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" | |
7d095257 | 46 | #include "xfs_quota.h" |
fe4fa4b8 | 47 | |
a167b17e DC |
48 | #include <linux/kthread.h> |
49 | #include <linux/freezer.h> | |
50 | ||
5a34d5cd DC |
51 | |
52 | STATIC int | |
53 | xfs_sync_inode_data( | |
54 | struct xfs_inode *ip, | |
55 | int flags) | |
56 | { | |
57 | struct inode *inode = VFS_I(ip); | |
58 | struct address_space *mapping = inode->i_mapping; | |
59 | int error = 0; | |
60 | ||
61 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
62 | goto out_wait; | |
63 | ||
64 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { | |
65 | if (flags & SYNC_TRYLOCK) | |
66 | goto out_wait; | |
67 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
68 | } | |
69 | ||
70 | error = xfs_flush_pages(ip, 0, -1, (flags & SYNC_WAIT) ? | |
71 | 0 : XFS_B_ASYNC, FI_NONE); | |
72 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
73 | ||
74 | out_wait: | |
75 | if (flags & SYNC_IOWAIT) | |
76 | xfs_ioend_wait(ip); | |
77 | return error; | |
78 | } | |
79 | ||
fe4fa4b8 | 80 | /* |
683a8970 DC |
81 | * Sync all the inodes in the given AG according to the |
82 | * direction given by the flags. | |
fe4fa4b8 | 83 | */ |
683a8970 DC |
84 | STATIC int |
85 | xfs_sync_inodes_ag( | |
fe4fa4b8 | 86 | xfs_mount_t *mp, |
683a8970 | 87 | int ag, |
2030b5ab | 88 | int flags) |
fe4fa4b8 | 89 | { |
683a8970 | 90 | xfs_perag_t *pag = &mp->m_perag[ag]; |
683a8970 | 91 | int nr_found; |
8c38ab03 | 92 | uint32_t first_index = 0; |
683a8970 DC |
93 | int error = 0; |
94 | int last_error = 0; | |
fe4fa4b8 | 95 | |
fe4fa4b8 | 96 | do { |
bc60a993 | 97 | struct inode *inode; |
bc60a993 | 98 | xfs_inode_t *ip = NULL; |
455486b9 | 99 | int lock_flags = XFS_ILOCK_SHARED; |
bc60a993 | 100 | |
fe4fa4b8 | 101 | /* |
683a8970 DC |
102 | * use a gang lookup to find the next inode in the tree |
103 | * as the tree is sparse and a gang lookup walks to find | |
104 | * the number of objects requested. | |
fe4fa4b8 | 105 | */ |
683a8970 DC |
106 | read_lock(&pag->pag_ici_lock); |
107 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, | |
108 | (void**)&ip, first_index, 1); | |
fe4fa4b8 | 109 | |
683a8970 DC |
110 | if (!nr_found) { |
111 | read_unlock(&pag->pag_ici_lock); | |
112 | break; | |
fe4fa4b8 DC |
113 | } |
114 | ||
8c38ab03 DC |
115 | /* |
116 | * Update the index for the next lookup. Catch overflows | |
117 | * into the next AG range which can occur if we have inodes | |
118 | * in the last block of the AG and we are currently | |
119 | * pointing to the last inode. | |
120 | */ | |
683a8970 | 121 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
8c38ab03 DC |
122 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) { |
123 | read_unlock(&pag->pag_ici_lock); | |
124 | break; | |
125 | } | |
fe4fa4b8 | 126 | |
683a8970 | 127 | /* nothing to sync during shutdown */ |
cb56a4b9 | 128 | if (XFS_FORCED_SHUTDOWN(mp)) { |
683a8970 | 129 | read_unlock(&pag->pag_ici_lock); |
fe4fa4b8 DC |
130 | return 0; |
131 | } | |
132 | ||
133 | /* | |
455486b9 DC |
134 | * If we can't get a reference on the inode, it must be |
