]>
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> |
1da177e4 | 8 | |
5467b34b | 9 | #include "xfs_shared.h" |
4fb6e8ad | 10 | #include "xfs_format.h" |
239880ef | 11 | #include "xfs_log_format.h" |
7fd36c44 | 12 | #include "xfs_trans_resv.h" |
239880ef | 13 | #include "xfs_sb.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" |
f593bf14 DC |
18 | #include "xfs_trans.h" |
19 | #include "xfs_buf_item.h" | |
e9e899a2 | 20 | #include "xfs_errortag.h" |
7561d27e | 21 | #include "xfs_error.h" |
b7963133 | 22 | |
7989cb8e | 23 | static kmem_zone_t *xfs_buf_zone; |
23ea4032 | 24 | |
ce8e922c | 25 | #define xb_to_gfp(flags) \ |
aa5c158e | 26 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN) |
1da177e4 | 27 | |
37fd1678 DC |
28 | /* |
29 | * Locking orders | |
30 | * | |
31 | * xfs_buf_ioacct_inc: | |
32 | * xfs_buf_ioacct_dec: | |
33 | * b_sema (caller holds) | |
34 | * b_lock | |
35 | * | |
36 | * xfs_buf_stale: | |
37 | * b_sema (caller holds) | |
38 | * b_lock | |
39 | * lru_lock | |
40 | * | |
41 | * xfs_buf_rele: | |
42 | * b_lock | |
43 | * pag_buf_lock | |
44 | * lru_lock | |
45 | * | |
46 | * xfs_buftarg_wait_rele | |
47 | * lru_lock | |
48 | * b_lock (trylock due to inversion) | |
49 | * | |
50 | * xfs_buftarg_isolate | |
51 | * lru_lock | |
52 | * b_lock (trylock due to inversion) | |
53 | */ | |
1da177e4 | 54 | |
73c77e2c JB |
55 | static inline int |
56 | xfs_buf_is_vmapped( | |
57 | struct xfs_buf *bp) | |
58 | { | |
59 | /* | |
60 | * Return true if the buffer is vmapped. | |
61 | * | |
611c9946 DC |
62 | * b_addr is null if the buffer is not mapped, but the code is clever |
63 | * enough to know it doesn't have to map a single page, so the check has | |
64 | * to be both for b_addr and bp->b_page_count > 1. | |
73c77e2c | 65 | */ |
611c9946 | 66 | return bp->b_addr && bp->b_page_count > 1; |
73c77e2c JB |
67 | } |
68 | ||
69 | static inline int | |
70 | xfs_buf_vmap_len( | |
71 | struct xfs_buf *bp) | |
72 | { | |
73 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
74 | } | |
75 | ||
9c7504aa BF |
76 | /* |
77 | * Bump the I/O in flight count on the buftarg if we haven't yet done so for | |
78 | * this buffer. The count is incremented once per buffer (per hold cycle) | |
79 | * because the corresponding decrement is deferred to buffer release. Buffers | |
80 | * can undergo I/O multiple times in a hold-release cycle and per buffer I/O | |
81 | * tracking adds unnecessary overhead. This is used for sychronization purposes | |
82 | * with unmount (see xfs_wait_buftarg()), so all we really need is a count of | |
83 | * in-flight buffers. | |
84 | * | |
85 | * Buffers that are never released (e.g., superblock, iclog buffers) must set | |
86 | * the XBF_NO_IOACCT flag before I/O submission. Otherwise, the buftarg count | |
87 | * never reaches zero and unmount hangs indefinitely. | |
88 | */ | |
89 | static inline void | |
90 | xfs_buf_ioacct_inc( | |
91 | struct xfs_buf *bp) | |
92 | { | |
63db7c81 | 93 | if (bp->b_flags & XBF_NO_IOACCT) |
9c7504aa BF |
94 | return; |
95 | ||
96 | ASSERT(bp->b_flags & XBF_ASYNC); | |
63db7c81 BF |
97 | spin_lock(&bp->b_lock); |
98 | if (!(bp->b_state & XFS_BSTATE_IN_FLIGHT)) { | |
99 | bp->b_state |= XFS_BSTATE_IN_FLIGHT; | |
100 | percpu_counter_inc(&bp->b_target->bt_io_count); | |
101 | } | |
102 | spin_unlock(&bp->b_lock); | |
9c7504aa BF |
103 | } |
104 | ||
105 | /* | |
