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 DC |
61 | int nr_found; |
62 | int first_index = 0; | |
63 | int error = 0; | |
64 | int last_error = 0; | |
65 | int fflag = XFS_B_ASYNC; | |
66 | int lock_flags = XFS_ILOCK_SHARED; | |
fe4fa4b8 | 67 | |
fe4fa4b8 DC |
68 | if (flags & SYNC_DELWRI) |
69 | fflag = XFS_B_DELWRI; | |
70 | if (flags & SYNC_WAIT) | |
71 | fflag = 0; /* synchronous overrides all */ | |
72 | ||
cb56a4b9 | 73 | if (flags & SYNC_DELWRI) { |
fe4fa4b8 DC |
74 | /* |
75 | * We need the I/O lock if we're going to call any of | |
76 | * the flush/inval routines. | |
77 | */ | |
683a8970 | 78 | lock_flags |= XFS_IOLOCK_SHARED; |
fe4fa4b8 DC |
79 | } |
80 | ||
fe4fa4b8 | 81 | do { |
bc60a993 DC |
82 | struct inode *inode; |
83 | boolean_t inode_refed; | |
84 | xfs_inode_t *ip = NULL; | |
85 | ||
fe4fa4b8 | 86 | /* |
683a8970 DC |
87 | * use a gang lookup to find the next inode in the tree |
88 | * as the tree is sparse and a gang lookup walks to find | |
89 | * the number of objects requested. | |
fe4fa4b8 | 90 | */ |
683a8970 DC |
91 | read_lock(&pag->pag_ici_lock); |
92 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, | |
93 | (void**)&ip, first_index, 1); | |
fe4fa4b8 | 94 | |
683a8970 DC |
95 | if (!nr_found) { |
96 | read_unlock(&pag->pag_ici_lock); | |
97 | break; | |
fe4fa4b8 DC |
98 | } |
99 | ||
683a8970 DC |
100 | /* update the index for the next lookup */ |
101 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); | |
fe4fa4b8 DC |
102 | |
103 | /* | |
683a8970 DC |
104 | * skip inodes in reclaim. Let xfs_syncsub do that for |
105 | * us so we don't need to worry. | |
fe4fa4b8 | 106 | */ |
bc60a993 | 107 | if (xfs_iflags_test(ip, (XFS_IRECLAIM|XFS_IRECLAIMABLE))) { |
683a8970 | 108 | read_unlock(&pag->pag_ici_lock); |
fe4fa4b8 DC |
109 | continue; |
110 | } | |
111 | ||
683a8970 | 112 | /* bad inodes are dealt with elsewhere */ |
bc60a993 DC |
113 | inode = VFS_I(ip); |
114 | if (is_bad_inode(inode)) { | |
683a8970 | 115 | read_unlock(&pag->pag_ici_lock); |
fe4fa4b8 DC |
116 | continue; |
117 | } | |
118 | ||
683a8970 | 119 | /* nothing to sync during shutdown */ |
cb56a4b9 | 120 | if (XFS_FORCED_SHUTDOWN(mp)) { |
683a8970 | 121 | read_unlock(&pag->pag_ici_lock); |
fe4fa4b8 DC |
122 | return 0; |
123 | } | |
124 | ||
125 | /* | |
bc60a993 DC |
126 | * If we can't get a reference on the VFS_I, the inode must be |
127 | * in reclaim. If we can get the inode lock without blocking, | |
128 | * it is safe to flush the inode because we hold the tree lock | |
129 | * and xfs_iextract will block right now. Hence if we lock the | |
130 | * inode while holding the tree lock, xfs_ireclaim() is | |
131 | * guaranteed to block on the inode lock we now hold and hence | |
132 | * it is safe to reference the inode until we drop the inode | |
133 | * locks completely. | |
fe4fa4b8 | 134 | */ |
bc60a993 DC |
135 | inode_refed = B_FALSE; |
136 | if (igrab(inode)) { | |
683a8970 | 137 | read_unlock(&pag->pag_ici_lock); |
fe4fa4b8 | 138 | xfs_ilock(ip, lock_flags); |
bc60a993 | 139 | inode_refed = B_TRUE; |
683a8970 | 140 | } else { |
bc60a993 DC |
141 | if (!xfs_ilock_nowait(ip, lock_flags)) { |
142 | /* leave it to reclaim */ | |
143 | read_unlock(&pag->pag_ici_lock); | |
144 | continue; | |
145 | } | |
683a8970 | 146 | read_unlock(&pag->pag_ici_lock); |
fe4fa4b8 | 147 | } |
bc60a993 | 148 | |
fe4fa4b8 DC |
149 | /* |
150 | * If we have to flush data or wait for I/O completion | |
151 | * we need to drop the ilock that we currently hold. | |
