]>
Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
7b718769 NS |
3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. | |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include "xfs.h" |
a844f451 | 7 | #include "xfs_fs.h" |
70a9883c | 8 | #include "xfs_shared.h" |
239880ef | 9 | #include "xfs_format.h" |
a4fbe6ab | 10 | #include "xfs_log_format.h" |
239880ef | 11 | #include "xfs_trans_resv.h" |
1da177e4 | 12 | #include "xfs_sb.h" |
1da177e4 | 13 | #include "xfs_mount.h" |
239880ef | 14 | #include "xfs_trans.h" |
1da177e4 LT |
15 | #include "xfs_error.h" |
16 | #include "xfs_alloc.h" | |
1da177e4 | 17 | #include "xfs_fsops.h" |
1da177e4 | 18 | #include "xfs_trans_space.h" |
239880ef | 19 | #include "xfs_log.h" |
41e63621 | 20 | #include "xfs_log_priv.h" |
b16817b6 | 21 | #include "xfs_ag.h" |
84d69619 | 22 | #include "xfs_ag_resv.h" |
7f89c838 | 23 | #include "xfs_trace.h" |
1da177e4 | 24 | |
c789c83c GX |
25 | /* |
26 | * Write new AG headers to disk. Non-transactional, but need to be | |
27 | * written and completed prior to the growfs transaction being logged. | |
28 | * To do this, we use a delayed write buffer list and wait for | |
29 | * submission and IO completion of the list as a whole. This allows the | |
30 | * IO subsystem to merge all the AG headers in a single AG into a single | |
31 | * IO and hide most of the latency of the IO from us. | |
32 | * | |
33 | * This also means that if we get an error whilst building the buffer | |
34 | * list to write, we can cancel the entire list without having written | |
35 | * anything. | |
36 | */ | |
37 | static int | |
38 | xfs_resizefs_init_new_ags( | |
39 | struct xfs_trans *tp, | |
40 | struct aghdr_init_data *id, | |
41 | xfs_agnumber_t oagcount, | |
42 | xfs_agnumber_t nagcount, | |
43 | xfs_rfsblock_t delta, | |
c6aee248 | 44 | struct xfs_perag *last_pag, |
c789c83c GX |
45 | bool *lastag_extended) |
46 | { | |
47 | struct xfs_mount *mp = tp->t_mountp; | |
48 | xfs_rfsblock_t nb = mp->m_sb.sb_dblocks + delta; | |
49 | int error; | |
50 | ||
51 | *lastag_extended = false; | |
52 | ||
53 | INIT_LIST_HEAD(&id->buffer_list); | |
54 | for (id->agno = nagcount - 1; | |
55 | id->agno >= oagcount; | |
56 | id->agno--, delta -= id->agsize) { | |
57 | ||
58 | if (id->agno == nagcount - 1) | |
59 | id->agsize = nb - (id->agno * | |
60 | (xfs_rfsblock_t)mp->m_sb.sb_agblocks); | |
61 | else | |
62 | id->agsize = mp->m_sb.sb_agblocks; | |
63 | ||
64 | error = xfs_ag_init_headers(mp, id); | |
65 | if (error) { | |
66 | xfs_buf_delwri_cancel(&id->buffer_list); | |
67 | return error; | |
68 | } | |
69 | } | |
70 | ||
71 | error = xfs_buf_delwri_submit(&id->buffer_list); | |
72 | if (error) | |
73 | return error; | |
74 | ||
c789c83c GX |
75 | if (delta) { |
76 | *lastag_extended = true; | |
c6aee248 | 77 | error = xfs_ag_extend_space(last_pag, tp, delta); |
c789c83c GX |
78 | } |
79 | return error; | |
80 | } | |
81 | ||
1da177e4 | 82 | /* |
b16817b6 | 83 | * growfs operations |
1da177e4 | 84 | */ |
1da177e4 LT |
85 | static int |
86 | xfs_growfs_data_private( | |
07aabd9c GX |
87 | struct xfs_mount *mp, /* mount point for filesystem */ |
88 | struct xfs_growfs_data *in) /* growfs data input struct */ | |
1da177e4 | 89 | { |
e8222613 | 90 | struct xfs_buf *bp; |
83a7f86e | 91 | int error; |