106 | * Clear the in-flight state on a buffer about to be released to the LRU or | |
107 | * freed and unaccount from the buftarg. | |
108 | */ | |
109 | static inline void | |
63db7c81 | 110 | __xfs_buf_ioacct_dec( |
9c7504aa BF |
111 | struct xfs_buf *bp) |
112 | { | |
95989c46 | 113 | lockdep_assert_held(&bp->b_lock); |
9c7504aa | 114 | |
63db7c81 BF |
115 | if (bp->b_state & XFS_BSTATE_IN_FLIGHT) { |
116 | bp->b_state &= ~XFS_BSTATE_IN_FLIGHT; | |
117 | percpu_counter_dec(&bp->b_target->bt_io_count); | |
118 | } | |
119 | } | |
120 | ||
121 | static inline void | |
122 | xfs_buf_ioacct_dec( | |
123 | struct xfs_buf *bp) | |
124 | { | |
125 | spin_lock(&bp->b_lock); | |
126 | __xfs_buf_ioacct_dec(bp); | |
127 | spin_unlock(&bp->b_lock); | |
9c7504aa BF |
128 | } |
129 | ||
430cbeb8 DC |
130 | /* |
131 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
132 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
133 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
134 | * to remove the reference that LRU holds on the buffer. | |
135 | * | |
136 | * This prevents build-up of stale buffers on the LRU. | |
137 | */ | |
138 | void | |
139 | xfs_buf_stale( | |
140 | struct xfs_buf *bp) | |
141 | { | |
43ff2122 CH |
142 | ASSERT(xfs_buf_islocked(bp)); |
143 | ||
430cbeb8 | 144 | bp->b_flags |= XBF_STALE; |
43ff2122 CH |
145 | |
146 | /* | |
147 | * Clear the delwri status so that a delwri queue walker will not | |
148 | * flush this buffer to disk now that it is stale. The delwri queue has | |
149 | * a reference to the buffer, so this is safe to do. | |
150 | */ | |
151 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
152 | ||
9c7504aa BF |
153 | /* |
154 | * Once the buffer is marked stale and unlocked, a subsequent lookup | |
155 | * could reset b_flags. There is no guarantee that the buffer is | |
156 | * unaccounted (released to LRU) before that occurs. Drop in-flight | |
157 | * status now to preserve accounting consistency. | |
158 | */ | |
a4082357 | 159 | spin_lock(&bp->b_lock); |
63db7c81 BF |
160 | __xfs_buf_ioacct_dec(bp); |
161 | ||
a4082357 DC |
162 | atomic_set(&bp->b_lru_ref, 0); |
163 | if (!(bp->b_state & XFS_BSTATE_DISPOSE) && | |
e80dfa19 DC |
164 | (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru))) |
165 | atomic_dec(&bp->b_hold); | |
166 | ||
430cbeb8 | 167 | ASSERT(atomic_read(&bp->b_hold) >= 1); |
a4082357 | 168 | spin_unlock(&bp->b_lock); |
430cbeb8 | 169 | } |
1da177e4 | 170 | |
3e85c868 DC |
171 | static int |
172 | xfs_buf_get_maps( | |
173 | struct xfs_buf *bp, | |
174 | int map_count) | |
175 | { | |
176 | ASSERT(bp->b_maps == NULL); | |
177 | bp->b_map_count = map_count; | |
178 | ||
179 | if (map_count == 1) { | |
f4b42421 | 180 | bp->b_maps = &bp->__b_map; |
3e85c868 DC |
181 | return 0; |
182 | } | |
183 | ||
184 | bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map), | |
185 | KM_NOFS); | |
186 | if (!bp->b_maps) | |
2451337d | 187 | return -ENOMEM; |
3e85c868 DC |
188 | return 0; |
189 | } | |
190 | ||
191 | /* | |
192 | * Frees b_pages if it was allocated. | |
193 | */ | |
194 | static void | |
195 | xfs_buf_free_maps( | |
196 | struct xfs_buf *bp) | |
197 | { | |
f4b42421 | 198 | if (bp->b_maps != &bp->__b_map) { |
3e85c868 DC |
199 | kmem_free(bp->b_maps); |
200 | bp->b_maps = NULL; | |
201 | } | |
202 | } | |
203 | ||
32dff5e5 | 204 | static int |
3e85c868 | 205 | _xfs_buf_alloc( |
4347b9d7 | 206 | struct xfs_buftarg *target, |
3e85c868 DC |
