zfs_range_reduce(rl, woff, n);
799 * XXX - should we really limit each write to z_max_blksz?
800 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
802 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
805 tx_bytes = uio->uio_resid;
806 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
808 tx_bytes -= uio->uio_resid;
811 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
813 * If this is not a full block write, but we are
814 * extending the file past EOF and this data starts
815 * block-aligned, use assign_arcbuf(). Otherwise,
816 * write via dmu_write().
818 if (tx_bytes < max_blksz && (!write_eof ||
819 aiov->iov_base != abuf->b_data)) {
821 dmu_write(zsb->z_os, zp->z_id, woff,
822 aiov->iov_len, aiov->iov_base, tx);
823 dmu_return_arcbuf(abuf);
824 xuio_stat_wbuf_copied();
826 ASSERT(xuio || tx_bytes == max_blksz);
827 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
830 ASSERT(tx_bytes <= uio->uio_resid);
831 uioskip(uio, tx_bytes);
834 if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT))
835 update_pages(ip, woff, tx_bytes, zsb->z_os, zp->z_id);
838 * If we made no progress, we're done. If we made even
839 * partial progress, update the znode and ZIL accordingly.
842 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
843 (void *)&zp->z_size, sizeof (uint64_t), tx);
850 * Clear Set-UID/Set-GID bits on successful write if not
851 * privileged and at least one of the excute bits is set.
853 * It would be nice to to this after all writes have
854 * been done, but that would still expose the ISUID/ISGID
855 * to another app after the partial write is committed.
857 * Note: we don't call zfs_fuid_map_id() here because
858 * user 0 is not an ephemeral uid.
860 mutex_enter(&zp->z_acl_lock);
861 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
862 (S_IXUSR >> 6))) != 0 &&
863 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
864 secpolicy_vnode_setid_retain(cr,
865 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
867 zp->z_mode &= ~(S_ISUID | S_ISGID);
868 newmode = zp->z_mode;
869 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zsb),
870 (void *)&newmode, sizeof (uint64_t), tx);
872 mutex_exit(&zp->z_acl_lock);
874 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
878 * Update the file size (zp_size) if it has changed;
879 * account for possible concurrent updates.
881 while ((end_size = zp->z_size) < uio->uio_loffset) {
882 (void) atomic_cas_64(&zp->z_size, end_size,
887 * If we are replaying and eof is non zero then force
888 * the file size to the specified eof. Note, there's no
889 * concurrency during replay.
891 if (zsb->z_replay && zsb->z_replay_eof != 0)
892 zp->z_size = zsb->z_replay_eof;
894 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
896 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag,
902 ASSERT(tx_bytes == nbytes);
906 uio_prefaultpages(MIN(n, max_blksz), uio);
909 zfs_range_unlock(rl);
912 * If we're in replay mode, or we made no progress, return error.
913 * Otherwise, it's at least a partial write, so it's successful.
915 if (zsb->z_replay || uio->uio_resid == start_resid) {
920 if (ioflag & (FSYNC | FDSYNC) ||
921 zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
922 zil_commit(zilog, zp->z_id);
924 zfs_inode_update(zp);
928 EXPORT_SYMBOL(zfs_write);
931 zfs_iput_async(struct inode *ip)
933 objset_t *os = ITOZSB(ip)->z_os;
935 ASSERT(atomic_read(&ip->i_count) > 0);
938 if (atomic_read(&ip->i_count) == 1)
939 taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os)),
940 (task_func_t *)iput, ip, TQ_PUSHPAGE);
946 zfs_get_done(zgd_t *zgd, int error)
948 znode_t *zp = zgd->zgd_private;
951 dmu_buf_rele(zgd->zgd_db, zgd);
953 zfs_range_unlock(zgd->zgd_rl);
956 * Release the vnode asynchronously as we currently have the
957 * txg stopped from syncing.
959 zfs_iput_async(ZTOI(zp));
961 if (error == 0 && zgd->zgd_bp)
962 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
964 kmem_free(zgd, sizeof (zgd_t));
968 static int zil_fault_io = 0;
972 * Get data to generate a TX_WRITE intent log record.
975 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
978 objset_t *os = zsb->z_os;
980 uint64_t object = lr->lr_foid;
981 uint64_t offset = lr->lr_offset;
982 uint64_t size = lr->lr_length;
983 blkptr_t *bp = &lr->lr_blkptr;
992 * Nothing to do if the file has been removed
994 if (zfs_zget(zsb, object, &zp) != 0)
995 return (SET_ERROR(ENOENT));
996 if (zp->z_unlinked) {
998 * Release the vnode asynchronously as we currently have the
999 * txg stopped from syncing.
1001 zfs_iput_async(ZTOI(zp));
1002 return (SET_ERROR(ENOENT));
1005 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_PUSHPAGE);
1006 zgd->zgd_zilog = zsb->z_log;
1007 zgd->zgd_private = zp;
1010 * Write records come in two flavors: immediate and indirect.
1011 * For small writes it's cheaper to store the data with the
1012 * log record (immediate); for large writes it's cheaper to
1013 * sync the data and get a pointer to it (indirect) so that
1014 * we don't have to write the data twice.
1016 if (buf != NULL) { /* immediate write */
1017 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1018 /* test for truncation needs to be done while range locked */
1019 if (offset >= zp->z_size) {
1020 error = SET_ERROR(ENOENT);
1022 error = dmu_read(os, object, offset, size, buf,
1023 DMU_READ_NO_PREFETCH);
1025 ASSERT(error == 0 || error == ENOENT);
1026 } else { /* indirect write */
1028 * Have to lock the whole block to ensure when it's
1029 * written out and it's checksum is being calculated
1030 * that no one can change the data. We need to re-check
1031 * blocksize after we get the lock in case it's changed!
1036 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1038 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1040 if (zp->z_blksz == size)
1043 zfs_range_unlock(zgd->zgd_rl);
1045 /* test for truncation needs to be done while range locked */
1046 if (lr->lr_offset >= zp->z_size)
1047 error = SET_ERROR(ENOENT);
1050 error = SET_ERROR(EIO);
1055 error = dmu_buf_hold(os, object, offset, zgd, &db,
1056 DMU_READ_NO_PREFETCH);
1059 blkptr_t *obp = dmu_buf_get_blkptr(db);
1061 ASSERT(BP_IS_HOLE(bp));
1068 ASSERT(db->db_offset == offset);
1069 ASSERT(db->db_size == size);
1071 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1073 ASSERT(error || lr->lr_length <= zp->z_blksz);
1076 * On success, we need to wait for the write I/O
1077 * initiated by dmu_sync() to complete before we can
1078 * release this dbuf. We will finish everything up
1079 * in the zfs_get_done() callback.
1084 if (error == EALREADY) {
1085 lr->lr_common.lrc_txtype = TX_WRITE2;
1091 zfs_get_done(zgd, error);
1098 zfs_access(struct inode *ip, int mode, int flag, cred_t *cr)
1100 znode_t *zp = ITOZ(ip);
1101 zfs_sb_t *zsb = ITOZSB(ip);
1107 if (flag & V_ACE_MASK)
1108 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1110 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1115 EXPORT_SYMBOL(zfs_access);
1118 * Lookup an entry in a directory, or an extended attribute directory.
1119 * If it exists, return a held inode reference for it.
1121 * IN: dip - inode of directory to search.
1122 * nm - name of entry to lookup.
1123 * flags - LOOKUP_XATTR set if looking for an attribute.
1124 * cr - credentials of caller.
1125 * direntflags - directory lookup flags
1126 * realpnp - returned pathname.
1128 * OUT: ipp - inode of located entry, NULL if not found.
1130 * RETURN: 0 on success, error code on failure.
1137 zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
1138 cred_t *cr, int *direntflags, pathname_t *realpnp)
1140 znode_t *zdp = ITOZ(dip);
1141 zfs_sb_t *zsb = ITOZSB(dip);
1145 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1147 if (!S_ISDIR(dip->i_mode)) {
1148 return (SET_ERROR(ENOTDIR));
1149 } else if (zdp->z_sa_hdl == NULL) {
1150 return (SET_ERROR(EIO));
1153 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1154 error = zfs_fastaccesschk_execute(zdp, cr);
1163 vnode_t *tvp = dnlc_lookup(dvp, nm);
1166 error = zfs_fastaccesschk_execute(zdp, cr);
1171 if (tvp == DNLC_NO_VNODE) {
1173 return (SET_ERROR(ENOENT));
1176 return (specvp_check(vpp, cr));
1179 #endif /* HAVE_DNLC */
1188 if (flags & LOOKUP_XATTR) {
1190 * We don't allow recursive attributes..
1191 * Maybe someday we will.
1193 if (zdp->z_pflags & ZFS_XATTR) {
1195 return (SET_ERROR(EINVAL));
1198 if ((error = zfs_get_xattrdir(zdp, ipp, cr, flags))) {
1204 * Do we have permission to get into attribute directory?
1207 if ((error = zfs_zaccess(ITOZ(*ipp), ACE_EXECUTE, 0,
1217 if (!S_ISDIR(dip->i_mode)) {
1219 return (SET_ERROR(ENOTDIR));
1223 * Check accessibility of directory.
