Commit | Line | Data |
---|---|---|
fe4fa4b8 DC |
1 | /* |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. | |
3 | * All Rights Reserved. | |
4 | * | |
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. | |
8 | * | |
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. | |
13 | * | |
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 | |
17 | */ | |
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_bmap_btree.h" | |
31 | #include "xfs_alloc_btree.h" | |
32 | #include "xfs_ialloc_btree.h" | |
33 | #include "xfs_btree.h" | |
34 | #include "xfs_dir2_sf.h" | |
35 | #include "xfs_attr_sf.h" | |
36 | #include "xfs_inode.h" | |
37 | #include "xfs_dinode.h" | |
38 | #include "xfs_error.h" | |
39 | #include "xfs_mru_cache.h" | |
40 | #include "xfs_filestream.h" | |
41 | #include "xfs_vnodeops.h" | |
42 | #include "xfs_utils.h" | |
43 | #include "xfs_buf_item.h" | |
44 | #include "xfs_inode_item.h" | |
45 | #include "xfs_rw.h" | |
46 | ||
a167b17e DC |
47 | #include <linux/kthread.h> |
48 | #include <linux/freezer.h> | |
49 | ||
fe4fa4b8 | 50 | /* |
683a8970 DC |
51 | * Sync all the inodes in the given AG according to the |
52 | * direction given by the flags. | |
fe4fa4b8 | 53 | */ |
683a8970 DC |
54 | STATIC int |
55 | xfs_sync_inodes_ag( | |
fe4fa4b8 | 56 | xfs_mount_t *mp, |
683a8970 | 57 | int ag, |
2030b5ab | 58 | int flags) |
fe4fa4b8 | 59 | { |
683a8970 | 60 | xfs_perag_t *pag = &mp->m_perag[ag]; |
683a8970 | 61 | int nr_found; |
8c38ab03 | 62 | uint32_t first_index = 0; |
683a8970 DC |
63 | int error = 0; |
64 | int last_error = 0; | |
65 | int fflag = XFS_B_ASYNC; | |
fe4fa4b8 | 66 | |
fe4fa4b8 DC |
67 | if (flags & SYNC_DELWRI) |
68 | fflag = XFS_B_DELWRI; | |
69 | if (flags & SYNC_WAIT) | |
70 | fflag = 0; /* synchronous overrides all */ | |
71 | ||
fe4fa4b8 | 72 | do { |
bc60a993 | 73 | struct inode *inode; |
bc60a993 | 74 | xfs_inode_t *ip = NULL; |
455486b9 | 75 | int lock_flags = XFS_ILOCK_SHARED; |
bc60a993 | 76 | |
fe4fa4b8 | 77 | /* |
683a8970 DC |
78 | * use a gang lookup to find the next inode in the tree |
79 | * as the tree is sparse and a gang lookup walks to find | |
80 | * the number of objects requested. | |
fe4fa4b8 | 81 | */ |
683a8970 DC |
82 | read_lock(&pag->pag_ici_lock); |
83 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, | |
84 | (void**)&ip, first_index, 1); | |
fe4fa4b8 | 85 | |
683a8970 DC |
86 | if (!nr_found) { |
87 | read_unlock(&pag->pag_ici_lock); | |
88 | break; | |
fe4fa4b8 DC |
89 | } |
90 | ||
8c38ab03 DC |
91 | /* |
92 | * Update the index for the next lookup. Catch overflows | |
93 | * into the next AG range which can occur if we have inodes | |
94 | * in the last block of the AG and we are currently | |
95 | * pointing to the last inode. | |
96 | */ | |
683a8970 | 97 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
8c38ab03 DC |
98 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) { |
99 | read_unlock(&pag->pag_ici_lock); | |
100 | break; | |
101 | } | |
fe4fa4b8 | 102 | |
683a8970 | 103 | /* nothing to sync during shutdown */ |
cb56a4b9 | 104 | if (XFS_FORCED_SHUTDOWN(mp)) { |
683a8970 | 105 | read_unlock(&pag->pag_ici_lock); |
fe4fa4b8 DC |
106 | return 0; |
