Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
f07c2250 | 3 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 4 | * All Rights Reserved. |
1da177e4 | 5 | */ |
93c189c1 | 6 | #include "xfs.h" |
3fcfab16 | 7 | #include <linux/backing-dev.h> |
6f643c57 | 8 | #include <linux/dax.h> |
1da177e4 | 9 | |
5467b34b | 10 | #include "xfs_shared.h" |
4fb6e8ad | 11 | #include "xfs_format.h" |
239880ef | 12 | #include "xfs_log_format.h" |
7fd36c44 | 13 | #include "xfs_trans_resv.h" |
b7963133 | 14 | #include "xfs_mount.h" |
0b1b213f | 15 | #include "xfs_trace.h" |
239880ef | 16 | #include "xfs_log.h" |
9fe5c77c | 17 | #include "xfs_log_recover.h" |
01728b44 | 18 | #include "xfs_log_priv.h" |
f593bf14 DC |
19 | #include "xfs_trans.h" |
20 | #include "xfs_buf_item.h" | |
e9e899a2 | 21 | #include "xfs_errortag.h" |
7561d27e | 22 | #include "xfs_error.h" |
9bbafc71 | 23 | #include "xfs_ag.h" |
b7963133 | 24 | |
231f91ab | 25 | struct kmem_cache *xfs_buf_cache; |
23ea4032 | 26 | |
37fd1678 DC |
27 | /* |
28 | * Locking orders | |
29 | * | |
30 | * xfs_buf_ioacct_inc: | |
31 | * xfs_buf_ioacct_dec: | |
32 | * b_sema (caller holds) | |
33 | * b_lock | |
34 | * | |
35 | * xfs_buf_stale: | |
36 | * b_sema (caller holds) | |
37 | * b_lock | |
38 | * lru_lock | |
39 | * | |
40 | * xfs_buf_rele: | |
41 | * b_lock | |
42 | * pag_buf_lock | |
43 | * lru_lock | |
44 | * | |
10fb9ac1 | 45 | * xfs_buftarg_drain_rele |
37fd1678 DC |
46 | * lru_lock |
47 | * b_lock (trylock due to inversion) | |
48 | * | |
49 | * xfs_buftarg_isolate | |
50 | * lru_lock | |
51 | * b_lock (trylock due to inversion) | |
52 | */ | |
1da177e4 | 53 | |
26e32875 CH |
54 | static int __xfs_buf_submit(struct xfs_buf *bp, bool wait); |
55 | ||
56 | static inline int | |
57 | xfs_buf_submit( | |
58 | struct xfs_buf *bp) | |
59 | { | |
60 | return __xfs_buf_submit(bp, !(bp->b_flags & XBF_ASYNC)); | |
61 | } | |
62 | ||
73c77e2c JB |
63 | static inline int |
64 | xfs_buf_is_vmapped( | |
65 | struct xfs_buf *bp) | |
66 | { | |
67 | /* | |
68 | * Return true if the buffer is vmapped. | |
69 | * | |
611c9946 DC |
70 | * b_addr is null if the buffer is not mapped, but the code is clever |
71 | * enough to know it doesn't have to map a single page, so the check has | |
72 | * to be both for b_addr and bp->b_page_count > 1. | |
73c77e2c | 73 | */ |
611c9946 | 74 | return bp->b_addr && bp->b_page_count > 1; |
73c77e2c JB |
75 | } |
76 | ||
77 | static inline int | |
78 | xfs_buf_vmap_len( | |
79 | struct xfs_buf *bp) | |
80 | { | |
54cd3aa6 | 81 | return (bp->b_page_count * PAGE_SIZE); |
73c77e2c JB |
82 | } |
83 | ||
9c7504aa BF |
84 | /* |
85 | * Bump the I/O in flight count on the buftarg if we haven't yet done so for | |
86 | * this buffer. The count is incremented once per buffer (per hold cycle) | |
87 | * because the corresponding decrement is deferred to buffer release. Buffers | |
88 | * can undergo I/O multiple times in a hold-release cycle and per buffer I/O | |
89 | * tracking adds unnecessary overhead. This is used for sychronization purposes | |
10fb9ac1 | 90 | * with unmount (see xfs_buftarg_drain()), so all we really need is a count of |
9c7504aa BF |
91 | * in-flight buffers. |
92 | * | |
93 | * Buffers that are never released (e.g., superblock, iclog buffers) must set | |
94 | * the XBF_NO_IOACCT flag before I/O submission. Otherwise, the buftarg count | |
95 | * never reaches zero and unmount hangs indefinitely. | |
96 | */ | |
97 | static inline void | |
98 | xfs_buf_ioacct_inc( | |
99 | struct xfs_buf *bp) | |
100 | { | |
63db7c81 | 101 | if (bp->b_flags & XBF_NO_IOACCT) |
9c7504aa BF |
102 | return; |
103 | ||
104 | ASSERT(bp->b_flags & XBF_ASYNC); | |
63db7c81 BF |
105 | spin_lock(&bp->b_lock); |
106 | if (!(bp->b_state & XFS_BSTATE_IN_FLIGHT)) { | |
107 | bp->b_state |= XFS_BSTATE_IN_FLIGHT; | |
108 | percpu_counter_inc(&bp->b_target->bt_io_count); | |
109 | } | |
110 | spin_unlock(&bp->b_lock); | |
9c7504aa BF |
111 | } |
112 | ||
113 | /* | |
114 | * Clear the in-flight state on a buffer about to be released to the LRU or | |
115 | * freed and unaccount from the buftarg. | |
116 | */ | |
117 | static inline void | |
63db7c81 | 118 | __xfs_buf_ioacct_dec( |
9c7504aa BF |
119 | struct xfs_buf *bp) |
120 | { | |
95989c46 | 121 | lockdep_assert_held(&bp->b_lock); |
9c7504aa | 122 | |
63db7c81 BF |
123 | if (bp->b_state & XFS_BSTATE_IN_FLIGHT) { |
124 | bp->b_state &= ~XFS_BSTATE_IN_FLIGHT; | |
125 | percpu_counter_dec(&bp->b_target->bt_io_count); | |
126 | } | |
127 | } | |
128 | ||
129 | static inline void | |
130 | xfs_buf_ioacct_dec( | |
131 | struct xfs_buf *bp) | |
132 | { | |
133 | spin_lock(&bp->b_lock); | |
134 | __xfs_buf_ioacct_dec(bp); | |
135 | spin_unlock(&bp->b_lock); | |
9c7504aa BF |
136 | } |
137 | ||
430cbeb8 DC |
138 | /* |
139 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
140 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
141 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
142 | * to remove the reference that LRU holds on the buffer. | |
143 | * | |
144 | * This prevents build-up of stale buffers on the LRU. | |
145 | */ | |
146 | void | |
147 | xfs_buf_stale( | |
148 | struct xfs_buf *bp) | |
149 | { | |
43ff2122 CH |
150 | ASSERT(xfs_buf_islocked(bp)); |
151 | ||
430cbeb8 | 152 | bp->b_flags |= XBF_STALE; |
43ff2122 CH |
153 | |
154 | /* | |
155 | * Clear the delwri status so that a delwri queue walker will not | |
156 | * flush this buffer to disk now that it is stale. The delwri queue has | |
157 | * a reference to the buffer, so this is safe to do. | |
158 | */ | |
159 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
160 | ||
9c7504aa BF |
161 | /* |
162 | * Once the buffer is marked stale and unlocked, a subsequent lookup | |
163 | * could reset b_flags. There is no guarantee that the buffer is | |
164 | * unaccounted (released to LRU) before that occurs. Drop in-flight | |
165 | * status now to preserve accounting consistency. | |
166 | */ | |
a4082357 | 167 | spin_lock(&bp->b_lock); |
63db7c81 BF |
168 | __xfs_buf_ioacct_dec(bp); |
169 | ||
a4082357 DC |
170 | atomic_set(&bp->b_lru_ref, 0); |
171 | if (!(bp->b_state & XFS_BSTATE_DISPOSE) && | |
e80dfa19 DC |
172 | (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru))) |
173 | atomic_dec(&bp->b_hold); | |
174 | ||
430cbeb8 | 175 | ASSERT(atomic_read(&bp->b_hold) >= 1); |
a4082357 | 176 | spin_unlock(&bp->b_lock); |
430cbeb8 | 177 | } |
1da177e4 | 178 | |
3e85c868 DC |
179 | static int |
180 | xfs_buf_get_maps( | |
181 | struct xfs_buf *bp, | |
182 | int map_count) | |
183 | { | |
184 | ASSERT(bp->b_maps == NULL); | |
185 | bp->b_map_count = map_count; | |
186 | ||
187 | if (map_count == 1) { | |
f4b42421 | 188 | bp->b_maps = &bp->__b_map; |
3e85c868 DC |
189 | return 0; |
190 | } | |
191 | ||
192 | bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map), | |
193 | KM_NOFS); | |
194 | if (!bp->b_maps) | |
2451337d | 195 | return -ENOMEM; |
3e85c868 DC |
196 | return 0; |
197 | } | |
198 | ||
199 | /* | |
200 | * Frees b_pages if it was allocated. | |
201 | */ | |
202 | static void | |
203 | xfs_buf_free_maps( | |
204 | struct xfs_buf *bp) | |
205 | { | |
f4b42421 | 206 | if (bp->b_maps != &bp->__b_map) { |
3e85c868 DC |
207 | kmem_free(bp->b_maps); |
208 | bp->b_maps = NULL; | |
209 | } | |
210 | } | |
211 | ||
32dff5e5 | 212 | static int |
3e85c868 | 213 | _xfs_buf_alloc( |
4347b9d7 | 214 | struct xfs_buftarg *target, |
3e85c868 DC |
215 | struct xfs_buf_map *map, |
216 | int nmaps, | |
32dff5e5 DW |
217 | xfs_buf_flags_t flags, |
218 | struct xfs_buf **bpp) | |
1da177e4 | 219 | { |
4347b9d7 | 220 | struct xfs_buf *bp; |
3e85c868 DC |
221 | int error; |
222 | int i; | |
4347b9d7 | 223 | |
32dff5e5 | 224 | *bpp = NULL; |
182696fb | 225 | bp = kmem_cache_zalloc(xfs_buf_cache, GFP_NOFS | __GFP_NOFAIL); |
4347b9d7 | 226 | |
1da177e4 | 227 | /* |
12bcb3f7 DC |
228 | * We don't want certain flags to appear in b_flags unless they are |
229 | * specifically set by later operations on the buffer. | |
1da177e4 | 230 | */ |
611c9946 | 231 | flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD); |
ce8e922c | 232 | |
ce8e922c | 233 | atomic_set(&bp->b_hold, 1); |
430cbeb8 | 234 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 235 | init_completion(&bp->b_iowait); |
430cbeb8 | 236 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 237 | INIT_LIST_HEAD(&bp->b_list); |
643c8c05 | 238 | INIT_LIST_HEAD(&bp->b_li_list); |
a731cd11 | 239 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
a4082357 | 240 | spin_lock_init(&bp->b_lock); |
ce8e922c | 241 | bp->b_target = target; |
dbd329f1 | 242 | bp->b_mount = target->bt_mount; |
3e85c868 | 243 | bp->b_flags = flags; |
de1cbee4 | 244 | |
1da177e4 | 245 | /* |
aa0e8833 DC |
246 | * Set length and io_length to the same value initially. |
247 | * I/O routines should use io_length, which will be the same in | |
1da177e4 LT |
248 | * most cases but may be reset (e.g. XFS recovery). |
249 | */ | |
3e85c868 DC |
250 | error = xfs_buf_get_maps(bp, nmaps); |
251 | if (error) { | |
182696fb | 252 | kmem_cache_free(xfs_buf_cache, bp); |
32dff5e5 | 253 | return error; |
3e85c868 DC |
254 | } |
255 | ||
4c7f65ae | 256 | bp->b_rhash_key = map[0].bm_bn; |
3e85c868 DC |
257 | bp->b_length = 0; |
258 | for (i = 0; i < nmaps; i++) { | |
259 | bp->b_maps[i].bm_bn = map[i].bm_bn; | |
260 | bp->b_maps[i].bm_len = map[i].bm_len; | |
261 | bp->b_length += map[i].bm_len; | |
262 | } | |
3e85c868 | 263 | |
ce8e922c NS |
264 | atomic_set(&bp->b_pin_count, 0); |
265 | init_waitqueue_head(&bp->b_waiters); | |
266 | ||
dbd329f1 | 267 | XFS_STATS_INC(bp->b_mount, xb_create); |
0b1b213f | 268 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 | 269 | |
32dff5e5 DW |
270 | *bpp = bp; |
271 | return 0; | |
1da177e4 LT |
272 | } |
273 | ||
e7d236a6 DC |
274 | static void |
275 | xfs_buf_free_pages( | |
e8222613 | 276 | struct xfs_buf *bp) |
1da177e4 | 277 | { |
e7d236a6 DC |
278 | uint i; |
279 | ||
280 | ASSERT(bp->b_flags & _XBF_PAGES); | |
281 | ||
282 | if (xfs_buf_is_vmapped(bp)) | |
54cd3aa6 | 283 | vm_unmap_ram(bp->b_addr, bp->b_page_count); |
e7d236a6 DC |
284 | |
285 | for (i = 0; i < bp->b_page_count; i++) { | |
286 | if (bp->b_pages[i]) | |
287 | __free_page(bp->b_pages[i]); | |
288 | } | |
c7b23b68 | 289 | mm_account_reclaimed_pages(bp->b_page_count); |
e7d236a6 | 290 | |
02c51173 | 291 | if (bp->b_pages != bp->b_page_array) |
f0e2d93c | 292 | kmem_free(bp->b_pages); |
02c51173 | 293 | bp->b_pages = NULL; |
e7d236a6 | 294 | bp->b_flags &= ~_XBF_PAGES; |
1da177e4 LT |
295 | } |
296 | ||
298f3422 DC |
297 | static void |
298 | xfs_buf_free_callback( | |
299 | struct callback_head *cb) | |
300 | { | |
301 | struct xfs_buf *bp = container_of(cb, struct xfs_buf, b_rcu); | |
302 | ||
303 | xfs_buf_free_maps(bp); | |
304 | kmem_cache_free(xfs_buf_cache, bp); | |
305 | } | |
306 | ||
25a40957 | 307 | static void |
ce8e922c | 308 | xfs_buf_free( |
e8222613 | 309 | struct xfs_buf *bp) |
1da177e4 | 310 | { |
0b1b213f | 311 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 312 | |
430cbeb8 DC |
313 | ASSERT(list_empty(&bp->b_lru)); |
314 | ||
e7d236a6 DC |
315 | if (bp->b_flags & _XBF_PAGES) |
316 | xfs_buf_free_pages(bp); | |
317 | else if (bp->b_flags & _XBF_KMEM) | |
0e6e847f | 318 | kmem_free(bp->b_addr); |
e7d236a6 | 319 | |
298f3422 | 320 | call_rcu(&bp->b_rcu, xfs_buf_free_callback); |
1da177e4 LT |
321 | } |
322 | ||
0a683794 DC |
323 | static int |