1226 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
1231 if (zsb->z_utf8 && u8_validate(nm, strlen(nm),
1232 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1234 return (SET_ERROR(EILSEQ));
1237 error = zfs_dirlook(zdp, nm, ipp, flags, direntflags, realpnp);
1238 if ((error == 0) && (*ipp))
1239 zfs_inode_update(ITOZ(*ipp));
1244 EXPORT_SYMBOL(zfs_lookup);
1247 * Attempt to create a new entry in a directory. If the entry
1248 * already exists, truncate the file if permissible, else return
1249 * an error. Return the ip of the created or trunc'd file.
1251 * IN: dip - inode of directory to put new file entry in.
1252 * name - name of new file entry.
1253 * vap - attributes of new file.
1254 * excl - flag indicating exclusive or non-exclusive mode.
1255 * mode - mode to open file with.
1256 * cr - credentials of caller.
1257 * flag - large file flag [UNUSED].
1258 * vsecp - ACL to be set
1260 * OUT: ipp - inode of created or trunc'd entry.
1262 * RETURN: 0 on success, error code on failure.
1265 * dip - ctime|mtime updated if new entry created
1266 * ip - ctime|mtime always, atime if new
1271 zfs_create(struct inode *dip, char *name, vattr_t *vap, int excl,
1272 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
1274 znode_t *zp, *dzp = ITOZ(dip);
1275 zfs_sb_t *zsb = ITOZSB(dip);
1283 zfs_acl_ids_t acl_ids;
1284 boolean_t fuid_dirtied;
1285 boolean_t have_acl = B_FALSE;
1286 boolean_t waited = B_FALSE;
1289 * If we have an ephemeral id, ACL, or XVATTR then
1290 * make sure file system is at proper version
1296 if (zsb->z_use_fuids == B_FALSE &&
1297 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1298 return (SET_ERROR(EINVAL));
1305 if (zsb->z_utf8 && u8_validate(name, strlen(name),
1306 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1308 return (SET_ERROR(EILSEQ));
1311 if (vap->va_mask & ATTR_XVATTR) {
1312 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1313 crgetuid(cr), cr, vap->va_mode)) != 0) {
1321 if (*name == '\0') {
1323 * Null component name refers to the directory itself.
1330 /* possible igrab(zp) */
1333 if (flag & FIGNORECASE)
1336 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1340 zfs_acl_ids_free(&acl_ids);
1341 if (strcmp(name, "..") == 0)
1342 error = SET_ERROR(EISDIR);
1352 * Create a new file object and update the directory
1355 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
1357 zfs_acl_ids_free(&acl_ids);
1362 * We only support the creation of regular files in
1363 * extended attribute directories.
1366 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
1368 zfs_acl_ids_free(&acl_ids);
1369 error = SET_ERROR(EINVAL);
1373 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1374 cr, vsecp, &acl_ids)) != 0)
1378 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1379 zfs_acl_ids_free(&acl_ids);
1380 error = SET_ERROR(EDQUOT);
1384 tx = dmu_tx_create(os);
1386 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1387 ZFS_SA_BASE_ATTR_SIZE);
1389 fuid_dirtied = zsb->z_fuid_dirty;
1391 zfs_fuid_txhold(zsb, tx);
1392 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1393 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1394 if (!zsb->z_use_sa &&
1395 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1396 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1397 0, acl_ids.z_aclp->z_acl_bytes);
1399 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1401 zfs_dirent_unlock(dl);
1402 if (error == ERESTART) {
1408 zfs_acl_ids_free(&acl_ids);
1413 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1416 zfs_fuid_sync(zsb, tx);
1418 (void) zfs_link_create(dl, zp, tx, ZNEW);
1419 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1420 if (flag & FIGNORECASE)
1422 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1423 vsecp, acl_ids.z_fuidp, vap);
1424 zfs_acl_ids_free(&acl_ids);
1427 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1430 zfs_acl_ids_free(&acl_ids);
1434 * A directory entry already exists for this name.
1437 * Can't truncate an existing file if in exclusive mode.
1440 error = SET_ERROR(EEXIST);
1444 * Can't open a directory for writing.
1446 if (S_ISDIR(ZTOI(zp)->i_mode)) {
1447 error = SET_ERROR(EISDIR);
1451 * Verify requested access to file.
1453 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1457 mutex_enter(&dzp->z_lock);
1459 mutex_exit(&dzp->z_lock);
1462 * Truncate regular files if requested.
1464 if (S_ISREG(ZTOI(zp)->i_mode) &&
1465 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
1466 /* we can't hold any locks when calling zfs_freesp() */
1467 zfs_dirent_unlock(dl);
1469 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1475 zfs_dirent_unlock(dl);
1481 zfs_inode_update(dzp);
1482 zfs_inode_update(zp);
1486 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1487 zil_commit(zilog, 0);
1492 EXPORT_SYMBOL(zfs_create);
1495 * Remove an entry from a directory.
1497 * IN: dip - inode of directory to remove entry from.
1498 * name - name of entry to remove.
1499 * cr - credentials of caller.
1501 * RETURN: 0 if success
1502 * error code if failure
1506 * ip - ctime (if nlink > 0)
1509 uint64_t null_xattr = 0;
1513 zfs_remove(struct inode *dip, char *name, cred_t *cr)
1515 znode_t *zp, *dzp = ITOZ(dip);
1518 zfs_sb_t *zsb = ITOZSB(dip);
1521 uint64_t xattr_obj_unlinked = 0;
1527 pathname_t *realnmp = NULL;
1528 #ifdef HAVE_PN_UTILS
1530 #endif /* HAVE_PN_UTILS */
1533 boolean_t waited = B_FALSE;
1539 #ifdef HAVE_PN_UTILS
1540 if (flags & FIGNORECASE) {
1545 #endif /* HAVE_PN_UTILS */
1551 * Attempt to lock directory; fail if entry doesn't exist.
1553 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1555 #ifdef HAVE_PN_UTILS
1558 #endif /* HAVE_PN_UTILS */
1565 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1570 * Need to use rmdir for removing directories.
1572 if (S_ISDIR(ip->i_mode)) {
1573 error = SET_ERROR(EPERM);
1579 dnlc_remove(dvp, realnmp->pn_buf);
1581 dnlc_remove(dvp, name);
1582 #endif /* HAVE_DNLC */
1585 * We never delete the znode and always place it in the unlinked
1586 * set. The dentry cache will always hold the last reference and
1587 * is responsible for safely freeing the znode.
1590 tx = dmu_tx_create(zsb->z_os);
1591 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1592 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1593 zfs_sa_upgrade_txholds(tx, zp);
1594 zfs_sa_upgrade_txholds(tx, dzp);
1596 /* are there any extended attributes? */
1597 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1598 &xattr_obj, sizeof (xattr_obj));
1599 if (error == 0 && xattr_obj) {
1600 error = zfs_zget(zsb, xattr_obj, &xzp);
1602 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1603 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1606 /* charge as an update -- would be nice not to charge at all */
1607 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
1609 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1611 zfs_dirent_unlock(dl);
1615 if (error == ERESTART) {
1621 #ifdef HAVE_PN_UTILS
1624 #endif /* HAVE_PN_UTILS */
1631 * Remove the directory entry.
1633 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1642 * Hold z_lock so that we can make sure that the ACL obj
1643 * hasn't changed. Could have been deleted due to
1646 mutex_enter(&zp->z_lock);
1647 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
1648 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1649 mutex_exit(&zp->z_lock);
1650 zfs_unlinked_add(zp, tx);
1654 #ifdef HAVE_PN_UTILS
1655 if (flags & FIGNORECASE)
1657 #endif /* HAVE_PN_UTILS */
1658 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1662 #ifdef HAVE_PN_UTILS
1665 #endif /* HAVE_PN_UTILS */
1667 zfs_dirent_unlock(dl);
1668 zfs_inode_update(dzp);
1669 zfs_inode_update(zp);
1671 zfs_inode_update(xzp);
1677 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1678 zil_commit(zilog, 0);
1683 EXPORT_SYMBOL(zfs_remove);
1686 * Create a new directory and insert it into dip using the name
1687 * provided. Return a pointer to the inserted directory.
1689 * IN: dip - inode of directory to add subdir to.
1690 * dirname - name of new directory.
1691 * vap - attributes of new directory.
1692 * cr - credentials of caller.
1693 * vsecp - ACL to be set
1695 * OUT: ipp - inode of created directory.
1697 * RETURN: 0 if success
1698 * error code if failure
1701 * dip - ctime|mtime updated
1702 * ipp - ctime|mtime|atime updated
1706 zfs_mkdir(struct inode *dip, char *dirname, vattr_t *vap, struct inode **ipp,
1707 cred_t *cr, int flags, vsecattr_t *vsecp)
1709 znode_t *zp, *dzp = ITOZ(dip);
1710 zfs_sb_t *zsb = ITOZSB(dip);
1718 gid_t gid = crgetgid(cr);
1719 zfs_acl_ids_t acl_ids;
1720 boolean_t fuid_dirtied;
1721 boolean_t waited = B_FALSE;
1723 ASSERT(S_ISDIR(vap->va_mode));
1726 * If we have an ephemeral id, ACL, or XVATTR then
1727 * make sure file system is at proper version
1731 if (zsb->z_use_fuids == B_FALSE &&
1732 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1733 return (SET_ERROR(EINVAL));
1739 if (dzp->z_pflags & ZFS_XATTR) {
1741 return (SET_ERROR(EINVAL));
1744 if (zsb->z_utf8 && u8_validate(dirname,
1745 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1747 return (SET_ERROR(EILSEQ));
1749 if (flags & FIGNORECASE)
1752 if (vap->va_mask & ATTR_XVATTR) {
1753 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1754 crgetuid(cr), cr, vap->va_mode)) != 0) {
1760 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1761 vsecp, &acl_ids)) != 0) {
1766 * First make sure the new directory doesn't exist.