1da177e4 LT |
92 | xfs_agnumber_t nagcount; |
93 | xfs_agnumber_t nagimax = 0; | |
ce5e1062 | 94 | xfs_rfsblock_t nb, nb_div, nb_mod; |
fb2fc172 | 95 | int64_t delta; |
c789c83c | 96 | bool lastag_extended; |
1da177e4 | 97 | xfs_agnumber_t oagcount; |
07aabd9c | 98 | struct xfs_trans *tp; |
0410c3bb | 99 | struct aghdr_init_data id = {}; |
c6aee248 | 100 | struct xfs_perag *last_pag; |
1da177e4 LT |
101 | |
102 | nb = in->newblocks; | |
fb2fc172 GX |
103 | error = xfs_sb_validate_fsb_count(&mp->m_sb, nb); |
104 | if (error) | |
4cc929ee | 105 | return error; |
fb2fc172 GX |
106 | |
107 | if (nb > mp->m_sb.sb_dblocks) { | |
108 | error = xfs_buf_read_uncached(mp->m_ddev_targp, | |
1922c949 | 109 | XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1), |
ba372674 | 110 | XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); |
fb2fc172 GX |
111 | if (error) |
112 | return error; | |
113 | xfs_buf_relse(bp); | |
114 | } | |
1da177e4 | 115 | |
ce5e1062 GX |
116 | nb_div = nb; |
117 | nb_mod = do_div(nb_div, mp->m_sb.sb_agblocks); | |
118 | nagcount = nb_div + (nb_mod != 0); | |
1da177e4 LT |
119 | if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) { |
120 | nagcount--; | |
e6da7c9f | 121 | nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks; |
1da177e4 | 122 | } |
ce5e1062 | 123 | delta = nb - mp->m_sb.sb_dblocks; |
fb2fc172 GX |
124 | /* |
125 | * Reject filesystems with a single AG because they are not | |
126 | * supported, and reject a shrink operation that would cause a | |
127 | * filesystem to become unsupported. | |
128 | */ | |
129 | if (delta < 0 && nagcount < 2) | |
130 | return -EINVAL; | |
131 | ||
1da177e4 | 132 | oagcount = mp->m_sb.sb_agcount; |
1c1c6ebc DC |
133 | /* allocate the new per-ag structures */ |
134 | if (nagcount > oagcount) { | |
0800169e | 135 | error = xfs_initialize_perag(mp, nagcount, nb, &nagimax); |
1c1c6ebc DC |
136 | if (error) |
137 | return error; | |
fb2fc172 GX |
138 | } else if (nagcount < oagcount) { |
139 | /* TODO: shrinking the entire AGs hasn't yet completed */ | |
140 | return -EINVAL; | |
1da177e4 | 141 | } |
1c1c6ebc | 142 | |
253f4911 | 143 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, |
fb2fc172 GX |
144 | (delta > 0 ? XFS_GROWFS_SPACE_RES(mp) : -delta), 0, |
145 | XFS_TRANS_RESERVE, &tp); | |
253f4911 | 146 | if (error) |
1da177e4 | 147 | return error; |
1da177e4 | 148 | |
c6aee248 | 149 | last_pag = xfs_perag_get(mp, oagcount - 1); |
fb2fc172 GX |
150 | if (delta > 0) { |
151 | error = xfs_resizefs_init_new_ags(tp, &id, oagcount, nagcount, | |
c6aee248 | 152 | delta, last_pag, &lastag_extended); |
fb2fc172 | 153 | } else { |
df5660cf | 154 | xfs_warn_mount(mp, XFS_OPSTATE_WARNED_SHRINK, |
fb2fc172 GX |
155 | "EXPERIMENTAL online shrink feature in use. Use at your own risk!"); |
156 | ||
c6aee248 | 157 | error = xfs_ag_shrink_space(last_pag, &tp, -delta); |
fb2fc172 | 158 | } |
c6aee248 | 159 | xfs_perag_put(last_pag); |
9aebe805 | 160 | if (error) |
83a7f86e | 161 | goto out_trans_cancel; |
9aebe805 | 162 | |
1c1c6ebc DC |
163 | /* |
164 | * Update changed superblock fields transactionally. These are not | |
165 | * seen by the rest of the world until the transaction commit applies | |
166 | * them atomically to the superblock. | |
167 | */ | |
1da177e4 LT |
168 | if (nagcount > oagcount) |
