]>
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; | |
fe4fa4b8 | 65 | |
fe4fa4b8 | 66 | do { |
bc60a993 | 67 | struct inode *inode; |
bc60a993 | 68 | xfs_inode_t *ip = NULL; |
455486b9 | 69 | int lock_flags = XFS_ILOCK_SHARED; |
bc60a993 | 70 | |
fe4fa4b8 | 71 | /* |
683a8970 DC |
72 | * 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 | |
74 | * the number of objects requested. | |
fe4fa4b8 | 75 | */ |
683a8970 DC |
76 | read_lock(&pag->pag_ici_lock); |
77 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, | |
78 | (void**)&ip, first_index, 1); | |
fe4fa4b8 | 79 | |
683a8970 DC |
80 | if (!nr_found) { |
81 | read_unlock(&pag->pag_ici_lock); | |
82 | break; | |
fe4fa4b8 DC |
83 | } |
84 | ||
8c38ab03 DC |
85 | /* |
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 | */ | |
683a8970 | 91 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
8c38ab03 DC |
92 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) { |
93 | read_unlock(&pag->pag_ici_lock); | |
94 | break; | |
95 | } | |
fe4fa4b8 | 96 | |
683a8970 | 97 | /* nothing to sync during shutdown */ |
cb56a4b9 | 98 | if (XFS_FORCED_SHUTDOWN(mp)) { |
683a8970 | 99 | read_unlock(&pag->pag_ici_lock); |
fe4fa4b8 DC |
100 | return 0; |
101 | } | |
102 | ||
103 | /* | |
455486b9 DC |
104 | * If we can't get a reference on the inode, it must be |
105 | * in reclaim. Leave it for the reclaim code to flush. | |
fe4fa4b8 | 106 | */ |
455486b9 DC |
107 | inode = VFS_I(ip); |
108 | if (!igrab(inode)) { | |
683a8970 | 109 | read_unlock(&pag->pag_ici_lock); |
455486b9 DC |
110 | continue; |
111 | } | |
112 | read_unlock(&pag->pag_ici_lock); | |
113 | ||
6307091f DC |
114 | /* avoid new or bad inodes */ |
115 | if (is_bad_inode(inode) || | |
116 | xfs_iflags_test(ip, XFS_INEW)) { | |
455486b9 DC |
117 | IRELE(ip); |
118 | continue; | |
fe4fa4b8 | 119 | } |
bc60a993 | 120 | |
fe4fa4b8 DC |
121 | /* |
122 | * If we have to flush data or wait for I/O completion | |
455486b9 | 123 | * we need to hold the iolock. |
fe4fa4b8 | 124 | */ |
a8d770d9 DC |
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)) | |
25e41b3d | 142 | xfs_ioend_wait(ip); |
683a8970 | 143 | } |
455486b9 | 144 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
fe4fa4b8 | 145 | |
683a8970 | 146 | if ((flags & SYNC_ATTR) && !xfs_inode_clean(ip)) { |
fe4fa4b8 DC |
147 | if (flags & SYNC_WAIT) { |
148 | xfs_iflock(ip); | |
683a8970 DC |
149 | if (!xfs_inode_clean(ip)) |
150 | error = xfs_iflush(ip, XFS_IFLUSH_SYNC); | |
151 | else | |
152 | xfs_ifunlock(ip); | |
fe4fa4b8 | 153 | } else if (xfs_iflock_nowait(ip)) { |
683a8970 DC |
154 | if (!xfs_inode_clean(ip)) |
155 | error = xfs_iflush(ip, XFS_IFLUSH_DELWRI); | |
156 | else | |
157 | xfs_ifunlock(ip); | |
fe4fa4b8 DC |
158 | } |
159 | } | |
455486b9 | 160 | xfs_iput(ip, lock_flags); |
fe4fa4b8 | 161 | |
683a8970 | 162 | if (error) |
fe4fa4b8 | 163 | last_error = error; |
fe4fa4b8 DC |
164 | /* |
165 | * bail out if the filesystem is corrupted. | |
166 | */ | |
683a8970 | 167 | if (error == EFSCORRUPTED) |
fe4fa4b8 | 168 | return XFS_ERROR(error); |
fe4fa4b8 | 169 | |
683a8970 | 170 | } while (nr_found); |
fe4fa4b8 | 171 | |
683a8970 DC |
172 | return last_error; |
173 | } | |
fe4fa4b8 | 174 | |
683a8970 DC |
175 | int |
176 | xfs_sync_inodes( | |
177 | xfs_mount_t *mp, | |