1768 * Existence is checked first to make sure we don't return
1769 * EACCES instead of EEXIST which can cause some applications
1775 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1777 zfs_acl_ids_free(&acl_ids);
1782 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
1783 zfs_acl_ids_free(&acl_ids);
1784 zfs_dirent_unlock(dl);
1789 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
1790 zfs_acl_ids_free(&acl_ids);
1791 zfs_dirent_unlock(dl);
1793 return (SET_ERROR(EDQUOT));
1797 * Add a new entry to the directory.
1799 tx = dmu_tx_create(zsb->z_os);
1800 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1801 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1802 fuid_dirtied = zsb->z_fuid_dirty;
1804 zfs_fuid_txhold(zsb, tx);
1805 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1806 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1807 acl_ids.z_aclp->z_acl_bytes);
1810 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1811 ZFS_SA_BASE_ATTR_SIZE);
1813 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1815 zfs_dirent_unlock(dl);
1816 if (error == ERESTART) {
1822 zfs_acl_ids_free(&acl_ids);
1831 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1834 zfs_fuid_sync(zsb, tx);
1837 * Now put new name in parent dir.
1839 (void) zfs_link_create(dl, zp, tx, ZNEW);
1843 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1844 if (flags & FIGNORECASE)
1846 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1847 acl_ids.z_fuidp, vap);
1849 zfs_acl_ids_free(&acl_ids);
1853 zfs_dirent_unlock(dl);
1855 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1856 zil_commit(zilog, 0);
1858 zfs_inode_update(dzp);
1859 zfs_inode_update(zp);
1863 EXPORT_SYMBOL(zfs_mkdir);
1866 * Remove a directory subdir entry. If the current working
1867 * directory is the same as the subdir to be removed, the
1870 * IN: dip - inode of directory to remove from.
1871 * name - name of directory to be removed.
1872 * cwd - inode of current working directory.
1873 * cr - credentials of caller.
1874 * flags - case flags
1876 * RETURN: 0 on success, error code on failure.
1879 * dip - ctime|mtime updated
1883 zfs_rmdir(struct inode *dip, char *name, struct inode *cwd, cred_t *cr,
1886 znode_t *dzp = ITOZ(dip);
1889 zfs_sb_t *zsb = ITOZSB(dip);
1895 boolean_t waited = B_FALSE;
1901 if (flags & FIGNORECASE)
1907 * Attempt to lock directory; fail if entry doesn't exist.
1909 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1917 if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1921 if (!S_ISDIR(ip->i_mode)) {
1922 error = SET_ERROR(ENOTDIR);
1927 error = SET_ERROR(EINVAL);
1932 * Grab a lock on the directory to make sure that noone is
1933 * trying to add (or lookup) entries while we are removing it.
1935 rw_enter(&zp->z_name_lock, RW_WRITER);
1938 * Grab a lock on the parent pointer to make sure we play well
1939 * with the treewalk and directory rename code.
1941 rw_enter(&zp->z_parent_lock, RW_WRITER);
1943 tx = dmu_tx_create(zsb->z_os);
1944 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1945 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1946 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
1947 zfs_sa_upgrade_txholds(tx, zp);
1948 zfs_sa_upgrade_txholds(tx, dzp);
1949 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1951 rw_exit(&zp->z_parent_lock);
1952 rw_exit(&zp->z_name_lock);
1953 zfs_dirent_unlock(dl);
1955 if (error == ERESTART) {
1966 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
1969 uint64_t txtype = TX_RMDIR;
1970 if (flags & FIGNORECASE)
1972 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
1977 rw_exit(&zp->z_parent_lock);
1978 rw_exit(&zp->z_name_lock);
1980 zfs_dirent_unlock(dl);
1982 zfs_inode_update(dzp);
1983 zfs_inode_update(zp);
1986 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
1987 zil_commit(zilog, 0);
1992 EXPORT_SYMBOL(zfs_rmdir);
1995 * Read as many directory entries as will fit into the provided
1996 * dirent buffer from the given directory cursor position.
1998 * IN: ip - inode of directory to read.
1999 * dirent - buffer for directory entries.
2001 * OUT: dirent - filler buffer of directory entries.
2003 * RETURN: 0 if success
2004 * error code if failure
2007 * ip - atime updated
2009 * Note that the low 4 bits of the cookie returned by zap is always zero.
2010 * This allows us to use the low range for "special" directory entries:
2011 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2012 * we use the offset 2 for the '.zfs' directory.
2016 zfs_readdir(struct inode *ip, struct dir_context *ctx, cred_t *cr)
2018 znode_t *zp = ITOZ(ip);
2019 zfs_sb_t *zsb = ITOZSB(ip);
2022 zap_attribute_t zap;
2028 uint64_t offset; /* must be unsigned; checks for < 1 */
2033 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zsb),
2034 &parent, sizeof (parent))) != 0)
2038 * Quit if directory has been removed (posix)
2046 prefetch = zp->z_zn_prefetch;
2049 * Initialize the iterator cursor.
2053 * Start iteration from the beginning of the directory.
2055 zap_cursor_init(&zc, os, zp->z_id);
2058 * The offset is a serialized cursor.
2060 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2064 * Transform to file-system independent format
2069 * Special case `.', `..', and `.zfs'.
2072 (void) strcpy(zap.za_name, ".");
2073 zap.za_normalization_conflict = 0;
2076 } else if (offset == 1) {
2077 (void) strcpy(zap.za_name, "..");
2078 zap.za_normalization_conflict = 0;
2081 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2082 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2083 zap.za_normalization_conflict = 0;
2084 objnum = ZFSCTL_INO_ROOT;
2090 if ((error = zap_cursor_retrieve(&zc, &zap))) {
2091 if (error == ENOENT)
2098 * Allow multiple entries provided the first entry is
2099 * the object id. Non-zpl consumers may safely make
2100 * use of the additional space.
2102 * XXX: This should be a feature flag for compatibility
2104 if (zap.za_integer_length != 8 ||
2105 zap.za_num_integers == 0) {
2106 cmn_err(CE_WARN, "zap_readdir: bad directory "
2107 "entry, obj = %lld, offset = %lld, "
2108 "length = %d, num = %lld\n",
2109 (u_longlong_t)zp->z_id,
2110 (u_longlong_t)offset,
2111 zap.za_integer_length,
2112 (u_longlong_t)zap.za_num_integers);
2113 error = SET_ERROR(ENXIO);
2117 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2118 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2121 done = !dir_emit(ctx, zap.za_name, strlen(zap.za_name),
2126 /* Prefetch znode */
2128 dmu_prefetch(os, objnum, 0, 0);
2132 * Move to the next entry, fill in the previous offset.
2134 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2135 zap_cursor_advance(&zc);
2136 offset = zap_cursor_serialize(&zc);
2142 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2145 zap_cursor_fini(&zc);
2146 if (error == ENOENT)
2149 ZFS_ACCESSTIME_STAMP(zsb, zp);
2156 EXPORT_SYMBOL(zfs_readdir);
2158 ulong_t zfs_fsync_sync_cnt = 4;
2161 zfs_fsync(struct inode *ip, int syncflag, cred_t *cr)
2163 znode_t *zp = ITOZ(ip);
2164 zfs_sb_t *zsb = ITOZSB(ip);
2166 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2168 if (zsb->z_os->os_sync != ZFS_SYNC_DISABLED) {
2171 zil_commit(zsb->z_log, zp->z_id);
2176 EXPORT_SYMBOL(zfs_fsync);
2180 * Get the requested file attributes and place them in the provided
2183 * IN: ip - inode of file.
2184 * vap - va_mask identifies requested attributes.
2185 * If ATTR_XVATTR set, then optional attrs are requested
2186 * flags - ATTR_NOACLCHECK (CIFS server context)
2187 * cr - credentials of caller.
2189 * OUT: vap - attribute values.
2191 * RETURN: 0 (always succeeds)
2195 zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2197 znode_t *zp = ITOZ(ip);
2198 zfs_sb_t *zsb = ITOZSB(ip);
2201 uint64_t mtime[2], ctime[2];
2202 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2203 xoptattr_t *xoap = NULL;
2204 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2205 sa_bulk_attr_t bulk[2];
2211 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2213 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
2214 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
2216 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2222 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2223 * Also, if we are the owner don't bother, since owner should
2224 * always be allowed to read basic attributes of file.
2226 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2227 (vap->va_uid != crgetuid(cr))) {
2228 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2236 * Return all attributes. It's cheaper to provide the answer
2237 * than to determine whether we were asked the question.
2240 mutex_enter(&zp->z_lock);
2241 vap->va_type = vn_mode_to_vtype(zp->z_mode);
2242 vap->va_mode = zp->z_mode;
2243 vap->va_fsid = ZTOI(zp)->i_sb->s_dev;
2244 vap->va_nodeid = zp->z_id;
2245 if ((zp->z_id == zsb->z_root) && zfs_show_ctldir(zp))
2246 links = zp->z_links + 1;
2248 links = zp->z_links;
2249 vap->va_nlink = MIN(links, ZFS_LINK_MAX);
2250 vap->va_size = i_size_read(ip);
2251 vap->va_rdev = ip->i_rdev;
2252 vap->va_seq = ip->i_generation;
2255 * Add in any requested optional attributes and the create time.