107 | } | |
108 | ||
109 | /* | |
455486b9 DC |
110 | * If we can't get a reference on the inode, it must be |
111 | * in reclaim. Leave it for the reclaim code to flush. | |
fe4fa4b8 | 112 | */ |
455486b9 DC |
113 | inode = VFS_I(ip); |
114 | if (!igrab(inode)) { | |
683a8970 | 115 | read_unlock(&pag->pag_ici_lock); |
455486b9 DC |
116 | continue; |
117 | } | |
118 | read_unlock(&pag->pag_ici_lock); | |
119 | ||
120 | /* bad inodes are dealt with elsewhere */ | |
121 | if (is_bad_inode(inode)) { | |
122 | IRELE(ip); | |
123 | continue; | |
fe4fa4b8 | 124 | } |
bc60a993 | 125 | |
fe4fa4b8 DC |
126 | /* |
127 | * If we have to flush data or wait for I/O completion | |
455486b9 | 128 | * we need to hold the iolock. |
fe4fa4b8 | 129 | */ |
bc60a993 | 130 | if ((flags & SYNC_DELWRI) && VN_DIRTY(inode)) { |
455486b9 DC |
131 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
132 | lock_flags |= XFS_IOLOCK_SHARED; | |
683a8970 DC |
133 | error = xfs_flush_pages(ip, 0, -1, fflag, FI_NONE); |
134 | if (flags & SYNC_IOWAIT) | |
135 | vn_iowait(ip); | |
683a8970 | 136 | } |
455486b9 | 137 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
fe4fa4b8 | 138 | |
683a8970 | 139 | if ((flags & SYNC_ATTR) && !xfs_inode_clean(ip)) { |
fe4fa4b8 DC |
140 | if (flags & SYNC_WAIT) { |
141 | xfs_iflock(ip); | |
683a8970 DC |
142 | if (!xfs_inode_clean(ip)) |
143 | error = xfs_iflush(ip, XFS_IFLUSH_SYNC); | |
144 | else | |
145 | xfs_ifunlock(ip); | |
fe4fa4b8 | 146 | } else if (xfs_iflock_nowait(ip)) { |
683a8970 DC |
147 | if (!xfs_inode_clean(ip)) |
148 | error = xfs_iflush(ip, XFS_IFLUSH_DELWRI); | |
149 | else | |
150 | xfs_ifunlock(ip); | |
fe4fa4b8 DC |
151 | } |
152 | } | |
455486b9 | 153 | xfs_iput(ip, lock_flags); |
fe4fa4b8 | 154 | |
683a8970 | 155 | if (error) |
fe4fa4b8 | 156 | last_error = error; |
fe4fa4b8 DC |
157 | /* |
158 | * bail out if the filesystem is corrupted. | |
159 | */ | |
683a8970 | 160 | if (error == EFSCORRUPTED) |
fe4fa4b8 | 161 | return XFS_ERROR(error); |
fe4fa4b8 | 162 | |
683a8970 | 163 | } while (nr_found); |
fe4fa4b8 | 164 | |
683a8970 DC |
165 | return last_error; |
166 | } | |
fe4fa4b8 | 167 | |
683a8970 DC |
168 | int |
169 | xfs_sync_inodes( | |
170 | xfs_mount_t *mp, | |
2030b5ab | 171 | int flags) |
683a8970 DC |
172 | { |
173 | int error; | |
174 | int last_error; | |
175 | int i; | |
e9f1c6ee | 176 | int lflags = XFS_LOG_FORCE; |
fe4fa4b8 | 177 | |
683a8970 DC |
178 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
179 | return 0; | |
180 | error = 0; | |
181 | last_error = 0; | |
fe4fa4b8 | 182 | |
e9f1c6ee DC |
183 | if (flags & SYNC_WAIT) |
184 | lflags |= XFS_LOG_SYNC; | |
185 | ||
683a8970 DC |
186 | for (i = 0; i < mp->m_sb.sb_agcount; i++) { |
187 | if (!mp->m_perag[i].pag_ici_init) | |
188 | continue; | |
2030b5ab | 189 | error = xfs_sync_inodes_ag(mp, i, flags); |
683a8970 DC |
190 | if (error) |
191 | last_error = error; | |
192 | if (error == EFSCORRUPTED) | |
193 | break; | |
194 | } | |
e9f1c6ee DC |
195 | if (flags & SYNC_DELWRI) |
196 | xfs_log_force(mp, 0, lflags); | |
197 | ||
fe4fa4b8 DC |
198 | return XFS_ERROR(last_error); |
199 | } | |
200 | ||
2af75df7 CH |
201 | STATIC int |
202 | xfs_commit_dummy_trans( | |