2030b5ab | 178 | int flags) |
683a8970 DC |
179 | { |
180 | int error; | |
181 | int last_error; | |
182 | int i; | |
e9f1c6ee | 183 | int lflags = XFS_LOG_FORCE; |
fe4fa4b8 | 184 | |
683a8970 DC |
185 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
186 | return 0; | |
187 | error = 0; | |
188 | last_error = 0; | |
fe4fa4b8 | 189 | |
e9f1c6ee DC |
190 | if (flags & SYNC_WAIT) |
191 | lflags |= XFS_LOG_SYNC; | |
192 | ||
683a8970 DC |
193 | for (i = 0; i < mp->m_sb.sb_agcount; i++) { |
194 | if (!mp->m_perag[i].pag_ici_init) | |
195 | continue; | |
2030b5ab | 196 | error = xfs_sync_inodes_ag(mp, i, flags); |
683a8970 DC |
197 | if (error) |
198 | last_error = error; | |
199 | if (error == EFSCORRUPTED) | |
200 | break; | |
201 | } | |
e9f1c6ee DC |
202 | if (flags & SYNC_DELWRI) |
203 | xfs_log_force(mp, 0, lflags); | |
204 | ||
fe4fa4b8 DC |
205 | return XFS_ERROR(last_error); |
206 | } | |
207 | ||
2af75df7 CH |
208 | STATIC int |
209 | xfs_commit_dummy_trans( | |
210 | struct xfs_mount *mp, | |
211 | uint log_flags) | |
212 | { | |
213 | struct xfs_inode *ip = mp->m_rootip; | |
214 | struct xfs_trans *tp; | |
215 | int error; | |
216 | ||
217 | /* | |
218 | * Put a dummy transaction in the log to tell recovery | |
219 | * that all others are OK. | |
220 | */ | |
221 | tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1); | |
222 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0); | |
223 | if (error) { | |
224 | xfs_trans_cancel(tp, 0); | |
225 | return error; | |
226 | } | |
227 | ||
228 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
229 | ||
230 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
231 | xfs_trans_ihold(tp, ip); | |
232 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
233 | /* XXX(hch): ignoring the error here.. */ | |
234 | error = xfs_trans_commit(tp, 0); | |
235 | ||
236 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
237 | ||
238 | xfs_log_force(mp, 0, log_flags); | |
239 | return 0; | |
240 | } | |
241 | ||
e9f1c6ee | 242 | int |
2af75df7 CH |
243 | xfs_sync_fsdata( |
244 | struct xfs_mount *mp, | |
245 | int flags) | |
246 | { | |
247 | struct xfs_buf *bp; | |
248 | struct xfs_buf_log_item *bip; | |
249 | int error = 0; | |
250 | ||
251 | /* | |
252 | * If this is xfssyncd() then only sync the superblock if we can | |
253 | * lock it without sleeping and it is not pinned. | |
254 | */ | |
255 | if (flags & SYNC_BDFLUSH) { | |
256 | ASSERT(!(flags & SYNC_WAIT)); | |
257 | ||
258 | bp = xfs_getsb(mp, XFS_BUF_TRYLOCK); | |
259 | if (!bp) | |
260 | goto out; | |
261 | ||
262 | bip = XFS_BUF_FSPRIVATE(bp, struct xfs_buf_log_item *); | |
263 | if (!bip || !xfs_buf_item_dirty(bip) || XFS_BUF_ISPINNED(bp)) | |
264 | goto out_brelse; | |
265 | } else { | |
266 | bp = xfs_getsb(mp, 0); | |
267 | ||
268 | /* | |
269 | * If the buffer is pinned then push on the log so we won't | |
270 | * get stuck waiting in the write for someone, maybe | |
271 | * ourselves, to flush the log. | |
272 | * | |
273 | * Even though we just pushed the log above, we did not have | |
274 | * the superblock buffer locked at that point so it can | |
275 | * become pinned in between there and here. | |
276 | */ | |
277 | if (XFS_BUF_ISPINNED(bp)) | |
278 | xfs_log_force(mp, 0, XFS_LOG_FORCE); | |