2256 * Also set the corresponding bits in the returned attribute bitmap.
2258 if ((xoap = xva_getxoptattr(xvap)) != NULL && zsb->z_use_fuids) {
2259 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2261 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2262 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2265 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2266 xoap->xoa_readonly =
2267 ((zp->z_pflags & ZFS_READONLY) != 0);
2268 XVA_SET_RTN(xvap, XAT_READONLY);
2271 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2273 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2274 XVA_SET_RTN(xvap, XAT_SYSTEM);
2277 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2279 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2280 XVA_SET_RTN(xvap, XAT_HIDDEN);
2283 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2284 xoap->xoa_nounlink =
2285 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2286 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2289 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2290 xoap->xoa_immutable =
2291 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2292 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2295 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2296 xoap->xoa_appendonly =
2297 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2298 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2301 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2303 ((zp->z_pflags & ZFS_NODUMP) != 0);
2304 XVA_SET_RTN(xvap, XAT_NODUMP);
2307 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2309 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2310 XVA_SET_RTN(xvap, XAT_OPAQUE);
2313 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2314 xoap->xoa_av_quarantined =
2315 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2316 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2319 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2320 xoap->xoa_av_modified =
2321 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2322 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2325 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2326 S_ISREG(ip->i_mode)) {
2327 zfs_sa_get_scanstamp(zp, xvap);
2330 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2333 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zsb),
2334 times, sizeof (times));
2335 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2336 XVA_SET_RTN(xvap, XAT_CREATETIME);
2339 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2340 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2341 XVA_SET_RTN(xvap, XAT_REPARSE);
2343 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2344 xoap->xoa_generation = zp->z_gen;
2345 XVA_SET_RTN(xvap, XAT_GEN);
2348 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2350 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2351 XVA_SET_RTN(xvap, XAT_OFFLINE);
2354 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2356 ((zp->z_pflags & ZFS_SPARSE) != 0);
2357 XVA_SET_RTN(xvap, XAT_SPARSE);
2361 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2362 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2363 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2365 mutex_exit(&zp->z_lock);
2367 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2369 if (zp->z_blksz == 0) {
2371 * Block size hasn't been set; suggest maximal I/O transfers.
2373 vap->va_blksize = zsb->z_max_blksz;
2379 EXPORT_SYMBOL(zfs_getattr);
2382 * Get the basic file attributes and place them in the provided kstat
2383 * structure. The inode is assumed to be the authoritative source
2384 * for most of the attributes. However, the znode currently has the
2385 * authoritative atime, blksize, and block count.
2387 * IN: ip - inode of file.
2389 * OUT: sp - kstat values.
2391 * RETURN: 0 (always succeeds)
2395 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
2397 znode_t *zp = ITOZ(ip);
2398 zfs_sb_t *zsb = ITOZSB(ip);
2400 u_longlong_t nblocks;
2405 mutex_enter(&zp->z_lock);
2407 generic_fillattr(ip, sp);
2408 ZFS_TIME_DECODE(&sp->atime, zp->z_atime);
2410 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2411 sp->blksize = blksize;
2412 sp->blocks = nblocks;
2414 if (unlikely(zp->z_blksz == 0)) {
2416 * Block size hasn't been set; suggest maximal I/O transfers.
2418 sp->blksize = zsb->z_max_blksz;
2421 mutex_exit(&zp->z_lock);
2427 EXPORT_SYMBOL(zfs_getattr_fast);
2430 * Set the file attributes to the values contained in the
2433 * IN: ip - inode of file to be modified.
2434 * vap - new attribute values.
2435 * If ATTR_XVATTR set, then optional attrs are being set
2436 * flags - ATTR_UTIME set if non-default time values provided.
2437 * - ATTR_NOACLCHECK (CIFS context only).
2438 * cr - credentials of caller.
2440 * RETURN: 0 if success
2441 * error code if failure
2444 * ip - ctime updated, mtime updated if size changed.
2448 zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
2450 znode_t *zp = ITOZ(ip);
2451 zfs_sb_t *zsb = ITOZSB(ip);
2455 xvattr_t *tmpxvattr;
2456 uint_t mask = vap->va_mask;
2457 uint_t saved_mask = 0;
2460 uint64_t new_uid, new_gid;
2462 uint64_t mtime[2], ctime[2];
2464 int need_policy = FALSE;
2466 zfs_fuid_info_t *fuidp = NULL;
2467 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2470 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2471 boolean_t fuid_dirtied = B_FALSE;
2472 sa_bulk_attr_t *bulk, *xattr_bulk;
2473 int count = 0, xattr_count = 0;
2484 * Make sure that if we have ephemeral uid/gid or xvattr specified
2485 * that file system is at proper version level
2488 if (zsb->z_use_fuids == B_FALSE &&
2489 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2490 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2491 (mask & ATTR_XVATTR))) {
2493 return (SET_ERROR(EINVAL));
2496 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
2498 return (SET_ERROR(EISDIR));
2501 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
2503 return (SET_ERROR(EINVAL));
2507 * If this is an xvattr_t, then get a pointer to the structure of
2508 * optional attributes. If this is NULL, then we have a vattr_t.
2510 xoap = xva_getxoptattr(xvap);
2512 tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP);
2513 xva_init(tmpxvattr);
2515 bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2516 xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * 7, KM_SLEEP);
2519 * Immutable files can only alter immutable bit and atime
2521 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2522 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
2523 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2528 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2534 * Verify timestamps doesn't overflow 32 bits.
2535 * ZFS can handle large timestamps, but 32bit syscalls can't
2536 * handle times greater than 2039. This check should be removed
2537 * once large timestamps are fully supported.
2539 if (mask & (ATTR_ATIME | ATTR_MTIME)) {
2540 if (((mask & ATTR_ATIME) &&
2541 TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2542 ((mask & ATTR_MTIME) &&
2543 TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2553 /* Can this be moved to before the top label? */
2554 if (zfs_is_readonly(zsb)) {
2560 * First validate permissions
2563 if (mask & ATTR_SIZE) {
2564 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2569 * XXX - Note, we are not providing any open
2570 * mode flags here (like FNDELAY), so we may
2571 * block if there are locks present... this
2572 * should be addressed in openat().
2574 /* XXX - would it be OK to generate a log record here? */
2575 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2580 if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2581 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2582 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2583 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2584 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2585 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2586 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2587 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2588 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2592 if (mask & (ATTR_UID|ATTR_GID)) {
2593 int idmask = (mask & (ATTR_UID|ATTR_GID));
2598 * NOTE: even if a new mode is being set,
2599 * we may clear S_ISUID/S_ISGID bits.
2602 if (!(mask & ATTR_MODE))
2603 vap->va_mode = zp->z_mode;
2606 * Take ownership or chgrp to group we are a member of
2609 take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
2610 take_group = (mask & ATTR_GID) &&
2611 zfs_groupmember(zsb, vap->va_gid, cr);
2614 * If both ATTR_UID and ATTR_GID are set then take_owner and
2615 * take_group must both be set in order to allow taking
2618 * Otherwise, send the check through secpolicy_vnode_setattr()
2622 if (((idmask == (ATTR_UID|ATTR_GID)) &&
2623 take_owner && take_group) ||
2624 ((idmask == ATTR_UID) && take_owner) ||
2625 ((idmask == ATTR_GID) && take_group)) {
2626 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2627 skipaclchk, cr) == 0) {
2629 * Remove setuid/setgid for non-privileged users
2631 (void) secpolicy_setid_clear(vap, cr);
2632 trim_mask = (mask & (ATTR_UID|ATTR_GID));
2641 mutex_enter(&zp->z_lock);
2642 oldva.va_mode = zp->z_mode;
2643 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2644 if (mask & ATTR_XVATTR) {
2646 * Update xvattr mask to include only those attributes
2647 * that are actually changing.
2649 * the bits will be restored prior to actually setting
2650 * the attributes so the caller thinks they were set.
2652 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2653 if (xoap->xoa_appendonly !=
2654 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2657 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2658 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
2662 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2663 if (xoap->xoa_nounlink !=
2664 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2667 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2668 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
2672 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2673 if (xoap->xoa_immutable !=
2674 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2677 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2678 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
2682 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2683 if (xoap->xoa_nodump !=
2684 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2687 XVA_CLR_REQ(xvap, XAT_NODUMP);
2688 XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
2692 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2693 if (xoap->xoa_av_modified !=
2694 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2697 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2698 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
2702 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2703 if ((!S_ISREG(ip->i_mode) &&
2704 xoap->xoa_av_quarantined) ||
2705 xoap->xoa_av_quarantined !=
2706 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2709 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2710 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
2714 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2715 mutex_exit(&zp->z_lock);
2720 if (need_policy == FALSE &&
2721 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2722 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2727 mutex_exit(&zp->z_lock);
2729 if (mask & ATTR_MODE) {
2730 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2731 err = secpolicy_setid_setsticky_clear(ip, vap,
2736 trim_mask |= ATTR_MODE;
2744 * If trim_mask is set then take ownership
2745 * has been granted or write_acl is present and user
2746 * has the ability to modify mode. In that case remove
2747 * UID|GID and or MODE from mask so that
2748 * secpolicy_vnode_setattr() doesn't revoke it.
