udp: Drop socket lock for encapsulated packets
[linux-2.6.git] / fs / xfs / xfs_vfsops.c
blob4a9a43315a8662262906292d0d3c21293644d015
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_btree.h"
40 #include "xfs_alloc.h"
41 #include "xfs_ialloc.h"
42 #include "xfs_quota.h"
43 #include "xfs_error.h"
44 #include "xfs_bmap.h"
45 #include "xfs_rw.h"
46 #include "xfs_buf_item.h"
47 #include "xfs_log_priv.h"
48 #include "xfs_dir2_trace.h"
49 #include "xfs_extfree_item.h"
50 #include "xfs_acl.h"
51 #include "xfs_attr.h"
52 #include "xfs_clnt.h"
53 #include "xfs_mru_cache.h"
54 #include "xfs_filestream.h"
55 #include "xfs_fsops.h"
56 #include "xfs_vnodeops.h"
57 #include "xfs_vfsops.h"
58 #include "xfs_utils.h"
61 STATIC void
62 xfs_quiesce_fs(
63 xfs_mount_t *mp)
65 int count = 0, pincount;
67 xfs_flush_buftarg(mp->m_ddev_targp, 0);
68 xfs_finish_reclaim_all(mp, 0);
70 /* This loop must run at least twice.
71 * The first instance of the loop will flush
72 * most meta data but that will generate more
73 * meta data (typically directory updates).
74 * Which then must be flushed and logged before
75 * we can write the unmount record.
77 do {
78 xfs_syncsub(mp, SYNC_INODE_QUIESCE, NULL);
79 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
80 if (!pincount) {
81 delay(50);
82 count++;
84 } while (count < 2);
88 * Second stage of a quiesce. The data is already synced, now we have to take
89 * care of the metadata. New transactions are already blocked, so we need to
90 * wait for any remaining transactions to drain out before proceding.
92 void
93 xfs_attr_quiesce(
94 xfs_mount_t *mp)
96 int error = 0;
98 /* wait for all modifications to complete */
99 while (atomic_read(&mp->m_active_trans) > 0)
100 delay(100);
102 /* flush inodes and push all remaining buffers out to disk */
103 xfs_quiesce_fs(mp);
105 ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0);
107 /* Push the superblock and write an unmount record */
108 error = xfs_log_sbcount(mp, 1);
109 if (error)
110 xfs_fs_cmn_err(CE_WARN, mp,
111 "xfs_attr_quiesce: failed to log sb changes. "
112 "Frozen image may not be consistent.");
113 xfs_log_unmount_write(mp);
114 xfs_unmountfs_writesb(mp);
118 * xfs_unmount_flush implements a set of flush operation on special
119 * inodes, which are needed as a separate set of operations so that
120 * they can be called as part of relocation process.
123 xfs_unmount_flush(
124 xfs_mount_t *mp, /* Mount structure we are getting
125 rid of. */
126 int relocation) /* Called from vfs relocation. */
128 xfs_inode_t *rip = mp->m_rootip;
129 xfs_inode_t *rbmip;
130 xfs_inode_t *rsumip = NULL;
131 bhv_vnode_t *rvp = XFS_ITOV(rip);
132 int error;
134 xfs_ilock(rip, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
135 xfs_iflock(rip);
138 * Flush out the real time inodes.
140 if ((rbmip = mp->m_rbmip) != NULL) {
141 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
142 xfs_iflock(rbmip);
143 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
144 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
146 if (error == EFSCORRUPTED)
147 goto fscorrupt_out;
149 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
151 rsumip = mp->m_rsumip;
152 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
153 xfs_iflock(rsumip);
154 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
155 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
157 if (error == EFSCORRUPTED)
158 goto fscorrupt_out;
160 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
164 * Synchronously flush root inode to disk
166 error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
167 if (error == EFSCORRUPTED)
168 goto fscorrupt_out2;
170 if (vn_count(rvp) != 1 && !relocation) {
171 xfs_iunlock(rip, XFS_ILOCK_EXCL);
172 return XFS_ERROR(EBUSY);
176 * Release dquot that rootinode, rbmino and rsumino might be holding,
177 * flush and purge the quota inodes.