135 | * in reclaim. Leave it for the reclaim code to flush. | |
fe4fa4b8 | 136 | */ |
455486b9 DC |
137 | inode = VFS_I(ip); |
138 | if (!igrab(inode)) { | |
683a8970 | 139 | read_unlock(&pag->pag_ici_lock); |
455486b9 DC |
140 | continue; |
141 | } | |
142 | read_unlock(&pag->pag_ici_lock); | |
143 | ||
6307091f DC |
144 | /* avoid new or bad inodes */ |
145 | if (is_bad_inode(inode) || | |
146 | xfs_iflags_test(ip, XFS_INEW)) { | |
455486b9 DC |
147 | IRELE(ip); |
148 | continue; | |
fe4fa4b8 | 149 | } |
bc60a993 | 150 | |
fe4fa4b8 DC |
151 | /* |
152 | * If we have to flush data or wait for I/O completion | |
455486b9 | 153 | * we need to hold the iolock. |
fe4fa4b8 | 154 | */ |
5a34d5cd DC |
155 | if (flags & SYNC_DELWRI) |
156 | error = xfs_sync_inode_data(ip, flags); | |
fe4fa4b8 | 157 | |
5a34d5cd | 158 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
683a8970 | 159 | if ((flags & SYNC_ATTR) && !xfs_inode_clean(ip)) { |
fe4fa4b8 DC |
160 | if (flags & SYNC_WAIT) { |
161 | xfs_iflock(ip); | |
683a8970 DC |
162 | if (!xfs_inode_clean(ip)) |
163 | error = xfs_iflush(ip, XFS_IFLUSH_SYNC); | |
164 | else | |
165 | xfs_ifunlock(ip); | |
fe4fa4b8 | 166 | } else if (xfs_iflock_nowait(ip)) { |
683a8970 DC |
167 | if (!xfs_inode_clean(ip)) |
168 | error = xfs_iflush(ip, XFS_IFLUSH_DELWRI); | |
169 | else | |
170 | xfs_ifunlock(ip); | |
fe4fa4b8 DC |
171 | } |
172 | } | |
455486b9 | 173 | xfs_iput(ip, lock_flags); |
fe4fa4b8 | 174 | |
683a8970 | 175 | if (error) |
fe4fa4b8 | 176 | last_error = error; |
fe4fa4b8 DC |
177 | /* |
178 | * bail out if the filesystem is corrupted. | |
179 | */ | |
683a8970 | 180 | if (error == EFSCORRUPTED) |
fe4fa4b8 | 181 | return XFS_ERROR(error); |
fe4fa4b8 | 182 | |
683a8970 | 183 | } while (nr_found); |
fe4fa4b8 | 184 | |
683a8970 DC |
185 | return last_error; |
186 | } | |
fe4fa4b8 | 187 | |
683a8970 DC |
188 | int |
189 | xfs_sync_inodes( | |
190 | xfs_mount_t *mp, | |
2030b5ab | 191 | int flags) |
683a8970 DC |
192 | { |
193 | int error; | |
194 | int last_error; | |
195 | int i; | |
e9f1c6ee | 196 | int lflags = XFS_LOG_FORCE; |
fe4fa4b8 | 197 | |
683a8970 DC |
198 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
199 | return 0; | |
200 | error = 0; | |
201 | last_error = 0; | |
fe4fa4b8 | 202 | |
e9f1c6ee DC |
203 | if (flags & SYNC_WAIT) |
204 | lflags |= XFS_LOG_SYNC; | |
205 | ||
683a8970 DC |
206 | for (i = 0; i < mp->m_sb.sb_agcount; i++) { |
207 | if (!mp->m_perag[i].pag_ici_init) | |
208 | continue; | |
2030b5ab | 209 | error = xfs_sync_inodes_ag(mp, i, flags); |
683a8970 DC |
210 | if (error) |
211 | last_error = error; | |
212 | if (error == EFSCORRUPTED) | |
213 | break; | |
214 | } | |
e9f1c6ee DC |
215 | if (flags & SYNC_DELWRI) |
216 | xfs_log_force(mp, 0, lflags); | |
217 | ||
fe4fa4b8 DC |
218 | return XFS_ERROR(last_error); |
219 | } | |
220 | ||
2af75df7 CH |
221 | STATIC int |
222 | xfs_commit_dummy_trans( | |
223 | struct xfs_mount *mp, | |
224 | uint log_flags) | |
225 | { | |
226 | struct xfs_inode *ip = mp->m_rootip; | |
227 | struct xfs_trans *tp; | |
228 | int error; | |
229 | ||
230 | /* | |
231 | * Put a dummy transaction in the log to tell recovery | |