152 | * If we need to drop the lock, insert a marker if we | |
153 | * have not already done so. | |
154 | */ | |
bc60a993 | 155 | if ((flags & SYNC_DELWRI) && VN_DIRTY(inode)) { |
683a8970 DC |
156 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
157 | error = xfs_flush_pages(ip, 0, -1, fflag, FI_NONE); | |
158 | if (flags & SYNC_IOWAIT) | |
159 | vn_iowait(ip); | |
160 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
161 | } | |
fe4fa4b8 | 162 | |
683a8970 | 163 | if ((flags & SYNC_ATTR) && !xfs_inode_clean(ip)) { |
fe4fa4b8 DC |
164 | if (flags & SYNC_WAIT) { |
165 | xfs_iflock(ip); | |
683a8970 DC |
166 | if (!xfs_inode_clean(ip)) |
167 | error = xfs_iflush(ip, XFS_IFLUSH_SYNC); | |
168 | else | |
169 | xfs_ifunlock(ip); | |
fe4fa4b8 | 170 | } else if (xfs_iflock_nowait(ip)) { |
683a8970 DC |
171 | if (!xfs_inode_clean(ip)) |
172 | error = xfs_iflush(ip, XFS_IFLUSH_DELWRI); | |
173 | else | |
174 | xfs_ifunlock(ip); | |
fe4fa4b8 DC |
175 | } |
176 | } | |
177 | ||
683a8970 | 178 | if (lock_flags) |
fe4fa4b8 | 179 | xfs_iunlock(ip, lock_flags); |
fe4fa4b8 | 180 | |
bc60a993 | 181 | if (inode_refed) { |
fe4fa4b8 | 182 | IRELE(ip); |
fe4fa4b8 DC |
183 | } |
184 | ||
683a8970 | 185 | if (error) |
fe4fa4b8 | 186 | last_error = error; |
fe4fa4b8 DC |
187 | /* |
188 | * bail out if the filesystem is corrupted. | |
189 | */ | |
683a8970 | 190 | if (error == EFSCORRUPTED) |
fe4fa4b8 | 191 | return XFS_ERROR(error); |
fe4fa4b8 | 192 | |
683a8970 | 193 | } while (nr_found); |
fe4fa4b8 | 194 | |
683a8970 DC |
195 | return last_error; |
196 | } | |
fe4fa4b8 | 197 | |
683a8970 DC |
198 | int |
199 | xfs_sync_inodes( | |
200 | xfs_mount_t *mp, | |
2030b5ab | 201 | int flags) |
683a8970 DC |
202 | { |
203 | int error; | |
204 | int last_error; | |
205 | int i; | |
e9f1c6ee | 206 | int lflags = XFS_LOG_FORCE; |
fe4fa4b8 | 207 | |
683a8970 DC |
208 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
209 | return 0; | |
210 | error = 0; | |
211 | last_error = 0; | |
fe4fa4b8 | 212 | |
e9f1c6ee DC |
213 | if (flags & SYNC_WAIT) |
214 | lflags |= XFS_LOG_SYNC; | |
215 | ||
683a8970 DC |
216 | for (i = 0; i < mp->m_sb.sb_agcount; i++) { |
217 | if (!mp->m_perag[i].pag_ici_init) | |
218 | continue; | |
2030b5ab | 219 | error = xfs_sync_inodes_ag(mp, i, flags); |
683a8970 DC |
220 | if (error) |
221 | last_error = error; | |
222 | if (error == EFSCORRUPTED) | |
223 | break; | |
224 | } | |
e9f1c6ee DC |
225 | if (flags & SYNC_DELWRI) |
226 | xfs_log_force(mp, 0, lflags); | |
227 | ||
fe4fa4b8 DC |
228 | return XFS_ERROR(last_error); |
229 | } | |
230 | ||
2af75df7 CH |
231 | STATIC int |
232 | xfs_commit_dummy_trans( | |
233 | struct xfs_mount *mp, | |
234 | uint log_flags) | |
235 | { | |
236 | struct xfs_inode *ip = mp->m_rootip; | |
237 | struct xfs_trans *tp; | |
238 | int error; | |
239 | ||
240 | /* | |
241 | * Put a dummy transaction in the log to tell recovery | |
242 | * that all others are OK. | |
243 | */ | |
244 | tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1); | |
245 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0); | |
246 | if (error) { | |
247 | xfs_trans_cancel(tp, 0); | |
248 | return error; | |
249 | } | |
250 | ||
251 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
252 | ||
253 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
254 | xfs_trans_ihold(tp, ip); | |
255 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
256 | /* XXX(hch): ignoring the error here.. */ | |
257 | error = xfs_trans_commit(tp, 0); | |
258 | ||