279 | } | |
280 | ||
281 | ||
282 | if (flags & SYNC_WAIT) | |
283 | XFS_BUF_UNASYNC(bp); | |
284 | else | |
285 | XFS_BUF_ASYNC(bp); | |
286 | ||
287 | return xfs_bwrite(mp, bp); | |
288 | ||
289 | out_brelse: | |
290 | xfs_buf_relse(bp); | |
291 | out: | |
292 | return error; | |
e9f1c6ee DC |
293 | } |
294 | ||
295 | /* | |
a4e4c4f4 DC |
296 | * When remounting a filesystem read-only or freezing the filesystem, we have |
297 | * two phases to execute. This first phase is syncing the data before we | |
298 | * quiesce the filesystem, and the second is flushing all the inodes out after | |
299 | * we've waited for all the transactions created by the first phase to | |
300 | * complete. The second phase ensures that the inodes are written to their | |
301 | * location on disk rather than just existing in transactions in the log. This | |
302 | * means after a quiesce there is no log replay required to write the inodes to | |
303 | * disk (this is the main difference between a sync and a quiesce). | |
304 | */ | |
305 | /* | |
306 | * First stage of freeze - no writers will make progress now we are here, | |
e9f1c6ee DC |
307 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
308 | * complete. Data is frozen at that point. Metadata is not frozen, | |
a4e4c4f4 DC |
309 | * transactions can still occur here so don't bother flushing the buftarg |
310 | * because it'll just get dirty again. | |
e9f1c6ee DC |
311 | */ |
312 | int | |
313 | xfs_quiesce_data( | |
314 | struct xfs_mount *mp) | |
315 | { | |
316 | int error; | |
317 | ||
318 | /* push non-blocking */ | |
319 | xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_BDFLUSH); | |
320 | XFS_QM_DQSYNC(mp, SYNC_BDFLUSH); | |
321 | xfs_filestream_flush(mp); | |
322 | ||
323 | /* push and block */ | |
324 | xfs_sync_inodes(mp, SYNC_DELWRI|SYNC_WAIT|SYNC_IOWAIT); | |
325 | XFS_QM_DQSYNC(mp, SYNC_WAIT); | |
326 | ||
a4e4c4f4 | 327 | /* write superblock and hoover up shutdown errors */ |
e9f1c6ee DC |
328 | error = xfs_sync_fsdata(mp, 0); |
329 | ||
a4e4c4f4 | 330 | /* flush data-only devices */ |
e9f1c6ee DC |
331 | if (mp->m_rtdev_targp) |
332 | XFS_bflush(mp->m_rtdev_targp); | |
333 | ||
334 | return error; | |
2af75df7 CH |
335 | } |
336 | ||
76bf105c DC |
337 | STATIC void |
338 | xfs_quiesce_fs( | |
339 | struct xfs_mount *mp) | |
340 | { | |
341 | int count = 0, pincount; | |
342 | ||
343 | xfs_flush_buftarg(mp->m_ddev_targp, 0); | |
1dc3318a | 344 | xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
76bf105c DC |
345 | |
346 | /* | |
347 | * This loop must run at least twice. The first instance of the loop | |
348 | * will flush most meta data but that will generate more meta data | |
349 | * (typically directory updates). Which then must be flushed and | |
350 | * logged before we can write the unmount record. | |
351 | */ | |
352 | do { | |
353 | xfs_sync_inodes(mp, SYNC_ATTR|SYNC_WAIT); | |
354 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); | |
355 | if (!pincount) { | |
356 | delay(50); | |
357 | count++; | |
358 | } | |
359 | } while (count < 2); | |
360 | } | |
361 | ||
362 | /* | |
363 | * Second stage of a quiesce. The data is already synced, now we have to take | |
364 | * care of the metadata. New transactions are already blocked, so we need to | |