179 error = XFS_QM_UNMOUNT(mp);
180 if (error == EFSCORRUPTED)
181 goto fscorrupt_out2;
183 if (rbmip) {
184 IRELE(rbmip);
185 IRELE(rsumip);
188 xfs_iunlock(rip, XFS_ILOCK_EXCL);
189 return 0;
191 fscorrupt_out:
192 xfs_ifunlock(rip);
194 fscorrupt_out2:
195 xfs_iunlock(rip, XFS_ILOCK_EXCL);
197 return XFS_ERROR(EFSCORRUPTED);
201 * xfs_sync flushes any pending I/O to file system vfsp.
203 * This routine is called by vfs_sync() to make sure that things make it
204 * out to disk eventually, on sync() system calls to flush out everything,
205 * and when the file system is unmounted. For the vfs_sync() case, all
206 * we really need to do is sync out the log to make all of our meta-data
207 * updates permanent (except for timestamps). For calls from pflushd(),
208 * dirty pages are kept moving by calling pdflush() on the inodes
209 * containing them. We also flush the inodes that we can lock without
210 * sleeping and the superblock if we can lock it without sleeping from
211 * vfs_sync() so that items at the tail of the log are always moving out.
213 * Flags:
214 * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
215 * to sleep if we can help it. All we really need
216 * to do is ensure that the log is synced at least
217 * periodically. We also push the inodes and
218 * superblock if we can lock them without sleeping
219 * and they are not pinned.
220 * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
221 * set, then we really want to lock each inode and flush
222 * it.
223 * SYNC_WAIT - All the flushes that take place in this call should
224 * be synchronous.
225 * SYNC_DELWRI - This tells us to push dirty pages associated with
226 * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
227 * determine if they should be flushed sync, async, or
228 * delwri.
229 * SYNC_CLOSE - This flag is passed when the system is being
230 * unmounted. We should sync and invalidate everything.
231 * SYNC_FSDATA - This indicates that the caller would like to make
232 * sure the superblock is safe on disk. We can ensure
233 * this by simply making sure the log gets flushed
234 * if SYNC_BDFLUSH is set, and by actually writing it
235 * out otherwise.
236 * SYNC_IOWAIT - The caller wants us to wait for all data I/O to complete
237 * before we return (including direct I/O). Forms the drain
238 * side of the write barrier needed to safely quiesce the
239 * filesystem.
243 xfs_sync(
244 xfs_mount_t *mp,
245 int flags)
247 int error;
250 * Get the Quota Manager to flush the dquots.
252 * If XFS quota support is not enabled or this filesystem
253 * instance does not use quotas XFS_QM_DQSYNC will always
254 * return zero.
256 error = XFS_QM_DQSYNC(mp, flags);
257 if (error) {
259 * If we got an IO error, we will be shutting down.
260 * So, there's nothing more for us to do here.
262 ASSERT(error != EIO || XFS_FORCED_SHUTDOWN(mp));
263 if (XFS_FORCED_SHUTDOWN(mp))
264 return XFS_ERROR(error);
267 if (flags & SYNC_IOWAIT)
268 xfs_filestream_flush(mp);
270 return xfs_syncsub(mp, flags, NULL);
274 * xfs sync routine for internal use
276 * This routine supports all of the flags defined for the generic vfs_sync
277 * interface as explained above under xfs_sync.
281 xfs_sync_inodes(
282 xfs_mount_t *mp,
283 int flags,
284 int *bypassed)
286 xfs_inode_t *ip = NULL;
287 bhv_vnode_t *vp = NULL;
288 int error;
289 int last_error;
290 uint64_t fflag;
291 uint lock_flags;
292 uint base_lock_flags;
293 boolean_t mount_locked;
294 boolean_t vnode_refed;
295 int preempt;
296 xfs_iptr_t *ipointer;
297 #ifdef DEBUG
298 boolean_t ipointer_in = B_FALSE;
300 #define IPOINTER_SET ipointer_in = B_TRUE
301 #define IPOINTER_CLR ipointer_in = B_FALSE
302 #else
303 #define IPOINTER_SET
304 #define IPOINTER_CLR
305 #endif
308 /* Insert a marker record into the inode list after inode ip. The list
309 * must be locked when this is called. After the call the list will no
310 * longer be locked.