324 | xfs_buf_alloc_kmem( | |
325 | struct xfs_buf *bp, | |
0a683794 | 326 | xfs_buf_flags_t flags) |
1da177e4 | 327 | { |
0a683794 | 328 | xfs_km_flags_t kmflag_mask = KM_NOFS; |
8bcac744 | 329 | size_t size = BBTOB(bp->b_length); |
3219e8cf | 330 | |
0a683794 DC |
331 | /* Assure zeroed buffer for non-read cases. */ |
332 | if (!(flags & XBF_READ)) | |
3219e8cf | 333 | kmflag_mask |= KM_ZERO; |
1da177e4 | 334 | |
98fe2c3c | 335 | bp->b_addr = kmem_alloc(size, kmflag_mask); |
0a683794 DC |
336 | if (!bp->b_addr) |
337 | return -ENOMEM; | |
0e6e847f | 338 | |
0a683794 DC |
339 | if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) != |
340 | ((unsigned long)bp->b_addr & PAGE_MASK)) { | |
341 | /* b_addr spans two pages - use alloc_page instead */ | |
342 | kmem_free(bp->b_addr); | |
343 | bp->b_addr = NULL; | |
344 | return -ENOMEM; | |
0e6e847f | 345 | } |
0a683794 DC |
346 | bp->b_offset = offset_in_page(bp->b_addr); |
347 | bp->b_pages = bp->b_page_array; | |
348 | bp->b_pages[0] = kmem_to_page(bp->b_addr); | |
349 | bp->b_page_count = 1; | |
350 | bp->b_flags |= _XBF_KMEM; | |
351 | return 0; | |
352 | } | |
353 | ||
354 | static int | |
355 | xfs_buf_alloc_pages( | |
356 | struct xfs_buf *bp, | |
0a683794 DC |
357 | xfs_buf_flags_t flags) |
358 | { | |
289ae7b4 | 359 | gfp_t gfp_mask = __GFP_NOWARN; |
c9fa5630 | 360 | long filled = 0; |
0a683794 | 361 | |
289ae7b4 DC |
362 | if (flags & XBF_READ_AHEAD) |
363 | gfp_mask |= __GFP_NORETRY; | |
364 | else | |
365 | gfp_mask |= GFP_NOFS; | |
366 | ||
02c51173 | 367 | /* Make sure that we have a page list */ |
934d1076 | 368 | bp->b_page_count = DIV_ROUND_UP(BBTOB(bp->b_length), PAGE_SIZE); |
02c51173 DC |
369 | if (bp->b_page_count <= XB_PAGES) { |
370 | bp->b_pages = bp->b_page_array; | |
371 | } else { | |
372 | bp->b_pages = kzalloc(sizeof(struct page *) * bp->b_page_count, | |
373 | gfp_mask); | |
374 | if (!bp->b_pages) | |
375 | return -ENOMEM; | |
376 | } | |
377 | bp->b_flags |= _XBF_PAGES; | |
378 | ||
0a683794 DC |
379 | /* Assure zeroed buffer for non-read cases. */ |
380 | if (!(flags & XBF_READ)) | |
381 | gfp_mask |= __GFP_ZERO; | |
0e6e847f | 382 | |
c9fa5630 DC |
383 | /* |
384 | * Bulk filling of pages can take multiple calls. Not filling the entire | |
385 | * array is not an allocation failure, so don't back off if we get at | |
386 | * least one extra page. | |
387 | */ | |
388 | for (;;) { | |
389 | long last = filled; | |
390 | ||
391 | filled = alloc_pages_bulk_array(gfp_mask, bp->b_page_count, | |
392 | bp->b_pages); | |
393 | if (filled == bp->b_page_count) { | |
394 | XFS_STATS_INC(bp->b_mount, xb_page_found); | |
395 | break; | |
1da177e4 LT |
396 | } |
397 | ||
c9fa5630 DC |
398 | if (filled != last) |
399 | continue; | |
400 | ||
401 | if (flags & XBF_READ_AHEAD) { | |
e7d236a6 DC |
402 | xfs_buf_free_pages(bp); |
403 | return -ENOMEM; | |
c9fa5630 | 404 | } |
1da177e4 | 405 | |
c9fa5630 | 406 | XFS_STATS_INC(bp->b_mount, xb_page_retries); |
4034247a | 407 | memalloc_retry_wait(gfp_mask); |
1da177e4 | 408 | } |
0e6e847f | 409 | return 0; |
1da177e4 LT |
410 | } |
411 | ||
412 | /* | |
25985edc | 413 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
414 | */ |
415 | STATIC int | |
ce8e922c | 416 | _xfs_buf_map_pages( |
e8222613 | 417 | struct xfs_buf *bp, |
b9b3fe15 | 418 | xfs_buf_flags_t flags) |
1da177e4 | 419 | { |
0e6e847f | 420 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 421 | if (bp->b_page_count == 1) { |
0e6e847f | 422 | /* A single page buffer is always mappable */ |
54cd3aa6 | 423 | bp->b_addr = page_address(bp->b_pages[0]); |
611c9946 DC |
424 | } else if (flags & XBF_UNMAPPED) { |
425 | bp->b_addr = NULL; | |
426 | } else { | |
a19fb380 | 427 | int retried = 0; |
9ba1fb2c | 428 | unsigned nofs_flag; |
ae687e58 DC |
429 | |
430 | /* | |
cf085a1b | 431 | * vm_map_ram() will allocate auxiliary structures (e.g. |
ae687e58 DC |
432 | * pagetables) with GFP_KERNEL, yet we are likely to be under |
433 | * GFP_NOFS context here. Hence we need to tell memory reclaim | |
9ba1fb2c | 434 | * that we are in such a context via PF_MEMALLOC_NOFS to prevent |
ae687e58 DC |
435 | * memory reclaim re-entering the filesystem here and |
436 | * potentially deadlocking. | |
437 | */ | |
9ba1fb2c | 438 | nofs_flag = memalloc_nofs_save(); |
a19fb380 DC |
439 | do { |
440 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
d4efd79a | 441 | -1); |
a19fb380 DC |
442 | if (bp->b_addr) |
443 | break; | |
444 | vm_unmap_aliases(); | |
445 | } while (retried++ <= 1); | |
9ba1fb2c | 446 | memalloc_nofs_restore(nofs_flag); |
a19fb380 DC |
447 | |
448 | if (!bp->b_addr) | |
1da177e4 | 449 | return -ENOMEM; |
1da177e4 LT |
450 | } |
451 | ||
452 | return 0; | |
453 | } | |
454 | ||
455 | /* | |
456 | * Finding and Reading Buffers | |
457 | */ | |
6031e73a LS |
458 | static int |
459 | _xfs_buf_obj_cmp( | |
460 | struct rhashtable_compare_arg *arg, | |
461 | const void *obj) | |
462 | { | |
463 | const struct xfs_buf_map *map = arg->key; | |
464 | const struct xfs_buf *bp = obj; | |
465 | ||
466 | /* | |
467 | * The key hashing in the lookup path depends on the key being the | |
468 | * first element of the compare_arg, make sure to assert this. | |
469 | */ | |
470 | BUILD_BUG_ON(offsetof(struct xfs_buf_map, bm_bn) != 0); | |
471 | ||
4c7f65ae | 472 | if (bp->b_rhash_key != map->bm_bn) |
6031e73a LS |
473 | return 1; |
474 | ||
475 | if (unlikely(bp->b_length != map->bm_len)) { | |
476 | /* | |
477 | * found a block number match. If the range doesn't | |
478 | * match, the only way this is allowed is if the buffer | |
479 | * in the cache is stale and the transaction that made | |
480 | * it stale has not yet committed. i.e. we are | |
481 | * reallocating a busy extent. Skip this buffer and | |
482 | * continue searching for an exact match. | |
483 | */ | |
9ed851f6 DW |
484 | if (!(map->bm_flags & XBM_LIVESCAN)) |
485 | ASSERT(bp->b_flags & XBF_STALE); | |
6031e73a LS |
486 | return 1; |
487 | } | |
488 | return 0; | |
489 | } | |
490 | ||
491 | static const struct rhashtable_params xfs_buf_hash_params = { | |
492 | .min_size = 32, /* empty AGs have minimal footprint */ | |
493 | .nelem_hint = 16, | |
494 | .key_len = sizeof(xfs_daddr_t), | |
4c7f65ae | 495 | .key_offset = offsetof(struct xfs_buf, b_rhash_key), |
6031e73a LS |
496 | .head_offset = offsetof(struct xfs_buf, b_rhash_head), |
497 | .automatic_shrinking = true, | |
498 | .obj_cmpfn = _xfs_buf_obj_cmp, | |
499 | }; | |
500 | ||
501 | int | |
502 | xfs_buf_hash_init( | |
503 | struct xfs_perag *pag) | |
504 | { | |
505 | spin_lock_init(&pag->pag_buf_lock); | |
506 | return rhashtable_init(&pag->pag_buf_hash, &xfs_buf_hash_params); | |
507 | } | |
508 | ||
509 | void | |
510 | xfs_buf_hash_destroy( | |
511 | struct xfs_perag *pag) | |
512 | { | |
513 | rhashtable_destroy(&pag->pag_buf_hash); | |
514 | } | |
1da177e4 | 515 | |
b027d4c9 | 516 | static int |
de67dc57 | 517 | xfs_buf_map_verify( |
e70b73f8 | 518 | struct xfs_buftarg *btp, |
de67dc57 | 519 | struct xfs_buf_map *map) |
1da177e4 | 520 | { |
10616b80 | 521 | xfs_daddr_t eofs; |
1da177e4 LT |
522 | |
523 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
de67dc57 DC |
524 | ASSERT(!(BBTOB(map->bm_len) < btp->bt_meta_sectorsize)); |
525 | ASSERT(!(BBTOB(map->bm_bn) & (xfs_off_t)btp->bt_meta_sectormask)); | |
1da177e4 | 526 | |
10616b80 DC |
527 | /* |
528 | * Corrupted block numbers can get through to here, unfortunately, so we | |
529 | * have to check that the buffer falls within the filesystem bounds. | |
530 | */ | |
531 | eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks); | |
de67dc57 | 532 | if (map->bm_bn < 0 || map->bm_bn >= eofs) { |
10616b80 | 533 | xfs_alert(btp->bt_mount, |
c219b015 | 534 | "%s: daddr 0x%llx out of range, EOFS 0x%llx", |
de67dc57 | 535 | __func__, map->bm_bn, eofs); |
7bc0dc27 | 536 | WARN_ON(1); |
b027d4c9 | 537 | return -EFSCORRUPTED; |
10616b80 | 538 | } |
b027d4c9 | 539 | return 0; |
de67dc57 | 540 | } |
1da177e4 | 541 | |
de67dc57 DC |
542 | static int |
543 | xfs_buf_find_lock( | |
544 | struct xfs_buf *bp, | |
545 | xfs_buf_flags_t flags) | |
546 | { | |
d8d9bbb0 DC |
547 | if (flags & XBF_TRYLOCK) { |
548 | if (!xfs_buf_trylock(bp)) { | |
de67dc57 | 549 | XFS_STATS_INC(bp->b_mount, xb_busy_locked); |
b027d4c9 | 550 | return -EAGAIN; |
1da177e4 | 551 | } |
d8d9bbb0 | 552 | } else { |
0c842ad4 | 553 | xfs_buf_lock(bp); |
de67dc57 | 554 | XFS_STATS_INC(bp->b_mount, xb_get_locked_waited); |
1da177e4 LT |
555 | } |
556 | ||
0e6e847f DC |
557 | /* |
558 | * if the buffer is stale, clear all the external state associated with | |
559 | * it. We need to keep flags such as how we allocated the buffer memory | |
560 | * intact here. | |
561 | */ | |
ce8e922c | 562 | if (bp->b_flags & XBF_STALE) { |
9ed851f6 DW |
563 | if (flags & XBF_LIVESCAN) { |
564 | xfs_buf_unlock(bp); | |
565 | return -ENOENT; | |
566 | } | |
ce8e922c | 567 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); |
611c9946 | 568 | bp->b_flags &= _XBF_KMEM | _XBF_PAGES; |
1813dd64 | 569 | bp->b_ops = NULL; |
2f926587 | 570 | } |
b027d4c9 | 571 | return 0; |
1da177e4 LT |
572 | } |
573 | ||
34800080 | 574 | static inline int |
de67dc57 DC |
575 | xfs_buf_lookup( |
576 | struct xfs_perag *pag, | |
34800080 DC |
577 | struct xfs_buf_map *map, |
578 | xfs_buf_flags_t flags, | |
579 | struct xfs_buf **bpp) | |
8925a3dc | 580 | { |
de67dc57 | 581 | struct xfs_buf *bp; |
b027d4c9 | 582 | int error; |
b027d4c9 | 583 | |
298f3422 | 584 | rcu_read_lock(); |
de67dc57 | 585 | bp = rhashtable_lookup(&pag->pag_buf_hash, map, xfs_buf_hash_params); |
298f3422 DC |
586 | if (!bp || !atomic_inc_not_zero(&bp->b_hold)) { |
587 | rcu_read_unlock(); | |
34800080 DC |
588 | return -ENOENT; |
589 | } | |
298f3422 | 590 | rcu_read_unlock(); |
de67dc57 | 591 | |
34800080 DC |
592 | error = xfs_buf_find_lock(bp, flags); |
593 | if (error) { | |
594 | xfs_buf_rele(bp); | |
595 | return error; | |
de67dc57 DC |
596 | } |
597 | ||
34800080 DC |
598 | trace_xfs_buf_find(bp, flags, _RET_IP_); |
599 | *bpp = bp; | |
de67dc57 | 600 | return 0; |
8925a3dc DC |
601 | } |
602 | ||
1da177e4 | 603 | /* |
34800080 DC |
604 | * Insert the new_bp into the hash table. This consumes the perag reference |
605 | * taken for the lookup regardless of the result of the insert. | |
1da177e4 | 606 | */ |
b027d4c9 | 607 | static int |
34800080 | 608 | xfs_buf_find_insert( |
e70b73f8 | 609 | struct xfs_buftarg *btp, |
34800080 DC |
610 | struct xfs_perag *pag, |
611 | struct xfs_buf_map *cmap, | |
6dde2707 DC |
612 | struct xfs_buf_map *map, |
613 | int nmaps, | |
3848b5f6 DW |
614 | xfs_buf_flags_t flags, |
615 | struct xfs_buf **bpp) | |
1da177e4 | 616 | { |
3815832a | 617 | struct xfs_buf *new_bp; |
e8222613 | 618 | struct xfs_buf *bp; |
9bb38aa0 | 619 | int error; |
1da177e4 | 620 | |
34800080 | 621 | error = _xfs_buf_alloc(btp, map, nmaps, flags, &new_bp); |
32dff5e5 | 622 | if (error) |
34800080 | 623 | goto out_drop_pag; |
1da177e4 | 624 | |
8bcac744 DC |
625 | /* |
626 | * For buffers that fit entirely within a single page, first attempt to | |
627 | * allocate the memory from the heap to minimise memory usage. If we | |
628 | * can't get heap memory for these small buffers, we fall back to using | |
629 | * the page allocator. | |
630 | */ | |
631 | if (BBTOB(new_bp->b_length) >= PAGE_SIZE || | |
632 | xfs_buf_alloc_kmem(new_bp, flags) < 0) { | |
633 | error = xfs_buf_alloc_pages(new_bp, flags); | |