365 | * wait for any remaining transactions to drain out before proceding. | |
366 | */ | |
367 | void | |
368 | xfs_quiesce_attr( | |
369 | struct xfs_mount *mp) | |
370 | { | |
371 | int error = 0; | |
372 | ||
373 | /* wait for all modifications to complete */ | |
374 | while (atomic_read(&mp->m_active_trans) > 0) | |
375 | delay(100); | |
376 | ||
377 | /* flush inodes and push all remaining buffers out to disk */ | |
378 | xfs_quiesce_fs(mp); | |
379 | ||
5e106572 FB |
380 | /* |
381 | * Just warn here till VFS can correctly support | |
382 | * read-only remount without racing. | |
383 | */ | |
384 | WARN_ON(atomic_read(&mp->m_active_trans) != 0); | |
76bf105c DC |
385 | |
386 | /* Push the superblock and write an unmount record */ | |
387 | error = xfs_log_sbcount(mp, 1); | |
388 | if (error) | |
389 | xfs_fs_cmn_err(CE_WARN, mp, | |
390 | "xfs_attr_quiesce: failed to log sb changes. " | |
391 | "Frozen image may not be consistent."); | |
392 | xfs_log_unmount_write(mp); | |
393 | xfs_unmountfs_writesb(mp); | |
394 | } | |
395 | ||
a167b17e DC |
396 | /* |
397 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. | |
398 | * Doing this has two advantages: | |
399 | * - It saves on stack space, which is tight in certain situations | |
400 | * - It can be used (with care) as a mechanism to avoid deadlocks. | |
401 | * Flushing while allocating in a full filesystem requires both. | |
402 | */ | |
403 | STATIC void | |
404 | xfs_syncd_queue_work( | |
405 | struct xfs_mount *mp, | |
406 | void *data, | |
407 | void (*syncer)(struct xfs_mount *, void *)) | |
408 | { | |
a8d770d9 | 409 | struct xfs_sync_work *work; |
a167b17e | 410 | |
a8d770d9 | 411 | work = kmem_alloc(sizeof(struct xfs_sync_work), KM_SLEEP); |
a167b17e DC |
412 | INIT_LIST_HEAD(&work->w_list); |
413 | work->w_syncer = syncer; | |
414 | work->w_data = data; | |
415 | work->w_mount = mp; | |
416 | spin_lock(&mp->m_sync_lock); | |
417 | list_add_tail(&work->w_list, &mp->m_sync_list); | |
418 | spin_unlock(&mp->m_sync_lock); | |
419 | wake_up_process(mp->m_sync_task); | |
420 | } | |
421 | ||
422 | /* | |
423 | * Flush delayed allocate data, attempting to free up reserved space | |
424 | * from existing allocations. At this point a new allocation attempt | |
425 | * has failed with ENOSPC and we are in the process of scratching our | |
426 | * heads, looking about for more room... | |
427 | */ | |
428 | STATIC void | |
a8d770d9 | 429 | xfs_flush_inodes_work( |
a167b17e DC |
430 | struct xfs_mount *mp, |
431 | void *arg) | |
432 | { | |
433 | struct inode *inode = arg; | |
a8d770d9 DC |
434 | xfs_sync_inodes(mp, SYNC_DELWRI | SYNC_TRYLOCK); |
435 | xfs_sync_inodes(mp, SYNC_DELWRI | SYNC_TRYLOCK | SYNC_IOWAIT); | |
a167b17e DC |
436 | iput(inode); |
437 | } | |
438 | ||
439 | void | |
a8d770d9 | 440 | xfs_flush_inodes( |
a167b17e DC |
441 | xfs_inode_t *ip) |
442 | { | |
443 | struct inode *inode = VFS_I(ip); | |
444 | ||
445 | igrab(inode); | |
a8d770d9 | 446 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inodes_work); |
a167b17e DC |
447 | delay(msecs_to_jiffies(500)); |
448 | xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC); | |
449 | } | |
450 | ||
aacaa880 DC |
451 | /* |
452 | * Every sync period we need to unpin all items, reclaim inodes, sync | |