203 | struct xfs_mount *mp, | |
204 | uint log_flags) | |
205 | { | |
206 | struct xfs_inode *ip = mp->m_rootip; | |
207 | struct xfs_trans *tp; | |
208 | int error; | |
209 | ||
210 | /* | |
211 | * Put a dummy transaction in the log to tell recovery | |
212 | * that all others are OK. | |
213 | */ | |
214 | tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1); | |
215 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0); | |
216 | if (error) { | |
217 | xfs_trans_cancel(tp, 0); | |
218 | return error; | |
219 | } | |
220 | ||
221 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
222 | ||
223 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
224 | xfs_trans_ihold(tp, ip); | |
225 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
226 | /* XXX(hch): ignoring the error here.. */ | |
227 | error = xfs_trans_commit(tp, 0); | |
228 | ||
229 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
230 | ||
231 | xfs_log_force(mp, 0, log_flags); | |
232 | return 0; | |
233 | } | |
234 | ||
e9f1c6ee | 235 | int |
2af75df7 CH |
236 | xfs_sync_fsdata( |
237 | struct xfs_mount *mp, | |
238 | int flags) | |
239 | { | |
240 | struct xfs_buf *bp; | |
241 | struct xfs_buf_log_item *bip; | |
242 | int error = 0; | |
243 | ||
244 | /* | |
245 | * If this is xfssyncd() then only sync the superblock if we can | |
246 | * lock it without sleeping and it is not pinned. | |
247 | */ | |
248 | if (flags & SYNC_BDFLUSH) { | |
249 | ASSERT(!(flags & SYNC_WAIT)); | |
250 | ||
251 | bp = xfs_getsb(mp, XFS_BUF_TRYLOCK); | |
252 | if (!bp) | |
253 | goto out; | |
254 | ||
255 | bip = XFS_BUF_FSPRIVATE(bp, struct xfs_buf_log_item *); | |
256 | if (!bip || !xfs_buf_item_dirty(bip) || XFS_BUF_ISPINNED(bp)) | |
257 | goto out_brelse; | |
258 | } else { | |
259 | bp = xfs_getsb(mp, 0); | |
260 | ||
261 | /* | |
262 | * If the buffer is pinned then push on the log so we won't | |
263 | * get stuck waiting in the write for someone, maybe | |
264 | * ourselves, to flush the log. | |
265 | * | |
266 | * Even though we just pushed the log above, we did not have | |
267 | * the superblock buffer locked at that point so it can | |
268 | * become pinned in between there and here. | |
269 | */ | |
270 | if (XFS_BUF_ISPINNED(bp)) | |
271 | xfs_log_force(mp, 0, XFS_LOG_FORCE); | |
272 | } | |
273 | ||
274 | ||
275 | if (flags & SYNC_WAIT) | |
276 | XFS_BUF_UNASYNC(bp); | |
277 | else | |
278 | XFS_BUF_ASYNC(bp); | |
279 | ||
280 | return xfs_bwrite(mp, bp); | |
281 | ||
282 | out_brelse: | |
283 | xfs_buf_relse(bp); | |
284 | out: | |
285 | return error; | |
e9f1c6ee DC |
286 | } |
287 | ||
288 | /* | |
a4e4c4f4 DC |
289 | * When remounting a filesystem read-only or freezing the filesystem, we have |
290 | * two phases to execute. This first phase is syncing the data before we | |
291 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
292 | * we've waited for all the transactions created by the first phase to | |
293 | * complete. The second phase ensures that the inodes are written to their | |
294 | * location on disk rather than just existing in transactions in the log. This | |
295 | * means after a quiesce there is no log replay required to write the inodes to | |