634 | if (error) | |
635 | goto out_free_buf; | |
636 | } | |
fe2429b0 | 637 | |
74f75a0c | 638 | spin_lock(&pag->pag_buf_lock); |
32dd4f9c DC |
639 | bp = rhashtable_lookup_get_insert_fast(&pag->pag_buf_hash, |
640 | &new_bp->b_rhash_head, xfs_buf_hash_params); | |
641 | if (IS_ERR(bp)) { | |
642 | error = PTR_ERR(bp); | |
643 | spin_unlock(&pag->pag_buf_lock); | |
170041f7 | 644 | goto out_free_buf; |
32dd4f9c | 645 | } |
34800080 | 646 | if (bp) { |
32dd4f9c | 647 | /* found an existing buffer */ |
34800080 DC |
648 | atomic_inc(&bp->b_hold); |
649 | spin_unlock(&pag->pag_buf_lock); | |
650 | error = xfs_buf_find_lock(bp, flags); | |
651 | if (error) | |
652 | xfs_buf_rele(bp); | |
653 | else | |
654 | *bpp = bp; | |
655 | goto out_free_buf; | |
656 | } | |
1da177e4 | 657 | |
32dd4f9c | 658 | /* The new buffer keeps the perag reference until it is freed. */ |
34800080 | 659 | new_bp->b_pag = pag; |
b027d4c9 | 660 | spin_unlock(&pag->pag_buf_lock); |
34800080 | 661 | *bpp = new_bp; |
b027d4c9 | 662 | return 0; |
3815832a | 663 | |
34800080 DC |
664 | out_free_buf: |
665 | xfs_buf_free(new_bp); | |
666 | out_drop_pag: | |
74f75a0c | 667 | xfs_perag_put(pag); |
34800080 | 668 | return error; |
1da177e4 | 669 | } |
1da177e4 | 670 | |
1da177e4 | 671 | /* |
3815832a DC |
672 | * Assembles a buffer covering the specified range. The code is optimised for |
673 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
674 | * more hits than misses. | |
1da177e4 | 675 | */ |
3848b5f6 | 676 | int |
6dde2707 | 677 | xfs_buf_get_map( |
34800080 | 678 | struct xfs_buftarg *btp, |
6dde2707 DC |
679 | struct xfs_buf_map *map, |
680 | int nmaps, | |
3848b5f6 DW |
681 | xfs_buf_flags_t flags, |
682 | struct xfs_buf **bpp) | |
1da177e4 | 683 | { |
34800080 DC |
684 | struct xfs_perag *pag; |
685 | struct xfs_buf *bp = NULL; | |
686 | struct xfs_buf_map cmap = { .bm_bn = map[0].bm_bn }; | |
9bb38aa0 | 687 | int error; |
34800080 | 688 | int i; |
1da177e4 | 689 | |
9ed851f6 DW |
690 | if (flags & XBF_LIVESCAN) |
691 | cmap.bm_flags |= XBM_LIVESCAN; | |
34800080 DC |
692 | for (i = 0; i < nmaps; i++) |
693 | cmap.bm_len += map[i].bm_len; | |
3815832a | 694 | |
34800080 | 695 | error = xfs_buf_map_verify(btp, &cmap); |
32dff5e5 | 696 | if (error) |
3848b5f6 | 697 | return error; |
1da177e4 | 698 | |
34800080 DC |
699 | pag = xfs_perag_get(btp->bt_mount, |
700 | xfs_daddr_to_agno(btp->bt_mount, cmap.bm_bn)); | |
fe2429b0 | 701 | |
34800080 DC |
702 | error = xfs_buf_lookup(pag, &cmap, flags, &bp); |
703 | if (error && error != -ENOENT) | |
704 | goto out_put_perag; | |
705 | ||
706 | /* cache hits always outnumber misses by at least 10:1 */ | |
707 | if (unlikely(!bp)) { | |
708 | XFS_STATS_INC(btp->bt_mount, xb_miss_locked); | |
3815832a | 709 | |
34800080 DC |
710 | if (flags & XBF_INCORE) |
711 | goto out_put_perag; | |
1da177e4 | 712 | |
34800080 DC |
713 | /* xfs_buf_find_insert() consumes the perag reference. */ |
714 | error = xfs_buf_find_insert(btp, pag, &cmap, map, nmaps, | |
715 | flags, &bp); | |
716 | if (error) | |
717 | return error; | |
718 | } else { | |
719 | XFS_STATS_INC(btp->bt_mount, xb_get_locked); | |
720 | xfs_perag_put(pag); | |
721 | } | |
722 | ||
723 | /* We do not hold a perag reference anymore. */ | |
611c9946 | 724 | if (!bp->b_addr) { |
ce8e922c | 725 | error = _xfs_buf_map_pages(bp, flags); |
1da177e4 | 726 | if (unlikely(error)) { |
34800080 | 727 | xfs_warn_ratelimited(btp->bt_mount, |
93baa55a DW |
728 | "%s: failed to map %u pages", __func__, |
729 | bp->b_page_count); | |
a8acad70 | 730 | xfs_buf_relse(bp); |
3848b5f6 | 731 | return error; |
1da177e4 LT |
732 | } |
733 | } | |
734 | ||
b79f4a1c DC |
735 | /* |
736 | * Clear b_error if this is a lookup from a caller that doesn't expect | |
737 | * valid data to be found in the buffer. | |
738 | */ | |
739 | if (!(flags & XBF_READ)) | |
740 | xfs_buf_ioerror(bp, 0); | |
741 | ||
34800080 | 742 | XFS_STATS_INC(btp->bt_mount, xb_get); |
0b1b213f | 743 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
3848b5f6 DW |
744 | *bpp = bp; |
745 | return 0; | |
34800080 DC |
746 | |
747 | out_put_perag: | |
748 | xfs_perag_put(pag); | |
170041f7 | 749 | return error; |
1da177e4 LT |
750 | } |
751 | ||
26e32875 | 752 | int |
5d765b97 | 753 | _xfs_buf_read( |
e8222613 | 754 | struct xfs_buf *bp, |
5d765b97 CH |
755 | xfs_buf_flags_t flags) |
756 | { | |
43ff2122 | 757 | ASSERT(!(flags & XBF_WRITE)); |
f4b42421 | 758 | ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL); |
5d765b97 | 759 | |
26e32875 | 760 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD | XBF_DONE); |
1d5ae5df | 761 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 762 | |
6af88cda | 763 | return xfs_buf_submit(bp); |
5d765b97 CH |
764 | } |
765 | ||
1aff5696 | 766 | /* |
75d02303 | 767 | * Reverify a buffer found in cache without an attached ->b_ops. |
add46b3b | 768 | * |
75d02303 BF |
769 | * If the caller passed an ops structure and the buffer doesn't have ops |
770 | * assigned, set the ops and use it to verify the contents. If verification | |
771 | * fails, clear XBF_DONE. We assume the buffer has no recorded errors and is | |
772 | * already in XBF_DONE state on entry. | |
add46b3b | 773 | * |
75d02303 BF |
774 | * Under normal operations, every in-core buffer is verified on read I/O |
775 | * completion. There are two scenarios that can lead to in-core buffers without | |
776 | * an assigned ->b_ops. The first is during log recovery of buffers on a V4 | |
777 | * filesystem, though these buffers are purged at the end of recovery. The | |
778 | * other is online repair, which intentionally reads with a NULL buffer ops to | |
779 | * run several verifiers across an in-core buffer in order to establish buffer | |
780 | * type. If repair can't establish that, the buffer will be left in memory | |
781 | * with NULL buffer ops. | |
1aff5696 DW |
782 | */ |
783 | int | |
75d02303 | 784 | xfs_buf_reverify( |
1aff5696 DW |
785 | struct xfs_buf *bp, |
786 | const struct xfs_buf_ops *ops) | |
787 | { | |
788 | ASSERT(bp->b_flags & XBF_DONE); | |
789 | ASSERT(bp->b_error == 0); | |
790 | ||
791 | if (!ops || bp->b_ops) | |
792 | return 0; | |
793 | ||
794 | bp->b_ops = ops; | |
795 | bp->b_ops->verify_read(bp); | |
796 | if (bp->b_error) | |
797 | bp->b_flags &= ~XBF_DONE; | |
798 | return bp->b_error; | |
799 | } | |
800 | ||
4ed8e27b | 801 | int |
6dde2707 DC |
802 | xfs_buf_read_map( |
803 | struct xfs_buftarg *target, | |
804 | struct xfs_buf_map *map, | |
805 | int nmaps, | |
c3f8fc73 | 806 | xfs_buf_flags_t flags, |
4ed8e27b | 807 | struct xfs_buf **bpp, |
cdbcf82b DW |
808 | const struct xfs_buf_ops *ops, |
809 | xfs_failaddr_t fa) | |
1da177e4 | 810 | { |
6dde2707 | 811 | struct xfs_buf *bp; |
3848b5f6 | 812 | int error; |
ce8e922c NS |
813 | |
814 | flags |= XBF_READ; | |
4ed8e27b | 815 | *bpp = NULL; |
ce8e922c | 816 | |
3848b5f6 DW |
817 | error = xfs_buf_get_map(target, map, nmaps, flags, &bp); |
818 | if (error) | |
4ed8e27b | 819 | return error; |
0b1b213f | 820 | |
1aff5696 DW |
821 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
822 | ||
823 | if (!(bp->b_flags & XBF_DONE)) { | |
4ed8e27b | 824 | /* Initiate the buffer read and wait. */ |
1aff5696 DW |
825 | XFS_STATS_INC(target->bt_mount, xb_get_read); |
826 | bp->b_ops = ops; | |
4ed8e27b DW |
827 | error = _xfs_buf_read(bp, flags); |
828 | ||
829 | /* Readahead iodone already dropped the buffer, so exit. */ | |
830 | if (flags & XBF_ASYNC) | |
831 | return 0; | |
832 | } else { | |
833 | /* Buffer already read; all we need to do is check it. */ | |
834 | error = xfs_buf_reverify(bp, ops); | |
835 | ||
836 | /* Readahead already finished; drop the buffer and exit. */ | |
837 | if (flags & XBF_ASYNC) { | |
838 | xfs_buf_relse(bp); | |
839 | return 0; | |
840 | } | |
841 | ||
842 | /* We do not want read in the flags */ | |
843 | bp->b_flags &= ~XBF_READ; | |
844 | ASSERT(bp->b_ops != NULL || ops == NULL); | |
1aff5696 DW |
845 | } |
846 | ||
4ed8e27b DW |
847 | /* |
848 | * If we've had a read error, then the contents of the buffer are | |
849 | * invalid and should not be used. To ensure that a followup read tries | |
850 | * to pull the buffer from disk again, we clear the XBF_DONE flag and | |
851 | * mark the buffer stale. This ensures that anyone who has a current | |
852 | * reference to the buffer will interpret it's contents correctly and | |
853 | * future cache lookups will also treat it as an empty, uninitialised | |
854 | * buffer. | |
855 | */ | |
856 | if (error) { | |
01728b44 DC |
857 | /* |
858 | * Check against log shutdown for error reporting because | |
859 | * metadata writeback may require a read first and we need to | |
860 | * report errors in metadata writeback until the log is shut | |
861 | * down. High level transaction read functions already check | |
862 | * against mount shutdown, anyway, so we only need to be | |
863 | * concerned about low level IO interactions here. | |
864 | */ | |
865 | if (!xlog_is_shutdown(target->bt_mount->m_log)) | |
cdbcf82b | 866 | xfs_buf_ioerror_alert(bp, fa); |
1aff5696 | 867 | |
4ed8e27b DW |
868 | bp->b_flags &= ~XBF_DONE; |
869 | xfs_buf_stale(bp); | |
1aff5696 | 870 | xfs_buf_relse(bp); |
4ed8e27b DW |
871 | |
872 | /* bad CRC means corrupted metadata */ | |
873 | if (error == -EFSBADCRC) | |
874 | error = -EFSCORRUPTED; | |
875 | return error; | |
1da177e4 LT |
876 | } |
877 | ||
4ed8e27b DW |
878 | *bpp = bp; |
879 | return 0; | |
1da177e4 LT |
880 | } |
881 | ||
1da177e4 | 882 | /* |
ce8e922c NS |
883 | * If we are not low on memory then do the readahead in a deadlock |
884 | * safe manner. | |
1da177e4 LT |
885 | */ |
886 | void | |
6dde2707 DC |
887 | xfs_buf_readahead_map( |
888 | struct xfs_buftarg *target, | |
889 | struct xfs_buf_map *map, | |
c3f8fc73 | 890 | int nmaps, |
1813dd64 | 891 | const struct xfs_buf_ops *ops) |
1da177e4 | 892 | { |
4ed8e27b DW |
893 | struct xfs_buf *bp; |
894 | ||
6dde2707 | 895 | xfs_buf_read_map(target, map, nmaps, |
cdbcf82b DW |
896 | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD, &bp, ops, |
897 | __this_address); | |
1da177e4 LT |
898 | } |
899 | ||
5adc94c2 DC |
900 | /* |
901 | * Read an uncached buffer from disk. Allocates and returns a locked | |
4c7f65ae DC |
902 | * buffer containing the disk contents or nothing. Uncached buffers always have |
903 | * a cache index of XFS_BUF_DADDR_NULL so we can easily determine if the buffer | |
904 | * is cached or uncached during fault diagnosis. | |
5adc94c2 | 905 | */ |
ba372674 | 906 | int |
5adc94c2 | 907 | xfs_buf_read_uncached( |
5adc94c2 DC |
908 | struct xfs_buftarg *target, |
909 | xfs_daddr_t daddr, | |
e70b73f8 | 910 | size_t numblks, |
b9b3fe15 | 911 | xfs_buf_flags_t flags, |
ba372674 | 912 | struct xfs_buf **bpp, |
1813dd64 | 913 | const struct xfs_buf_ops *ops) |
5adc94c2 | 914 | { |
eab4e633 | 915 | struct xfs_buf *bp; |
2842b6db | 916 | int error; |
5adc94c2 | 917 | |
ba372674 DC |
918 | *bpp = NULL; |
919 | ||
2842b6db DW |
920 | error = xfs_buf_get_uncached(target, numblks, flags, &bp); |
921 | if (error) | |
922 | return error; | |
5adc94c2 DC |
923 | |
924 | /* set up the buffer for a read IO */ | |
3e85c868 | 925 | ASSERT(bp->b_map_count == 1); |
4c7f65ae | 926 | bp->b_rhash_key = XFS_BUF_DADDR_NULL; |
3e85c868 | 927 | bp->b_maps[0].bm_bn = daddr; |
cbb7baab | 928 | bp->b_flags |= XBF_READ; |
1813dd64 | 929 | bp->b_ops = ops; |