2752 saved_mask = vap->va_mask;
2753 vap->va_mask &= ~trim_mask;
2755 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
2756 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2761 vap->va_mask |= saved_mask;
2765 * secpolicy_vnode_setattr, or take ownership may have
2768 mask = vap->va_mask;
2770 if ((mask & (ATTR_UID | ATTR_GID))) {
2771 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
2772 &xattr_obj, sizeof (xattr_obj));
2774 if (err == 0 && xattr_obj) {
2775 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
2779 if (mask & ATTR_UID) {
2780 new_uid = zfs_fuid_create(zsb,
2781 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2782 if (new_uid != zp->z_uid &&
2783 zfs_fuid_overquota(zsb, B_FALSE, new_uid)) {
2791 if (mask & ATTR_GID) {
2792 new_gid = zfs_fuid_create(zsb, (uint64_t)vap->va_gid,
2793 cr, ZFS_GROUP, &fuidp);
2794 if (new_gid != zp->z_gid &&
2795 zfs_fuid_overquota(zsb, B_TRUE, new_gid)) {
2803 tx = dmu_tx_create(zsb->z_os);
2805 if (mask & ATTR_MODE) {
2806 uint64_t pmode = zp->z_mode;
2808 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2810 zfs_acl_chmod_setattr(zp, &aclp, new_mode);
2812 mutex_enter(&zp->z_lock);
2813 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2815 * Are we upgrading ACL from old V0 format
2818 if (zsb->z_version >= ZPL_VERSION_FUID &&
2819 zfs_znode_acl_version(zp) ==
2820 ZFS_ACL_VERSION_INITIAL) {
2821 dmu_tx_hold_free(tx, acl_obj, 0,
2823 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2824 0, aclp->z_acl_bytes);
2826 dmu_tx_hold_write(tx, acl_obj, 0,
2829 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2830 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2831 0, aclp->z_acl_bytes);
2833 mutex_exit(&zp->z_lock);
2834 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2836 if ((mask & ATTR_XVATTR) &&
2837 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2838 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2840 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2844 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
2847 fuid_dirtied = zsb->z_fuid_dirty;
2849 zfs_fuid_txhold(zsb, tx);
2851 zfs_sa_upgrade_txholds(tx, zp);
2853 err = dmu_tx_assign(tx, TXG_WAIT);
2859 * Set each attribute requested.
2860 * We group settings according to the locks they need to acquire.
2862 * Note: you cannot set ctime directly, although it will be
2863 * updated as a side-effect of calling this function.
2867 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2868 mutex_enter(&zp->z_acl_lock);
2869 mutex_enter(&zp->z_lock);
2871 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
2872 &zp->z_pflags, sizeof (zp->z_pflags));
2875 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2876 mutex_enter(&attrzp->z_acl_lock);
2877 mutex_enter(&attrzp->z_lock);
2878 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2879 SA_ZPL_FLAGS(zsb), NULL, &attrzp->z_pflags,
2880 sizeof (attrzp->z_pflags));
2883 if (mask & (ATTR_UID|ATTR_GID)) {
2885 if (mask & ATTR_UID) {
2886 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zsb), NULL,
2887 &new_uid, sizeof (new_uid));
2888 zp->z_uid = new_uid;
2890 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2891 SA_ZPL_UID(zsb), NULL, &new_uid,
2893 attrzp->z_uid = new_uid;
2897 if (mask & ATTR_GID) {
2898 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zsb),
2899 NULL, &new_gid, sizeof (new_gid));
2900 zp->z_gid = new_gid;
2902 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2903 SA_ZPL_GID(zsb), NULL, &new_gid,
2905 attrzp->z_gid = new_gid;
2908 if (!(mask & ATTR_MODE)) {
2909 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb),
2910 NULL, &new_mode, sizeof (new_mode));
2911 new_mode = zp->z_mode;
2913 err = zfs_acl_chown_setattr(zp);
2916 err = zfs_acl_chown_setattr(attrzp);
2921 if (mask & ATTR_MODE) {
2922 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb), NULL,
2923 &new_mode, sizeof (new_mode));
2924 zp->z_mode = new_mode;
2925 ASSERT3P(aclp, !=, NULL);
2926 err = zfs_aclset_common(zp, aclp, cr, tx);
2928 if (zp->z_acl_cached)
2929 zfs_acl_free(zp->z_acl_cached);
2930 zp->z_acl_cached = aclp;
2935 if (mask & ATTR_ATIME) {
2936 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
2937 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zsb), NULL,
2938 &zp->z_atime, sizeof (zp->z_atime));
2941 if (mask & ATTR_MTIME) {
2942 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
2943 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL,
2944 mtime, sizeof (mtime));
2947 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2948 if (mask & ATTR_SIZE && !(mask & ATTR_MTIME)) {
2949 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb),
2950 NULL, mtime, sizeof (mtime));
2951 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
2952 &ctime, sizeof (ctime));
2953 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
2955 } else if (mask != 0) {
2956 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
2957 &ctime, sizeof (ctime));
2958 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
2961 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2962 SA_ZPL_CTIME(zsb), NULL,
2963 &ctime, sizeof (ctime));
2964 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
2965 mtime, ctime, B_TRUE);
2969 * Do this after setting timestamps to prevent timestamp
2970 * update from toggling bit
2973 if (xoap && (mask & ATTR_XVATTR)) {
2976 * restore trimmed off masks
2977 * so that return masks can be set for caller.
2980 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
2981 XVA_SET_REQ(xvap, XAT_APPENDONLY);
2983 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
2984 XVA_SET_REQ(xvap, XAT_NOUNLINK);
2986 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
2987 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
2989 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
2990 XVA_SET_REQ(xvap, XAT_NODUMP);
2992 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
2993 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
2995 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
2996 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
2999 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3000 ASSERT(S_ISREG(ip->i_mode));
3002 zfs_xvattr_set(zp, xvap, tx);
3006 zfs_fuid_sync(zsb, tx);
3009 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3011 mutex_exit(&zp->z_lock);
3012 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3013 mutex_exit(&zp->z_acl_lock);
3016 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
3017 mutex_exit(&attrzp->z_acl_lock);
3018 mutex_exit(&attrzp->z_lock);
3021 if (err == 0 && attrzp) {
3022 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3033 zfs_fuid_info_free(fuidp);
3039 if (err == ERESTART)
3042 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3044 zfs_inode_update(zp);
3048 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3049 zil_commit(zilog, 0);
3052 kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * 7);
3053 kmem_free(bulk, sizeof (sa_bulk_attr_t) * 7);
3054 kmem_free(tmpxvattr, sizeof (xvattr_t));
3058 EXPORT_SYMBOL(zfs_setattr);
3060 typedef struct zfs_zlock {
3061 krwlock_t *zl_rwlock; /* lock we acquired */
3062 znode_t *zl_znode; /* znode we held */
3063 struct zfs_zlock *zl_next; /* next in list */
3067 * Drop locks and release vnodes that were held by zfs_rename_lock().
3070 zfs_rename_unlock(zfs_zlock_t **zlpp)
3074 while ((zl = *zlpp) != NULL) {
3075 if (zl->zl_znode != NULL)
3076 iput(ZTOI(zl->zl_znode));
3077 rw_exit(zl->zl_rwlock);
3078 *zlpp = zl->zl_next;
3079 kmem_free(zl, sizeof (*zl));
3084 * Search back through the directory tree, using the ".." entries.
3085 * Lock each directory in the chain to prevent concurrent renames.
3086 * Fail any attempt to move a directory into one of its own descendants.
3087 * XXX - z_parent_lock can overlap with map or grow locks
3090 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3094 uint64_t rootid = ZTOZSB(zp)->z_root;
3095 uint64_t oidp = zp->z_id;
3096 krwlock_t *rwlp = &szp->z_parent_lock;
3097 krw_t rw = RW_WRITER;
3100 * First pass write-locks szp and compares to zp->z_id.
3101 * Later passes read-lock zp and compare to zp->z_parent.
3104 if (!rw_tryenter(rwlp, rw)) {
3106 * Another thread is renaming in this path.
3107 * Note that if we are a WRITER, we don't have any
3108 * parent_locks held yet.
3110 if (rw == RW_READER && zp->z_id > szp->z_id) {
3112 * Drop our locks and restart
3114 zfs_rename_unlock(&zl);
3118 rwlp = &szp->z_parent_lock;
3123 * Wait for other thread to drop its locks
3129 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3130 zl->zl_rwlock = rwlp;
3131 zl->zl_znode = NULL;
3132 zl->zl_next = *zlpp;
3135 if (oidp == szp->z_id) /* We're a descendant of szp */
3136 return (SET_ERROR(EINVAL));
3138 if (oidp == rootid) /* We've hit the top */
3141 if (rw == RW_READER) { /* i.e. not the first pass */
3142 int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
3147 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
3148 &oidp, sizeof (oidp));
3149 rwlp = &zp->z_parent_lock;
3152 } while (zp->z_id != sdzp->z_id);
3158 * Move an entry from the provided source directory to the target
3159 * directory. Change the entry name as indicated.
3161 * IN: sdip - Source directory containing the "old entry".
3162 * snm - Old entry name.
3163 * tdip - Target directory to contain the "new entry".
3164 * tnm - New entry name.
3165 * cr - credentials of caller.
3166 * flags - case flags
3168 * RETURN: 0 on success, error code on failure.