259 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
260 | ||
261 | xfs_log_force(mp, 0, log_flags); | |
262 | return 0; | |
263 | } | |
264 | ||
e9f1c6ee | 265 | int |
2af75df7 CH |
266 | xfs_sync_fsdata( |
267 | struct xfs_mount *mp, | |
268 | int flags) | |
269 | { | |
270 | struct xfs_buf *bp; | |
271 | struct xfs_buf_log_item *bip; | |
272 | int error = 0; | |
273 | ||
274 | /* | |
275 | * If this is xfssyncd() then only sync the superblock if we can | |
276 | * lock it without sleeping and it is not pinned. | |
277 | */ | |
278 | if (flags & SYNC_BDFLUSH) { | |
279 | ASSERT(!(flags & SYNC_WAIT)); | |
280 | ||
281 | bp = xfs_getsb(mp, XFS_BUF_TRYLOCK); | |
282 | if (!bp) | |
283 | goto out; | |
284 | ||
285 | bip = XFS_BUF_FSPRIVATE(bp, struct xfs_buf_log_item *); | |
286 | if (!bip || !xfs_buf_item_dirty(bip) || XFS_BUF_ISPINNED(bp)) | |
287 | goto out_brelse; | |
288 | } else { | |
289 | bp = xfs_getsb(mp, 0); | |
290 | ||
291 | /* | |
292 | * If the buffer is pinned then push on the log so we won't | |
293 | * get stuck waiting in the write for someone, maybe | |
294 | * ourselves, to flush the log. | |
295 | * | |
296 | * Even though we just pushed the log above, we did not have | |
297 | * the superblock buffer locked at that point so it can | |
298 | * become pinned in between there and here. | |
299 | */ | |
300 | if (XFS_BUF_ISPINNED(bp)) | |
301 | xfs_log_force(mp, 0, XFS_LOG_FORCE); | |
302 | } | |
303 | ||
304 | ||
305 | if (flags & SYNC_WAIT) | |
306 | XFS_BUF_UNASYNC(bp); | |
307 | else | |
308 | XFS_BUF_ASYNC(bp); | |
309 | ||
310 | return xfs_bwrite(mp, bp); | |
311 | ||
312 | out_brelse: | |
313 | xfs_buf_relse(bp); | |
314 | out: | |
315 | return error; | |
e9f1c6ee DC |
316 | } |
317 | ||
318 | /* | |
a4e4c4f4 DC |
319 | * When remounting a filesystem read-only or freezing the filesystem, we have |
320 | * two phases to execute. This first phase is syncing the data before we | |
321 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
322 | * we've waited for all the transactions created by the first phase to | |
323 | * complete. The second phase ensures that the inodes are written to their | |
324 | * location on disk rather than just existing in transactions in the log. This | |
325 | * means after a quiesce there is no log replay required to write the inodes to | |
326 | * disk (this is the main difference between a sync and a quiesce). | |
327 | */ | |
328 | /* | |
329 | * First stage of freeze - no writers will make progress now we are here, | |
e9f1c6ee DC |
330 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
331 | * complete. Data is frozen at that point. Metadata is not frozen, | |
a4e4c4f4 DC |
332 | * transactions can still occur here so don't bother flushing the buftarg |
333 | * because it'll just get dirty again. | |
e9f1c6ee DC |
334 | */ |
335 | int | |
336 | xfs_quiesce_data( | |
337 | struct xfs_mount *mp) | |
338 | { | |
339 | int error; | |
340 | ||
341 | /* push non-blocking */ | |
342 | xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_BDFLUSH); | |
343 | XFS_QM_DQSYNC(mp, SYNC_BDFLUSH); | |
344 | xfs_filestream_flush(mp); | |
345 | ||
346 | /* push and block */ | |
347 | xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_WAIT|SYNC_IOWAIT); | |
348 | XFS_QM_DQSYNC(mp, SYNC_WAIT); | |
349 | ||
a4e4c4f4 | 350 | /* write superblock and hoover up shutdown errors */ |
e9f1c6ee DC |
351 | error = xfs_sync_fsdata(mp, 0); |
352 | ||
a4e4c4f4 | 353 | /* flush data-only devices */ |
e9f1c6ee DC |
