aboutsummaryrefslogtreecommitdiffstats
path: root/fs/xfs/linux-2.6/xfs_sync.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/xfs/linux-2.6/xfs_sync.c')
-rw-r--r--fs/xfs/linux-2.6/xfs_sync.c479
1 files changed, 253 insertions, 226 deletions
diff --git a/fs/xfs/linux-2.6/xfs_sync.c b/fs/xfs/linux-2.6/xfs_sync.c
index f7ba76633c29..b619d6b8ca43 100644
--- a/fs/xfs/linux-2.6/xfs_sync.c
+++ b/fs/xfs/linux-2.6/xfs_sync.c
@@ -43,166 +43,267 @@
43#include "xfs_buf_item.h" 43#include "xfs_buf_item.h"
44#include "xfs_inode_item.h" 44#include "xfs_inode_item.h"
45#include "xfs_rw.h" 45#include "xfs_rw.h"
46#include "xfs_quota.h"
46 47
47#include <linux/kthread.h> 48#include <linux/kthread.h>
48#include <linux/freezer.h> 49#include <linux/freezer.h>
49 50
50/*
51 * Sync all the inodes in the given AG according to the
52 * direction given by the flags.
53 */
54STATIC int
55xfs_sync_inodes_ag(
56 xfs_mount_t *mp,
57 int ag,
58 int flags)
59{
60 xfs_perag_t *pag = &mp->m_perag[ag];
61 int nr_found;
62 uint32_t first_index = 0;
63 int error = 0;
64 int last_error = 0;
65 51
66 do { 52STATIC xfs_inode_t *
67 struct inode *inode; 53xfs_inode_ag_lookup(
68 xfs_inode_t *ip = NULL; 54 struct xfs_mount *mp,
69 int lock_flags = XFS_ILOCK_SHARED; 55 struct xfs_perag *pag,
56 uint32_t *first_index,
57 int tag)
58{
59 int nr_found;
60 struct xfs_inode *ip;
70 61
71 /* 62 /*
72 * use a gang lookup to find the next inode in the tree 63 * use a gang lookup to find the next inode in the tree
73 * as the tree is sparse and a gang lookup walks to find 64 * as the tree is sparse and a gang lookup walks to find
74 * the number of objects requested. 65 * the number of objects requested.
75 */ 66 */
76 read_lock(&pag->pag_ici_lock); 67 read_lock(&pag->pag_ici_lock);
68 if (tag == XFS_ICI_NO_TAG) {
77 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, 69 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
78 (void**)&ip, first_index, 1); 70 (void **)&ip, *first_index, 1);
71 } else {
72 nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root,
73 (void **)&ip, *first_index, 1, tag);
74 }
75 if (!nr_found)
76 goto unlock;
79 77
80 if (!nr_found) { 78 /*
81 read_unlock(&pag->pag_ici_lock); 79 * Update the index for the next lookup. Catch overflows
82 break; 80 * into the next AG range which can occur if we have inodes
83 } 81 * in the last block of the AG and we are currently
82 * pointing to the last inode.
83 */
84 *first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
85 if (*first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
86 goto unlock;
84 87
85 /* 88 return ip;
86 * Update the index for the next lookup. Catch overflows
87 * into the next AG range which can occur if we have inodes
88 * in the last block of the AG and we are currently
89 * pointing to the last inode.