169 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount); | |
c789c83c GX |
170 | if (delta) |
171 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS, delta); | |
0410c3bb DC |
172 | if (id.nfree) |
173 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree); | |
014695c0 GX |
174 | |
175 | /* | |
176 | * Sync sb counters now to reflect the updated values. This is | |
177 | * particularly important for shrink because the write verifier | |
178 | * will fail if sb_fdblocks is ever larger than sb_dblocks. | |
179 | */ | |
38c26bfd | 180 | if (xfs_has_lazysbcount(mp)) |
014695c0 GX |
181 | xfs_log_sb(tp); |
182 | ||
f8079b85 | 183 | xfs_trans_set_sync(tp); |
70393313 | 184 | error = xfs_trans_commit(tp); |
1c1c6ebc | 185 | if (error) |
1da177e4 | 186 | return error; |
1c1c6ebc | 187 | |
1da177e4 LT |
188 | /* New allocation groups fully initialized, so update mount struct */ |
189 | if (nagimax) | |
190 | mp->m_maxagi = nagimax; | |
055388a3 | 191 | xfs_set_low_space_thresholds(mp); |
52548852 | 192 | mp->m_alloc_set_aside = xfs_alloc_set_aside(mp); |
1c1c6ebc | 193 | |
fb2fc172 GX |
194 | if (delta > 0) { |
195 | /* | |
196 | * If we expanded the last AG, free the per-AG reservation | |
197 | * so we can reinitialize it with the new size. | |
198 | */ | |
199 | if (lastag_extended) { | |
200 | struct xfs_perag *pag; | |
201 | ||
202 | pag = xfs_perag_get(mp, id.agno); | |
203 | error = xfs_ag_resv_free(pag); | |
204 | xfs_perag_put(pag); | |
205 | if (error) | |
206 | return error; | |
207 | } | |
208 | /* | |
209 | * Reserve AG metadata blocks. ENOSPC here does not mean there | |
210 | * was a growfs failure, just that there still isn't space for | |
211 | * new user data after the grow has been run. | |
212 | */ | |
213 | error = xfs_fs_reserve_ag_blocks(mp); | |
214 | if (error == -ENOSPC) | |
215 | error = 0; | |
20e73b00 | 216 | } |
83a7f86e DC |
217 | return error; |
218 | ||
219 | out_trans_cancel: | |
220 | xfs_trans_cancel(tp); | |
221 | return error; | |
222 | } | |
223 | ||
224 | static int | |
225 | xfs_growfs_log_private( | |
07aabd9c GX |
226 | struct xfs_mount *mp, /* mount point for filesystem */ |
227 | struct xfs_growfs_log *in) /* growfs log input struct */ | |
83a7f86e DC |
228 | { |
229 | xfs_extlen_t nb; | |
230 | ||
231 | nb = in->newblocks; | |
232 | if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES)) | |
233 | return -EINVAL; | |
234 | if (nb == mp->m_sb.sb_logblocks && | |
235 | in->isint == (mp->m_sb.sb_logstart != 0)) | |
236 | return -EINVAL; | |
237 | /* | |
238 | * Moving the log is hard, need new interfaces to sync | |
239 | * the log first, hold off all activity while moving it. | |
240 | * Can have shorter or longer log in the same space, | |
241 | * or transform internal to external log or vice versa. | |
242 | */ | |
243 | return -ENOSYS; | |
244 | } | |
245 | ||
246 | static int | |
247 | xfs_growfs_imaxpct( | |
248 | struct xfs_mount *mp, | |
249 | __u32 imaxpct) | |
250 | { | |
251 | struct xfs_trans *tp; | |
252 | int dpct; | |
253 | int error; | |
254 | ||
255 | if (imaxpct > 100) | |
256 | return -EINVAL; | |
257 | ||
258 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, | |
259 | XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp); | |
260 | if (error) | |
261 | return error; | |