207 | struct xfs_buf_map *map, |
208 | int nmaps, | |
32dff5e5 DW |
209 | xfs_buf_flags_t flags, |
210 | struct xfs_buf **bpp) | |
1da177e4 | 211 | { |
4347b9d7 | 212 | struct xfs_buf *bp; |
3e85c868 DC |
213 | int error; |
214 | int i; | |
4347b9d7 | 215 | |
32dff5e5 | 216 | *bpp = NULL; |
aa5c158e | 217 | bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS); |
4347b9d7 | 218 | if (unlikely(!bp)) |
32dff5e5 | 219 | return -ENOMEM; |
4347b9d7 | 220 | |
1da177e4 | 221 | /* |
12bcb3f7 DC |
222 | * We don't want certain flags to appear in b_flags unless they are |
223 | * specifically set by later operations on the buffer. | |
1da177e4 | 224 | */ |
611c9946 | 225 | flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD); |
ce8e922c | 226 | |
ce8e922c | 227 | atomic_set(&bp->b_hold, 1); |
430cbeb8 | 228 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 229 | init_completion(&bp->b_iowait); |
430cbeb8 | 230 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 231 | INIT_LIST_HEAD(&bp->b_list); |
643c8c05 | 232 | INIT_LIST_HEAD(&bp->b_li_list); |
a731cd11 | 233 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
a4082357 | 234 | spin_lock_init(&bp->b_lock); |
ce8e922c | 235 | bp->b_target = target; |
dbd329f1 | 236 | bp->b_mount = target->bt_mount; |
3e85c868 | 237 | bp->b_flags = flags; |
de1cbee4 | 238 | |
1da177e4 | 239 | /* |
aa0e8833 DC |
240 | * Set length and io_length to the same value initially. |
241 | * I/O routines should use io_length, which will be the same in | |
1da177e4 LT |
242 | * most cases but may be reset (e.g. XFS recovery). |
243 | */ | |
3e85c868 DC |
244 | error = xfs_buf_get_maps(bp, nmaps); |
245 | if (error) { | |
377bcd5f | 246 | kmem_cache_free(xfs_buf_zone, bp); |
32dff5e5 | 247 | return error; |
3e85c868 DC |
248 | } |
249 | ||
250 | bp->b_bn = map[0].bm_bn; | |
251 | bp->b_length = 0; | |
252 | for (i = 0; i < nmaps; i++) { | |
253 | bp->b_maps[i].bm_bn = map[i].bm_bn; | |
254 | bp->b_maps[i].bm_len = map[i].bm_len; | |
255 | bp->b_length += map[i].bm_len; | |
256 | } | |
3e85c868 | 257 | |
ce8e922c NS |
258 | atomic_set(&bp->b_pin_count, 0); |
259 | init_waitqueue_head(&bp->b_waiters); | |
260 | ||
dbd329f1 | 261 | XFS_STATS_INC(bp->b_mount, xb_create); |
0b1b213f | 262 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 | 263 | |
32dff5e5 DW |
264 | *bpp = bp; |
265 | return 0; | |
1da177e4 LT |
266 | } |
267 | ||
268 | /* | |
ce8e922c NS |
269 | * Allocate a page array capable of holding a specified number |
270 | * of pages, and point the page buf at it. | |
1da177e4 LT |
271 | */ |
272 | STATIC int | |
ce8e922c NS |
273 | _xfs_buf_get_pages( |
274 | xfs_buf_t *bp, | |
87937bf8 | 275 | int page_count) |
1da177e4 LT |
276 | { |
277 | /* Make sure that we have a page list */ | |
ce8e922c | 278 | if (bp->b_pages == NULL) { |
ce8e922c NS |
279 | bp->b_page_count = page_count; |
280 | if (page_count <= XB_PAGES) { | |
281 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 282 | } else { |
ce8e922c | 283 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
aa5c158e | 284 | page_count, KM_NOFS); |
ce8e922c | 285 | if (bp->b_pages == NULL) |
1da177e4 LT |
286 | return -ENOMEM; |
287 | } | |
ce8e922c | 288 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
289 | } |
290 | return 0; | |
291 | } | |
292 | ||
293 | /* | |
ce8e922c | 294 | * Frees b_pages if it was allocated. |
1da177e4 LT |
295 | */ |
296 | STATIC void | |