296 | * disk (this is the main difference between a sync and a quiesce). | |
297 | */ | |
298 | /* | |
299 | * First stage of freeze - no writers will make progress now we are here, | |
e9f1c6ee DC |
300 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
301 | * complete. Data is frozen at that point. Metadata is not frozen, | |
a4e4c4f4 DC |
302 | * transactions can still occur here so don't bother flushing the buftarg |
303 | * because it'll just get dirty again. | |
e9f1c6ee DC |
304 | */ |
305 | int | |
306 | xfs_quiesce_data( | |
307 | struct xfs_mount *mp) | |
308 | { | |
309 | int error; | |
310 | ||
311 | /* push non-blocking */ | |
312 | xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_BDFLUSH); | |
313 | XFS_QM_DQSYNC(mp, SYNC_BDFLUSH); | |
314 | xfs_filestream_flush(mp); | |
315 | ||
316 | /* push and block */ | |
317 | xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_WAIT|SYNC_IOWAIT); | |
318 | XFS_QM_DQSYNC(mp, SYNC_WAIT); | |
319 | ||
a4e4c4f4 | 320 | /* write superblock and hoover up shutdown errors */ |
e9f1c6ee DC |
321 | error = xfs_sync_fsdata(mp, 0); |
322 | ||
a4e4c4f4 | 323 | /* flush data-only devices */ |
e9f1c6ee DC |
324 | if (mp->m_rtdev_targp) |
325 | XFS_bflush(mp->m_rtdev_targp); | |
326 | ||
327 | return error; | |
2af75df7 CH |
328 | } |
329 | ||
76bf105c DC |
330 | STATIC void |
331 | xfs_quiesce_fs( | |
332 | struct xfs_mount *mp) | |
333 | { | |
334 | int count = 0, pincount; | |
335 | ||
336 | xfs_flush_buftarg(mp->m_ddev_targp, 0); | |
1dc3318a | 337 | xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
76bf105c DC |
338 | |
339 | /* | |
340 | * This loop must run at least twice. The first instance of the loop | |
341 | * will flush most meta data but that will generate more meta data | |
342 | * (typically directory updates). Which then must be flushed and | |
343 | * logged before we can write the unmount record. | |
344 | */ | |
345 | do { | |
346 | xfs_sync_inodes(mp, SYNC_ATTR|SYNC_WAIT); | |
347 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
348 | if (!pincount) { | |
349 | delay(50); | |
350 | count++; | |
351 | } | |
352 | } while (count < 2); | |
353 | } | |
354 | ||
355 | /* | |
356 | * Second stage of a quiesce. The data is already synced, now we have to take | |
357 | * care of the metadata. New transactions are already blocked, so we need to | |
358 | * wait for any remaining transactions to drain out before proceding. | |
359 | */ | |
360 | void | |
361 | xfs_quiesce_attr( | |
362 | struct xfs_mount *mp) | |
363 | { | |
364 | int error = 0; | |
365 | ||
366 | /* wait for all modifications to complete */ | |
367 | while (atomic_read(&mp->m_active_trans) > 0) | |
368 | delay(100); | |
369 | ||
370 | /* flush inodes and push all remaining buffers out to disk */ | |
371 | xfs_quiesce_fs(mp); | |
372 | ||
373 | ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0); | |
374 | ||
375 | /* Push the superblock and write an unmount record */ | |
376 | error = xfs_log_sbcount(mp, 1); | |
377 | if (error) | |
378 | xfs_fs_cmn_err(CE_WARN, mp, | |
379 | "xfs_attr_quiesce: failed to log sb changes. " | |
380 | "Frozen image may not be consistent."); | |