312 #define IPOINTER_INSERT(ip, mp) { \
313 ASSERT(ipointer_in == B_FALSE); \
314 ipointer->ip_mnext = ip->i_mnext; \
315 ipointer->ip_mprev = ip; \
316 ip->i_mnext = (xfs_inode_t *)ipointer; \
317 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
318 preempt = 0; \
319 XFS_MOUNT_IUNLOCK(mp); \
320 mount_locked = B_FALSE; \
321 IPOINTER_SET; \
324 /* Remove the marker from the inode list. If the marker was the only item
325 * in the list then there are no remaining inodes and we should zero out
326 * the whole list. If we are the current head of the list then move the head
327 * past us.
329 #define IPOINTER_REMOVE(ip, mp) { \
330 ASSERT(ipointer_in == B_TRUE); \
331 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
332 ip = ipointer->ip_mnext; \
333 ip->i_mprev = ipointer->ip_mprev; \
334 ipointer->ip_mprev->i_mnext = ip; \
335 if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
336 mp->m_inodes = ip; \
338 } else { \
339 ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
340 mp->m_inodes = NULL; \
341 ip = NULL; \
343 IPOINTER_CLR; \
346 #define XFS_PREEMPT_MASK 0x7f
348 ASSERT(!(flags & SYNC_BDFLUSH));
350 if (bypassed)
351 *bypassed = 0;
352 if (mp->m_flags & XFS_MOUNT_RDONLY)
353 return 0;
354 error = 0;
355 last_error = 0;
356 preempt = 0;
358 /* Allocate a reference marker */
359 ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
361 fflag = XFS_B_ASYNC; /* default is don't wait */
362 if (flags & SYNC_DELWRI)
363 fflag = XFS_B_DELWRI;
364 if (flags & SYNC_WAIT)
365 fflag = 0; /* synchronous overrides all */
367 base_lock_flags = XFS_ILOCK_SHARED;
368 if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
370 * We need the I/O lock if we're going to call any of
371 * the flush/inval routines.
373 base_lock_flags |= XFS_IOLOCK_SHARED;
376 XFS_MOUNT_ILOCK(mp);
378 ip = mp->m_inodes;
380 mount_locked = B_TRUE;
381 vnode_refed = B_FALSE;
383 IPOINTER_CLR;
385 do {
386 ASSERT(ipointer_in == B_FALSE);
387 ASSERT(vnode_refed == B_FALSE);
389 lock_flags = base_lock_flags;
392 * There were no inodes in the list, just break out
393 * of the loop.
395 if (ip == NULL) {
396 break;
400 * We found another sync thread marker - skip it
402 if (ip->i_mount == NULL) {
403 ip = ip->i_mnext;
404 continue;
407 vp = XFS_ITOV_NULL(ip);
410 * If the vnode is gone then this is being torn down,
411 * call reclaim if it is flushed, else let regular flush
412 * code deal with it later in the loop.
415 if (vp == NULL) {
416 /* Skip ones already in reclaim */
417 if (ip->i_flags & XFS_IRECLAIM) {
418 ip = ip->i_mnext;
419 continue;
421 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
422 ip = ip->i_mnext;
423 } else if ((xfs_ipincount(ip) == 0) &&
424 xfs_iflock_nowait(ip)) {
425 IPOINTER_INSERT(ip, mp);
427 xfs_finish_reclaim(ip, 1,
428 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
430 XFS_MOUNT_ILOCK(mp);
431 mount_locked = B_TRUE;
432 IPOINTER_REMOVE(ip, mp);
433 } else {
434 xfs_iunlock(ip, XFS_ILOCK_EXCL);
435 ip = ip->i_mnext;
437 continue;
440 if (VN_BAD(vp)) {
441 ip = ip->i_mnext;
442 continue;
445 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
446 XFS_MOUNT_IUNLOCK(mp);
447 kmem_free(ipointer);
448 return 0;
452 * Try to lock without sleeping. We're out of order with
453 * the inode list lock here, so if we fail we need to drop
454 * the mount lock and try again. If we're called from
455 * bdflush() here, then don't bother.