354 | if (mp->m_rtdev_targp) |
355 | XFS_bflush(mp->m_rtdev_targp); | |
356 | ||
357 | return error; | |
2af75df7 CH |
358 | } |
359 | ||
76bf105c DC |
360 | STATIC void |
361 | xfs_quiesce_fs( | |
362 | struct xfs_mount *mp) | |
363 | { | |
364 | int count = 0, pincount; | |
365 | ||
366 | xfs_flush_buftarg(mp->m_ddev_targp, 0); | |
1dc3318a | 367 | xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
76bf105c DC |
368 | |
369 | /* | |
370 | * This loop must run at least twice. The first instance of the loop | |
371 | * will flush most meta data but that will generate more meta data | |
372 | * (typically directory updates). Which then must be flushed and | |
373 | * logged before we can write the unmount record. | |
374 | */ | |
375 | do { | |
376 | xfs_sync_inodes(mp, SYNC_ATTR|SYNC_WAIT); | |
377 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
378 | if (!pincount) { | |
379 | delay(50); | |
380 | count++; | |
381 | } | |
382 | } while (count < 2); | |
383 | } | |
384 | ||
385 | /* | |
386 | * Second stage of a quiesce. The data is already synced, now we have to take | |
387 | * care of the metadata. New transactions are already blocked, so we need to | |
388 | * wait for any remaining transactions to drain out before proceding. | |
389 | */ | |
390 | void | |
391 | xfs_quiesce_attr( | |
392 | struct xfs_mount *mp) | |
393 | { | |
394 | int error = 0; | |
395 | ||
396 | /* wait for all modifications to complete */ | |
397 | while (atomic_read(&mp->m_active_trans) > 0) | |
398 | delay(100); | |
399 | ||
400 | /* flush inodes and push all remaining buffers out to disk */ | |
401 | xfs_quiesce_fs(mp); | |
402 | ||
403 | ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0); | |
404 | ||
405 | /* Push the superblock and write an unmount record */ | |
406 | error = xfs_log_sbcount(mp, 1); | |
407 | if (error) | |
408 | xfs_fs_cmn_err(CE_WARN, mp, | |
409 | "xfs_attr_quiesce: failed to log sb changes. " | |
410 | "Frozen image may not be consistent."); | |
411 | xfs_log_unmount_write(mp); | |
412 | xfs_unmountfs_writesb(mp); | |
413 | } | |
414 | ||
a167b17e DC |
415 | /* |
416 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. | |
417 | * Doing this has two advantages: | |
418 | * - It saves on stack space, which is tight in certain situations | |
419 | * - It can be used (with care) as a mechanism to avoid deadlocks. | |
420 | * Flushing while allocating in a full filesystem requires both. | |
421 | */ | |
422 | STATIC void | |
423 | xfs_syncd_queue_work( | |
424 | struct xfs_mount *mp, | |
425 | void *data, | |
426 | void (*syncer)(struct xfs_mount *, void *)) | |
427 | { | |
428 | struct bhv_vfs_sync_work *work; | |
429 | ||
430 | work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP); | |
431 | INIT_LIST_HEAD(&work->w_list); | |
432 | work->w_syncer = syncer; | |
433 | work->w_data = data; | |
434 | work->w_mount = mp; | |
435 | spin_lock(&mp->m_sync_lock); | |
436 | list_add_tail(&work->w_list, &mp->m_sync_list); | |
437 | spin_unlock(&mp->m_sync_lock); | |
438 | wake_up_process(mp->m_sync_task); | |
439 | } | |
440 | ||
441 | /* | |
442 | * Flush delayed allocate data, attempting to free up reserved space | |
443 | * from existing allocations. At this point a new allocation attempt | |
444 | * has failed with ENOSPC and we are in the process of scratching our | |
445 | * heads, looking about for more room... | |
446 | */ | |
447 | STATIC void | |
448 | xfs_flush_inode_work( | |
449 | struct xfs_mount *mp, | |
450 | void *arg) | |
451 | { | |