381 | xfs_log_unmount_write(mp); | |
382 | xfs_unmountfs_writesb(mp); | |
383 | } | |
384 | ||
a167b17e DC |
385 | /* |
386 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. | |
387 | * Doing this has two advantages: | |
388 | * - It saves on stack space, which is tight in certain situations | |
389 | * - It can be used (with care) as a mechanism to avoid deadlocks. | |
390 | * Flushing while allocating in a full filesystem requires both. | |
391 | */ | |
392 | STATIC void | |
393 | xfs_syncd_queue_work( | |
394 | struct xfs_mount *mp, | |
395 | void *data, | |
396 | void (*syncer)(struct xfs_mount *, void *)) | |
397 | { | |
398 | struct bhv_vfs_sync_work *work; | |
399 | ||
400 | work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP); | |
401 | INIT_LIST_HEAD(&work->w_list); | |
402 | work->w_syncer = syncer; | |
403 | work->w_data = data; | |
404 | work->w_mount = mp; | |
405 | spin_lock(&mp->m_sync_lock); | |
406 | list_add_tail(&work->w_list, &mp->m_sync_list); | |
407 | spin_unlock(&mp->m_sync_lock); | |
408 | wake_up_process(mp->m_sync_task); | |
409 | } | |
410 | ||
411 | /* | |
412 | * Flush delayed allocate data, attempting to free up reserved space | |
413 | * from existing allocations. At this point a new allocation attempt | |
414 | * has failed with ENOSPC and we are in the process of scratching our | |
415 | * heads, looking about for more room... | |
416 | */ | |
417 | STATIC void | |
418 | xfs_flush_inode_work( | |
419 | struct xfs_mount *mp, | |
420 | void *arg) | |
421 | { | |
422 | struct inode *inode = arg; | |
423 | filemap_flush(inode->i_mapping); | |
424 | iput(inode); | |
425 | } | |
426 | ||
427 | void | |
428 | xfs_flush_inode( | |
429 | xfs_inode_t *ip) | |
430 | { | |
431 | struct inode *inode = VFS_I(ip); | |
432 | ||
433 | igrab(inode); | |
434 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inode_work); | |
435 | delay(msecs_to_jiffies(500)); | |
436 | } | |
437 | ||
438 | /* | |
439 | * This is the "bigger hammer" version of xfs_flush_inode_work... | |
440 | * (IOW, "If at first you don't succeed, use a Bigger Hammer"). | |
441 | */ | |
442 | STATIC void | |
443 | xfs_flush_device_work( | |
444 | struct xfs_mount *mp, | |
445 | void *arg) | |
446 | { | |
447 | struct inode *inode = arg; | |
448 | sync_blockdev(mp->m_super->s_bdev); | |
449 | iput(inode); | |
450 | } | |
451 | ||
452 | void | |
453 | xfs_flush_device( | |
454 | xfs_inode_t *ip) | |
455 | { | |
456 | struct inode *inode = VFS_I(ip); | |
457 | ||
458 | igrab(inode); | |
459 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_device_work); | |
460 | delay(msecs_to_jiffies(500)); | |
461 | xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC); | |
462 | } | |
463 | ||
aacaa880 DC |
464 | /* |
465 | * Every sync period we need to unpin all items, reclaim inodes, sync | |
466 | * quota and write out the superblock. We might need to cover the log | |
467 | * to indicate it is idle. | |
468 | */ | |
a167b17e DC |
469 | STATIC void |
470 | xfs_sync_worker( | |
471 | struct xfs_mount *mp, | |
472 | void *unused) | |
473 | { | |
474 | int error; | |
475 | ||
aacaa880 DC |
476 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