5adc94c2 | 930 | |
6af88cda | 931 | xfs_buf_submit(bp); |
ba372674 | 932 | if (bp->b_error) { |
2842b6db | 933 | error = bp->b_error; |
83a0adc3 | 934 | xfs_buf_relse(bp); |
ba372674 | 935 | return error; |
83a0adc3 | 936 | } |
ba372674 DC |
937 | |
938 | *bpp = bp; | |
939 | return 0; | |
1da177e4 LT |
940 | } |
941 | ||
2842b6db | 942 | int |
686865f7 DC |
943 | xfs_buf_get_uncached( |
944 | struct xfs_buftarg *target, | |
e70b73f8 | 945 | size_t numblks, |
b9b3fe15 | 946 | xfs_buf_flags_t flags, |
2842b6db | 947 | struct xfs_buf **bpp) |
1da177e4 | 948 | { |
07b5c5ad | 949 | int error; |
3e85c868 DC |
950 | struct xfs_buf *bp; |
951 | DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks); | |
1da177e4 | 952 | |
2842b6db DW |
953 | *bpp = NULL; |
954 | ||
c891c30a | 955 | /* flags might contain irrelevant bits, pass only what we care about */ |
32dff5e5 DW |
956 | error = _xfs_buf_alloc(target, &map, 1, flags & XBF_NO_IOACCT, &bp); |
957 | if (error) | |
07b5c5ad | 958 | return error; |
1da177e4 | 959 | |
934d1076 | 960 | error = xfs_buf_alloc_pages(bp, flags); |
1fa40b01 | 961 | if (error) |
1da177e4 LT |
962 | goto fail_free_buf; |
963 | ||
611c9946 | 964 | error = _xfs_buf_map_pages(bp, 0); |
1fa40b01 | 965 | if (unlikely(error)) { |
4f10700a | 966 | xfs_warn(target->bt_mount, |
08e96e1a | 967 | "%s: failed to map pages", __func__); |
07b5c5ad | 968 | goto fail_free_buf; |
1fa40b01 | 969 | } |
1da177e4 | 970 | |
686865f7 | 971 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
2842b6db DW |
972 | *bpp = bp; |
973 | return 0; | |
1fa40b01 | 974 | |
07b5c5ad DC |
975 | fail_free_buf: |
976 | xfs_buf_free(bp); | |
2842b6db | 977 | return error; |
1da177e4 LT |
978 | } |
979 | ||
980 | /* | |
1da177e4 LT |
981 | * Increment reference count on buffer, to hold the buffer concurrently |
982 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
983 | * Must hold the buffer already to call this function. |
984 | */ | |
985 | void | |
ce8e922c | 986 | xfs_buf_hold( |
e8222613 | 987 | struct xfs_buf *bp) |
1da177e4 | 988 | { |
0b1b213f | 989 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 990 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
991 | } |
992 | ||
993 | /* | |
9c7504aa BF |
994 | * Release a hold on the specified buffer. If the hold count is 1, the buffer is |
995 | * placed on LRU or freed (depending on b_lru_ref). | |
1da177e4 LT |
996 | */ |
997 | void | |
ce8e922c | 998 | xfs_buf_rele( |
e8222613 | 999 | struct xfs_buf *bp) |
1da177e4 | 1000 | { |
74f75a0c | 1001 | struct xfs_perag *pag = bp->b_pag; |
9c7504aa BF |
1002 | bool release; |
1003 | bool freebuf = false; | |
1da177e4 | 1004 | |
0b1b213f | 1005 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 1006 | |
74f75a0c | 1007 | if (!pag) { |
430cbeb8 | 1008 | ASSERT(list_empty(&bp->b_lru)); |
9c7504aa BF |
1009 | if (atomic_dec_and_test(&bp->b_hold)) { |
1010 | xfs_buf_ioacct_dec(bp); | |
fad3aa1e | 1011 | xfs_buf_free(bp); |
9c7504aa | 1012 | } |
fad3aa1e NS |
1013 | return; |
1014 | } | |
1015 | ||
3790689f | 1016 | ASSERT(atomic_read(&bp->b_hold) > 0); |
a4082357 | 1017 | |
37fd1678 DC |
1018 | /* |
1019 | * We grab the b_lock here first to serialise racing xfs_buf_rele() | |
1020 | * calls. The pag_buf_lock being taken on the last reference only | |
1021 | * serialises against racing lookups in xfs_buf_find(). IOWs, the second | |
1022 | * to last reference we drop here is not serialised against the last | |
1023 | * reference until we take bp->b_lock. Hence if we don't grab b_lock | |
1024 | * first, the last "release" reference can win the race to the lock and | |
1025 | * free the buffer before the second-to-last reference is processed, | |
1026 | * leading to a use-after-free scenario. | |
1027 | */ | |
9c7504aa | 1028 | spin_lock(&bp->b_lock); |
37fd1678 | 1029 | release = atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock); |
9c7504aa BF |
1030 | if (!release) { |
1031 | /* | |
1032 | * Drop the in-flight state if the buffer is already on the LRU | |
1033 | * and it holds the only reference. This is racy because we | |
1034 | * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT | |
1035 | * ensures the decrement occurs only once per-buf. | |
1036 | */ | |
1037 | if ((atomic_read(&bp->b_hold) == 1) && !list_empty(&bp->b_lru)) | |
63db7c81 | 1038 | __xfs_buf_ioacct_dec(bp); |
9c7504aa BF |
1039 | goto out_unlock; |
1040 | } | |
1041 | ||
1042 | /* the last reference has been dropped ... */ | |
63db7c81 | 1043 | __xfs_buf_ioacct_dec(bp); |
9c7504aa BF |
1044 | if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) { |
1045 | /* | |
1046 | * If the buffer is added to the LRU take a new reference to the | |
1047 | * buffer for the LRU and clear the (now stale) dispose list | |
1048 | * state flag | |
1049 | */ | |
1050 | if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) { | |
1051 | bp->b_state &= ~XFS_BSTATE_DISPOSE; | |
1052 | atomic_inc(&bp->b_hold); | |
1da177e4 | 1053 | } |
9c7504aa BF |
1054 | spin_unlock(&pag->pag_buf_lock); |
1055 | } else { | |
1056 | /* | |
1057 | * most of the time buffers will already be removed from the | |
1058 | * LRU, so optimise that case by checking for the | |
1059 | * XFS_BSTATE_DISPOSE flag indicating the last list the buffer | |
1060 | * was on was the disposal list | |
1061 | */ | |
1062 | if (!(bp->b_state & XFS_BSTATE_DISPOSE)) { | |
1063 | list_lru_del(&bp->b_target->bt_lru, &bp->b_lru); | |
1064 | } else { | |
1065 | ASSERT(list_empty(&bp->b_lru)); | |
1da177e4 | 1066 | } |
9c7504aa BF |
1067 | |
1068 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); | |
6031e73a LS |
1069 | rhashtable_remove_fast(&pag->pag_buf_hash, &bp->b_rhash_head, |
1070 | xfs_buf_hash_params); | |
9c7504aa BF |
1071 | spin_unlock(&pag->pag_buf_lock); |
1072 | xfs_perag_put(pag); | |
1073 | freebuf = true; | |
1da177e4 | 1074 | } |
9c7504aa BF |
1075 | |
1076 | out_unlock: | |
1077 | spin_unlock(&bp->b_lock); | |
1078 | ||
1079 | if (freebuf) | |
1080 | xfs_buf_free(bp); | |
1da177e4 LT |
1081 | } |
1082 | ||
1083 | ||
1084 | /* | |
0e6e847f | 1085 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
1086 | * |
1087 | * If we come across a stale, pinned, locked buffer, we know that we are | |
1088 | * being asked to lock a buffer that has been reallocated. Because it is | |
1089 | * pinned, we know that the log has not been pushed to disk and hence it | |
1090 | * will still be locked. Rather than continuing to have trylock attempts | |
1091 | * fail until someone else pushes the log, push it ourselves before | |
1092 | * returning. This means that the xfsaild will not get stuck trying | |
1093 | * to push on stale inode buffers. | |
1da177e4 LT |
1094 | */ |
1095 | int | |
0c842ad4 CH |
1096 | xfs_buf_trylock( |
1097 | struct xfs_buf *bp) | |
1da177e4 LT |
1098 | { |
1099 | int locked; | |
1100 | ||
ce8e922c | 1101 | locked = down_trylock(&bp->b_sema) == 0; |
fa6c668d | 1102 | if (locked) |
479c6412 | 1103 | trace_xfs_buf_trylock(bp, _RET_IP_); |
fa6c668d | 1104 | else |
479c6412 | 1105 | trace_xfs_buf_trylock_fail(bp, _RET_IP_); |
0c842ad4 | 1106 | return locked; |
1da177e4 | 1107 | } |
1da177e4 LT |
1108 | |
1109 | /* | |
0e6e847f | 1110 | * Lock a buffer object. |
ed3b4d6c DC |
1111 | * |
1112 | * If we come across a stale, pinned, locked buffer, we know that we | |
1113 | * are being asked to lock a buffer that has been reallocated. Because | |
1114 | * it is pinned, we know that the log has not been pushed to disk and | |
1115 | * hence it will still be locked. Rather than sleeping until someone | |
1116 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 1117 | */ |
ce8e922c NS |
1118 | void |
1119 | xfs_buf_lock( | |
0c842ad4 | 1120 | struct xfs_buf *bp) |
1da177e4 | 1121 | { |
0b1b213f CH |
1122 | trace_xfs_buf_lock(bp, _RET_IP_); |
1123 | ||
ed3b4d6c | 1124 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
dbd329f1 | 1125 | xfs_log_force(bp->b_mount, 0); |
ce8e922c | 1126 | down(&bp->b_sema); |
0b1b213f CH |
1127 | |
1128 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
1129 | } |
1130 | ||
1da177e4 | 1131 | void |
ce8e922c | 1132 | xfs_buf_unlock( |
0c842ad4 | 1133 | struct xfs_buf *bp) |
1da177e4 | 1134 | { |
20e8a063 BF |
1135 | ASSERT(xfs_buf_islocked(bp)); |
1136 | ||
ce8e922c | 1137 | up(&bp->b_sema); |
0b1b213f | 1138 | trace_xfs_buf_unlock(bp, _RET_IP_); |
1da177e4 LT |
1139 | } |
1140 | ||
ce8e922c NS |
1141 | STATIC void |
1142 | xfs_buf_wait_unpin( | |
e8222613 | 1143 | struct xfs_buf *bp) |
1da177e4 LT |
1144 | { |
1145 | DECLARE_WAITQUEUE (wait, current); | |
1146 | ||
ce8e922c | 1147 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
1148 | return; |
1149 | ||
ce8e922c | 1150 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1151 | for (;;) { |
1152 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 1153 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 1154 | break; |
7eaceacc | 1155 | io_schedule(); |
1da177e4 | 1156 | } |
ce8e922c | 1157 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1158 | set_current_state(TASK_RUNNING); |
1159 | } | |
1160 | ||
f58d0ea9 CH |
1161 | static void |
1162 | xfs_buf_ioerror_alert_ratelimited( | |
664ffb8a CH |
1163 | struct xfs_buf *bp) |
1164 | { | |
664ffb8a CH |
1165 | static unsigned long lasttime; |
1166 | static struct xfs_buftarg *lasttarg; | |
1167 | ||
664ffb8a CH |
1168 | if (bp->b_target != lasttarg || |
1169 | time_after(jiffies, (lasttime + 5*HZ))) { | |
1170 | lasttime = jiffies; | |
1171 | xfs_buf_ioerror_alert(bp, __this_address); | |
1172 | } | |
1173 | lasttarg = bp->b_target; | |
664ffb8a CH |
1174 | } |
1175 | ||
664ffb8a CH |
1176 | /* |
1177 | * Account for this latest trip around the retry handler, and decide if | |
1178 | * we've failed enough times to constitute a permanent failure. | |
1179 | */ | |
1180 | static bool | |
1181 | xfs_buf_ioerror_permanent( | |
1182 | struct xfs_buf *bp, | |
1183 | struct xfs_error_cfg *cfg) | |
1184 | { | |
1185 | struct xfs_mount *mp = bp->b_mount; | |
1186 | ||
1187 | if (cfg->max_retries != XFS_ERR_RETRY_FOREVER && | |
1188 | ++bp->b_retries > cfg->max_retries) | |
1189 | return true; | |
1190 | if (cfg->retry_timeout != XFS_ERR_RETRY_FOREVER && | |
1191 | time_after(jiffies, cfg->retry_timeout + bp->b_first_retry_time)) | |
1192 | return true; | |
1193 | ||
1194 | /* At unmount we may treat errors differently */ | |
2e973b2c | 1195 | if (xfs_is_unmounting(mp) && mp->m_fail_unmount) |
664ffb8a CH |
1196 | return true; |
1197 | ||
1198 | return false; | |
1199 | } | |
1200 | ||
1201 | /* | |
1202 | * On a sync write or shutdown we just want to stale the buffer and let the | |
1203 | * caller handle the error in bp->b_error appropriately. | |
1204 | * | |
1205 | * If the write was asynchronous then no one will be looking for the error. If | |
1206 | * this is the first failure of this type, clear the error state and write the | |
1207 | * buffer out again. This means we always retry an async write failure at least | |
1208 | * once, but we also need to set the buffer up to behave correctly now for | |
1209 | * repeated failures. | |
1210 | * | |
1211 | * If we get repeated async write failures, then we take action according to the | |
1212 | * error configuration we have been set up to use. | |
1213 | * | |
70796c6b CH |
1214 | * Returns true if this function took care of error handling and the caller must |
1215 | * not touch the buffer again. Return false if the caller should proceed with | |