453 | * quota and write out the superblock. We might need to cover the log | |
454 | * to indicate it is idle. | |
455 | */ | |
a167b17e DC |
456 | STATIC void |
457 | xfs_sync_worker( | |
458 | struct xfs_mount *mp, | |
459 | void *unused) | |
460 | { | |
461 | int error; | |
462 | ||
aacaa880 DC |
463 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
464 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
1dc3318a | 465 | xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_DELWRI_ELSE_ASYNC); |
aacaa880 DC |
466 | /* dgc: errors ignored here */ |
467 | error = XFS_QM_DQSYNC(mp, SYNC_BDFLUSH); | |
468 | error = xfs_sync_fsdata(mp, SYNC_BDFLUSH); | |
469 | if (xfs_log_need_covered(mp)) | |
470 | error = xfs_commit_dummy_trans(mp, XFS_LOG_FORCE); | |
471 | } | |
a167b17e DC |
472 | mp->m_sync_seq++; |
473 | wake_up(&mp->m_wait_single_sync_task); | |
474 | } | |
475 | ||
476 | STATIC int | |
477 | xfssyncd( | |
478 | void *arg) | |
479 | { | |
480 | struct xfs_mount *mp = arg; | |
481 | long timeleft; | |
a8d770d9 | 482 | xfs_sync_work_t *work, *n; |
a167b17e DC |
483 | LIST_HEAD (tmp); |
484 | ||
485 | set_freezable(); | |
486 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); | |
487 | for (;;) { | |
488 | timeleft = schedule_timeout_interruptible(timeleft); | |
489 | /* swsusp */ | |
490 | try_to_freeze(); | |
491 | if (kthread_should_stop() && list_empty(&mp->m_sync_list)) | |
492 | break; | |
493 | ||
494 | spin_lock(&mp->m_sync_lock); | |
495 | /* | |
496 | * We can get woken by laptop mode, to do a sync - | |
497 | * that's the (only!) case where the list would be | |
498 | * empty with time remaining. | |
499 | */ | |
500 | if (!timeleft || list_empty(&mp->m_sync_list)) { | |
501 | if (!timeleft) | |
502 | timeleft = xfs_syncd_centisecs * | |
503 | msecs_to_jiffies(10); | |
504 | INIT_LIST_HEAD(&mp->m_sync_work.w_list); | |
505 | list_add_tail(&mp->m_sync_work.w_list, | |
506 | &mp->m_sync_list); | |
507 | } | |
508 | list_for_each_entry_safe(work, n, &mp->m_sync_list, w_list) | |
509 | list_move(&work->w_list, &tmp); | |
510 | spin_unlock(&mp->m_sync_lock); | |
511 | ||
512 | list_for_each_entry_safe(work, n, &tmp, w_list) { | |
513 | (*work->w_syncer)(mp, work->w_data); | |
514 | list_del(&work->w_list); | |
515 | if (work == &mp->m_sync_work) | |
516 | continue; | |
517 | kmem_free(work); | |
518 | } | |
519 | } | |
520 | ||
521 | return 0; | |
522 | } | |
523 | ||
524 | int | |
525 | xfs_syncd_init( | |
526 | struct xfs_mount *mp) | |
527 | { | |
528 | mp->m_sync_work.w_syncer = xfs_sync_worker; | |
529 | mp->m_sync_work.w_mount = mp; | |
530 | mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd"); | |
531 | if (IS_ERR(mp->m_sync_task)) | |
532 | return -PTR_ERR(mp->m_sync_task); | |
533 | return 0; | |
534 | } | |
535 | ||
536 | void | |
537 | xfs_syncd_stop( | |
538 | struct xfs_mount *mp) | |
539 | { | |
540 | kthread_stop(mp->m_sync_task); | |
541 | } | |
542 | ||
fce08f2f | 543 | int |
1dc3318a | 544 | xfs_reclaim_inode( |
fce08f2f DC |
545 | xfs_inode_t *ip, |
546 | int locked, | |
547 | int sync_mode) | |
548 | { | |
549 | xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino); | |
550 | ||
551 | /* The hash lock here protects a thread in xfs_iget_core from | |
552 | * racing with us on linking the inode back with a vnode. | |