477 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
1dc3318a | 478 | xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
aacaa880 DC |
479 | /* dgc: errors ignored here */ |
480 | error = XFS_QM_DQSYNC(mp, SYNC_BDFLUSH); | |
481 | error = xfs_sync_fsdata(mp, SYNC_BDFLUSH); | |
482 | if (xfs_log_need_covered(mp)) | |
483 | error = xfs_commit_dummy_trans(mp, XFS_LOG_FORCE); | |
484 | } | |
a167b17e DC |
485 | mp->m_sync_seq++; |
486 | wake_up(&mp->m_wait_single_sync_task); | |
487 | } | |
488 | ||
489 | STATIC int | |
490 | xfssyncd( | |
491 | void *arg) | |
492 | { | |
493 | struct xfs_mount *mp = arg; | |
494 | long timeleft; | |
495 | bhv_vfs_sync_work_t *work, *n; | |
496 | LIST_HEAD (tmp); | |
497 | ||
498 | set_freezable(); | |
499 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); | |
500 | for (;;) { | |
501 | timeleft = schedule_timeout_interruptible(timeleft); | |
502 | /* swsusp */ | |
503 | try_to_freeze(); | |
504 | if (kthread_should_stop() && list_empty(&mp->m_sync_list)) | |
505 | break; | |
506 | ||
507 | spin_lock(&mp->m_sync_lock); | |
508 | /* | |
509 | * We can get woken by laptop mode, to do a sync - | |
510 | * that's the (only!) case where the list would be | |
511 | * empty with time remaining. | |
512 | */ | |
513 | if (!timeleft || list_empty(&mp->m_sync_list)) { | |
514 | if (!timeleft) | |
515 | timeleft = xfs_syncd_centisecs * | |
516 | msecs_to_jiffies(10); | |
517 | INIT_LIST_HEAD(&mp->m_sync_work.w_list); | |
518 | list_add_tail(&mp->m_sync_work.w_list, | |
519 | &mp->m_sync_list); | |
520 | } | |
521 | list_for_each_entry_safe(work, n, &mp->m_sync_list, w_list) | |
522 | list_move(&work->w_list, &tmp); | |
523 | spin_unlock(&mp->m_sync_lock); | |
524 | ||
525 | list_for_each_entry_safe(work, n, &tmp, w_list) { | |
526 | (*work->w_syncer)(mp, work->w_data); | |
527 | list_del(&work->w_list); | |
528 | if (work == &mp->m_sync_work) | |
529 | continue; | |
530 | kmem_free(work); | |
531 | } | |
532 | } | |
533 | ||
534 | return 0; | |
535 | } | |
536 | ||
537 | int | |
538 | xfs_syncd_init( | |
539 | struct xfs_mount *mp) | |
540 | { | |
541 | mp->m_sync_work.w_syncer = xfs_sync_worker; | |
542 | mp->m_sync_work.w_mount = mp; | |
543 | mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd"); | |
544 | if (IS_ERR(mp->m_sync_task)) | |
545 | return -PTR_ERR(mp->m_sync_task); | |
546 | return 0; | |
547 | } | |
548 | ||
549 | void | |
550 | xfs_syncd_stop( | |
551 | struct xfs_mount *mp) | |
552 | { | |
553 | kthread_stop(mp->m_sync_task); | |
554 | } | |
555 | ||
fce08f2f | 556 | int |
1dc3318a | 557 | xfs_reclaim_inode( |
fce08f2f DC |
558 | xfs_inode_t *ip, |
559 | int locked, | |
560 | int sync_mode) | |
561 | { | |
562 | xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino); | |
563 | ||
564 | /* The hash lock here protects a thread in xfs_iget_core from | |
565 | * racing with us on linking the inode back with a vnode. | |
566 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
567 | * us. | |
568 | */ | |
569 | write_lock(&pag->pag_ici_lock); | |
570 | spin_lock(&ip->i_flags_lock); | |
571 | if (__xfs_iflags_test(ip, XFS_IRECLAIM) || | |
572 | !__xfs_iflags_test(ip, XFS_IRECLAIMABLE)) { | |