232 | * that all others are OK. | |
233 | */ | |
234 | tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1); | |
235 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0); | |
236 | if (error) { | |
237 | xfs_trans_cancel(tp, 0); | |
238 | return error; | |
239 | } | |
240 | ||
241 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
242 | ||
243 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
244 | xfs_trans_ihold(tp, ip); | |
245 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
246 | /* XXX(hch): ignoring the error here.. */ | |
247 | error = xfs_trans_commit(tp, 0); | |
248 | ||
249 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
250 | ||
251 | xfs_log_force(mp, 0, log_flags); | |
252 | return 0; | |
253 | } | |
254 | ||
e9f1c6ee | 255 | int |
2af75df7 CH |
256 | xfs_sync_fsdata( |
257 | struct xfs_mount *mp, | |
258 | int flags) | |
259 | { | |
260 | struct xfs_buf *bp; | |
261 | struct xfs_buf_log_item *bip; | |
262 | int error = 0; | |
263 | ||
264 | /* | |
265 | * If this is xfssyncd() then only sync the superblock if we can | |
266 | * lock it without sleeping and it is not pinned. | |
267 | */ | |
268 | if (flags & SYNC_BDFLUSH) { | |
269 | ASSERT(!(flags & SYNC_WAIT)); | |
270 | ||
271 | bp = xfs_getsb(mp, XFS_BUF_TRYLOCK); | |
272 | if (!bp) | |
273 | goto out; | |
274 | ||
275 | bip = XFS_BUF_FSPRIVATE(bp, struct xfs_buf_log_item *); | |
276 | if (!bip || !xfs_buf_item_dirty(bip) || XFS_BUF_ISPINNED(bp)) | |
277 | goto out_brelse; | |
278 | } else { | |
279 | bp = xfs_getsb(mp, 0); | |
280 | ||
281 | /* | |
282 | * If the buffer is pinned then push on the log so we won't | |
283 | * get stuck waiting in the write for someone, maybe | |
284 | * ourselves, to flush the log. | |
285 | * | |
286 | * Even though we just pushed the log above, we did not have | |
287 | * the superblock buffer locked at that point so it can | |
288 | * become pinned in between there and here. | |
289 | */ | |
290 | if (XFS_BUF_ISPINNED(bp)) | |
291 | xfs_log_force(mp, 0, XFS_LOG_FORCE); | |
292 | } | |
293 | ||
294 | ||
295 | if (flags & SYNC_WAIT) | |
296 | XFS_BUF_UNASYNC(bp); | |
297 | else | |
298 | XFS_BUF_ASYNC(bp); | |
299 | ||
300 | return xfs_bwrite(mp, bp); | |
301 | ||
302 | out_brelse: | |
303 | xfs_buf_relse(bp); | |
304 | out: | |
305 | return error; | |
e9f1c6ee DC |
306 | } |
307 | ||
308 | /* | |
a4e4c4f4 DC |
309 | * When remounting a filesystem read-only or freezing the filesystem, we have |
310 | * two phases to execute. This first phase is syncing the data before we | |
311 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
312 | * we've waited for all the transactions created by the first phase to | |
313 | * complete. The second phase ensures that the inodes are written to their | |
314 | * location on disk rather than just existing in transactions in the log. This | |
315 | * means after a quiesce there is no log replay required to write the inodes to | |
316 | * disk (this is the main difference between a sync and a quiesce). | |
317 | */ | |
318 | /* | |
319 | * First stage of freeze - no writers will make progress now we are here, | |
e9f1c6ee DC |