90 */
91 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
92 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) {
93 read_unlock(&pag->pag_ici_lock);
94 break;
95 }
96 89
97 /* nothing to sync during shutdown */ 90unlock:
98 if (XFS_FORCED_SHUTDOWN(mp)) { 91 read_unlock(&pag->pag_ici_lock);
99 read_unlock(&pag->pag_ici_lock); 92 return NULL;
100 return 0; 93}
101 }
102 94
103 /* 95STATIC int
104 * If we can't get a reference on the inode, it must be 96xfs_inode_ag_walk(
105 * in reclaim. Leave it for the reclaim code to flush. 97 struct xfs_mount *mp,
106 */ 98 xfs_agnumber_t ag,
107 inode = VFS_I(ip); 99 int (*execute)(struct xfs_inode *ip,
108 if (!igrab(inode)) { 100 struct xfs_perag *pag, int flags),
109 read_unlock(&pag->pag_ici_lock); 101 int flags,
110 continue; 102 int tag)
111 } 103{
112 read_unlock(&pag->pag_ici_lock); 104 struct xfs_perag *pag = &mp->m_perag[ag];
105 uint32_t first_index;
106 int last_error = 0;
107 int skipped;
113 108
114 /* avoid new or bad inodes */ 109restart:
115 if (is_bad_inode(inode) || 110 skipped = 0;
116 xfs_iflags_test(ip, XFS_INEW)) { 111 first_index = 0;
117 IRELE(ip); 112 do {
118 continue; 113 int error = 0;
119 } 114 xfs_inode_t *ip;
120 115
121 /* 116 ip = xfs_inode_ag_lookup(mp, pag, &first_index, tag);
122 * If we have to flush data or wait for I/O completion 117 if (!ip)
123 * we need to hold the iolock. 118 break;
124 */
125 if (flags & SYNC_DELWRI) {
126 if (VN_DIRTY(inode)) {
127 if (flags & SYNC_TRYLOCK) {
128 if (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
129 lock_flags |= XFS_IOLOCK_SHARED;
130 } else {
131 xfs_ilock(ip, XFS_IOLOCK_SHARED);
132 lock_flags |= XFS_IOLOCK_SHARED;
133 }
134 if (lock_flags & XFS_IOLOCK_SHARED) {
135 error = xfs_flush_pages(ip, 0, -1,
136 (flags & SYNC_WAIT) ? 0
137 : XFS_B_ASYNC,
138 FI_NONE);
139 }
140 }
141 if (VN_CACHED(inode) && (flags & SYNC_IOWAIT))
142 xfs_ioend_wait(ip);
143 }
144 xfs_ilock(ip, XFS_ILOCK_SHARED);
145
146 if ((flags & SYNC_ATTR) && !xfs_inode_clean(ip)) {
147 if (flags & SYNC_WAIT) {
148 xfs_iflock(ip);
149 if (!xfs_inode_clean(ip))
150 error = xfs_iflush(ip, XFS_IFLUSH_SYNC);
151 else
152 xfs_ifunlock(ip);
153 } else if (xfs_iflock_nowait(ip)) {
154 if (!xfs_inode_clean(ip))
155 error = xfs_iflush(ip, XFS_IFLUSH_DELWRI);
156 else
157 xfs_ifunlock(ip);
158 }
159 }
160 xfs_iput(ip, lock_flags);
161 119
120 error = execute(ip, pag, flags);
121 if (error == EAGAIN) {
122 skipped++;
123 continue;
124 }
162 if (error) 125 if (error)
163 last_error = error; 126 last_error = error;
164 /* 127 /*
165 * bail out if the filesystem is corrupted. 128 * bail out if the filesystem is corrupted.
166 */ 129 */
167 if (error == EFSCORRUPTED) 130 if (error == EFSCORRUPTED)
168 return XFS_ERROR(error); 131 break;
169 132
170 } while (nr_found); 133 } while (1);
134
135 if (skipped) {
136 delay(1);
137 goto restart;
138 }
171 139
140 xfs_put_perag(mp, pag);
172 return last_error; 141 return last_error;
173} 142}
174 143
175int 144int
176xfs_sync_inodes( 145xfs_inode_ag_iterator(
177 xfs_mount_t *mp, 146 struct xfs_mount *mp,
178 int flags) 147 int (*execute)(struct xfs_inode *ip,
148 struct xfs_perag *pag, int flags),
149 int flags,
150 int tag)
179{ 151{
180 int error; 152 int error = 0;
181 int last_error; 153 int last_error = 0;
182 int i; 154 xfs_agnumber_t ag;
183 int lflags = XFS_LOG_FORCE;
184 155
185 if (mp->m_flags & XFS_MOUNT_RDONLY) 156 for (ag = 0; ag < mp->m_sb.sb_agcount; ag++) {
186 return 0; 157 if (!mp->m_perag[ag].pag_ici_init)
187 error = 0; 158 continue;
188 last_error = 0; 159 error = xfs_inode_ag_walk(mp, ag, execute, flags, tag);
160 if (error) {
161 last_error = error;
162 if (error == EFSCORRUPTED)
163 break;
164 }
165 }
166 return XFS_ERROR(last_error);
167}
168
169/* must be called with pag_ici_lock held and releases it */
170int
171xfs_sync_inode_valid(
172 struct xfs_inode *ip,
173 struct xfs_perag *pag)
174{
175 struct inode *inode = VFS_I(ip);
176
177 /* nothing to sync during shutdown */
178 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
179 read_unlock(&pag->pag_ici_lock);
180 return EFSCORRUPTED;
181 }
182
183 /*
184 * If we can't get a reference on the inode, it must be in reclaim.