573 | spin_unlock(&ip->i_flags_lock); | |
574 | write_unlock(&pag->pag_ici_lock); | |
575 | if (locked) { | |
576 | xfs_ifunlock(ip); | |
577 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
578 | } | |
579 | return 1; | |
580 | } | |
581 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
582 | spin_unlock(&ip->i_flags_lock); | |
583 | write_unlock(&pag->pag_ici_lock); | |
584 | xfs_put_perag(ip->i_mount, pag); | |
585 | ||
586 | /* | |
587 | * If the inode is still dirty, then flush it out. If the inode | |
588 | * is not in the AIL, then it will be OK to flush it delwri as | |
589 | * long as xfs_iflush() does not keep any references to the inode. | |
590 | * We leave that decision up to xfs_iflush() since it has the | |
591 | * knowledge of whether it's OK to simply do a delwri flush of | |
592 | * the inode or whether we need to wait until the inode is | |
593 | * pulled from the AIL. | |
594 | * We get the flush lock regardless, though, just to make sure | |
595 | * we don't free it while it is being flushed. | |
596 | */ | |
597 | if (!locked) { | |
598 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
599 | xfs_iflock(ip); | |
600 | } | |
601 | ||
602 | /* | |
603 | * In the case of a forced shutdown we rely on xfs_iflush() to | |
604 | * wait for the inode to be unpinned before returning an error. | |
605 | */ | |
606 | if (!is_bad_inode(VFS_I(ip)) && xfs_iflush(ip, sync_mode) == 0) { | |
607 | /* synchronize with xfs_iflush_done */ | |
608 | xfs_iflock(ip); | |
609 | xfs_ifunlock(ip); | |
610 | } | |
611 | ||
612 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
613 | xfs_ireclaim(ip); | |
614 | return 0; | |
615 | } | |
616 | ||
11654513 DC |
617 | /* |
618 | * We set the inode flag atomically with the radix tree tag. | |
619 | * Once we get tag lookups on the radix tree, this inode flag | |
620 | * can go away. | |
621 | */ | |
396beb85 DC |
622 | void |
623 | xfs_inode_set_reclaim_tag( | |
624 | xfs_inode_t *ip) | |
625 | { | |
626 | xfs_mount_t *mp = ip->i_mount; | |
627 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
628 | ||
629 | read_lock(&pag->pag_ici_lock); | |
630 | spin_lock(&ip->i_flags_lock); | |
631 | radix_tree_tag_set(&pag->pag_ici_root, | |
632 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
11654513 | 633 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
396beb85 DC |
634 | spin_unlock(&ip->i_flags_lock); |
635 | read_unlock(&pag->pag_ici_lock); | |
636 | xfs_put_perag(mp, pag); | |
637 | } | |
638 | ||
639 | void | |
640 | __xfs_inode_clear_reclaim_tag( | |
641 | xfs_mount_t *mp, | |
642 | xfs_perag_t *pag, | |
643 | xfs_inode_t *ip) | |
644 | { | |
645 | radix_tree_tag_clear(&pag->pag_ici_root, | |
646 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
647 | } | |
648 | ||
649 | void | |
650 | xfs_inode_clear_reclaim_tag( | |
651 | xfs_inode_t *ip) | |
652 | { | |
653 | xfs_mount_t *mp = ip->i_mount; | |
654 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
655 | ||
656 | read_lock(&pag->pag_ici_lock); | |
657 | spin_lock(&ip->i_flags_lock); | |
658 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
659 | spin_unlock(&ip->i_flags_lock); | |
660 | read_unlock(&pag->pag_ici_lock); | |