452 | struct inode *inode = arg; | |
453 | filemap_flush(inode->i_mapping); | |
454 | iput(inode); | |
455 | } | |
456 | ||
457 | void | |
458 | xfs_flush_inode( | |
459 | xfs_inode_t *ip) | |
460 | { | |
461 | struct inode *inode = VFS_I(ip); | |
462 | ||
463 | igrab(inode); | |
464 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inode_work); | |
465 | delay(msecs_to_jiffies(500)); | |
466 | } | |
467 | ||
468 | /* | |
469 | * This is the "bigger hammer" version of xfs_flush_inode_work... | |
470 | * (IOW, "If at first you don't succeed, use a Bigger Hammer"). | |
471 | */ | |
472 | STATIC void | |
473 | xfs_flush_device_work( | |
474 | struct xfs_mount *mp, | |
475 | void *arg) | |
476 | { | |
477 | struct inode *inode = arg; | |
478 | sync_blockdev(mp->m_super->s_bdev); | |
479 | iput(inode); | |
480 | } | |
481 | ||
482 | void | |
483 | xfs_flush_device( | |
484 | xfs_inode_t *ip) | |
485 | { | |
486 | struct inode *inode = VFS_I(ip); | |
487 | ||
488 | igrab(inode); | |
489 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_device_work); | |
490 | delay(msecs_to_jiffies(500)); | |
491 | xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC); | |
492 | } | |
493 | ||
aacaa880 DC |
494 | /* |
495 | * Every sync period we need to unpin all items, reclaim inodes, sync | |
496 | * quota and write out the superblock. We might need to cover the log | |
497 | * to indicate it is idle. | |
498 | */ | |
a167b17e DC |
499 | STATIC void |
500 | xfs_sync_worker( | |
501 | struct xfs_mount *mp, | |
502 | void *unused) | |
503 | { | |
504 | int error; | |
505 | ||
aacaa880 DC |
506 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
507 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
1dc3318a | 508 | xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
aacaa880 DC |
509 | /* dgc: errors ignored here */ |
510 | error = XFS_QM_DQSYNC(mp, SYNC_BDFLUSH); | |
511 | error = xfs_sync_fsdata(mp, SYNC_BDFLUSH); | |
512 | if (xfs_log_need_covered(mp)) | |
513 | error = xfs_commit_dummy_trans(mp, XFS_LOG_FORCE); | |
514 | } | |
a167b17e DC |
515 | mp->m_sync_seq++; |
516 | wake_up(&mp->m_wait_single_sync_task); | |
517 | } | |
518 | ||
519 | STATIC int | |
520 | xfssyncd( | |
521 | void *arg) | |
522 | { | |
523 | struct xfs_mount *mp = arg; | |
524 | long timeleft; | |
525 | bhv_vfs_sync_work_t *work, *n; | |
526 | LIST_HEAD (tmp); | |
527 | ||
528 | set_freezable(); | |
529 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); | |
530 | for (;;) { | |
531 | timeleft = schedule_timeout_interruptible(timeleft); | |
532 | /* swsusp */ | |
533 | try_to_freeze(); | |
534 | if (kthread_should_stop() && list_empty(&mp->m_sync_list)) | |
535 | break; | |
536 | ||
537 | spin_lock(&mp->m_sync_lock); | |
538 | /* | |
539 | * We can get woken by laptop mode, to do a sync - | |
540 | * that's the (only!) case where the list would be | |
541 | * empty with time remaining. | |
542 | */ | |
543 | if (!timeleft || list_empty(&mp->m_sync_list)) { | |
544 | if (!timeleft) | |
545 | timeleft = xfs_syncd_centisecs * | |
546 | msecs_to_jiffies(10); | |
547 | INIT_LIST_HEAD(&mp->m_sync_work.w_list); | |
548 | list_add_tail(&mp->m_sync_work.w_list, | |
549 | &mp->m_sync_list); | |
550 | } | |
551 | list_for_each_entry_safe(work, n, &mp->m_sync_list, w_list) | |
552 | list_move(&work->w_list, &tmp); | |
553 | spin_unlock(&mp->m_sync_lock); | |
554 | ||
555 | list_for_each_entry_safe(work, n, &tmp, w_list) { | |
556 | (*work->w_syncer)(mp, work->w_data); | |
557 | list_del(&work->w_list); | |