262 | ||
263 | dpct = imaxpct - mp->m_sb.sb_imax_pct; | |
264 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct); | |
265 | xfs_trans_set_sync(tp); | |
266 | return xfs_trans_commit(tp); | |
267 | } | |
268 | ||
1da177e4 LT |
269 | /* |
270 | * protected versions of growfs function acquire and release locks on the mount | |
271 | * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG, | |
272 | * XFS_IOC_FSGROWFSRT | |
273 | */ | |
1da177e4 LT |
274 | int |
275 | xfs_growfs_data( | |
87444b8c DC |
276 | struct xfs_mount *mp, |
277 | struct xfs_growfs_data *in) | |
1da177e4 | 278 | { |
87444b8c | 279 | int error = 0; |
743bb465 | 280 | |
281 | if (!capable(CAP_SYS_ADMIN)) | |
2451337d | 282 | return -EPERM; |
cc92e7ac | 283 | if (!mutex_trylock(&mp->m_growlock)) |
2451337d | 284 | return -EWOULDBLOCK; |
87444b8c DC |
285 | |
286 | /* update imaxpct separately to the physical grow of the filesystem */ | |
287 | if (in->imaxpct != mp->m_sb.sb_imax_pct) { | |
288 | error = xfs_growfs_imaxpct(mp, in->imaxpct); | |
289 | if (error) | |
290 | goto out_error; | |
291 | } | |
292 | ||
293 | if (in->newblocks != mp->m_sb.sb_dblocks) { | |
294 | error = xfs_growfs_data_private(mp, in); | |
295 | if (error) | |
296 | goto out_error; | |
297 | } | |
298 | ||
299 | /* Post growfs calculations needed to reflect new state in operations */ | |
300 | if (mp->m_sb.sb_imax_pct) { | |
301 | uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct; | |
302 | do_div(icount, 100); | |
ef325959 | 303 | M_IGEO(mp)->maxicount = XFS_FSB_TO_INO(mp, icount); |
87444b8c | 304 | } else |
ef325959 | 305 | M_IGEO(mp)->maxicount = 0; |
87444b8c | 306 | |
83a7f86e | 307 | /* Update secondary superblocks now the physical grow has completed */ |
b16817b6 | 308 | error = xfs_update_secondary_sbs(mp); |
83a7f86e | 309 | |
87444b8c | 310 | out_error: |
52785112 CH |
311 | /* |
312 | * Increment the generation unconditionally, the error could be from | |
313 | * updating the secondary superblocks, in which case the new size | |
314 | * is live already. | |
315 | */ | |
316 | mp->m_generation++; | |
cc92e7ac | 317 | mutex_unlock(&mp->m_growlock); |
1da177e4 LT |
318 | return error; |
319 | } | |
320 | ||
321 | int | |
322 | xfs_growfs_log( | |
323 | xfs_mount_t *mp, | |
07aabd9c | 324 | struct xfs_growfs_log *in) |
1da177e4 LT |
325 | { |
326 | int error; | |
743bb465 | 327 | |
328 | if (!capable(CAP_SYS_ADMIN)) | |
2451337d | 329 | return -EPERM; |
cc92e7ac | 330 | if (!mutex_trylock(&mp->m_growlock)) |
2451337d | 331 | return -EWOULDBLOCK; |
1da177e4 | 332 | error = xfs_growfs_log_private(mp, in); |
cc92e7ac | 333 | mutex_unlock(&mp->m_growlock); |
1da177e4 LT |
334 | return error; |
335 | } | |
336 | ||
337 | /* | |
338 | * exported through ioctl XFS_IOC_FSCOUNTS | |
339 | */ | |
340 | ||
91083269 | 341 | void |
1da177e4 LT |
342 | xfs_fs_counts( |
343 | xfs_mount_t *mp, | |
344 | xfs_fsop_counts_t *cnt) | |
345 | { | |
501ab323 | 346 | cnt->allocino = percpu_counter_read_positive(&mp->m_icount); |
e88b64ea | 347 | cnt->freeino = percpu_counter_read_positive(&mp->m_ifree); |
0d485ada | 348 | cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) - |
85bcfa26 | 349 | xfs_fdblocks_unavailable(mp); |
2229276c | 350 | cnt->freertx = percpu_counter_read_positive(&mp->m_frextents); |