1216 | * normal I/O completion handling. | |
664ffb8a | 1217 | */ |
70796c6b CH |
1218 | static bool |
1219 | xfs_buf_ioend_handle_error( | |
664ffb8a CH |
1220 | struct xfs_buf *bp) |
1221 | { | |
1222 | struct xfs_mount *mp = bp->b_mount; | |
1223 | struct xfs_error_cfg *cfg; | |
1224 | ||
f58d0ea9 | 1225 | /* |
01728b44 DC |
1226 | * If we've already shutdown the journal because of I/O errors, there's |
1227 | * no point in giving this a retry. | |
f58d0ea9 | 1228 | */ |
01728b44 | 1229 | if (xlog_is_shutdown(mp->m_log)) |
f58d0ea9 CH |
1230 | goto out_stale; |
1231 | ||
1232 | xfs_buf_ioerror_alert_ratelimited(bp); | |
1233 | ||
22c10589 CH |
1234 | /* |
1235 | * We're not going to bother about retrying this during recovery. | |
1236 | * One strike! | |
1237 | */ | |
1238 | if (bp->b_flags & _XBF_LOGRECOVERY) { | |
1239 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1240 | return false; | |
1241 | } | |
1242 | ||
f58d0ea9 CH |
1243 | /* |
1244 | * Synchronous writes will have callers process the error. | |
1245 | */ | |
1246 | if (!(bp->b_flags & XBF_ASYNC)) | |
664ffb8a CH |
1247 | goto out_stale; |
1248 | ||
1249 | trace_xfs_buf_iodone_async(bp, _RET_IP_); | |
1250 | ||
1251 | cfg = xfs_error_get_cfg(mp, XFS_ERR_METADATA, bp->b_error); | |
3cc49884 CH |
1252 | if (bp->b_last_error != bp->b_error || |
1253 | !(bp->b_flags & (XBF_STALE | XBF_WRITE_FAIL))) { | |
1254 | bp->b_last_error = bp->b_error; | |
1255 | if (cfg->retry_timeout != XFS_ERR_RETRY_FOREVER && | |
1256 | !bp->b_first_retry_time) | |
1257 | bp->b_first_retry_time = jiffies; | |
1258 | goto resubmit; | |
664ffb8a CH |
1259 | } |
1260 | ||
1261 | /* | |
1262 | * Permanent error - we need to trigger a shutdown if we haven't already | |
1263 | * to indicate that inconsistency will result from this action. | |
1264 | */ | |
1265 | if (xfs_buf_ioerror_permanent(bp, cfg)) { | |
1266 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1267 | goto out_stale; | |
1268 | } | |
1269 | ||
1270 | /* Still considered a transient error. Caller will schedule retries. */ | |
844c9358 CH |
1271 | if (bp->b_flags & _XBF_INODES) |
1272 | xfs_buf_inode_io_fail(bp); | |
1273 | else if (bp->b_flags & _XBF_DQUOTS) | |
1274 | xfs_buf_dquot_io_fail(bp); | |
1275 | else | |
1276 | ASSERT(list_empty(&bp->b_li_list)); | |
1277 | xfs_buf_ioerror(bp, 0); | |
1278 | xfs_buf_relse(bp); | |
70796c6b | 1279 | return true; |
664ffb8a | 1280 | |
3cc49884 CH |
1281 | resubmit: |
1282 | xfs_buf_ioerror(bp, 0); | |
55b7d711 | 1283 | bp->b_flags |= (XBF_DONE | XBF_WRITE_FAIL); |
3cc49884 | 1284 | xfs_buf_submit(bp); |
70796c6b | 1285 | return true; |
664ffb8a CH |
1286 | out_stale: |
1287 | xfs_buf_stale(bp); | |
1288 | bp->b_flags |= XBF_DONE; | |
55b7d711 | 1289 | bp->b_flags &= ~XBF_WRITE; |
664ffb8a | 1290 | trace_xfs_buf_error_relse(bp, _RET_IP_); |
70796c6b | 1291 | return false; |
664ffb8a | 1292 | } |
1da177e4 | 1293 | |
76b2d323 | 1294 | static void |
e8aaba9a DC |
1295 | xfs_buf_ioend( |
1296 | struct xfs_buf *bp) | |
1da177e4 | 1297 | { |
e8aaba9a | 1298 | trace_xfs_buf_iodone(bp, _RET_IP_); |
1813dd64 | 1299 | |
61be9c52 DC |
1300 | /* |
1301 | * Pull in IO completion errors now. We are guaranteed to be running | |
1302 | * single threaded, so we don't need the lock to read b_io_error. | |
1303 | */ | |
1304 | if (!bp->b_error && bp->b_io_error) | |
1305 | xfs_buf_ioerror(bp, bp->b_io_error); | |
1306 | ||
55b7d711 | 1307 | if (bp->b_flags & XBF_READ) { |
b01d1461 DC |
1308 | if (!bp->b_error && bp->b_ops) |
1309 | bp->b_ops->verify_read(bp); | |
1310 | if (!bp->b_error) | |
1311 | bp->b_flags |= XBF_DONE; | |
23fb5a93 CH |
1312 | } else { |
1313 | if (!bp->b_error) { | |
1314 | bp->b_flags &= ~XBF_WRITE_FAIL; | |
1315 | bp->b_flags |= XBF_DONE; | |
1316 | } | |
f593bf14 | 1317 | |
70796c6b | 1318 | if (unlikely(bp->b_error) && xfs_buf_ioend_handle_error(bp)) |
664ffb8a | 1319 | return; |
664ffb8a CH |
1320 | |
1321 | /* clear the retry state */ | |
1322 | bp->b_last_error = 0; | |
1323 | bp->b_retries = 0; | |
1324 | bp->b_first_retry_time = 0; | |
1325 | ||
1326 | /* | |
1327 | * Note that for things like remote attribute buffers, there may | |
1328 | * not be a buffer log item here, so processing the buffer log | |
1329 | * item must remain optional. | |
1330 | */ | |
1331 | if (bp->b_log_item) | |
1332 | xfs_buf_item_done(bp); | |
1333 | ||
23fb5a93 CH |
1334 | if (bp->b_flags & _XBF_INODES) |
1335 | xfs_buf_inode_iodone(bp); | |
1336 | else if (bp->b_flags & _XBF_DQUOTS) | |
1337 | xfs_buf_dquot_iodone(bp); | |
22c10589 | 1338 | |
0c7e5afb | 1339 | } |
6a7584b1 | 1340 | |
22c10589 CH |
1341 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD | |
1342 | _XBF_LOGRECOVERY); | |
55b7d711 | 1343 | |
6a7584b1 CH |
1344 | if (bp->b_flags & XBF_ASYNC) |
1345 | xfs_buf_relse(bp); | |
1346 | else | |
1347 | complete(&bp->b_iowait); | |
1da177e4 LT |
1348 | } |
1349 | ||
e8aaba9a DC |
1350 | static void |
1351 | xfs_buf_ioend_work( | |
1352 | struct work_struct *work) | |
1da177e4 | 1353 | { |
e8aaba9a | 1354 | struct xfs_buf *bp = |
e8222613 | 1355 | container_of(work, struct xfs_buf, b_ioend_work); |
0b1b213f | 1356 | |
e8aaba9a DC |
1357 | xfs_buf_ioend(bp); |
1358 | } | |
1da177e4 | 1359 | |
211fe1a4 | 1360 | static void |
e8aaba9a DC |
1361 | xfs_buf_ioend_async( |
1362 | struct xfs_buf *bp) | |
1363 | { | |
b29c70f5 | 1364 | INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work); |
dbd329f1 | 1365 | queue_work(bp->b_mount->m_buf_workqueue, &bp->b_ioend_work); |
1da177e4 LT |
1366 | } |
1367 | ||
1da177e4 | 1368 | void |
31ca03c9 | 1369 | __xfs_buf_ioerror( |
e8222613 | 1370 | struct xfs_buf *bp, |
31ca03c9 DW |
1371 | int error, |
1372 | xfs_failaddr_t failaddr) | |
1da177e4 | 1373 | { |
2451337d DC |
1374 | ASSERT(error <= 0 && error >= -1000); |
1375 | bp->b_error = error; | |
31ca03c9 | 1376 | trace_xfs_buf_ioerror(bp, error, failaddr); |
1da177e4 LT |
1377 | } |
1378 | ||
901796af CH |
1379 | void |
1380 | xfs_buf_ioerror_alert( | |
1381 | struct xfs_buf *bp, | |
cdbcf82b | 1382 | xfs_failaddr_t func) |
901796af | 1383 | { |
f9bccfcc BF |
1384 | xfs_buf_alert_ratelimited(bp, "XFS: metadata IO error", |
1385 | "metadata I/O error in \"%pS\" at daddr 0x%llx len %d error %d", | |
04fcad80 | 1386 | func, (uint64_t)xfs_buf_daddr(bp), |
f9bccfcc | 1387 | bp->b_length, -bp->b_error); |
901796af CH |
1388 | } |
1389 | ||
54b3b1f6 BF |
1390 | /* |
1391 | * To simulate an I/O failure, the buffer must be locked and held with at least | |
1392 | * three references. The LRU reference is dropped by the stale call. The buf | |
1393 | * item reference is dropped via ioend processing. The third reference is owned | |
1394 | * by the caller and is dropped on I/O completion if the buffer is XBF_ASYNC. | |
1395 | */ | |
1396 | void | |
1397 | xfs_buf_ioend_fail( | |
1398 | struct xfs_buf *bp) | |
1399 | { | |
1400 | bp->b_flags &= ~XBF_DONE; | |
1401 | xfs_buf_stale(bp); | |
1402 | xfs_buf_ioerror(bp, -EIO); | |
1403 | xfs_buf_ioend(bp); | |
901796af CH |
1404 | } |
1405 | ||
a2dcf5df CH |
1406 | int |
1407 | xfs_bwrite( | |
1408 | struct xfs_buf *bp) | |
1409 | { | |
1410 | int error; | |
1411 | ||
1412 | ASSERT(xfs_buf_islocked(bp)); | |
1413 | ||
1414 | bp->b_flags |= XBF_WRITE; | |
27187754 | 1415 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q | |
b6983e80 | 1416 | XBF_DONE); |
a2dcf5df | 1417 | |
6af88cda | 1418 | error = xfs_buf_submit(bp); |
dbd329f1 CH |
1419 | if (error) |
1420 | xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR); | |
a2dcf5df CH |
1421 | return error; |
1422 | } | |
1423 | ||
9bdd9bd6 | 1424 | static void |
ce8e922c | 1425 | xfs_buf_bio_end_io( |
4246a0b6 | 1426 | struct bio *bio) |
1da177e4 | 1427 | { |
9bdd9bd6 | 1428 | struct xfs_buf *bp = (struct xfs_buf *)bio->bi_private; |
1da177e4 | 1429 | |
7376d745 BF |
1430 | if (!bio->bi_status && |
1431 | (bp->b_flags & XBF_WRITE) && (bp->b_flags & XBF_ASYNC) && | |
43dc0aa8 | 1432 | XFS_TEST_ERROR(false, bp->b_mount, XFS_ERRTAG_BUF_IOERROR)) |
7376d745 | 1433 | bio->bi_status = BLK_STS_IOERR; |
1da177e4 | 1434 | |
37eb17e6 DC |
1435 | /* |
1436 | * don't overwrite existing errors - otherwise we can lose errors on | |
1437 | * buffers that require multiple bios to complete. | |
1438 | */ | |
4e4cbee9 CH |
1439 | if (bio->bi_status) { |
1440 | int error = blk_status_to_errno(bio->bi_status); | |
1441 | ||
1442 | cmpxchg(&bp->b_io_error, 0, error); | |
1443 | } | |
1da177e4 | 1444 | |
37eb17e6 | 1445 | if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
73c77e2c JB |
1446 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); |
1447 | ||
e8aaba9a DC |
1448 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
1449 | xfs_buf_ioend_async(bp); | |
1da177e4 | 1450 | bio_put(bio); |
1da177e4 LT |
1451 | } |
1452 | ||
3e85c868 DC |
1453 | static void |
1454 | xfs_buf_ioapply_map( | |
1455 | struct xfs_buf *bp, | |
1456 | int map, | |
1457 | int *buf_offset, | |
1458 | int *count, | |
d03025ae | 1459 | blk_opf_t op) |
1da177e4 | 1460 | { |
3e85c868 | 1461 | int page_index; |
5f7136db | 1462 | unsigned int total_nr_pages = bp->b_page_count; |
3e85c868 DC |
1463 | int nr_pages; |
1464 | struct bio *bio; | |
1465 | sector_t sector = bp->b_maps[map].bm_bn; | |
1466 | int size; | |
1467 | int offset; | |
1da177e4 | 1468 | |
3e85c868 DC |
1469 | /* skip the pages in the buffer before the start offset */ |
1470 | page_index = 0; | |
1471 | offset = *buf_offset; | |
1472 | while (offset >= PAGE_SIZE) { | |
1473 | page_index++; | |
1474 | offset -= PAGE_SIZE; | |
f538d4da CH |
1475 | } |
1476 | ||
3e85c868 DC |
1477 | /* |
1478 | * Limit the IO size to the length of the current vector, and update the | |
1479 | * remaining IO count for the next time around. | |
1480 | */ | |
1481 | size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count); | |
1482 | *count -= size; | |
1483 | *buf_offset += size; | |
34951f5c | 1484 | |
1da177e4 | 1485 | next_chunk: |
ce8e922c | 1486 | atomic_inc(&bp->b_io_remaining); |
5f7136db | 1487 | nr_pages = bio_max_segs(total_nr_pages); |
1da177e4 | 1488 | |
07888c66 | 1489 | bio = bio_alloc(bp->b_target->bt_bdev, nr_pages, op, GFP_NOIO); |
4f024f37 | 1490 | bio->bi_iter.bi_sector = sector; |
ce8e922c NS |
1491 | bio->bi_end_io = xfs_buf_bio_end_io; |
1492 | bio->bi_private = bp; | |
0e6e847f | 1493 | |
3e85c868 | 1494 | for (; size && nr_pages; nr_pages--, page_index++) { |
0e6e847f | 1495 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1496 | |
1497 | if (nbytes > size) | |
1498 | nbytes = size; | |
1499 | ||
3e85c868 DC |
1500 | rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes, |
1501 | offset); | |
ce8e922c | 1502 | if (rbytes < nbytes) |
1da177e4 LT |
1503 | break; |
1504 | ||
1505 | offset = 0; | |
aa0e8833 | 1506 | sector += BTOBB(nbytes); |
1da177e4 LT |
1507 | size -= nbytes; |
1508 | total_nr_pages--; | |
1509 | } | |
1510 | ||
4f024f37 | 1511 | if (likely(bio->bi_iter.bi_size)) { |
73c77e2c JB |
1512 | if (xfs_buf_is_vmapped(bp)) { |
1513 | flush_kernel_vmap_range(bp->b_addr, | |
1514 | xfs_buf_vmap_len(bp)); | |
1515 | } | |
4e49ea4a | 1516 | submit_bio(bio); |
1da177e4 LT |
1517 | if (size) |
1518 | goto next_chunk; | |
1519 | } else { | |
37eb17e6 DC |
1520 | /* |
1521 | * This is guaranteed not to be the last io reference count | |
595bff75 | 1522 | * because the caller (xfs_buf_submit) holds a count itself. |
37eb17e6 DC |
1523 | */ |