558 | if (work == &mp->m_sync_work) | |
559 | continue; | |
560 | kmem_free(work); | |
561 | } | |
562 | } | |
563 | ||
564 | return 0; | |
565 | } | |
566 | ||
567 | int | |
568 | xfs_syncd_init( | |
569 | struct xfs_mount *mp) | |
570 | { | |
571 | mp->m_sync_work.w_syncer = xfs_sync_worker; | |
572 | mp->m_sync_work.w_mount = mp; | |
573 | mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd"); | |
574 | if (IS_ERR(mp->m_sync_task)) | |
575 | return -PTR_ERR(mp->m_sync_task); | |
576 | return 0; | |
577 | } | |
578 | ||
579 | void | |
580 | xfs_syncd_stop( | |
581 | struct xfs_mount *mp) | |
582 | { | |
583 | kthread_stop(mp->m_sync_task); | |
584 | } | |
585 | ||
fce08f2f | 586 | int |
1dc3318a | 587 | xfs_reclaim_inode( |
fce08f2f DC |
588 | xfs_inode_t *ip, |
589 | int locked, | |
590 | int sync_mode) | |
591 | { | |
592 | xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino); | |
593 | ||
594 | /* The hash lock here protects a thread in xfs_iget_core from | |
595 | * racing with us on linking the inode back with a vnode. | |
596 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
597 | * us. | |
598 | */ | |
599 | write_lock(&pag->pag_ici_lock); | |
600 | spin_lock(&ip->i_flags_lock); | |
601 | if (__xfs_iflags_test(ip, XFS_IRECLAIM) || | |
602 | !__xfs_iflags_test(ip, XFS_IRECLAIMABLE)) { | |
603 | spin_unlock(&ip->i_flags_lock); | |
604 | write_unlock(&pag->pag_ici_lock); | |
605 | if (locked) { | |
606 | xfs_ifunlock(ip); | |
607 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
608 | } | |
609 | return 1; | |
610 | } | |
611 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
612 | spin_unlock(&ip->i_flags_lock); | |
613 | write_unlock(&pag->pag_ici_lock); | |
614 | xfs_put_perag(ip->i_mount, pag); | |
615 | ||
616 | /* | |
617 | * If the inode is still dirty, then flush it out. If the inode | |
618 | * is not in the AIL, then it will be OK to flush it delwri as | |
619 | * long as xfs_iflush() does not keep any references to the inode. | |
620 | * We leave that decision up to xfs_iflush() since it has the | |
621 | * knowledge of whether it's OK to simply do a delwri flush of | |
622 | * the inode or whether we need to wait until the inode is | |
623 | * pulled from the AIL. | |
624 | * We get the flush lock regardless, though, just to make sure | |
625 | * we don't free it while it is being flushed. | |
626 | */ | |
627 | if (!locked) { | |
628 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
629 | xfs_iflock(ip); | |
630 | } | |
631 | ||
632 | /* | |
633 | * In the case of a forced shutdown we rely on xfs_iflush() to | |
634 | * wait for the inode to be unpinned before returning an error. | |
635 | */ | |
636 | if (!is_bad_inode(VFS_I(ip)) && xfs_iflush(ip, sync_mode) == 0) { | |
637 | /* synchronize with xfs_iflush_done */ | |
638 | xfs_iflock(ip); | |
639 | xfs_ifunlock(ip); | |
640 | } | |
641 | ||
642 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
643 | xfs_ireclaim(ip); | |
644 | return 0; | |
645 | } | |
646 | ||
396beb85 DC |
647 | void |
648 | xfs_inode_set_reclaim_tag( | |
649 | xfs_inode_t *ip) | |
650 | { | |
651 | xfs_mount_t *mp = ip->i_mount; | |
652 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
653 | ||
654 | read_lock(&pag->pag_ici_lock); | |
655 | spin_lock(&ip->i_flags_lock); | |
656 | radix_tree_tag_set(&pag->pag_ici_root, | |
657 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
658 | spin_unlock(&ip->i_flags_lock); | |
659 | read_unlock(&pag->pag_ici_lock); | |