185 * Leave it for the reclaim code to flush. Also avoid inodes that
186 * haven't been fully initialised.
187 */
188 if (!igrab(inode)) {
189 read_unlock(&pag->pag_ici_lock);
190 return ENOENT;
191 }
192 read_unlock(&pag->pag_ici_lock);
193
194 if (is_bad_inode(inode) || xfs_iflags_test(ip, XFS_INEW)) {
195 IRELE(ip);
196 return ENOENT;
197 }
198
199 return 0;
200}
201
202STATIC int
203xfs_sync_inode_data(
204 struct xfs_inode *ip,
205 struct xfs_perag *pag,
206 int flags)
207{
208 struct inode *inode = VFS_I(ip);
209 struct address_space *mapping = inode->i_mapping;
210 int error = 0;
211
212 error = xfs_sync_inode_valid(ip, pag);
213 if (error)
214 return error;
215
216 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
217 goto out_wait;
218
219 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) {
220 if (flags & SYNC_TRYLOCK)
221 goto out_wait;
222 xfs_ilock(ip, XFS_IOLOCK_SHARED);
223 }
224
225 error = xfs_flush_pages(ip, 0, -1, (flags & SYNC_WAIT) ?
226 0 : XFS_B_ASYNC, FI_NONE);
227 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
189 228
229 out_wait:
190 if (flags & SYNC_WAIT) 230 if (flags & SYNC_WAIT)
191 lflags |= XFS_LOG_SYNC; 231 xfs_ioend_wait(ip);
232 IRELE(ip);
233 return error;
234}
192 235
193 for (i = 0; i < mp->m_sb.sb_agcount; i++) { 236STATIC int
194 if (!mp->m_perag[i].pag_ici_init) 237xfs_sync_inode_attr(
195 continue; 238 struct xfs_inode *ip,
196 error = xfs_sync_inodes_ag(mp, i, flags); 239 struct xfs_perag *pag,
197 if (error) 240 int flags)
198 last_error = error; 241{
199 if (error == EFSCORRUPTED) 242 int error = 0;
200 break; 243
244 error = xfs_sync_inode_valid(ip, pag);
245 if (error)
246 return error;
247
248 xfs_ilock(ip, XFS_ILOCK_SHARED);
249 if (xfs_inode_clean(ip))
250 goto out_unlock;
251 if (!xfs_iflock_nowait(ip)) {
252 if (!(flags & SYNC_WAIT))
253 goto out_unlock;
254 xfs_iflock(ip);
201 } 255 }
202 if (flags & SYNC_DELWRI)
203 xfs_log_force(mp, 0, lflags);
204 256
205 return XFS_ERROR(last_error); 257 if (xfs_inode_clean(ip)) {
258 xfs_ifunlock(ip);
259 goto out_unlock;
260 }
261
262 error = xfs_iflush(ip, (flags & SYNC_WAIT) ?
263 XFS_IFLUSH_SYNC : XFS_IFLUSH_DELWRI);
264
265 out_unlock:
266 xfs_iunlock(ip, XFS_ILOCK_SHARED);
267 IRELE(ip);
268 return error;
269}
270
271/*
272 * Write out pagecache data for the whole filesystem.
273 */
274int
275xfs_sync_data(
276 struct xfs_mount *mp,
277 int flags)
278{
279 int error;
280
281 ASSERT((flags & ~(SYNC_TRYLOCK|SYNC_WAIT)) == 0);
282
283 error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags,
284 XFS_ICI_NO_TAG);
285 if (error)
286 return XFS_ERROR(error);
287
288 xfs_log_force(mp, 0,
289 (flags & SYNC_WAIT) ?
290 XFS_LOG_FORCE | XFS_LOG_SYNC :
291 XFS_LOG_FORCE);
292 return 0;
293}
294
295/*
296 * Write out inode metadata (attributes) for the whole filesystem.