320 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
321 | * complete. Data is frozen at that point. Metadata is not frozen, | |
a4e4c4f4 DC |
322 | * transactions can still occur here so don't bother flushing the buftarg |
323 | * because it'll just get dirty again. | |
e9f1c6ee DC |
324 | */ |
325 | int | |
326 | xfs_quiesce_data( | |
327 | struct xfs_mount *mp) | |
328 | { | |
329 | int error; | |
330 | ||
331 | /* push non-blocking */ | |
332 | xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_BDFLUSH); | |
7d095257 | 333 | xfs_qm_sync(mp, SYNC_BDFLUSH); |
e9f1c6ee DC |
334 | xfs_filestream_flush(mp); |
335 | ||
336 | /* push and block */ | |
337 | xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_WAIT|SYNC_IOWAIT); | |
7d095257 | 338 | xfs_qm_sync(mp, SYNC_WAIT); |
e9f1c6ee | 339 | |
a4e4c4f4 | 340 | /* write superblock and hoover up shutdown errors */ |
e9f1c6ee DC |
341 | error = xfs_sync_fsdata(mp, 0); |
342 | ||
a4e4c4f4 | 343 | /* flush data-only devices */ |
e9f1c6ee DC |
344 | if (mp->m_rtdev_targp) |
345 | XFS_bflush(mp->m_rtdev_targp); | |
346 | ||
347 | return error; | |
2af75df7 CH |
348 | } |
349 | ||
76bf105c DC |
350 | STATIC void |
351 | xfs_quiesce_fs( | |
352 | struct xfs_mount *mp) | |
353 | { | |
354 | int count = 0, pincount; | |
355 | ||
356 | xfs_flush_buftarg(mp->m_ddev_targp, 0); | |
1dc3318a | 357 | xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
76bf105c DC |
358 | |
359 | /* | |
360 | * This loop must run at least twice. The first instance of the loop | |
361 | * will flush most meta data but that will generate more meta data | |
362 | * (typically directory updates). Which then must be flushed and | |
363 | * logged before we can write the unmount record. | |
364 | */ | |
365 | do { | |
366 | xfs_sync_inodes(mp, SYNC_ATTR|SYNC_WAIT); | |
367 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
368 | if (!pincount) { | |
369 | delay(50); | |
370 | count++; | |
371 | } | |
372 | } while (count < 2); | |
373 | } | |
374 | ||
375 | /* | |
376 | * Second stage of a quiesce. The data is already synced, now we have to take | |
377 | * care of the metadata. New transactions are already blocked, so we need to | |
378 | * wait for any remaining transactions to drain out before proceding. | |
379 | */ | |
380 | void | |
381 | xfs_quiesce_attr( | |
382 | struct xfs_mount *mp) | |
383 | { | |
384 | int error = 0; | |
385 | ||
386 | /* wait for all modifications to complete */ | |
387 | while (atomic_read(&mp->m_active_trans) > 0) | |
388 | delay(100); | |
389 | ||
390 | /* flush inodes and push all remaining buffers out to disk */ | |
391 | xfs_quiesce_fs(mp); | |
392 | ||
5e106572 FB |
393 | /* |
394 | * Just warn here till VFS can correctly support | |
395 | * read-only remount without racing. | |
396 | */ | |
397 | WARN_ON(atomic_read(&mp->m_active_trans) != 0); | |
76bf105c DC |
398 | |
399 | /* Push the superblock and write an unmount record */ | |
400 | error = xfs_log_sbcount(mp, 1); | |
401 | if (error) | |
402 | xfs_fs_cmn_err(CE_WARN, mp, | |
403 | "xfs_attr_quiesce: failed to log sb changes. " | |
404 | "Frozen image may not be consistent."); | |