660 | xfs_put_perag(mp, pag); | |
661 | } | |
662 | ||
663 | void | |
664 | __xfs_inode_clear_reclaim_tag( | |
665 | xfs_mount_t *mp, | |
666 | xfs_perag_t *pag, | |
667 | xfs_inode_t *ip) | |
668 | { | |
669 | radix_tree_tag_clear(&pag->pag_ici_root, | |
670 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
671 | } | |
672 | ||
673 | void | |
674 | xfs_inode_clear_reclaim_tag( | |
675 | xfs_inode_t *ip) | |
676 | { | |
677 | xfs_mount_t *mp = ip->i_mount; | |
678 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
679 | ||
680 | read_lock(&pag->pag_ici_lock); | |
681 | spin_lock(&ip->i_flags_lock); | |
682 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
683 | spin_unlock(&ip->i_flags_lock); | |
684 | read_unlock(&pag->pag_ici_lock); | |
685 | xfs_put_perag(mp, pag); | |
686 | } | |
687 | ||
7a3be02b DC |
688 | |
689 | STATIC void | |
690 | xfs_reclaim_inodes_ag( | |
fce08f2f | 691 | xfs_mount_t *mp, |
7a3be02b DC |
692 | int ag, |
693 | int noblock, | |
fce08f2f DC |
694 | int mode) |
695 | { | |
7a3be02b DC |
696 | xfs_inode_t *ip = NULL; |
697 | xfs_perag_t *pag = &mp->m_perag[ag]; | |
698 | int nr_found; | |
699 | int first_index; | |
700 | int skipped; | |
fce08f2f DC |
701 | |
702 | restart: | |
7a3be02b DC |
703 | first_index = 0; |
704 | skipped = 0; | |
705 | do { | |
706 | /* | |
707 | * use a gang lookup to find the next inode in the tree | |
708 | * as the tree is sparse and a gang lookup walks to find | |
709 | * the number of objects requested. | |
710 | */ | |
711 | read_lock(&pag->pag_ici_lock); | |
712 | nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root, | |
713 | (void**)&ip, first_index, 1, | |
714 | XFS_ICI_RECLAIM_TAG); | |
715 | ||
716 | if (!nr_found) { | |
717 | read_unlock(&pag->pag_ici_lock); | |
718 | break; | |
719 | } | |
720 | ||
721 | /* update the index for the next lookup */ | |
722 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); | |
723 | ||
724 | ASSERT(xfs_iflags_test(ip, (XFS_IRECLAIMABLE|XFS_IRECLAIM))); | |
725 | ||
726 | /* ignore if already under reclaim */ | |
727 | if (xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
728 | read_unlock(&pag->pag_ici_lock); | |
729 | continue; | |
730 | } | |
731 | ||
fce08f2f | 732 | if (noblock) { |
7a3be02b DC |
733 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { |
734 | read_unlock(&pag->pag_ici_lock); | |
fce08f2f | 735 | continue; |
7a3be02b | 736 | } |
fce08f2f DC |
737 | if (xfs_ipincount(ip) || |
738 | !xfs_iflock_nowait(ip)) { | |
739 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
7a3be02b | 740 | read_unlock(&pag->pag_ici_lock); |
fce08f2f DC |
741 | continue; |
742 | } | |
743 | } | |
7a3be02b DC |
744 | read_unlock(&pag->pag_ici_lock); |
745 | ||
746 | /* | |
747 | * hmmm - this is an inode already in reclaim. Do | |
748 | * we even bother catching it here? | |
749 | */ | |
1dc3318a | 750 | if (xfs_reclaim_inode(ip, noblock, mode)) |
7a3be02b DC |
751 | skipped++; |
752 | } while (nr_found); | |
753 | ||
754 | if (skipped) { | |
755 | delay(1); | |
fce08f2f DC |
756 | goto restart; |
757 | } | |
7a3be02b DC |
758 | return; |
759 | ||
760 | } | |
761 | ||
762 | int | |
763 | xfs_reclaim_inodes( | |
764 | xfs_mount_t *mp, | |
765 | int noblock, | |
766 | int mode) | |
767 | { | |
768 | int i; | |
769 | ||
770 | for (i = 0; i < mp->m_sb.sb_agcount; i++) { | |
771 | if (!mp->m_perag[i].pag_ici_init) | |
772 | continue; | |
773 | xfs_reclaim_inodes_ag(mp, i, noblock, mode); | |
774 | } | |
fce08f2f DC |
775 | return 0; |
776 | } | |
777 | ||
778 |