1da177e4 LT |
351 | } |
352 | ||
353 | /* | |
354 | * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS | |
355 | * | |
356 | * xfs_reserve_blocks is called to set m_resblks | |
357 | * in the in-core mount table. The number of unused reserved blocks | |
c41564b5 | 358 | * is kept in m_resblks_avail. |
1da177e4 LT |
359 | * |
360 | * Reserve the requested number of blocks if available. Otherwise return | |
361 | * as many as possible to satisfy the request. The actual number | |
362 | * reserved are returned in outval | |
363 | * | |
364 | * A null inval pointer indicates that only the current reserved blocks | |
365 | * available should be returned no settings are changed. | |
366 | */ | |
367 | ||
368 | int | |
369 | xfs_reserve_blocks( | |
370 | xfs_mount_t *mp, | |
c8ce540d | 371 | uint64_t *inval, |
1da177e4 LT |
372 | xfs_fsop_resblks_t *outval) |
373 | { | |
c8ce540d DW |
374 | int64_t lcounter, delta; |
375 | int64_t fdblks_delta = 0; | |
376 | uint64_t request; | |
377 | int64_t free; | |
408fd484 | 378 | int error = 0; |
1da177e4 LT |
379 | |
380 | /* If inval is null, report current values and return */ | |
c8ce540d | 381 | if (inval == (uint64_t *)NULL) { |
84e1e99f | 382 | if (!outval) |
2451337d | 383 | return -EINVAL; |
1da177e4 LT |
384 | outval->resblks = mp->m_resblks; |
385 | outval->resblks_avail = mp->m_resblks_avail; | |
014c2544 | 386 | return 0; |
1da177e4 LT |
387 | } |
388 | ||
389 | request = *inval; | |
dbcabad1 DC |
390 | |
391 | /* | |
408fd484 BF |
392 | * With per-cpu counters, this becomes an interesting problem. we need |
393 | * to work out if we are freeing or allocation blocks first, then we can | |
394 | * do the modification as necessary. | |
dbcabad1 | 395 | * |
408fd484 BF |
396 | * We do this under the m_sb_lock so that if we are near ENOSPC, we will |
397 | * hold out any changes while we work out what to do. This means that | |
398 | * the amount of free space can change while we do this, so we need to | |
399 | * retry if we end up trying to reserve more space than is available. | |
dbcabad1 | 400 | */ |
3685c2a1 | 401 | spin_lock(&mp->m_sb_lock); |
1da177e4 LT |
402 | |
403 | /* | |
404 | * If our previous reservation was larger than the current value, | |
408fd484 BF |
405 | * then move any unused blocks back to the free pool. Modify the resblks |
406 | * counters directly since we shouldn't have any problems unreserving | |
407 | * space. | |
1da177e4 | 408 | */ |
1da177e4 LT |
409 | if (mp->m_resblks > request) { |
410 | lcounter = mp->m_resblks_avail - request; | |
411 | if (lcounter > 0) { /* release unused blocks */ | |
dbcabad1 | 412 | fdblks_delta = lcounter; |
1da177e4 LT |
413 | mp->m_resblks_avail -= lcounter; |
414 | } | |
415 | mp->m_resblks = request; | |
408fd484 BF |
416 | if (fdblks_delta) { |
417 | spin_unlock(&mp->m_sb_lock); | |
418 | error = xfs_mod_fdblocks(mp, fdblks_delta, 0); | |
419 | spin_lock(&mp->m_sb_lock); | |
420 | } | |
421 | ||
422 | goto out; | |
423 | } | |
4be536de | 424 | |
408fd484 BF |
425 | /* |
426 | * If the request is larger than the current reservation, reserve the | |
427 | * blocks before we update the reserve counters. Sample m_fdblocks and | |
428 | * perform a partial reservation if the request exceeds free space. | |
15f04fdc DW |
429 | * |