297 */
298int
299xfs_sync_attr(
300 struct xfs_mount *mp,
301 int flags)
302{
303 ASSERT((flags & ~SYNC_WAIT) == 0);
304
305 return xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags,
306 XFS_ICI_NO_TAG);
206} 307}
207 308
208STATIC int 309STATIC int
@@ -252,7 +353,7 @@ xfs_sync_fsdata(
252 * If this is xfssyncd() then only sync the superblock if we can 353 * If this is xfssyncd() then only sync the superblock if we can
253 * lock it without sleeping and it is not pinned. 354 * lock it without sleeping and it is not pinned.
254 */ 355 */
255 if (flags & SYNC_BDFLUSH) { 356 if (flags & SYNC_TRYLOCK) {
256 ASSERT(!(flags & SYNC_WAIT)); 357 ASSERT(!(flags & SYNC_WAIT));
257 358
258 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK); 359 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
@@ -316,13 +417,13 @@ xfs_quiesce_data(
316 int error; 417 int error;
317 418
318 /* push non-blocking */ 419 /* push non-blocking */
319 xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_BDFLUSH); 420 xfs_sync_data(mp, 0);
320 XFS_QM_DQSYNC(mp, SYNC_BDFLUSH); 421 xfs_qm_sync(mp, SYNC_TRYLOCK);
321 xfs_filestream_flush(mp); 422 xfs_filestream_flush(mp);
322 423
323 /* push and block */ 424 /* push and block */
324 xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_WAIT|SYNC_IOWAIT); 425 xfs_sync_data(mp, SYNC_WAIT);
325 XFS_QM_DQSYNC(mp, SYNC_WAIT); 426 xfs_qm_sync(mp, SYNC_WAIT);
326 427
327 /* write superblock and hoover up shutdown errors */ 428 /* write superblock and hoover up shutdown errors */
328 error = xfs_sync_fsdata(mp, 0); 429 error = xfs_sync_fsdata(mp, 0);
@@ -341,7 +442,7 @@ xfs_quiesce_fs(
341 int count = 0, pincount; 442 int count = 0, pincount;
342 443
343 xfs_flush_buftarg(mp->m_ddev_targp, 0); 444 xfs_flush_buftarg(mp->m_ddev_targp, 0);
344 xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC); 445 xfs_reclaim_inodes(mp, XFS_IFLUSH_DELWRI_ELSE_ASYNC);
345 446
346 /* 447 /*
347 * This loop must run at least twice. The first instance of the loop 448 * This loop must run at least twice. The first instance of the loop
@@ -350,7 +451,7 @@ xfs_quiesce_fs(
350 * logged before we can write the unmount record. 451 * logged before we can write the unmount record.
351 */ 452 */
352 do { 453 do {
353 xfs_sync_inodes(mp, SYNC_ATTR|SYNC_WAIT); 454 xfs_sync_attr(mp, SYNC_WAIT);
354 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); 455 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
355 if (!pincount) { 456 if (!pincount) {
356 delay(50); 457 delay(50);
@@ -433,8 +534,8 @@ xfs_flush_inodes_work(
433 void *arg) 534 void *arg)
434{ 535{
435 struct inode *inode = arg; 536 struct inode *inode = arg;
436 xfs_sync_inodes(mp, SYNC_DELWRI | SYNC_TRYLOCK); 537 xfs_sync_data(mp, SYNC_TRYLOCK);
437 xfs_sync_inodes(mp, SYNC_DELWRI | SYNC_TRYLOCK | SYNC_IOWAIT); 538 xfs_sync_data(mp, SYNC_TRYLOCK | SYNC_WAIT);
438 iput(inode); 539 iput(inode);
439} 540}
440 541
@@ -465,10 +566,10 @@ xfs_sync_worker(
465 566
466 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { 567 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
467 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); 568 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
468 xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC); 569 xfs_reclaim_inodes(mp, XFS_IFLUSH_DELWRI_ELSE_ASYNC);
469 /* dgc: errors ignored here */ 570 /* dgc: errors ignored here */
470 error = XFS_QM_DQSYNC(mp, SYNC_BDFLUSH); 571 error = xfs_qm_sync(mp, SYNC_TRYLOCK);