3171 * sdip,tdip - ctime|mtime updated
3175 zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
3176 cred_t *cr, int flags)
3178 znode_t *tdzp, *szp, *tzp;
3179 znode_t *sdzp = ITOZ(sdip);
3180 zfs_sb_t *zsb = ITOZSB(sdip);
3182 zfs_dirlock_t *sdl, *tdl;
3185 int cmp, serr, terr;
3188 boolean_t waited = B_FALSE;
3191 ZFS_VERIFY_ZP(sdzp);
3194 if (tdip->i_sb != sdip->i_sb || zfsctl_is_node(tdip)) {
3196 return (SET_ERROR(EXDEV));
3200 ZFS_VERIFY_ZP(tdzp);
3201 if (zsb->z_utf8 && u8_validate(tnm,
3202 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3204 return (SET_ERROR(EILSEQ));
3207 if (flags & FIGNORECASE)
3216 * This is to prevent the creation of links into attribute space
3217 * by renaming a linked file into/outof an attribute directory.
3218 * See the comment in zfs_link() for why this is considered bad.
3220 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3222 return (SET_ERROR(EINVAL));
3226 * Lock source and target directory entries. To prevent deadlock,
3227 * a lock ordering must be defined. We lock the directory with
3228 * the smallest object id first, or if it's a tie, the one with
3229 * the lexically first name.
3231 if (sdzp->z_id < tdzp->z_id) {
3233 } else if (sdzp->z_id > tdzp->z_id) {
3237 * First compare the two name arguments without
3238 * considering any case folding.
3240 int nofold = (zsb->z_norm & ~U8_TEXTPREP_TOUPPER);
3242 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3243 ASSERT(error == 0 || !zsb->z_utf8);
3246 * POSIX: "If the old argument and the new argument
3247 * both refer to links to the same existing file,
3248 * the rename() function shall return successfully
3249 * and perform no other action."
3255 * If the file system is case-folding, then we may
3256 * have some more checking to do. A case-folding file
3257 * system is either supporting mixed case sensitivity
3258 * access or is completely case-insensitive. Note
3259 * that the file system is always case preserving.
3261 * In mixed sensitivity mode case sensitive behavior
3262 * is the default. FIGNORECASE must be used to
3263 * explicitly request case insensitive behavior.
3265 * If the source and target names provided differ only
3266 * by case (e.g., a request to rename 'tim' to 'Tim'),
3267 * we will treat this as a special case in the
3268 * case-insensitive mode: as long as the source name
3269 * is an exact match, we will allow this to proceed as
3270 * a name-change request.
3272 if ((zsb->z_case == ZFS_CASE_INSENSITIVE ||
3273 (zsb->z_case == ZFS_CASE_MIXED &&
3274 flags & FIGNORECASE)) &&
3275 u8_strcmp(snm, tnm, 0, zsb->z_norm, U8_UNICODE_LATEST,
3278 * case preserving rename request, require exact
3287 * If the source and destination directories are the same, we should
3288 * grab the z_name_lock of that directory only once.
3292 rw_enter(&sdzp->z_name_lock, RW_READER);
3296 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3297 ZEXISTS | zflg, NULL, NULL);
3298 terr = zfs_dirent_lock(&tdl,
3299 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3301 terr = zfs_dirent_lock(&tdl,
3302 tdzp, tnm, &tzp, zflg, NULL, NULL);
3303 serr = zfs_dirent_lock(&sdl,
3304 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3310 * Source entry invalid or not there.
3313 zfs_dirent_unlock(tdl);
3319 rw_exit(&sdzp->z_name_lock);
3321 if (strcmp(snm, "..") == 0)
3327 zfs_dirent_unlock(sdl);
3331 rw_exit(&sdzp->z_name_lock);
3333 if (strcmp(tnm, "..") == 0)
3340 * Must have write access at the source to remove the old entry
3341 * and write access at the target to create the new entry.
3342 * Note that if target and source are the same, this can be
3343 * done in a single check.
3346 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
3349 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3351 * Check to make sure rename is valid.
3352 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3354 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
3359 * Does target exist?
3363 * Source and target must be the same type.
3365 if (S_ISDIR(ZTOI(szp)->i_mode)) {
3366 if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
3367 error = SET_ERROR(ENOTDIR);
3371 if (S_ISDIR(ZTOI(tzp)->i_mode)) {
3372 error = SET_ERROR(EISDIR);
3377 * POSIX dictates that when the source and target
3378 * entries refer to the same file object, rename
3379 * must do nothing and exit without error.
3381 if (szp->z_id == tzp->z_id) {
3387 tx = dmu_tx_create(zsb->z_os);
3388 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3389 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3390 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3391 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3393 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3394 zfs_sa_upgrade_txholds(tx, tdzp);
3397 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3398 zfs_sa_upgrade_txholds(tx, tzp);
3401 zfs_sa_upgrade_txholds(tx, szp);
3402 dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
3403 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3406 zfs_rename_unlock(&zl);
3407 zfs_dirent_unlock(sdl);
3408 zfs_dirent_unlock(tdl);
3411 rw_exit(&sdzp->z_name_lock);
3416 if (error == ERESTART) {
3427 if (tzp) /* Attempt to remove the existing target */
3428 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3431 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3433 szp->z_pflags |= ZFS_AV_MODIFIED;
3435 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zsb),
3436 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3439 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3441 zfs_log_rename(zilog, tx, TX_RENAME |
3442 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3443 sdl->dl_name, tdzp, tdl->dl_name, szp);
3446 * At this point, we have successfully created
3447 * the target name, but have failed to remove
3448 * the source name. Since the create was done
3449 * with the ZRENAMING flag, there are
3450 * complications; for one, the link count is
3451 * wrong. The easiest way to deal with this
3452 * is to remove the newly created target, and
3453 * return the original error. This must
3454 * succeed; fortunately, it is very unlikely to
3455 * fail, since we just created it.
3457 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3458 ZRENAMING, NULL), ==, 0);
3466 zfs_rename_unlock(&zl);
3468 zfs_dirent_unlock(sdl);
3469 zfs_dirent_unlock(tdl);
3471 zfs_inode_update(sdzp);
3473 rw_exit(&sdzp->z_name_lock);
3476 zfs_inode_update(tdzp);
3478 zfs_inode_update(szp);
3481 zfs_inode_update(tzp);
3485 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3486 zil_commit(zilog, 0);
3491 EXPORT_SYMBOL(zfs_rename);
3494 * Insert the indicated symbolic reference entry into the directory.
3496 * IN: dip - Directory to contain new symbolic link.
3497 * link - Name for new symlink entry.
3498 * vap - Attributes of new entry.
3499 * target - Target path of new symlink.
3501 * cr - credentials of caller.
3502 * flags - case flags
3504 * RETURN: 0 on success, error code on failure.
3507 * dip - ctime|mtime updated
3511 zfs_symlink(struct inode *dip, char *name, vattr_t *vap, char *link,
3512 struct inode **ipp, cred_t *cr, int flags)
3514 znode_t *zp, *dzp = ITOZ(dip);
3517 zfs_sb_t *zsb = ITOZSB(dip);
3519 uint64_t len = strlen(link);
3522 zfs_acl_ids_t acl_ids;
3523 boolean_t fuid_dirtied;
3524 uint64_t txtype = TX_SYMLINK;
3525 boolean_t waited = B_FALSE;
3527 ASSERT(S_ISLNK(vap->va_mode));
3533 if (zsb->z_utf8 && u8_validate(name, strlen(name),
3534 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3536 return (SET_ERROR(EILSEQ));
3538 if (flags & FIGNORECASE)
3541 if (len > MAXPATHLEN) {
3543 return (SET_ERROR(ENAMETOOLONG));
3546 if ((error = zfs_acl_ids_create(dzp, 0,
3547 vap, cr, NULL, &acl_ids)) != 0) {
3555 * Attempt to lock directory; fail if entry already exists.
3557 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3559 zfs_acl_ids_free(&acl_ids);
3564 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3565 zfs_acl_ids_free(&acl_ids);
3566 zfs_dirent_unlock(dl);
3571 if (zfs_acl_ids_overquota(zsb, &acl_ids)) {
3572 zfs_acl_ids_free(&acl_ids);
3573 zfs_dirent_unlock(dl);
3575 return (SET_ERROR(EDQUOT));
3577 tx = dmu_tx_create(zsb->z_os);
3578 fuid_dirtied = zsb->z_fuid_dirty;
3579 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3580 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3581 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3582 ZFS_SA_BASE_ATTR_SIZE + len);
3583 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3584 if (!zsb->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3585 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3586 acl_ids.z_aclp->z_acl_bytes);
3589 zfs_fuid_txhold(zsb, tx);
3590 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3592 zfs_dirent_unlock(dl);
3593 if (error == ERESTART) {
3599 zfs_acl_ids_free(&acl_ids);
3606 * Create a new object for the symlink.
3607 * for version 4 ZPL datsets the symlink will be an SA attribute
3609 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3612 zfs_fuid_sync(zsb, tx);
3614 mutex_enter(&zp->z_lock);
3616 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zsb),
3619 zfs_sa_symlink(zp, link, len, tx);
3620 mutex_exit(&zp->z_lock);
3623 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zsb),
3624 &zp->z_size, sizeof (zp->z_size), tx);
3626 * Insert the new object into the directory.
3628 (void) zfs_link_create(dl, zp, tx, ZNEW);
3630 if (flags & FIGNORECASE)
3632 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3634 zfs_inode_update(dzp);
3635 zfs_inode_update(zp);
3637 zfs_acl_ids_free(&acl_ids);
3641 zfs_dirent_unlock(dl);
3645 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3646 zil_commit(zilog, 0);
3651 EXPORT_SYMBOL(zfs_symlink);
3654 * Return, in the buffer contained in the provided uio structure,
3655 * the symbolic path referred to by ip.