430 | * The code below estimates how many blocks it can request from | |
431 | * fdblocks to stash in the reserve pool. This is a classic TOCTOU | |
432 | * race since fdblocks updates are not always coordinated via | |
0baa2657 DW |
433 | * m_sb_lock. Set the reserve size even if there's not enough free |
434 | * space to fill it because mod_fdblocks will refill an undersized | |
435 | * reserve when it can. | |
408fd484 | 436 | */ |
15f04fdc | 437 | free = percpu_counter_sum(&mp->m_fdblocks) - |
c8c56825 | 438 | xfs_fdblocks_unavailable(mp); |
15f04fdc | 439 | delta = request - mp->m_resblks; |
0baa2657 | 440 | mp->m_resblks = request; |
15f04fdc | 441 | if (delta > 0 && free > 0) { |
dbcabad1 | 442 | /* |
408fd484 | 443 | * We'll either succeed in getting space from the free block |
15f04fdc DW |
444 | * count or we'll get an ENOSPC. Don't set the reserved flag |
445 | * here - we don't want to reserve the extra reserve blocks | |
446 | * from the reserve. | |
82be38bc DW |
447 | * |
448 | * The desired reserve size can change after we drop the lock. | |
449 | * Use mod_fdblocks to put the space into the reserve or into | |
450 | * fdblocks as appropriate. | |
dbcabad1 | 451 | */ |
15f04fdc | 452 | fdblks_delta = min(free, delta); |
408fd484 BF |
453 | spin_unlock(&mp->m_sb_lock); |
454 | error = xfs_mod_fdblocks(mp, -fdblks_delta, 0); | |
0baa2657 | 455 | if (!error) |
82be38bc DW |
456 | xfs_mod_fdblocks(mp, fdblks_delta, 0); |
457 | spin_lock(&mp->m_sb_lock); | |
dbcabad1 | 458 | } |
408fd484 BF |
459 | out: |
460 | if (outval) { | |
461 | outval->resblks = mp->m_resblks; | |
462 | outval->resblks_avail = mp->m_resblks_avail; | |
463 | } | |
464 | ||
465 | spin_unlock(&mp->m_sb_lock); | |
466 | return error; | |
1da177e4 LT |
467 | } |
468 | ||
1da177e4 LT |
469 | int |
470 | xfs_fs_goingdown( | |
471 | xfs_mount_t *mp, | |
c8ce540d | 472 | uint32_t inflags) |
1da177e4 LT |
473 | { |
474 | switch (inflags) { | |
475 | case XFS_FSOP_GOING_FLAGS_DEFAULT: { | |
040f04bd | 476 | if (!freeze_bdev(mp->m_super->s_bdev)) { |
7d04a335 | 477 | xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT); |
040f04bd | 478 | thaw_bdev(mp->m_super->s_bdev); |
1da177e4 | 479 | } |
1da177e4 LT |
480 | break; |
481 | } | |
482 | case XFS_FSOP_GOING_FLAGS_LOGFLUSH: | |
7d04a335 | 483 | xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT); |
1da177e4 LT |
484 | break; |
485 | case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH: | |
7d04a335 NS |
486 | xfs_force_shutdown(mp, |
487 | SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR); | |
1da177e4 LT |
488 | break; |
489 | default: | |
2451337d | 490 | return -EINVAL; |
1da177e4 LT |
491 | } |
492 | ||
493 | return 0; | |
494 | } | |
2af51f3a DC |
495 | |
496 | /* | |
497 | * Force a shutdown of the filesystem instantly while keeping the filesystem | |
498 | * consistent. We don't do an unmount here; just shutdown the shop, make sure | |
499 | * that absolutely nothing persistent happens to this filesystem after this | |
500 | * point. | |
b36d4651 DC |
501 | * |
502 | * The shutdown state change is atomic, resulting in the first and only the | |
503 | * first shutdown call processing the shutdown. This means we only shutdown the | |
504 | * log once as it requires, and we don't spam the logs when multiple concurrent | |