1524 | atomic_dec(&bp->b_io_remaining); | |
2451337d | 1525 | xfs_buf_ioerror(bp, -EIO); |
ec53d1db | 1526 | bio_put(bio); |
1da177e4 | 1527 | } |
3e85c868 DC |
1528 | |
1529 | } | |
1530 | ||
1531 | STATIC void | |
1532 | _xfs_buf_ioapply( | |
1533 | struct xfs_buf *bp) | |
1534 | { | |
1535 | struct blk_plug plug; | |
d03025ae | 1536 | blk_opf_t op; |
3e85c868 DC |
1537 | int offset; |
1538 | int size; | |
1539 | int i; | |
1540 | ||
c163f9a1 DC |
1541 | /* |
1542 | * Make sure we capture only current IO errors rather than stale errors | |
1543 | * left over from previous use of the buffer (e.g. failed readahead). | |
1544 | */ | |
1545 | bp->b_error = 0; | |
1546 | ||
3e85c868 | 1547 | if (bp->b_flags & XBF_WRITE) { |
50bfcd0c | 1548 | op = REQ_OP_WRITE; |
1813dd64 DC |
1549 | |
1550 | /* | |
1551 | * Run the write verifier callback function if it exists. If | |
1552 | * this function fails it will mark the buffer with an error and | |
1553 | * the IO should not be dispatched. | |
1554 | */ | |
1555 | if (bp->b_ops) { | |
1556 | bp->b_ops->verify_write(bp); | |
1557 | if (bp->b_error) { | |
dbd329f1 | 1558 | xfs_force_shutdown(bp->b_mount, |
1813dd64 DC |
1559 | SHUTDOWN_CORRUPT_INCORE); |
1560 | return; | |
1561 | } | |
4c7f65ae | 1562 | } else if (bp->b_rhash_key != XFS_BUF_DADDR_NULL) { |
dbd329f1 | 1563 | struct xfs_mount *mp = bp->b_mount; |
400b9d88 DC |
1564 | |
1565 | /* | |
1566 | * non-crc filesystems don't attach verifiers during | |
1567 | * log recovery, so don't warn for such filesystems. | |
1568 | */ | |
38c26bfd | 1569 | if (xfs_has_crc(mp)) { |
400b9d88 | 1570 | xfs_warn(mp, |
c219b015 | 1571 | "%s: no buf ops on daddr 0x%llx len %d", |
4c7f65ae DC |
1572 | __func__, xfs_buf_daddr(bp), |
1573 | bp->b_length); | |
9c712a13 DW |
1574 | xfs_hex_dump(bp->b_addr, |
1575 | XFS_CORRUPTION_DUMP_LEN); | |
400b9d88 DC |
1576 | dump_stack(); |
1577 | } | |
1813dd64 | 1578 | } |
3e85c868 | 1579 | } else { |
50bfcd0c | 1580 | op = REQ_OP_READ; |
2123ef85 CH |
1581 | if (bp->b_flags & XBF_READ_AHEAD) |
1582 | op |= REQ_RAHEAD; | |
3e85c868 DC |
1583 | } |
1584 | ||
1585 | /* we only use the buffer cache for meta-data */ | |
2123ef85 | 1586 | op |= REQ_META; |
3e85c868 DC |
1587 | |
1588 | /* | |
1589 | * Walk all the vectors issuing IO on them. Set up the initial offset | |
1590 | * into the buffer and the desired IO size before we start - | |
1591 | * _xfs_buf_ioapply_vec() will modify them appropriately for each | |
1592 | * subsequent call. | |
1593 | */ | |
1594 | offset = bp->b_offset; | |
8124b9b6 | 1595 | size = BBTOB(bp->b_length); |
3e85c868 DC |
1596 | blk_start_plug(&plug); |
1597 | for (i = 0; i < bp->b_map_count; i++) { | |
2123ef85 | 1598 | xfs_buf_ioapply_map(bp, i, &offset, &size, op); |
3e85c868 DC |
1599 | if (bp->b_error) |
1600 | break; | |
1601 | if (size <= 0) | |
1602 | break; /* all done */ | |
1603 | } | |
1604 | blk_finish_plug(&plug); | |
1da177e4 LT |
1605 | } |
1606 | ||
595bff75 | 1607 | /* |
bb00b6f1 | 1608 | * Wait for I/O completion of a sync buffer and return the I/O error code. |
595bff75 | 1609 | */ |
eaebb515 | 1610 | static int |
bb00b6f1 | 1611 | xfs_buf_iowait( |
595bff75 | 1612 | struct xfs_buf *bp) |
1da177e4 | 1613 | { |
bb00b6f1 BF |
1614 | ASSERT(!(bp->b_flags & XBF_ASYNC)); |
1615 | ||
1616 | trace_xfs_buf_iowait(bp, _RET_IP_); | |
1617 | wait_for_completion(&bp->b_iowait); | |
1618 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
1619 | ||
1620 | return bp->b_error; | |
1621 | } | |
1622 | ||
1623 | /* | |
1624 | * Buffer I/O submission path, read or write. Asynchronous submission transfers | |
1625 | * the buffer lock ownership and the current reference to the IO. It is not | |
1626 | * safe to reference the buffer after a call to this function unless the caller | |
1627 | * holds an additional reference itself. | |
1628 | */ | |
26e32875 | 1629 | static int |
bb00b6f1 BF |
1630 | __xfs_buf_submit( |
1631 | struct xfs_buf *bp, | |
1632 | bool wait) | |
1633 | { | |
1634 | int error = 0; | |
1635 | ||
595bff75 | 1636 | trace_xfs_buf_submit(bp, _RET_IP_); |
1da177e4 | 1637 | |
43ff2122 | 1638 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
595bff75 | 1639 | |
01728b44 DC |
1640 | /* |
1641 | * On log shutdown we stale and complete the buffer immediately. We can | |
1642 | * be called to read the superblock before the log has been set up, so | |
1643 | * be careful checking the log state. | |
1644 | * | |
1645 | * Checking the mount shutdown state here can result in the log tail | |
1646 | * moving inappropriately on disk as the log may not yet be shut down. | |
1647 | * i.e. failing this buffer on mount shutdown can remove it from the AIL | |
1648 | * and move the tail of the log forwards without having written this | |
1649 | * buffer to disk. This corrupts the log tail state in memory, and | |
1650 | * because the log may not be shut down yet, it can then be propagated | |
1651 | * to disk before the log is shutdown. Hence we check log shutdown | |
1652 | * state here rather than mount state to avoid corrupting the log tail | |
1653 | * on shutdown. | |
1654 | */ | |
1655 | if (bp->b_mount->m_log && | |
1656 | xlog_is_shutdown(bp->b_mount->m_log)) { | |
54b3b1f6 | 1657 | xfs_buf_ioend_fail(bp); |
eaebb515 | 1658 | return -EIO; |
595bff75 | 1659 | } |
1da177e4 | 1660 | |
bb00b6f1 BF |
1661 | /* |
1662 | * Grab a reference so the buffer does not go away underneath us. For | |
1663 | * async buffers, I/O completion drops the callers reference, which | |
1664 | * could occur before submission returns. | |
1665 | */ | |
1666 | xfs_buf_hold(bp); | |
1667 | ||
375ec69d | 1668 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1669 | xfs_buf_wait_unpin(bp); |
e11bb805 | 1670 | |
61be9c52 DC |
1671 | /* clear the internal error state to avoid spurious errors */ |
1672 | bp->b_io_error = 0; | |
1673 | ||
8d6c1210 | 1674 | /* |
e11bb805 DC |
1675 | * Set the count to 1 initially, this will stop an I/O completion |
1676 | * callout which happens before we have started all the I/O from calling | |
1677 | * xfs_buf_ioend too early. | |
1da177e4 | 1678 | */ |
ce8e922c | 1679 | atomic_set(&bp->b_io_remaining, 1); |
eaebb515 BF |
1680 | if (bp->b_flags & XBF_ASYNC) |
1681 | xfs_buf_ioacct_inc(bp); | |
ce8e922c | 1682 | _xfs_buf_ioapply(bp); |
e11bb805 | 1683 | |
8d6c1210 | 1684 | /* |
595bff75 DC |
1685 | * If _xfs_buf_ioapply failed, we can get back here with only the IO |
1686 | * reference we took above. If we drop it to zero, run completion so | |
1687 | * that we don't return to the caller with completion still pending. | |
8d6c1210 | 1688 | */ |
e8aaba9a | 1689 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
eaebb515 | 1690 | if (bp->b_error || !(bp->b_flags & XBF_ASYNC)) |
e8aaba9a DC |
1691 | xfs_buf_ioend(bp); |
1692 | else | |
1693 | xfs_buf_ioend_async(bp); | |
1694 | } | |
1da177e4 | 1695 | |
6af88cda BF |
1696 | if (wait) |
1697 | error = xfs_buf_iowait(bp); | |
bb00b6f1 | 1698 | |
595bff75 | 1699 | /* |
6af88cda BF |
1700 | * Release the hold that keeps the buffer referenced for the entire |
1701 | * I/O. Note that if the buffer is async, it is not safe to reference | |
1702 | * after this release. | |
595bff75 DC |
1703 | */ |
1704 | xfs_buf_rele(bp); | |
1705 | return error; | |
1da177e4 LT |
1706 | } |
1707 | ||
88ee2df7 | 1708 | void * |
ce8e922c | 1709 | xfs_buf_offset( |
88ee2df7 | 1710 | struct xfs_buf *bp, |
1da177e4 LT |
1711 | size_t offset) |
1712 | { | |
1713 | struct page *page; | |
1714 | ||
611c9946 | 1715 | if (bp->b_addr) |
62926044 | 1716 | return bp->b_addr + offset; |
1da177e4 | 1717 | |
0e6e847f | 1718 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
88ee2df7 | 1719 | return page_address(page) + (offset & (PAGE_SIZE-1)); |
1da177e4 LT |
1720 | } |
1721 | ||
1da177e4 | 1722 | void |
f9a196ee CH |
1723 | xfs_buf_zero( |
1724 | struct xfs_buf *bp, | |
1725 | size_t boff, | |
1726 | size_t bsize) | |
1da177e4 | 1727 | { |
795cac72 | 1728 | size_t bend; |
1da177e4 LT |
1729 | |
1730 | bend = boff + bsize; | |
1731 | while (boff < bend) { | |
795cac72 DC |
1732 | struct page *page; |
1733 | int page_index, page_offset, csize; | |
1734 | ||
1735 | page_index = (boff + bp->b_offset) >> PAGE_SHIFT; | |
1736 | page_offset = (boff + bp->b_offset) & ~PAGE_MASK; | |
1737 | page = bp->b_pages[page_index]; | |
1738 | csize = min_t(size_t, PAGE_SIZE - page_offset, | |
8124b9b6 | 1739 | BBTOB(bp->b_length) - boff); |
1da177e4 | 1740 | |
795cac72 | 1741 | ASSERT((csize + page_offset) <= PAGE_SIZE); |
1da177e4 | 1742 | |
f9a196ee | 1743 | memset(page_address(page) + page_offset, 0, csize); |
1da177e4 LT |
1744 | |
1745 | boff += csize; | |
1da177e4 LT |
1746 | } |
1747 | } | |
1748 | ||
8d57c216 DW |
1749 | /* |
1750 | * Log a message about and stale a buffer that a caller has decided is corrupt. | |
1751 | * | |
1752 | * This function should be called for the kinds of metadata corruption that | |
1753 | * cannot be detect from a verifier, such as incorrect inter-block relationship | |
1754 | * data. Do /not/ call this function from a verifier function. | |
1755 | * | |
1756 | * The buffer must be XBF_DONE prior to the call. Afterwards, the buffer will | |
1757 | * be marked stale, but b_error will not be set. The caller is responsible for | |
1758 | * releasing the buffer or fixing it. | |
1759 | */ | |
1760 | void | |
1761 | __xfs_buf_mark_corrupt( | |
1762 | struct xfs_buf *bp, | |
1763 | xfs_failaddr_t fa) | |
1764 | { | |
1765 | ASSERT(bp->b_flags & XBF_DONE); | |
1766 | ||
e83cf875 | 1767 | xfs_buf_corruption_error(bp, fa); |
8d57c216 DW |
1768 | xfs_buf_stale(bp); |
1769 | } | |
1770 | ||
1da177e4 | 1771 | /* |
ce8e922c | 1772 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1773 | */ |
1774 | ||
1775 | /* | |
430cbeb8 DC |
1776 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1777 | * returned. These buffers will have an elevated hold count, so wait on those | |
1778 | * while freeing all the buffers only held by the LRU. | |
1da177e4 | 1779 | */ |
e80dfa19 | 1780 | static enum lru_status |
10fb9ac1 | 1781 | xfs_buftarg_drain_rele( |
e80dfa19 | 1782 | struct list_head *item, |
3f97b163 | 1783 | struct list_lru_one *lru, |
e80dfa19 DC |
1784 | spinlock_t *lru_lock, |
1785 | void *arg) | |
1786 | ||
1da177e4 | 1787 | { |
e80dfa19 | 1788 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); |
a4082357 | 1789 | struct list_head *dispose = arg; |
430cbeb8 | 1790 | |
e80dfa19 | 1791 | if (atomic_read(&bp->b_hold) > 1) { |
a4082357 | 1792 | /* need to wait, so skip it this pass */ |
10fb9ac1 | 1793 | trace_xfs_buf_drain_buftarg(bp, _RET_IP_); |
a4082357 | 1794 | return LRU_SKIP; |
1da177e4 | 1795 | } |
a4082357 DC |
1796 | if (!spin_trylock(&bp->b_lock)) |
1797 | return LRU_SKIP; | |
e80dfa19 | 1798 | |
a4082357 DC |
1799 | /* |
1800 | * clear the LRU reference count so the buffer doesn't get | |
1801 | * ignored in xfs_buf_rele(). | |
1802 | */ | |
1803 | atomic_set(&bp->b_lru_ref, 0); | |
1804 | bp->b_state |= XFS_BSTATE_DISPOSE; | |
3f97b163 | 1805 | list_lru_isolate_move(lru, item, dispose); |
a4082357 DC |
1806 | spin_unlock(&bp->b_lock); |
1807 | return LRU_REMOVED; | |
1da177e4 LT |
1808 | } |
1809 | ||
8321ddb2 BF |
1810 | /* |
1811 | * Wait for outstanding I/O on the buftarg to complete. | |
1812 | */ | |
e80dfa19 | 1813 | void |
8321ddb2 | 1814 | xfs_buftarg_wait( |
e80dfa19 DC |
1815 | struct xfs_buftarg *btp) |
1816 | { | |
85bec546 | 1817 | /* |