3657 * IN: ip - inode of symbolic link
3658 * uio - structure to contain the link path.
3659 * cr - credentials of caller.
3661 * RETURN: 0 if success
3662 * error code if failure
3665 * ip - atime updated
3669 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
3671 znode_t *zp = ITOZ(ip);
3672 zfs_sb_t *zsb = ITOZSB(ip);
3678 mutex_enter(&zp->z_lock);
3680 error = sa_lookup_uio(zp->z_sa_hdl,
3681 SA_ZPL_SYMLINK(zsb), uio);
3683 error = zfs_sa_readlink(zp, uio);
3684 mutex_exit(&zp->z_lock);
3686 ZFS_ACCESSTIME_STAMP(zsb, zp);
3690 EXPORT_SYMBOL(zfs_readlink);
3693 * Insert a new entry into directory tdip referencing sip.
3695 * IN: tdip - Directory to contain new entry.
3696 * sip - inode of new entry.
3697 * name - name of new entry.
3698 * cr - credentials of caller.
3700 * RETURN: 0 if success
3701 * error code if failure
3704 * tdip - ctime|mtime updated
3705 * sip - ctime updated
3709 zfs_link(struct inode *tdip, struct inode *sip, char *name, cred_t *cr)
3711 znode_t *dzp = ITOZ(tdip);
3713 zfs_sb_t *zsb = ITOZSB(tdip);
3721 boolean_t waited = B_FALSE;
3723 ASSERT(S_ISDIR(tdip->i_mode));
3730 * POSIX dictates that we return EPERM here.
3731 * Better choices include ENOTSUP or EISDIR.
3733 if (S_ISDIR(sip->i_mode)) {
3735 return (SET_ERROR(EPERM));
3738 if (sip->i_sb != tdip->i_sb || zfsctl_is_node(sip)) {
3740 return (SET_ERROR(EXDEV));
3746 /* Prevent links to .zfs/shares files */
3748 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zsb),
3749 &parent, sizeof (uint64_t))) != 0) {
3753 if (parent == zsb->z_shares_dir) {
3755 return (SET_ERROR(EPERM));
3758 if (zsb->z_utf8 && u8_validate(name,
3759 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3761 return (SET_ERROR(EILSEQ));
3763 #ifdef HAVE_PN_UTILS
3764 if (flags & FIGNORECASE)
3766 #endif /* HAVE_PN_UTILS */
3769 * We do not support links between attributes and non-attributes
3770 * because of the potential security risk of creating links
3771 * into "normal" file space in order to circumvent restrictions
3772 * imposed in attribute space.
3774 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3776 return (SET_ERROR(EINVAL));
3779 owner = zfs_fuid_map_id(zsb, szp->z_uid, cr, ZFS_OWNER);
3780 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3782 return (SET_ERROR(EPERM));
3785 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3792 * Attempt to lock directory; fail if entry already exists.
3794 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3800 tx = dmu_tx_create(zsb->z_os);
3801 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3802 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3803 zfs_sa_upgrade_txholds(tx, szp);
3804 zfs_sa_upgrade_txholds(tx, dzp);
3805 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3807 zfs_dirent_unlock(dl);
3808 if (error == ERESTART) {
3819 error = zfs_link_create(dl, szp, tx, 0);
3822 uint64_t txtype = TX_LINK;
3823 #ifdef HAVE_PN_UTILS
3824 if (flags & FIGNORECASE)
3826 #endif /* HAVE_PN_UTILS */
3827 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3832 zfs_dirent_unlock(dl);
3834 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
3835 zil_commit(zilog, 0);
3837 zfs_inode_update(dzp);
3838 zfs_inode_update(szp);
3842 EXPORT_SYMBOL(zfs_link);
3845 zfs_putpage_commit_cb(void *arg)
3847 struct page *pp = arg;
3850 end_page_writeback(pp);
3854 * Push a page out to disk, once the page is on stable storage the
3855 * registered commit callback will be run as notification of completion.
3857 * IN: ip - page mapped for inode.
3858 * pp - page to push (page is locked)
3859 * wbc - writeback control data
3861 * RETURN: 0 if success
3862 * error code if failure
3865 * ip - ctime|mtime updated
3869 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
3871 znode_t *zp = ITOZ(ip);
3872 zfs_sb_t *zsb = ITOZSB(ip);
3880 uint64_t mtime[2], ctime[2];
3881 sa_bulk_attr_t bulk[3];
3887 ASSERT(PageLocked(pp));
3889 pgoff = page_offset(pp); /* Page byte-offset in file */
3890 offset = i_size_read(ip); /* File length in bytes */
3891 pglen = MIN(PAGE_CACHE_SIZE, /* Page length in bytes */
3892 P2ROUNDUP(offset, PAGE_CACHE_SIZE)-pgoff);
3894 /* Page is beyond end of file */
3895 if (pgoff >= offset) {
3901 /* Truncate page length to end of file */
3902 if (pgoff + pglen > offset)
3903 pglen = offset - pgoff;
3907 * FIXME: Allow mmap writes past its quota. The correct fix
3908 * is to register a page_mkwrite() handler to count the page
3909 * against its quota when it is about to be dirtied.
3911 if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
3912 zfs_owner_overquota(zsb, zp, B_TRUE)) {
3917 set_page_writeback(pp);
3920 rl = zfs_range_lock(zp, pgoff, pglen, RL_WRITER);
3921 tx = dmu_tx_create(zsb->z_os);
3923 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
3925 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3926 zfs_sa_upgrade_txholds(tx, zp);
3927 err = dmu_tx_assign(tx, TXG_NOWAIT);
3929 if (err == ERESTART)
3933 __set_page_dirty_nobuffers(pp);
3935 end_page_writeback(pp);
3936 zfs_range_unlock(rl);
3942 ASSERT3U(pglen, <=, PAGE_CACHE_SIZE);
3943 dmu_write(zsb->z_os, zp->z_id, pgoff, pglen, va, tx);
3946 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
3947 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
3948 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zsb), NULL, &zp->z_pflags, 8);
3950 /* Preserve the mtime and ctime provided by the inode */
3951 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
3952 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
3953 zp->z_atime_dirty = 0;
3956 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
3958 zfs_log_write(zsb->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0,
3959 zfs_putpage_commit_cb, pp);
3962 zfs_range_unlock(rl);
3964 if (wbc->sync_mode != WB_SYNC_NONE) {
3966 * Note that this is rarely called under writepages(), because
3967 * writepages() normally handles the entire commit for
3968 * performance reasons.
3970 if (zsb->z_log != NULL)
3971 zil_commit(zsb->z_log, zp->z_id);
3979 * Update the system attributes when the inode has been dirtied. For the
3980 * moment we only update the mode, atime, mtime, and ctime.
3983 zfs_dirty_inode(struct inode *ip, int flags)
3985 znode_t *zp = ITOZ(ip);
3986 zfs_sb_t *zsb = ITOZSB(ip);
3988 uint64_t mode, atime[2], mtime[2], ctime[2];
3989 sa_bulk_attr_t bulk[4];
3996 tx = dmu_tx_create(zsb->z_os);
3998 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3999 zfs_sa_upgrade_txholds(tx, zp);
4001 error = dmu_tx_assign(tx, TXG_WAIT);
4007 mutex_enter(&zp->z_lock);
4008 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zsb), NULL, &mode, 8);
4009 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zsb), NULL, &atime, 16);
4010 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
4011 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
4013 /* Preserve the mode, mtime and ctime provided by the inode */
4014 ZFS_TIME_ENCODE(&ip->i_atime, atime);
4015 ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
4016 ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
4020 zp->z_atime_dirty = 0;
4022 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
4023 mutex_exit(&zp->z_lock);
4030 EXPORT_SYMBOL(zfs_dirty_inode);
4034 zfs_inactive(struct inode *ip)
4036 znode_t *zp = ITOZ(ip);
4037 zfs_sb_t *zsb = ITOZSB(ip);
4040 if (zfsctl_is_node(ip)) {
4041 zfsctl_inode_inactive(ip);
4045 rw_enter(&zsb->z_teardown_inactive_lock, RW_READER);
4046 if (zp->z_sa_hdl == NULL) {
4047 rw_exit(&zsb->z_teardown_inactive_lock);
4051 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4052 dmu_tx_t *tx = dmu_tx_create(zsb->z_os);
4054 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4055 zfs_sa_upgrade_txholds(tx, zp);
4056 error = dmu_tx_assign(tx, TXG_WAIT);
4060 mutex_enter(&zp->z_lock);
4061 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zsb),
4062 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4063 zp->z_atime_dirty = 0;
4064 mutex_exit(&zp->z_lock);
4070 rw_exit(&zsb->z_teardown_inactive_lock);
4072 EXPORT_SYMBOL(zfs_inactive);
4075 * Bounds-check the seek operation.
4077 * IN: ip - inode seeking within
4078 * ooff - old file offset
4079 * noffp - pointer to new file offset
4080 * ct - caller context
4082 * RETURN: 0 if success
4083 * EINVAL if new offset invalid
4087 zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp)
4089 if (S_ISDIR(ip->i_mode))
4091 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4093 EXPORT_SYMBOL(zfs_seek);
4096 * Fill pages with data from the disk.