405 | xfs_log_unmount_write(mp); | |
406 | xfs_unmountfs_writesb(mp); | |
407 | } | |
408 | ||
a167b17e DC |
409 | /* |
410 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. | |
411 | * Doing this has two advantages: | |
412 | * - It saves on stack space, which is tight in certain situations | |
413 | * - It can be used (with care) as a mechanism to avoid deadlocks. | |
414 | * Flushing while allocating in a full filesystem requires both. | |
415 | */ | |
416 | STATIC void | |
417 | xfs_syncd_queue_work( | |
418 | struct xfs_mount *mp, | |
419 | void *data, | |
e43afd72 DC |
420 | void (*syncer)(struct xfs_mount *, void *), |
421 | struct completion *completion) | |
a167b17e | 422 | { |
a8d770d9 | 423 | struct xfs_sync_work *work; |
a167b17e | 424 | |
a8d770d9 | 425 | work = kmem_alloc(sizeof(struct xfs_sync_work), KM_SLEEP); |
a167b17e DC |
426 | INIT_LIST_HEAD(&work->w_list); |
427 | work->w_syncer = syncer; | |
428 | work->w_data = data; | |
429 | work->w_mount = mp; | |
e43afd72 | 430 | work->w_completion = completion; |
a167b17e DC |
431 | spin_lock(&mp->m_sync_lock); |
432 | list_add_tail(&work->w_list, &mp->m_sync_list); | |
433 | spin_unlock(&mp->m_sync_lock); | |
434 | wake_up_process(mp->m_sync_task); | |
435 | } | |
436 | ||
437 | /* | |
438 | * Flush delayed allocate data, attempting to free up reserved space | |
439 | * from existing allocations. At this point a new allocation attempt | |
440 | * has failed with ENOSPC and we are in the process of scratching our | |
441 | * heads, looking about for more room... | |
442 | */ | |
443 | STATIC void | |
a8d770d9 | 444 | xfs_flush_inodes_work( |
a167b17e DC |
445 | struct xfs_mount *mp, |
446 | void *arg) | |
447 | { | |
448 | struct inode *inode = arg; | |
a8d770d9 DC |
449 | xfs_sync_inodes(mp, SYNC_DELWRI | SYNC_TRYLOCK); |
450 | xfs_sync_inodes(mp, SYNC_DELWRI | SYNC_TRYLOCK | SYNC_IOWAIT); | |
a167b17e DC |
451 | iput(inode); |
452 | } | |
453 | ||
454 | void | |
a8d770d9 | 455 | xfs_flush_inodes( |
a167b17e DC |
456 | xfs_inode_t *ip) |
457 | { | |
458 | struct inode *inode = VFS_I(ip); | |
e43afd72 | 459 | DECLARE_COMPLETION_ONSTACK(completion); |
a167b17e DC |
460 | |
461 | igrab(inode); | |
e43afd72 DC |
462 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inodes_work, &completion); |
463 | wait_for_completion(&completion); | |
a167b17e DC |
464 | xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC); |
465 | } | |
466 | ||
aacaa880 DC |
467 | /* |
468 | * Every sync period we need to unpin all items, reclaim inodes, sync | |
469 | * quota and write out the superblock. We might need to cover the log | |
470 | * to indicate it is idle. | |
471 | */ | |
a167b17e DC |
472 | STATIC void |
473 | xfs_sync_worker( | |
474 | struct xfs_mount *mp, | |
475 | void *unused) | |
476 | { | |
477 | int error; | |
478 | ||
aacaa880 DC |
479 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
480 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
1dc3318a | 481 | xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
aacaa880 | 482 | /* dgc: errors ignored here */ |
7d095257 | 483 | error = xfs_qm_sync(mp, SYNC_BDFLUSH); |
aacaa880 DC |