553 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
554 | * us. | |
555 | */ | |
556 | write_lock(&pag->pag_ici_lock); | |
557 | spin_lock(&ip->i_flags_lock); | |
558 | if (__xfs_iflags_test(ip, XFS_IRECLAIM) || | |
559 | !__xfs_iflags_test(ip, XFS_IRECLAIMABLE)) { | |
560 | spin_unlock(&ip->i_flags_lock); | |
561 | write_unlock(&pag->pag_ici_lock); | |
562 | if (locked) { | |
563 | xfs_ifunlock(ip); | |
564 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
565 | } | |
566 | return 1; | |
567 | } | |
568 | __xfs_iflags_set(ip, XFS_IRECLAIM); | |
569 | spin_unlock(&ip->i_flags_lock); | |
570 | write_unlock(&pag->pag_ici_lock); | |
571 | xfs_put_perag(ip->i_mount, pag); | |
572 | ||
573 | /* | |
574 | * If the inode is still dirty, then flush it out. If the inode | |
575 | * is not in the AIL, then it will be OK to flush it delwri as | |
576 | * long as xfs_iflush() does not keep any references to the inode. | |
577 | * We leave that decision up to xfs_iflush() since it has the | |
578 | * knowledge of whether it's OK to simply do a delwri flush of | |
579 | * the inode or whether we need to wait until the inode is | |
580 | * pulled from the AIL. | |
581 | * We get the flush lock regardless, though, just to make sure | |
582 | * we don't free it while it is being flushed. | |
583 | */ | |
584 | if (!locked) { | |
585 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
586 | xfs_iflock(ip); | |
587 | } | |
588 | ||
589 | /* | |
590 | * In the case of a forced shutdown we rely on xfs_iflush() to | |
591 | * wait for the inode to be unpinned before returning an error. | |
592 | */ | |
593 | if (!is_bad_inode(VFS_I(ip)) && xfs_iflush(ip, sync_mode) == 0) { | |
594 | /* synchronize with xfs_iflush_done */ | |
595 | xfs_iflock(ip); | |
596 | xfs_ifunlock(ip); | |
597 | } | |
598 | ||
599 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
600 | xfs_ireclaim(ip); | |
601 | return 0; | |
602 | } | |
603 | ||
11654513 DC |
604 | /* |
605 | * We set the inode flag atomically with the radix tree tag. | |
606 | * Once we get tag lookups on the radix tree, this inode flag | |
607 | * can go away. | |
608 | */ | |
396beb85 DC |
609 | void |
610 | xfs_inode_set_reclaim_tag( | |
611 | xfs_inode_t *ip) | |
612 | { | |
613 | xfs_mount_t *mp = ip->i_mount; | |
614 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
615 | ||
616 | read_lock(&pag->pag_ici_lock); | |
617 | spin_lock(&ip->i_flags_lock); | |
618 | radix_tree_tag_set(&pag->pag_ici_root, | |
619 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
11654513 | 620 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
396beb85 DC |
621 | spin_unlock(&ip->i_flags_lock); |
622 | read_unlock(&pag->pag_ici_lock); | |
623 | xfs_put_perag(mp, pag); | |
624 | } | |
625 | ||
626 | void | |
627 | __xfs_inode_clear_reclaim_tag( | |
628 | xfs_mount_t *mp, | |
629 | xfs_perag_t *pag, | |
630 | xfs_inode_t *ip) | |
631 | { | |
632 | radix_tree_tag_clear(&pag->pag_ici_root, | |
633 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); | |
634 | } | |
635 | ||
636 | void | |
637 | xfs_inode_clear_reclaim_tag( | |
638 | xfs_inode_t *ip) | |
639 | { | |
640 | xfs_mount_t *mp = ip->i_mount; | |
641 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
642 | ||
643 | read_lock(&pag->pag_ici_lock); | |