457 * The inode lock here actually coordinates with the
458 * almost spurious inode lock in xfs_ireclaim() to prevent
459 * the vnode we handle here without a reference from
460 * being freed while we reference it. If we lock the inode
461 * while it's on the mount list here, then the spurious inode
462 * lock in xfs_ireclaim() after the inode is pulled from
463 * the mount list will sleep until we release it here.
464 * This keeps the vnode from being freed while we reference
465 * it.
467 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
468 if (vp == NULL) {
469 ip = ip->i_mnext;
470 continue;
473 vp = vn_grab(vp);
474 if (vp == NULL) {
475 ip = ip->i_mnext;
476 continue;
479 IPOINTER_INSERT(ip, mp);
480 xfs_ilock(ip, lock_flags);
482 ASSERT(vp == XFS_ITOV(ip));
483 ASSERT(ip->i_mount == mp);
485 vnode_refed = B_TRUE;
488 /* From here on in the loop we may have a marker record
489 * in the inode list.
493 * If we have to flush data or wait for I/O completion
494 * we need to drop the ilock that we currently hold.
495 * If we need to drop the lock, insert a marker if we
496 * have not already done so.
498 if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) ||
499 ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) {
500 if (mount_locked) {
501 IPOINTER_INSERT(ip, mp);
503 xfs_iunlock(ip, XFS_ILOCK_SHARED);
505 if (flags & SYNC_CLOSE) {
506 /* Shutdown case. Flush and invalidate. */
507 if (XFS_FORCED_SHUTDOWN(mp))
508 xfs_tosspages(ip, 0, -1,
509 FI_REMAPF);
510 else
511 error = xfs_flushinval_pages(ip,
512 0, -1, FI_REMAPF);
513 } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
514 error = xfs_flush_pages(ip, 0,
515 -1, fflag, FI_NONE);
519 * When freezing, we need to wait ensure all I/O (including direct
520 * I/O) is complete to ensure no further data modification can take
521 * place after this point
523 if (flags & SYNC_IOWAIT)
524 vn_iowait(ip);
526 xfs_ilock(ip, XFS_ILOCK_SHARED);
529 if ((flags & SYNC_ATTR) &&
530 (ip->i_update_core ||
531 (ip->i_itemp && ip->i_itemp->ili_format.ilf_fields))) {
532 if (mount_locked)
533 IPOINTER_INSERT(ip, mp);
535 if (flags & SYNC_WAIT) {
536 xfs_iflock(ip);
537 error = xfs_iflush(ip, XFS_IFLUSH_SYNC);
540 * If we can't acquire the flush lock, then the inode
541 * is already being flushed so don't bother waiting.
543 * If we can lock it then do a delwri flush so we can
544 * combine multiple inode flushes in each disk write.
546 } else if (xfs_iflock_nowait(ip)) {
547 error = xfs_iflush(ip, XFS_IFLUSH_DELWRI);
548 } else if (bypassed) {
549 (*bypassed)++;
553 if (lock_flags != 0) {
554 xfs_iunlock(ip, lock_flags);
557 if (vnode_refed) {
559 * If we had to take a reference on the vnode
560 * above, then wait until after we've unlocked
561 * the inode to release the reference. This is
562 * because we can be already holding the inode
563 * lock when IRELE() calls xfs_inactive().
565 * Make sure to drop the mount lock before calling
566 * IRELE() so that we don't trip over ourselves if
567 * we have to go for the mount lock again in the
568 * inactive code.
570 if (mount_locked) {
571 IPOINTER_INSERT(ip, mp);
574 IRELE(ip);
576 vnode_refed = B_FALSE;
579 if (error) {
580 last_error = error;
584 * bail out if the filesystem is corrupted.
586 if (error == EFSCORRUPTED) {
587 if (!mount_locked) {
588 XFS_MOUNT_ILOCK(mp);
589 IPOINTER_REMOVE(ip, mp);
591 XFS_MOUNT_IUNLOCK(mp);
592 ASSERT(ipointer_in == B_FALSE);
593 kmem_free(ipointer);
594 return XFS_ERROR(error);
597 /* Let other threads have a chance at the mount lock
598 * if we have looped many times without dropping the
599 * lock.