484 | error = xfs_sync_fsdata(mp, SYNC_BDFLUSH); |
485 | if (xfs_log_need_covered(mp)) | |
486 | error = xfs_commit_dummy_trans(mp, XFS_LOG_FORCE); | |
487 | } | |
a167b17e DC |
488 | mp->m_sync_seq++; |
489 | wake_up(&mp->m_wait_single_sync_task); | |
490 | } | |
491 | ||
492 | STATIC int | |
493 | xfssyncd( | |
494 | void *arg) | |
495 | { | |
496 | struct xfs_mount *mp = arg; | |
497 | long timeleft; | |
a8d770d9 | 498 | xfs_sync_work_t *work, *n; |
a167b17e DC |
499 | LIST_HEAD (tmp); |
500 | ||
501 | set_freezable(); | |
502 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); | |
503 | for (;;) { | |
504 | timeleft = schedule_timeout_interruptible(timeleft); | |
505 | /* swsusp */ | |
506 | try_to_freeze(); | |
507 | if (kthread_should_stop() && list_empty(&mp->m_sync_list)) | |
508 | break; | |
509 | ||
510 | spin_lock(&mp->m_sync_lock); | |
511 | /* | |
512 | * We can get woken by laptop mode, to do a sync - | |
513 | * that's the (only!) case where the list would be | |
514 | * empty with time remaining. | |
515 | */ | |
516 | if (!timeleft || list_empty(&mp->m_sync_list)) { | |
517 | if (!timeleft) | |
518 | timeleft = xfs_syncd_centisecs * | |
519 | msecs_to_jiffies(10); | |
520 | INIT_LIST_HEAD(&mp->m_sync_work.w_list); | |
521 | list_add_tail(&mp->m_sync_work.w_list, | |
522 | &mp->m_sync_list); | |
523 | } | |
524 | list_for_each_entry_safe(work, n, &mp->m_sync_list, w_list) | |
525 | list_move(&work->w_list, &tmp); | |
526 | spin_unlock(&mp->m_sync_lock); | |
527 | ||
528 | list_for_each_entry_safe(work, n, &tmp, w_list) { | |
529 | (*work->w_syncer)(mp, work->w_data); | |
530 | list_del(&work->w_list); | |
531 | if (work == &mp->m_sync_work) | |
532 | continue; | |
e43afd72 DC |
533 | if (work->w_completion) |
534 | complete(work->w_completion); | |
a167b17e DC |
535 | kmem_free(work); |
536 | } | |
537 | } | |
538 | ||
539 | return 0; | |
540 | } | |
541 | ||
542 | int | |
543 | xfs_syncd_init( | |
544 | struct xfs_mount *mp) | |
545 | { | |
546 | mp->m_sync_work.w_syncer = xfs_sync_worker; | |
547 | mp->m_sync_work.w_mount = mp; | |
e43afd72 | 548 | mp->m_sync_work.w_completion = NULL; |
a167b17e DC |
549 | mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd"); |
550 | if (IS_ERR(mp->m_sync_task)) | |
551 | return -PTR_ERR(mp->m_sync_task); | |
552 | return 0; | |
553 | } | |
554 | ||
555 | void | |
556 | xfs_syncd_stop( | |
557 | struct xfs_mount *mp) | |
558 | { | |
559 | kthread_stop(mp->m_sync_task); | |
560 | } | |
561 | ||
fce08f2f | 562 | int |
1dc3318a | 563 | xfs_reclaim_inode( |
fce08f2f DC |
564 | xfs_inode_t *ip, |
565 | int locked, | |
566 | int sync_mode) | |
567 | { | |
568 | xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino); | |
569 | ||
570 | /* The hash lock here protects a thread in xfs_iget_core from | |
571 | * racing with us on linking the inode back with a vnode. | |
572 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
573 | * us. | |
574 | */ | |
575 | write_lock(&pag->pag_ici_lock); | |
576 | spin_lock(&ip->i_flags_lock); | |
577 | if (__xfs_iflags_test(ip, XFS_IRECLAIM) || | |
578 | !__xfs_iflags_test(ip, XFS_IRECLAIMABLE)) { | |