644 | spin_lock(&ip->i_flags_lock); | |
645 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
646 | spin_unlock(&ip->i_flags_lock); | |
647 | read_unlock(&pag->pag_ici_lock); | |
648 | xfs_put_perag(mp, pag); | |
649 | } | |
650 | ||
7a3be02b DC |
651 | |
652 | STATIC void | |
653 | xfs_reclaim_inodes_ag( | |
fce08f2f | 654 | xfs_mount_t *mp, |
7a3be02b DC |
655 | int ag, |
656 | int noblock, | |
fce08f2f DC |
657 | int mode) |
658 | { | |
7a3be02b DC |
659 | xfs_inode_t *ip = NULL; |
660 | xfs_perag_t *pag = &mp->m_perag[ag]; | |
661 | int nr_found; | |
8c38ab03 | 662 | uint32_t first_index; |
7a3be02b | 663 | int skipped; |
fce08f2f DC |
664 | |
665 | restart: | |
7a3be02b DC |
666 | first_index = 0; |
667 | skipped = 0; | |
668 | do { | |
669 | /* | |
670 | * use a gang lookup to find the next inode in the tree | |
671 | * as the tree is sparse and a gang lookup walks to find | |
672 | * the number of objects requested. | |
673 | */ | |
674 | read_lock(&pag->pag_ici_lock); | |
675 | nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root, | |
676 | (void**)&ip, first_index, 1, | |
677 | XFS_ICI_RECLAIM_TAG); | |
678 | ||
679 | if (!nr_found) { | |
680 | read_unlock(&pag->pag_ici_lock); | |
681 | break; | |
682 | } | |
683 | ||
8c38ab03 DC |
684 | /* |
685 | * Update the index for the next lookup. Catch overflows | |
686 | * into the next AG range which can occur if we have inodes | |
687 | * in the last block of the AG and we are currently | |
688 | * pointing to the last inode. | |
689 | */ | |
7a3be02b | 690 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
8c38ab03 DC |
691 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) { |
692 | read_unlock(&pag->pag_ici_lock); | |
693 | break; | |
694 | } | |
7a3be02b | 695 | |
7a3be02b DC |
696 | /* ignore if already under reclaim */ |
697 | if (xfs_iflags_test(ip, XFS_IRECLAIM)) { | |
698 | read_unlock(&pag->pag_ici_lock); | |
699 | continue; | |
700 | } | |
701 | ||
fce08f2f | 702 | if (noblock) { |
7a3be02b DC |
703 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { |
704 | read_unlock(&pag->pag_ici_lock); | |
fce08f2f | 705 | continue; |
7a3be02b | 706 | } |
fce08f2f DC |
707 | if (xfs_ipincount(ip) || |
708 | !xfs_iflock_nowait(ip)) { | |
709 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
7a3be02b | 710 | read_unlock(&pag->pag_ici_lock); |
fce08f2f DC |
711 | continue; |
712 | } | |
713 | } | |
7a3be02b DC |
714 | read_unlock(&pag->pag_ici_lock); |
715 | ||
716 | /* | |
717 | * hmmm - this is an inode already in reclaim. Do | |
718 | * we even bother catching it here? | |
719 | */ | |
1dc3318a | 720 | if (xfs_reclaim_inode(ip, noblock, mode)) |
7a3be02b DC |
721 | skipped++; |
722 | } while (nr_found); | |
723 | ||
724 | if (skipped) { | |
725 | delay(1); | |
fce08f2f DC |
726 | goto restart; |
727 | } | |
7a3be02b DC |
728 | return; |
729 | ||
730 | } | |
731 | ||
732 | int | |
733 | xfs_reclaim_inodes( | |
734 | xfs_mount_t *mp, | |
735 | int noblock, | |
736 | int mode) | |
737 | { | |
738 | int i; | |
739 | ||
740 | for (i = 0; i < mp->m_sb.sb_agcount; i++) { | |
741 | if (!mp->m_perag[i].pag_ici_init) | |
742 | continue; | |
743 | xfs_reclaim_inodes_ag(mp, i, noblock, mode); | |
744 | } | |
fce08f2f DC |
745 | return 0; |
746 | } | |
747 | ||
748 |