4099 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
4101 znode_t *zp = ITOZ(ip);
4102 zfs_sb_t *zsb = ITOZSB(ip);
4104 struct page *cur_pp;
4105 u_offset_t io_off, total;
4112 io_len = nr_pages << PAGE_CACHE_SHIFT;
4113 i_size = i_size_read(ip);
4114 io_off = page_offset(pl[0]);
4116 if (io_off + io_len > i_size)
4117 io_len = i_size - io_off;
4120 * Iterate over list of pages and read each page individually.
4124 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4128 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4132 /* convert checksum errors into IO errors */
4134 err = SET_ERROR(EIO);
4137 cur_pp = pl[++page_idx];
4144 * Uses zfs_fillpage to read data from the file and fill the pages.
4146 * IN: ip - inode of file to get data from.
4147 * pl - list of pages to read
4148 * nr_pages - number of pages to read
4150 * RETURN: 0 on success, error code on failure.
4153 * vp - atime updated
4157 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
4159 znode_t *zp = ITOZ(ip);
4160 zfs_sb_t *zsb = ITOZSB(ip);
4169 err = zfs_fillpage(ip, pl, nr_pages);
4172 ZFS_ACCESSTIME_STAMP(zsb, zp);
4177 EXPORT_SYMBOL(zfs_getpage);
4180 * Check ZFS specific permissions to memory map a section of a file.
4182 * IN: ip - inode of the file to mmap
4184 * addrp - start address in memory region
4185 * len - length of memory region
4186 * vm_flags- address flags
4188 * RETURN: 0 if success
4189 * error code if failure
4193 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
4194 unsigned long vm_flags)
4196 znode_t *zp = ITOZ(ip);
4197 zfs_sb_t *zsb = ITOZSB(ip);
4202 if ((vm_flags & VM_WRITE) && (zp->z_pflags &
4203 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4205 return (SET_ERROR(EPERM));
4208 if ((vm_flags & (VM_READ | VM_EXEC)) &&
4209 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4211 return (SET_ERROR(EACCES));
4214 if (off < 0 || len > MAXOFFSET_T - off) {
4216 return (SET_ERROR(ENXIO));
4222 EXPORT_SYMBOL(zfs_map);
4225 * convoff - converts the given data (start, whence) to the
4229 convoff(struct inode *ip, flock64_t *lckdat, int whence, offset_t offset)
4234 if ((lckdat->l_whence == 2) || (whence == 2)) {
4235 if ((error = zfs_getattr(ip, &vap, 0, CRED()) != 0))
4239 switch (lckdat->l_whence) {
4241 lckdat->l_start += offset;
4244 lckdat->l_start += vap.va_size;
4249 return (SET_ERROR(EINVAL));
4252 if (lckdat->l_start < 0)
4253 return (SET_ERROR(EINVAL));
4257 lckdat->l_start -= offset;
4260 lckdat->l_start -= vap.va_size;
4265 return (SET_ERROR(EINVAL));
4268 lckdat->l_whence = (short)whence;
4273 * Free or allocate space in a file. Currently, this function only
4274 * supports the `F_FREESP' command. However, this command is somewhat
4275 * misnamed, as its functionality includes the ability to allocate as
4276 * well as free space.
4278 * IN: ip - inode of file to free data in.
4279 * cmd - action to take (only F_FREESP supported).
4280 * bfp - section of file to free/alloc.
4281 * flag - current file open mode flags.
4282 * offset - current file offset.
4283 * cr - credentials of caller [UNUSED].
4285 * RETURN: 0 on success, error code on failure.
4288 * ip - ctime|mtime updated
4292 zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
4293 offset_t offset, cred_t *cr)
4295 znode_t *zp = ITOZ(ip);
4296 zfs_sb_t *zsb = ITOZSB(ip);
4303 if (cmd != F_FREESP) {
4305 return (SET_ERROR(EINVAL));
4308 if ((error = convoff(ip, bfp, 0, offset))) {
4313 if (bfp->l_len < 0) {
4315 return (SET_ERROR(EINVAL));
4319 * Permissions aren't checked on Solaris because on this OS
4320 * zfs_space() can only be called with an opened file handle.
4321 * On Linux we can get here through truncate_range() which
4322 * operates directly on inodes, so we need to check access rights.
4324 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
4330 len = bfp->l_len; /* 0 means from off to end of file */
4332 error = zfs_freesp(zp, off, len, flag, TRUE);
4337 EXPORT_SYMBOL(zfs_space);
4341 zfs_fid(struct inode *ip, fid_t *fidp)
4343 znode_t *zp = ITOZ(ip);
4344 zfs_sb_t *zsb = ITOZSB(ip);
4347 uint64_t object = zp->z_id;
4354 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zsb),
4355 &gen64, sizeof (uint64_t))) != 0) {
4360 gen = (uint32_t)gen64;
4362 size = (zsb->z_parent != zsb) ? LONG_FID_LEN : SHORT_FID_LEN;
4363 if (fidp->fid_len < size) {
4364 fidp->fid_len = size;
4366 return (SET_ERROR(ENOSPC));
4369 zfid = (zfid_short_t *)fidp;
4371 zfid->zf_len = size;
4373 for (i = 0; i < sizeof (zfid->zf_object); i++)
4374 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4376 /* Must have a non-zero generation number to distinguish from .zfs */
4379 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4380 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4382 if (size == LONG_FID_LEN) {
4383 uint64_t objsetid = dmu_objset_id(zsb->z_os);
4386 zlfid = (zfid_long_t *)fidp;
4388 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4389 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4391 /* XXX - this should be the generation number for the objset */
4392 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4393 zlfid->zf_setgen[i] = 0;
4399 EXPORT_SYMBOL(zfs_fid);
4403 zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4405 znode_t *zp = ITOZ(ip);
4406 zfs_sb_t *zsb = ITOZSB(ip);
4408 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4412 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4417 EXPORT_SYMBOL(zfs_getsecattr);
4421 zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
4423 znode_t *zp = ITOZ(ip);
4424 zfs_sb_t *zsb = ITOZSB(ip);
4426 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4427 zilog_t *zilog = zsb->z_log;
4432 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4434 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
4435 zil_commit(zilog, 0);
4440 EXPORT_SYMBOL(zfs_setsecattr);
4442 #ifdef HAVE_UIO_ZEROCOPY
4444 * Tunable, both must be a power of 2.
4446 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4447 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4448 * an arcbuf for a partial block read
4450 int zcr_blksz_min = (1 << 10); /* 1K */
4451 int zcr_blksz_max = (1 << 17); /* 128K */
4455 zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
4457 znode_t *zp = ITOZ(ip);
4458 zfs_sb_t *zsb = ITOZSB(ip);
4459 int max_blksz = zsb->z_max_blksz;
4460 uio_t *uio = &xuio->xu_uio;
4461 ssize_t size = uio->uio_resid;
4462 offset_t offset = uio->uio_loffset;
4467 int preamble, postamble;
4469 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4470 return (SET_ERROR(EINVAL));
4477 * Loan out an arc_buf for write if write size is bigger than
4478 * max_blksz, and the file's block size is also max_blksz.
4481 if (size < blksz || zp->z_blksz != blksz) {
4483 return (SET_ERROR(EINVAL));
4486 * Caller requests buffers for write before knowing where the
4487 * write offset might be (e.g. NFS TCP write).
4492 preamble = P2PHASE(offset, blksz);
4494 preamble = blksz - preamble;
4499 postamble = P2PHASE(size, blksz);
4502 fullblk = size / blksz;
4503 (void) dmu_xuio_init(xuio,
4504 (preamble != 0) + fullblk + (postamble != 0));
4507 * Have to fix iov base/len for partial buffers. They
4508 * currently represent full arc_buf's.
4511 /* data begins in the middle of the arc_buf */
4512 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4515 (void) dmu_xuio_add(xuio, abuf,
4516 blksz - preamble, preamble);
4519 for (i = 0; i < fullblk; i++) {
4520 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4523 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
4527 /* data ends in the middle of the arc_buf */
4528 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4531 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
4536 * Loan out an arc_buf for read if the read size is larger than
4537 * the current file block size. Block alignment is not
4538 * considered. Partial arc_buf will be loaned out for read.
4540 blksz = zp->z_blksz;
4541 if (blksz < zcr_blksz_min)
4542 blksz = zcr_blksz_min;
4543 if (blksz > zcr_blksz_max)
4544 blksz = zcr_blksz_max;
4545 /* avoid potential complexity of dealing with it */
4546 if (blksz > max_blksz) {
4548 return (SET_ERROR(EINVAL));
4551 maxsize = zp->z_size - uio->uio_loffset;
4557 return (SET_ERROR(EINVAL));
4562 return (SET_ERROR(EINVAL));
4565 uio->uio_extflg = UIO_XUIO;
4566 XUIO_XUZC_RW(xuio) = ioflag;
4573 zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
4577 int ioflag = XUIO_XUZC_RW(xuio);
4579 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
4581 i = dmu_xuio_cnt(xuio);
4583 abuf = dmu_xuio_arcbuf(xuio, i);
4585 * if abuf == NULL, it must be a write buffer
4586 * that has been returned in zfs_write().
4589 dmu_return_arcbuf(abuf);
4590 ASSERT(abuf || ioflag == UIO_WRITE);
4593 dmu_xuio_fini(xuio);
4596 #endif /* HAVE_UIO_ZEROCOPY */
4598 #if defined(_KERNEL) && defined(HAVE_SPL)
4599 module_param(zfs_read_chunk_size, long, 0644);
4600 MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");