505 | * shutdowns race to set the shutdown flags. | |
2af51f3a DC |
506 | */ |
507 | void | |
508 | xfs_do_force_shutdown( | |
56668a5c | 509 | struct xfs_mount *mp, |
2eb7550d | 510 | uint32_t flags, |
2af51f3a DC |
511 | char *fname, |
512 | int lnnum) | |
513 | { | |
b36d4651 DC |
514 | int tag; |
515 | const char *why; | |
2af51f3a | 516 | |
41e63621 DC |
517 | |
518 | if (test_and_set_bit(XFS_OPSTATE_SHUTDOWN, &mp->m_opstate)) { | |
519 | xlog_shutdown_wait(mp->m_log); | |
56668a5c | 520 | return; |
41e63621 | 521 | } |
b36d4651 DC |
522 | if (mp->m_sb_bp) |
523 | mp->m_sb_bp->b_flags |= XBF_DONE; | |
b36d4651 DC |
524 | |
525 | if (flags & SHUTDOWN_FORCE_UMOUNT) | |
526 | xfs_alert(mp, "User initiated shutdown received."); | |
56668a5c | 527 | |
b36d4651 DC |
528 | if (xlog_force_shutdown(mp->m_log, flags)) { |
529 | tag = XFS_PTAG_SHUTDOWN_LOGERROR; | |
530 | why = "Log I/O Error"; | |
531 | } else if (flags & SHUTDOWN_CORRUPT_INCORE) { | |
532 | tag = XFS_PTAG_SHUTDOWN_CORRUPT; | |
533 | why = "Corruption of in-memory data"; | |
6f643c57 SR |
534 | } else if (flags & SHUTDOWN_CORRUPT_ONDISK) { |
535 | tag = XFS_PTAG_SHUTDOWN_CORRUPT; | |
536 | why = "Corruption of on-disk metadata"; | |
28d84620 | 537 | } else { |
b36d4651 DC |
538 | tag = XFS_PTAG_SHUTDOWN_IOERROR; |
539 | why = "Metadata I/O Error"; | |
2af51f3a | 540 | } |
56668a5c | 541 | |
7f89c838 DW |
542 | trace_xfs_force_shutdown(mp, tag, flags, fname, lnnum); |
543 | ||
b36d4651 DC |
544 | xfs_alert_tag(mp, tag, |
545 | "%s (0x%x) detected at %pS (%s:%d). Shutting down filesystem.", | |
546 | why, flags, __return_address, fname, lnnum); | |
56668a5c DC |
547 | xfs_alert(mp, |
548 | "Please unmount the filesystem and rectify the problem(s)"); | |
b36d4651 DC |
549 | if (xfs_error_level >= XFS_ERRLEVEL_HIGH) |
550 | xfs_stack_trace(); | |
2af51f3a | 551 | } |
84d69619 DW |
552 | |
553 | /* | |
554 | * Reserve free space for per-AG metadata. | |
555 | */ | |
556 | int | |
557 | xfs_fs_reserve_ag_blocks( | |
558 | struct xfs_mount *mp) | |
559 | { | |
560 | xfs_agnumber_t agno; | |
561 | struct xfs_perag *pag; | |
562 | int error = 0; | |
563 | int err2; | |
564 | ||
15a268d9 | 565 | mp->m_finobt_nores = false; |
f250eedc | 566 | for_each_perag(mp, agno, pag) { |
ebcbef3a | 567 | err2 = xfs_ag_resv_init(pag, NULL); |
84d69619 DW |
568 | if (err2 && !error) |
569 | error = err2; | |
570 | } | |
571 | ||
572 | if (error && error != -ENOSPC) { | |
573 | xfs_warn(mp, | |
574 | "Error %d reserving per-AG metadata reserve pool.", error); | |
575 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
576 | } | |
577 | ||
578 | return error; | |
579 | } | |
580 | ||
581 | /* | |
582 | * Free space reserved for per-AG metadata. | |
583 | */ | |
584 | int | |
585 | xfs_fs_unreserve_ag_blocks( | |
586 | struct xfs_mount *mp) | |
587 | { | |
588 | xfs_agnumber_t agno; | |
589 | struct xfs_perag *pag; | |
590 | int error = 0; | |
591 | int err2; | |
592 | ||
f250eedc | 593 | for_each_perag(mp, agno, pag) { |
84d69619 | 594 | err2 = xfs_ag_resv_free(pag); |
84d69619 DW |
595 | if (err2 && !error) |
596 | error = err2; | |
597 | } | |
598 | ||
599 | if (error) | |
600 | xfs_warn(mp, | |
601 | "Error %d freeing per-AG metadata reserve pool.", error); | |
602 | ||
603 | return error; | |
604 | } |