579 | spin_unlock(&ip->i_flags_lock); | |
580 | write_unlock(&pag->pag_ici_lock); | |
581 | if (locked) { | |
582 | xfs_ifunlock(ip); | |
583 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
584 | } | |
585 | return 1; | |
586 | } | |
587 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
588 | spin_unlock(&ip->i_flags_lock); | |
589 | write_unlock(&pag->pag_ici_lock); | |
590 | xfs_put_perag(ip->i_mount, pag); | |
591 | ||
592 | /* | |
593 | * If the inode is still dirty, then flush it out. If the inode | |
594 | * is not in the AIL, then it will be OK to flush it delwri as | |
595 | * long as xfs_iflush() does not keep any references to the inode. | |
596 | * We leave that decision up to xfs_iflush() since it has the | |
597 | * knowledge of whether it's OK to simply do a delwri flush of | |
598 | * the inode or whether we need to wait until the inode is | |
599 | * pulled from the AIL. | |
600 | * We get the flush lock regardless, though, just to make sure | |
601 | * we don't free it while it is being flushed. | |
602 | */ | |
603 | if (!locked) { | |
604 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
605 | xfs_iflock(ip); | |
606 | } | |
607 | ||
608 | /* | |
609 | * In the case of a forced shutdown we rely on xfs_iflush() to | |
610 | * wait for the inode to be unpinned before returning an error. | |
611 | */ | |
612 | if (!is_bad_inode(VFS_I(ip)) && xfs_iflush(ip, sync_mode) == 0) { | |
613 | /* synchronize with xfs_iflush_done */ | |
614 | xfs_iflock(ip); | |
615 | xfs_ifunlock(ip); | |
616 | } | |
617 | ||
618 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
619 | xfs_ireclaim(ip); | |
620 | return 0; | |
621 | } | |
622 | ||
11654513 DC |
623 | /* |
624 | * We set the inode flag atomically with the radix tree tag. | |
625 | * Once we get tag lookups on the radix tree, this inode flag | |
626 | * can go away. | |
627 | */ | |
396beb85 DC |
628 | void |
629 | xfs_inode_set_reclaim_tag( | |
630 | xfs_inode_t *ip) | |
631 | { | |
632 | xfs_mount_t *mp = ip->i_mount; | |
633 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
634 | ||
635 | read_lock(&pag->pag_ici_lock); | |
636 | spin_lock(&ip->i_flags_lock); | |
637 | radix_tree_tag_set(&pag->pag_ici_root, | |
638 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
11654513 | 639 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
396beb85 DC |
640 | spin_unlock(&ip->i_flags_lock); |
641 | read_unlock(&pag->pag_ici_lock); | |
642 | xfs_put_perag(mp, pag); | |
643 | } | |
644 | ||
645 | void | |
646 | __xfs_inode_clear_reclaim_tag( | |
647 | xfs_mount_t *mp, | |
648 | xfs_perag_t *pag, | |
649 | xfs_inode_t *ip) | |
650 | { | |
651 | radix_tree_tag_clear(&pag->pag_ici_root, | |
652 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
653 | } | |
654 | ||
655 | void | |
656 | xfs_inode_clear_reclaim_tag( | |
657 | xfs_inode_t *ip) | |
658 | { | |
659 | xfs_mount_t *mp = ip->i_mount; | |
660 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
661 | ||
662 | read_lock(&pag->pag_ici_lock); | |
663 | spin_lock(&ip->i_flags_lock); | |
664 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
665 | spin_unlock(&ip->i_flags_lock); | |