661 | xfs_put_perag(mp, pag); | |
662 | } | |
663 | ||
7a3be02b DC |
664 | |
665 | STATIC void | |
666 | xfs_reclaim_inodes_ag( | |
fce08f2f | 667 | xfs_mount_t *mp, |
7a3be02b DC |
668 | int ag, |
669 | int noblock, | |
fce08f2f DC |
670 | int mode) |
671 | { | |
7a3be02b DC |
672 | xfs_inode_t *ip = NULL; |
673 | xfs_perag_t *pag = &mp->m_perag[ag]; | |
674 | int nr_found; | |
8c38ab03 | 675 | uint32_t first_index; |
7a3be02b | 676 | int skipped; |
fce08f2f DC |
677 | |
678 | restart: | |
7a3be02b DC |
679 | first_index = 0; |
680 | skipped = 0; | |
681 | do { | |
682 | /* | |
683 | * use a gang lookup to find the next inode in the tree | |
684 | * as the tree is sparse and a gang lookup walks to find | |
685 | * the number of objects requested. | |
686 | */ | |
687 | read_lock(&pag->pag_ici_lock); | |
688 | nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root, | |
689 | (void**)&ip, first_index, 1, | |
690 | XFS_ICI_RECLAIM_TAG); | |
691 | ||
692 | if (!nr_found) { | |
693 | read_unlock(&pag->pag_ici_lock); | |
694 | break; | |
695 | } | |
696 | ||
8c38ab03 DC |
697 | /* |
698 | * Update the index for the next lookup. Catch overflows | |
699 | * into the next AG range which can occur if we have inodes | |
700 | * in the last block of the AG and we are currently | |
701 | * pointing to the last inode. | |
702 | */ | |
7a3be02b | 703 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
8c38ab03 DC |
704 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) { |
705 | read_unlock(&pag->pag_ici_lock); | |
706 | break; | |
707 | } | |
7a3be02b DC |
708 | |
709 | ASSERT(xfs_iflags_test(ip, (XFS_IRECLAIMABLE|XFS_IRECLAIM))); | |
710 | ||
711 | /* ignore if already under reclaim */ | |
712 | if (xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
713 | read_unlock(&pag->pag_ici_lock); | |
714 | continue; | |
715 | } | |
716 | ||
fce08f2f | 717 | if (noblock) { |
7a3be02b DC |
718 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { |
719 | read_unlock(&pag->pag_ici_lock); | |
fce08f2f | 720 | continue; |
7a3be02b | 721 | } |
fce08f2f DC |
722 | if (xfs_ipincount(ip) || |
723 | !xfs_iflock_nowait(ip)) { | |
724 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
7a3be02b | 725 | read_unlock(&pag->pag_ici_lock); |
fce08f2f DC |
726 | continue; |
727 | } | |
728 | } | |
7a3be02b DC |
729 | read_unlock(&pag->pag_ici_lock); |
730 | ||
731 | /* | |
732 | * hmmm - this is an inode already in reclaim. Do | |
733 | * we even bother catching it here? | |
734 | */ | |
1dc3318a | 735 | if (xfs_reclaim_inode(ip, noblock, mode)) |
7a3be02b DC |
736 | skipped++; |
737 | } while (nr_found); | |
738 | ||
739 | if (skipped) { | |
740 | delay(1); | |
fce08f2f DC |
741 | goto restart; |
742 | } | |
7a3be02b DC |
743 | return; |
744 | ||
745 | } | |
746 | ||
747 | int | |
748 | xfs_reclaim_inodes( | |
749 | xfs_mount_t *mp, | |
750 | int noblock, | |
751 | int mode) | |
752 | { | |
753 | int i; | |
754 | ||
755 | for (i = 0; i < mp->m_sb.sb_agcount; i++) { | |
756 | if (!mp->m_perag[i].pag_ici_init) | |
757 | continue; | |
758 | xfs_reclaim_inodes_ag(mp, i, noblock, mode); | |
759 | } | |
fce08f2f DC |
760 | return 0; |
761 | } | |
762 | ||
763 |