9c7504aa BF |
1818 | * First wait on the buftarg I/O count for all in-flight buffers to be |
1819 | * released. This is critical as new buffers do not make the LRU until | |
1820 | * they are released. | |
1821 | * | |
1822 | * Next, flush the buffer workqueue to ensure all completion processing | |
1823 | * has finished. Just waiting on buffer locks is not sufficient for | |
1824 | * async IO as the reference count held over IO is not released until | |
1825 | * after the buffer lock is dropped. Hence we need to ensure here that | |
1826 | * all reference counts have been dropped before we start walking the | |
1827 | * LRU list. | |
85bec546 | 1828 | */ |
9c7504aa BF |
1829 | while (percpu_counter_sum(&btp->bt_io_count)) |
1830 | delay(100); | |
800b2694 | 1831 | flush_workqueue(btp->bt_mount->m_buf_workqueue); |
8321ddb2 BF |
1832 | } |
1833 | ||
1834 | void | |
1835 | xfs_buftarg_drain( | |
1836 | struct xfs_buftarg *btp) | |
1837 | { | |
1838 | LIST_HEAD(dispose); | |
1839 | int loop = 0; | |
1840 | bool write_fail = false; | |
1841 | ||
1842 | xfs_buftarg_wait(btp); | |
85bec546 | 1843 | |
a4082357 DC |
1844 | /* loop until there is nothing left on the lru list. */ |
1845 | while (list_lru_count(&btp->bt_lru)) { | |
10fb9ac1 | 1846 | list_lru_walk(&btp->bt_lru, xfs_buftarg_drain_rele, |
a4082357 DC |
1847 | &dispose, LONG_MAX); |
1848 | ||
1849 | while (!list_empty(&dispose)) { | |
1850 | struct xfs_buf *bp; | |
1851 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1852 | list_del_init(&bp->b_lru); | |
ac8809f9 | 1853 | if (bp->b_flags & XBF_WRITE_FAIL) { |
61948b6f BF |
1854 | write_fail = true; |
1855 | xfs_buf_alert_ratelimited(bp, | |
1856 | "XFS: Corruption Alert", | |
c219b015 | 1857 | "Corruption Alert: Buffer at daddr 0x%llx had permanent write failures!", |
4c7f65ae | 1858 | (long long)xfs_buf_daddr(bp)); |
ac8809f9 | 1859 | } |
a4082357 DC |
1860 | xfs_buf_rele(bp); |
1861 | } | |
1862 | if (loop++ != 0) | |
1863 | delay(100); | |
1864 | } | |
61948b6f BF |
1865 | |
1866 | /* | |
1867 | * If one or more failed buffers were freed, that means dirty metadata | |
1868 | * was thrown away. This should only ever happen after I/O completion | |
1869 | * handling has elevated I/O error(s) to permanent failures and shuts | |
01728b44 | 1870 | * down the journal. |
61948b6f BF |
1871 | */ |
1872 | if (write_fail) { | |
01728b44 | 1873 | ASSERT(xlog_is_shutdown(btp->bt_mount->m_log)); |
61948b6f BF |
1874 | xfs_alert(btp->bt_mount, |
1875 | "Please run xfs_repair to determine the extent of the problem."); | |
1876 | } | |
e80dfa19 DC |
1877 | } |
1878 | ||
1879 | static enum lru_status | |
1880 | xfs_buftarg_isolate( | |
1881 | struct list_head *item, | |
3f97b163 | 1882 | struct list_lru_one *lru, |
e80dfa19 DC |
1883 | spinlock_t *lru_lock, |
1884 | void *arg) | |
1885 | { | |
1886 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); | |
1887 | struct list_head *dispose = arg; | |
1888 | ||
a4082357 DC |
1889 | /* |
1890 | * we are inverting the lru lock/bp->b_lock here, so use a trylock. | |
1891 | * If we fail to get the lock, just skip it. | |
1892 | */ | |
1893 | if (!spin_trylock(&bp->b_lock)) | |
1894 | return LRU_SKIP; | |
e80dfa19 DC |
1895 | /* |
1896 | * Decrement the b_lru_ref count unless the value is already | |
1897 | * zero. If the value is already zero, we need to reclaim the | |
1898 | * buffer, otherwise it gets another trip through the LRU. | |
1899 | */ | |
19957a18 | 1900 | if (atomic_add_unless(&bp->b_lru_ref, -1, 0)) { |
a4082357 | 1901 | spin_unlock(&bp->b_lock); |
e80dfa19 | 1902 | return LRU_ROTATE; |
a4082357 | 1903 | } |
e80dfa19 | 1904 | |
a4082357 | 1905 | bp->b_state |= XFS_BSTATE_DISPOSE; |
3f97b163 | 1906 | list_lru_isolate_move(lru, item, dispose); |
a4082357 | 1907 | spin_unlock(&bp->b_lock); |
e80dfa19 DC |
1908 | return LRU_REMOVED; |
1909 | } | |
1910 | ||
addbda40 | 1911 | static unsigned long |
e80dfa19 | 1912 | xfs_buftarg_shrink_scan( |
ff57ab21 | 1913 | struct shrinker *shrink, |
1495f230 | 1914 | struct shrink_control *sc) |
a6867a68 | 1915 | { |
ff57ab21 DC |
1916 | struct xfs_buftarg *btp = container_of(shrink, |
1917 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1918 | LIST_HEAD(dispose); |
addbda40 | 1919 | unsigned long freed; |
430cbeb8 | 1920 | |
503c358c VD |
1921 | freed = list_lru_shrink_walk(&btp->bt_lru, sc, |
1922 | xfs_buftarg_isolate, &dispose); | |
430cbeb8 DC |
1923 | |
1924 | while (!list_empty(&dispose)) { | |
e80dfa19 | 1925 | struct xfs_buf *bp; |
430cbeb8 DC |
1926 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); |
1927 | list_del_init(&bp->b_lru); | |
1928 | xfs_buf_rele(bp); | |
1929 | } | |
1930 | ||
e80dfa19 DC |
1931 | return freed; |
1932 | } | |
1933 | ||
addbda40 | 1934 | static unsigned long |
e80dfa19 DC |
1935 | xfs_buftarg_shrink_count( |
1936 | struct shrinker *shrink, | |
1937 | struct shrink_control *sc) | |
1938 | { | |
1939 | struct xfs_buftarg *btp = container_of(shrink, | |
1940 | struct xfs_buftarg, bt_shrinker); | |
503c358c | 1941 | return list_lru_shrink_count(&btp->bt_lru, sc); |
a6867a68 DC |
1942 | } |
1943 | ||
1da177e4 LT |
1944 | void |
1945 | xfs_free_buftarg( | |
b7963133 | 1946 | struct xfs_buftarg *btp) |
1da177e4 | 1947 | { |
ff57ab21 | 1948 | unregister_shrinker(&btp->bt_shrinker); |
9c7504aa BF |
1949 | ASSERT(percpu_counter_sum(&btp->bt_io_count) == 0); |
1950 | percpu_counter_destroy(&btp->bt_io_count); | |
f5e1dd34 | 1951 | list_lru_destroy(&btp->bt_lru); |
ff57ab21 | 1952 | |
6f643c57 | 1953 | fs_put_dax(btp->bt_daxdev, btp->bt_mount); |
41233576 | 1954 | /* the main block device is closed by kill_block_super */ |
e340dd63 JK |
1955 | if (btp->bt_bdev != btp->bt_mount->m_super->s_bdev) |
1956 | bdev_release(btp->bt_bdev_handle); | |
a6867a68 | 1957 | |
f0e2d93c | 1958 | kmem_free(btp); |
1da177e4 LT |
1959 | } |
1960 | ||
3fefdeee ES |
1961 | int |
1962 | xfs_setsize_buftarg( | |
1da177e4 | 1963 | xfs_buftarg_t *btp, |
3fefdeee | 1964 | unsigned int sectorsize) |
1da177e4 | 1965 | { |
7c71ee78 | 1966 | /* Set up metadata sector size info */ |
6da54179 ES |
1967 | btp->bt_meta_sectorsize = sectorsize; |
1968 | btp->bt_meta_sectormask = sectorsize - 1; | |
1da177e4 | 1969 | |
ce8e922c | 1970 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
4f10700a | 1971 | xfs_warn(btp->bt_mount, |
a1c6f057 DM |
1972 | "Cannot set_blocksize to %u on device %pg", |
1973 | sectorsize, btp->bt_bdev); | |
2451337d | 1974 | return -EINVAL; |
1da177e4 LT |
1975 | } |
1976 | ||
7c71ee78 ES |
1977 | /* Set up device logical sector size mask */ |
1978 | btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev); | |
1979 | btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1; | |
1980 | ||
1da177e4 LT |
1981 | return 0; |
1982 | } | |
1983 | ||
1984 | /* | |
3fefdeee ES |
1985 | * When allocating the initial buffer target we have not yet |
1986 | * read in the superblock, so don't know what sized sectors | |
1987 | * are being used at this early stage. Play safe. | |
ce8e922c | 1988 | */ |
1da177e4 LT |
1989 | STATIC int |
1990 | xfs_setsize_buftarg_early( | |
e340dd63 | 1991 | xfs_buftarg_t *btp) |
1da177e4 | 1992 | { |
e340dd63 | 1993 | return xfs_setsize_buftarg(btp, bdev_logical_block_size(btp->bt_bdev)); |
1da177e4 LT |
1994 | } |
1995 | ||
5b5abbef | 1996 | struct xfs_buftarg * |
1da177e4 | 1997 | xfs_alloc_buftarg( |
ebad861b | 1998 | struct xfs_mount *mp, |
e340dd63 | 1999 | struct bdev_handle *bdev_handle) |
1da177e4 LT |
2000 | { |
2001 | xfs_buftarg_t *btp; | |
6f643c57 | 2002 | const struct dax_holder_operations *ops = NULL; |
1da177e4 | 2003 | |
6f643c57 SR |
2004 | #if defined(CONFIG_FS_DAX) && defined(CONFIG_MEMORY_FAILURE) |
2005 | ops = &xfs_dax_holder_operations; | |
2006 | #endif | |
707e0dda | 2007 | btp = kmem_zalloc(sizeof(*btp), KM_NOFS); |
1da177e4 | 2008 | |
ebad861b | 2009 | btp->bt_mount = mp; |
e340dd63 JK |
2010 | btp->bt_bdev_handle = bdev_handle; |
2011 | btp->bt_dev = bdev_handle->bdev->bd_dev; | |
2012 | btp->bt_bdev = bdev_handle->bdev; | |
2013 | btp->bt_daxdev = fs_dax_get_by_bdev(btp->bt_bdev, &btp->bt_dax_part_off, | |
6f643c57 | 2014 | mp, ops); |
0e6e847f | 2015 | |
f9bccfcc BF |
2016 | /* |
2017 | * Buffer IO error rate limiting. Limit it to no more than 10 messages | |
2018 | * per 30 seconds so as to not spam logs too much on repeated errors. | |
2019 | */ | |
2020 | ratelimit_state_init(&btp->bt_ioerror_rl, 30 * HZ, | |
2021 | DEFAULT_RATELIMIT_BURST); | |
2022 | ||
e340dd63 | 2023 | if (xfs_setsize_buftarg_early(btp)) |
d210a987 | 2024 | goto error_free; |
5ca302c8 GC |
2025 | |
2026 | if (list_lru_init(&btp->bt_lru)) | |
d210a987 | 2027 | goto error_free; |
5ca302c8 | 2028 | |
9c7504aa | 2029 | if (percpu_counter_init(&btp->bt_io_count, 0, GFP_KERNEL)) |
d210a987 | 2030 | goto error_lru; |
9c7504aa | 2031 | |
e80dfa19 DC |
2032 | btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count; |
2033 | btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan; | |
ff57ab21 | 2034 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; |
e80dfa19 | 2035 | btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE; |
e33c267a RG |
2036 | if (register_shrinker(&btp->bt_shrinker, "xfs-buf:%s", |
2037 | mp->m_super->s_id)) | |
d210a987 | 2038 | goto error_pcpu; |
1da177e4 LT |
2039 | return btp; |
2040 | ||
d210a987 MH |
2041 | error_pcpu: |
2042 | percpu_counter_destroy(&btp->bt_io_count); | |
2043 | error_lru: | |
2044 | list_lru_destroy(&btp->bt_lru); | |
2045 | error_free: | |
f0e2d93c | 2046 | kmem_free(btp); |
1da177e4 LT |
2047 | return NULL; |
2048 | } | |
2049 | ||
20e8a063 BF |
2050 | /* |
2051 | * Cancel a delayed write list. | |
2052 | * | |
2053 | * Remove each buffer from the list, clear the delwri queue flag and drop the | |
2054 | * associated buffer reference. | |
2055 | */ | |
2056 | void | |
2057 | xfs_buf_delwri_cancel( | |
2058 | struct list_head *list) | |
2059 | { | |
2060 | struct xfs_buf *bp; | |
2061 | ||
2062 | while (!list_empty(list)) { | |
2063 | bp = list_first_entry(list, struct xfs_buf, b_list); | |
2064 | ||
2065 | xfs_buf_lock(bp); | |
2066 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
2067 | list_del_init(&bp->b_list); | |
2068 | xfs_buf_relse(bp); | |
2069 | } | |
2070 | } | |
2071 | ||
1da177e4 | 2072 | /* |
43ff2122 CH |
2073 | * Add a buffer to the delayed write list. |
2074 | * | |
2075 | * This queues a buffer for writeout if it hasn't already been. Note that | |
2076 | * neither this routine nor the buffer list submission functions perform | |
2077 | * any internal synchronization. It is expected that the lists are thread-local | |
2078 | * to the callers. | |
2079 | * | |
2080 | * Returns true if we queued up the buffer, or false if it already had | |
2081 | * been on the buffer list. | |
1da177e4 | 2082 | */ |
43ff2122 | 2083 | bool |
ce8e922c | 2084 | xfs_buf_delwri_queue( |
43ff2122 CH |
2085 | struct xfs_buf *bp, |
2086 | struct list_head *list) | |
1da177e4 | 2087 | { |
43ff2122 | 2088 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 2089 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 2090 | |
43ff2122 CH |
2091 | /* |
2092 | * If the buffer is already marked delwri it already is queued up | |
2093 | * by someone else for imediate writeout. Just ignore it in that | |
2094 | * case. | |
2095 | */ | |
2096 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
2097 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