666 | read_unlock(&pag->pag_ici_lock); | |
667 | xfs_put_perag(mp, pag); | |
668 | } | |
669 | ||
7a3be02b DC |
670 | |
671 | STATIC void | |
672 | xfs_reclaim_inodes_ag( | |
fce08f2f | 673 | xfs_mount_t *mp, |
7a3be02b DC |
674 | int ag, |
675 | int noblock, | |
fce08f2f DC |
676 | int mode) |
677 | { | |
7a3be02b DC |
678 | xfs_inode_t *ip = NULL; |
679 | xfs_perag_t *pag = &mp->m_perag[ag]; | |
680 | int nr_found; | |
8c38ab03 | 681 | uint32_t first_index; |
7a3be02b | 682 | int skipped; |
fce08f2f DC |
683 | |
684 | restart: | |
7a3be02b DC |
685 | first_index = 0; |
686 | skipped = 0; | |
687 | do { | |
688 | /* | |
689 | * use a gang lookup to find the next inode in the tree | |
690 | * as the tree is sparse and a gang lookup walks to find | |
691 | * the number of objects requested. | |
692 | */ | |
693 | read_lock(&pag->pag_ici_lock); | |
694 | nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root, | |
695 | (void**)&ip, first_index, 1, | |
696 | XFS_ICI_RECLAIM_TAG); | |
697 | ||
698 | if (!nr_found) { | |
699 | read_unlock(&pag->pag_ici_lock); | |
700 | break; | |
701 | } | |
702 | ||
8c38ab03 DC |
703 | /* |
704 | * Update the index for the next lookup. Catch overflows | |
705 | * into the next AG range which can occur if we have inodes | |
706 | * in the last block of the AG and we are currently | |
707 | * pointing to the last inode. | |
708 | */ | |
7a3be02b | 709 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
8c38ab03 DC |
710 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) { |
711 | read_unlock(&pag->pag_ici_lock); | |
712 | break; | |
713 | } | |
7a3be02b | 714 | |
7a3be02b DC |
715 | /* ignore if already under reclaim */ |
716 | if (xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
717 | read_unlock(&pag->pag_ici_lock); | |
718 | continue; | |
719 | } | |
720 | ||
fce08f2f | 721 | if (noblock) { |
7a3be02b DC |
722 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { |
723 | read_unlock(&pag->pag_ici_lock); | |
fce08f2f | 724 | continue; |
7a3be02b | 725 | } |
fce08f2f DC |
726 | if (xfs_ipincount(ip) || |
727 | !xfs_iflock_nowait(ip)) { | |
728 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
7a3be02b | 729 | read_unlock(&pag->pag_ici_lock); |
fce08f2f DC |
730 | continue; |
731 | } | |
732 | } | |
7a3be02b DC |
733 | read_unlock(&pag->pag_ici_lock); |
734 | ||
735 | /* | |
736 | * hmmm - this is an inode already in reclaim. Do | |
737 | * we even bother catching it here? | |
738 | */ | |
1dc3318a | 739 | if (xfs_reclaim_inode(ip, noblock, mode)) |
7a3be02b DC |
740 | skipped++; |
741 | } while (nr_found); | |
742 | ||
743 | if (skipped) { | |
744 | delay(1); | |
fce08f2f DC |
745 | goto restart; |
746 | } | |
7a3be02b DC |
747 | return; |
748 | ||
749 | } | |
750 | ||
751 | int | |
752 | xfs_reclaim_inodes( | |
753 | xfs_mount_t *mp, | |
754 | int noblock, | |
755 | int mode) | |
756 | { | |
757 | int i; | |
758 | ||
759 | for (i = 0; i < mp->m_sb.sb_agcount; i++) { | |
760 | if (!mp->m_perag[i].pag_ici_init) | |
761 | continue; | |
762 | xfs_reclaim_inodes_ag(mp, i, noblock, mode); | |
763 | } | |
fce08f2f DC |
764 | return 0; |
765 | } | |
766 | ||
767 |