601 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
602 if (mount_locked) {
603 IPOINTER_INSERT(ip, mp);
607 if (mount_locked == B_FALSE) {
608 XFS_MOUNT_ILOCK(mp);
609 mount_locked = B_TRUE;
610 IPOINTER_REMOVE(ip, mp);
611 continue;
614 ASSERT(ipointer_in == B_FALSE);
615 ip = ip->i_mnext;
617 } while (ip != mp->m_inodes);
619 XFS_MOUNT_IUNLOCK(mp);
621 ASSERT(ipointer_in == B_FALSE);
623 kmem_free(ipointer);
624 return XFS_ERROR(last_error);
628 * xfs sync routine for internal use
630 * This routine supports all of the flags defined for the generic vfs_sync
631 * interface as explained above under xfs_sync.
635 xfs_syncsub(
636 xfs_mount_t *mp,
637 int flags,
638 int *bypassed)
640 int error = 0;
641 int last_error = 0;
642 uint log_flags = XFS_LOG_FORCE;
643 xfs_buf_t *bp;
644 xfs_buf_log_item_t *bip;
647 * Sync out the log. This ensures that the log is periodically
648 * flushed even if there is not enough activity to fill it up.
650 if (flags & SYNC_WAIT)
651 log_flags |= XFS_LOG_SYNC;
653 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
655 if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
656 if (flags & SYNC_BDFLUSH)
657 xfs_finish_reclaim_all(mp, 1);
658 else
659 error = xfs_sync_inodes(mp, flags, bypassed);
663 * Flushing out dirty data above probably generated more
664 * log activity, so if this isn't vfs_sync() then flush
665 * the log again.
667 if (flags & SYNC_DELWRI) {
668 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
671 if (flags & SYNC_FSDATA) {
673 * If this is vfs_sync() then only sync the superblock
674 * if we can lock it without sleeping and it is not pinned.
676 if (flags & SYNC_BDFLUSH) {
677 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
678 if (bp != NULL) {
679 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
680 if ((bip != NULL) &&
681 xfs_buf_item_dirty(bip)) {
682 if (!(XFS_BUF_ISPINNED(bp))) {
683 XFS_BUF_ASYNC(bp);
684 error = xfs_bwrite(mp, bp);
685 } else {
686 xfs_buf_relse(bp);
688 } else {
689 xfs_buf_relse(bp);
692 } else {
693 bp = xfs_getsb(mp, 0);
695 * If the buffer is pinned then push on the log so
696 * we won't get stuck waiting in the write for
697 * someone, maybe ourselves, to flush the log.
698 * Even though we just pushed the log above, we
699 * did not have the superblock buffer locked at
700 * that point so it can become pinned in between
701 * there and here.
703 if (XFS_BUF_ISPINNED(bp))
704 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
705 if (flags & SYNC_WAIT)
706 XFS_BUF_UNASYNC(bp);
707 else
708 XFS_BUF_ASYNC(bp);
709 error = xfs_bwrite(mp, bp);
711 if (error) {
712 last_error = error;
717 * Now check to see if the log needs a "dummy" transaction.
719 if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
720 xfs_trans_t *tp;
721 xfs_inode_t *ip;
724 * Put a dummy transaction in the log to tell
725 * recovery that all others are OK.
727 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
728 if ((error = xfs_trans_reserve(tp, 0,
729 XFS_ICHANGE_LOG_RES(mp),
730 0, 0, 0))) {
731 xfs_trans_cancel(tp, 0);
732 return error;
735 ip = mp->m_rootip;
736 xfs_ilock(ip, XFS_ILOCK_EXCL);
738 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
739 xfs_trans_ihold(tp, ip);
740 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
741 error = xfs_trans_commit(tp, 0);
742 xfs_iunlock(ip, XFS_ILOCK_EXCL);
743 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
747 * When shutting down, we need to insure that the AIL is pushed
748 * to disk or the filesystem can appear corrupt from the PROM.
750 if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
751 XFS_bflush(mp->m_ddev_targp);
752 if (mp->m_rtdev_targp) {
753 XFS_bflush(mp->m_rtdev_targp);
757 return XFS_ERROR(last_error);