2098 | return false; | |
1da177e4 | 2099 | } |
1da177e4 | 2100 | |
43ff2122 | 2101 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
2102 | |
2103 | /* | |
43ff2122 CH |
2104 | * If a buffer gets written out synchronously or marked stale while it |
2105 | * is on a delwri list we lazily remove it. To do this, the other party | |
2106 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
2107 | * It remains referenced and on the list. In a rare corner case it | |
2108 | * might get readded to a delwri list after the synchronous writeout, in | |
2109 | * which case we need just need to re-add the flag here. | |
d808f617 | 2110 | */ |
43ff2122 CH |
2111 | bp->b_flags |= _XBF_DELWRI_Q; |
2112 | if (list_empty(&bp->b_list)) { | |
2113 | atomic_inc(&bp->b_hold); | |
2114 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 2115 | } |
585e6d88 | 2116 | |
43ff2122 | 2117 | return true; |
585e6d88 DC |
2118 | } |
2119 | ||
089716aa DC |
2120 | /* |
2121 | * Compare function is more complex than it needs to be because | |
2122 | * the return value is only 32 bits and we are doing comparisons | |
2123 | * on 64 bit values | |
2124 | */ | |
2125 | static int | |
2126 | xfs_buf_cmp( | |
4f0f586b ST |
2127 | void *priv, |
2128 | const struct list_head *a, | |
2129 | const struct list_head *b) | |
089716aa DC |
2130 | { |
2131 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
2132 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
2133 | xfs_daddr_t diff; | |
2134 | ||
f4b42421 | 2135 | diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn; |
089716aa DC |
2136 | if (diff < 0) |
2137 | return -1; | |
2138 | if (diff > 0) | |
2139 | return 1; | |
2140 | return 0; | |
2141 | } | |
2142 | ||
26f1fe85 | 2143 | /* |
e339dd8d BF |
2144 | * Submit buffers for write. If wait_list is specified, the buffers are |
2145 | * submitted using sync I/O and placed on the wait list such that the caller can | |
2146 | * iowait each buffer. Otherwise async I/O is used and the buffers are released | |
2147 | * at I/O completion time. In either case, buffers remain locked until I/O | |
2148 | * completes and the buffer is released from the queue. | |
26f1fe85 | 2149 | */ |
43ff2122 | 2150 | static int |
26f1fe85 | 2151 | xfs_buf_delwri_submit_buffers( |
43ff2122 | 2152 | struct list_head *buffer_list, |
26f1fe85 | 2153 | struct list_head *wait_list) |
1da177e4 | 2154 | { |
43ff2122 CH |
2155 | struct xfs_buf *bp, *n; |
2156 | int pinned = 0; | |
26f1fe85 | 2157 | struct blk_plug plug; |
43ff2122 | 2158 | |
26f1fe85 | 2159 | list_sort(NULL, buffer_list, xfs_buf_cmp); |
43ff2122 | 2160 | |
26f1fe85 | 2161 | blk_start_plug(&plug); |
43ff2122 | 2162 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { |
26f1fe85 | 2163 | if (!wait_list) { |
dbd0f529 DC |
2164 | if (!xfs_buf_trylock(bp)) |
2165 | continue; | |
43ff2122 | 2166 | if (xfs_buf_ispinned(bp)) { |
dbd0f529 | 2167 | xfs_buf_unlock(bp); |
43ff2122 CH |
2168 | pinned++; |
2169 | continue; | |
2170 | } | |
43ff2122 CH |
2171 | } else { |
2172 | xfs_buf_lock(bp); | |
2173 | } | |
978c7b2f | 2174 | |
43ff2122 CH |
2175 | /* |
2176 | * Someone else might have written the buffer synchronously or | |
2177 | * marked it stale in the meantime. In that case only the | |
2178 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
2179 | * reference and remove it from the list here. | |
2180 | */ | |
2181 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
2182 | list_del_init(&bp->b_list); | |
2183 | xfs_buf_relse(bp); | |
2184 | continue; | |
2185 | } | |
c9c12971 | 2186 | |
43ff2122 | 2187 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
a1b7ea5d | 2188 | |
cf53e99d | 2189 | /* |
e339dd8d BF |
2190 | * If we have a wait list, each buffer (and associated delwri |
2191 | * queue reference) transfers to it and is submitted | |
2192 | * synchronously. Otherwise, drop the buffer from the delwri | |
2193 | * queue and submit async. | |
cf53e99d | 2194 | */ |
b6983e80 | 2195 | bp->b_flags &= ~_XBF_DELWRI_Q; |
e339dd8d | 2196 | bp->b_flags |= XBF_WRITE; |
26f1fe85 | 2197 | if (wait_list) { |
e339dd8d | 2198 | bp->b_flags &= ~XBF_ASYNC; |
26f1fe85 | 2199 | list_move_tail(&bp->b_list, wait_list); |
e339dd8d BF |
2200 | } else { |
2201 | bp->b_flags |= XBF_ASYNC; | |
ce8e922c | 2202 | list_del_init(&bp->b_list); |
e339dd8d | 2203 | } |
6af88cda | 2204 | __xfs_buf_submit(bp, false); |
43ff2122 CH |
2205 | } |
2206 | blk_finish_plug(&plug); | |
1da177e4 | 2207 | |
43ff2122 | 2208 | return pinned; |
1da177e4 LT |
2209 | } |
2210 | ||
2211 | /* | |
43ff2122 CH |
2212 | * Write out a buffer list asynchronously. |
2213 | * | |
2214 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
2215 | * out and not wait for I/O completion on any of the buffers. This interface | |
2216 | * is only safely useable for callers that can track I/O completion by higher | |
2217 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
2218 | * function. | |
efc3289c BF |
2219 | * |
2220 | * Note: this function will skip buffers it would block on, and in doing so | |
2221 | * leaves them on @buffer_list so they can be retried on a later pass. As such, | |
2222 | * it is up to the caller to ensure that the buffer list is fully submitted or | |
2223 | * cancelled appropriately when they are finished with the list. Failure to | |
2224 | * cancel or resubmit the list until it is empty will result in leaked buffers | |
2225 | * at unmount time. | |
1da177e4 LT |
2226 | */ |
2227 | int | |
43ff2122 CH |
2228 | xfs_buf_delwri_submit_nowait( |
2229 | struct list_head *buffer_list) | |
1da177e4 | 2230 | { |
26f1fe85 | 2231 | return xfs_buf_delwri_submit_buffers(buffer_list, NULL); |
43ff2122 | 2232 | } |
1da177e4 | 2233 | |
43ff2122 CH |
2234 | /* |
2235 | * Write out a buffer list synchronously. | |
2236 | * | |
2237 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
2238 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
2239 | * so callers must have some other way of tracking buffers if they require such | |
2240 | * functionality. | |
2241 | */ | |
2242 | int | |
2243 | xfs_buf_delwri_submit( | |
2244 | struct list_head *buffer_list) | |
2245 | { | |
26f1fe85 | 2246 | LIST_HEAD (wait_list); |
43ff2122 CH |
2247 | int error = 0, error2; |
2248 | struct xfs_buf *bp; | |
1da177e4 | 2249 | |
26f1fe85 | 2250 | xfs_buf_delwri_submit_buffers(buffer_list, &wait_list); |
1da177e4 | 2251 | |
43ff2122 | 2252 | /* Wait for IO to complete. */ |
26f1fe85 DC |
2253 | while (!list_empty(&wait_list)) { |
2254 | bp = list_first_entry(&wait_list, struct xfs_buf, b_list); | |
a1b7ea5d | 2255 | |
089716aa | 2256 | list_del_init(&bp->b_list); |
cf53e99d | 2257 | |
e339dd8d BF |
2258 | /* |
2259 | * Wait on the locked buffer, check for errors and unlock and | |
2260 | * release the delwri queue reference. | |
2261 | */ | |
2262 | error2 = xfs_buf_iowait(bp); | |
43ff2122 CH |
2263 | xfs_buf_relse(bp); |
2264 | if (!error) | |
2265 | error = error2; | |
1da177e4 LT |
2266 | } |
2267 | ||
43ff2122 | 2268 | return error; |
1da177e4 LT |
2269 | } |
2270 | ||
7912e7fe BF |
2271 | /* |
2272 | * Push a single buffer on a delwri queue. | |
2273 | * | |
2274 | * The purpose of this function is to submit a single buffer of a delwri queue | |
2275 | * and return with the buffer still on the original queue. The waiting delwri | |
2276 | * buffer submission infrastructure guarantees transfer of the delwri queue | |
2277 | * buffer reference to a temporary wait list. We reuse this infrastructure to | |
2278 | * transfer the buffer back to the original queue. | |
2279 | * | |
2280 | * Note the buffer transitions from the queued state, to the submitted and wait | |
2281 | * listed state and back to the queued state during this call. The buffer | |
2282 | * locking and queue management logic between _delwri_pushbuf() and | |
2283 | * _delwri_queue() guarantee that the buffer cannot be queued to another list | |
2284 | * before returning. | |
2285 | */ | |
2286 | int | |
2287 | xfs_buf_delwri_pushbuf( | |
2288 | struct xfs_buf *bp, | |
2289 | struct list_head *buffer_list) | |
2290 | { | |
2291 | LIST_HEAD (submit_list); | |
2292 | int error; | |
2293 | ||
2294 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
2295 | ||
2296 | trace_xfs_buf_delwri_pushbuf(bp, _RET_IP_); | |
2297 | ||
2298 | /* | |
2299 | * Isolate the buffer to a new local list so we can submit it for I/O | |
2300 | * independently from the rest of the original list. | |
2301 | */ | |
2302 | xfs_buf_lock(bp); | |
2303 | list_move(&bp->b_list, &submit_list); | |
2304 | xfs_buf_unlock(bp); | |
2305 | ||
2306 | /* | |
2307 | * Delwri submission clears the DELWRI_Q buffer flag and returns with | |
e339dd8d | 2308 | * the buffer on the wait list with the original reference. Rather than |
7912e7fe BF |
2309 | * bounce the buffer from a local wait list back to the original list |
2310 | * after I/O completion, reuse the original list as the wait list. | |
2311 | */ | |
2312 | xfs_buf_delwri_submit_buffers(&submit_list, buffer_list); | |
2313 | ||
2314 | /* | |
e339dd8d BF |
2315 | * The buffer is now locked, under I/O and wait listed on the original |
2316 | * delwri queue. Wait for I/O completion, restore the DELWRI_Q flag and | |
2317 | * return with the buffer unlocked and on the original queue. | |
7912e7fe | 2318 | */ |
e339dd8d | 2319 | error = xfs_buf_iowait(bp); |
7912e7fe BF |
2320 | bp->b_flags |= _XBF_DELWRI_Q; |
2321 | xfs_buf_unlock(bp); | |
2322 | ||
2323 | return error; | |
2324 | } | |
2325 | ||
7561d27e BF |
2326 | void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref) |
2327 | { | |
7561d27e BF |
2328 | /* |
2329 | * Set the lru reference count to 0 based on the error injection tag. | |
2330 | * This allows userspace to disrupt buffer caching for debug/testing | |
2331 | * purposes. | |
2332 | */ | |
dbd329f1 | 2333 | if (XFS_TEST_ERROR(false, bp->b_mount, XFS_ERRTAG_BUF_LRU_REF)) |
7561d27e BF |
2334 | lru_ref = 0; |
2335 | ||
2336 | atomic_set(&bp->b_lru_ref, lru_ref); | |
2337 | } | |
8473fee3 BF |
2338 | |
2339 | /* | |
2340 | * Verify an on-disk magic value against the magic value specified in the | |
2341 | * verifier structure. The verifier magic is in disk byte order so the caller is | |
2342 | * expected to pass the value directly from disk. | |
2343 | */ | |
2344 | bool | |
2345 | xfs_verify_magic( | |
2346 | struct xfs_buf *bp, | |
15baadf7 | 2347 | __be32 dmagic) |
8473fee3 | 2348 | { |
dbd329f1 | 2349 | struct xfs_mount *mp = bp->b_mount; |
8473fee3 BF |
2350 | int idx; |
2351 | ||
38c26bfd | 2352 | idx = xfs_has_crc(mp); |
14ed8688 | 2353 | if (WARN_ON(!bp->b_ops || !bp->b_ops->magic[idx])) |
8473fee3 BF |
2354 | return false; |
2355 | return dmagic == bp->b_ops->magic[idx]; | |
2356 | } | |
15baadf7 DW |
2357 | /* |
2358 | * Verify an on-disk magic value against the magic value specified in the | |
2359 | * verifier structure. The verifier magic is in disk byte order so the caller is | |
2360 | * expected to pass the value directly from disk. | |
2361 | */ | |
2362 | bool | |
2363 | xfs_verify_magic16( | |
2364 | struct xfs_buf *bp, | |
2365 | __be16 dmagic) | |
2366 | { | |
dbd329f1 | 2367 | struct xfs_mount *mp = bp->b_mount; |
15baadf7 DW |
2368 | int idx; |
2369 | ||
38c26bfd | 2370 | idx = xfs_has_crc(mp); |
14ed8688 | 2371 | if (WARN_ON(!bp->b_ops || !bp->b_ops->magic16[idx])) |
15baadf7 DW |
2372 | return false; |
2373 | return dmagic == bp->b_ops->magic16[idx]; | |
2374 | } |