471 error = xfs_sync_fsdata(mp, SYNC_BDFLUSH); 572 error = xfs_sync_fsdata(mp, SYNC_TRYLOCK);
472 if (xfs_log_need_covered(mp)) 573 if (xfs_log_need_covered(mp))
473 error = xfs_commit_dummy_trans(mp, XFS_LOG_FORCE); 574 error = xfs_commit_dummy_trans(mp, XFS_LOG_FORCE);
474 } 575 }
@@ -569,7 +670,7 @@ xfs_reclaim_inode(
569 xfs_ifunlock(ip); 670 xfs_ifunlock(ip);
570 xfs_iunlock(ip, XFS_ILOCK_EXCL); 671 xfs_iunlock(ip, XFS_ILOCK_EXCL);
571 } 672 }
572 return 1; 673 return -EAGAIN;
573 } 674 }
574 __xfs_iflags_set(ip, XFS_IRECLAIM); 675 __xfs_iflags_set(ip, XFS_IRECLAIM);
575 spin_unlock(&ip->i_flags_lock); 676 spin_unlock(&ip->i_flags_lock);
@@ -654,101 +755,27 @@ xfs_inode_clear_reclaim_tag(
654 xfs_put_perag(mp, pag); 755 xfs_put_perag(mp, pag);
655} 756}
656 757
657 758STATIC int
658STATIC void 759xfs_reclaim_inode_now(
659xfs_reclaim_inodes_ag( 760 struct xfs_inode *ip,
660 xfs_mount_t *mp, 761 struct xfs_perag *pag,
661 int ag, 762 int flags)
662 int noblock,
663 int mode)
664{ 763{
665 xfs_inode_t *ip = NULL; 764 /* ignore if already under reclaim */
666 xfs_perag_t *pag = &mp->m_perag[ag]; 765 if (xfs_iflags_test(ip, XFS_IRECLAIM)) {
667 int nr_found;
668 uint32_t first_index;
669 int skipped;
670
671restart:
672 first_index = 0;
673 skipped = 0;
674 do {
675 /*
676 * use a gang lookup to find the next inode in the tree
677 * as the tree is sparse and a gang lookup walks to find
678 * the number of objects requested.
679 */
680 read_lock(&pag->pag_ici_lock);
681 nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root,
682 (void**)&ip, first_index, 1,
683 XFS_ICI_RECLAIM_TAG);
684
685 if (!nr_found) {
686 read_unlock(&pag->pag_ici_lock);
687 break;
688 }
689
690 /*
691 * Update the index for the next lookup. Catch overflows
692 * into the next AG range which can occur if we have inodes
693 * in the last block of the AG and we are currently
694 * pointing to the last inode.
695 */
696 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
697 if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) {
698 read_unlock(&pag->pag_ici_lock);
699 break;
700 }
701
702 /* ignore if already under reclaim */
703 if (xfs_iflags_test(ip, XFS_IRECLAIM)) {
704 read_unlock(&pag->pag_ici_lock);
705 continue;
706 }
707
708 if (noblock) {
709 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
710 read_unlock(&pag->pag_ici_lock);
711 continue;
712 }
713 if (xfs_ipincount(ip) ||
714 !xfs_iflock_nowait(ip)) {
715 xfs_iunlock(ip, XFS_ILOCK_EXCL);
716 read_unlock(&pag->pag_ici_lock);
717 continue;
718 }
719 }
720 read_unlock(&pag->pag_ici_lock); 766 read_unlock(&pag->pag_ici_lock);
721 767 return 0;
722 /*
723 * hmmm - this is an inode already in reclaim. Do
724 * we even bother catching it here?
725 */
726 if (xfs_reclaim_inode(ip, noblock, mode))
727 skipped++;
728 } while (nr_found);
729
730 if (skipped) {
731 delay(1);
732 goto restart;
733 } 768 }
734 return; 769 read_unlock(&pag->pag_ici_lock);
735 770
771 return xfs_reclaim_inode(ip, 0, flags);
736} 772}
737 773
738int 774int
739xfs_reclaim_inodes( 775xfs_reclaim_inodes(
740 xfs_mount_t *mp, 776 xfs_mount_t *mp,
741 int noblock,
742 int mode) 777 int mode)
743{ 778{
744 int i; 779 return xfs_inode_ag_iterator(mp, xfs_reclaim_inode_now, mode,
745 780 XFS_ICI_RECLAIM_TAG);
746 for (i = 0; i < mp->m_sb.sb_agcount; i++) {
747 if (!mp->m_perag[i].pag_ici_init)
748 continue;
749 xfs_reclaim_inodes_ag(mp, i, noblock, mode);
750 }
751 return 0;
752} 781}
753
754
generated by cgit v1.2.2 (git 2.25.0) at 2025-12-07 22:26:34 -0500