ce8e922c | 297 | _xfs_buf_free_pages( |
1da177e4 LT |
298 | xfs_buf_t *bp) |
299 | { | |
ce8e922c | 300 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 301 | kmem_free(bp->b_pages); |
3fc98b1a | 302 | bp->b_pages = NULL; |
1da177e4 LT |
303 | } |
304 | } | |
305 | ||
306 | /* | |
307 | * Releases the specified buffer. | |
308 | * | |
309 | * The modification state of any associated pages is left unchanged. | |
b46fe825 | 310 | * The buffer must not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
311 | * hashed and refcounted buffers |
312 | */ | |
25a40957 | 313 | static void |
ce8e922c | 314 | xfs_buf_free( |
1da177e4 LT |
315 | xfs_buf_t *bp) |
316 | { | |
0b1b213f | 317 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 318 | |
430cbeb8 DC |
319 | ASSERT(list_empty(&bp->b_lru)); |
320 | ||
0e6e847f | 321 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
322 | uint i; |
323 | ||
73c77e2c | 324 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
325 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
326 | bp->b_page_count); | |
1da177e4 | 327 | |
948ecdb4 NS |
328 | for (i = 0; i < bp->b_page_count; i++) { |
329 | struct page *page = bp->b_pages[i]; | |
330 | ||
0e6e847f | 331 | __free_page(page); |
948ecdb4 | 332 | } |
12eba65b DC |
333 | if (current->reclaim_state) |
334 | current->reclaim_state->reclaimed_slab += | |
335 | bp->b_page_count; | |
0e6e847f DC |
336 | } else if (bp->b_flags & _XBF_KMEM) |
337 | kmem_free(bp->b_addr); | |
3fc98b1a | 338 | _xfs_buf_free_pages(bp); |
3e85c868 | 339 | xfs_buf_free_maps(bp); |
377bcd5f | 340 | kmem_cache_free(xfs_buf_zone, bp); |
1da177e4 LT |
341 | } |
342 | ||
343 | /* | |
0e6e847f | 344 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
345 | */ |
346 | STATIC int | |
0e6e847f | 347 | xfs_buf_allocate_memory( |
1da177e4 LT |
348 | xfs_buf_t *bp, |
349 | uint flags) | |
350 | { | |
aa0e8833 | 351 | size_t size; |
1da177e4 | 352 | size_t nbytes, offset; |
ce8e922c | 353 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 354 | unsigned short page_count, i; |
795cac72 | 355 | xfs_off_t start, end; |
1da177e4 | 356 | int error; |
3219e8cf BD |
357 | xfs_km_flags_t kmflag_mask = 0; |
358 | ||
359 | /* | |
360 | * assure zeroed buffer for non-read cases. | |
361 | */ | |
362 | if (!(flags & XBF_READ)) { | |
363 | kmflag_mask |= KM_ZERO; | |
364 | gfp_mask |= __GFP_ZERO; | |
365 | } | |
1da177e4 | 366 | |
0e6e847f DC |
367 | /* |
368 | * for buffers that are contained within a single page, just allocate | |
369 | * the memory from the heap - there's no need for the complexity of | |
370 | * page arrays to keep allocation down to order 0. | |
371 | */ | |
795cac72 DC |
372 | size = BBTOB(bp->b_length); |
373 | if (size < PAGE_SIZE) { | |
f8f9ee47 | 374 | int align_mask = xfs_buftarg_dma_alignment(bp->b_target); |
3219e8cf BD |
375 | bp->b_addr = kmem_alloc_io(size, align_mask, |
376 | KM_NOFS | kmflag_mask); | |
0e6e847f DC |
377 | if (!bp->b_addr) { |
378 | /* low memory - use alloc_page loop instead */ | |
379 | goto use_alloc_page; | |
380 | } | |
381 | ||
795cac72 | 382 | if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) != |
0e6e847f DC |
383 | ((unsigned long)bp->b_addr & PAGE_MASK)) { |
384 | /* b_addr spans two pages - use alloc_page instead */ | |
385 | kmem_free(bp->b_addr); | |
386 | bp->b_addr = NULL; | |
387 | goto use_alloc_page; | |
388 | } | |
389 | bp->b_offset = offset_in_page(bp->b_addr); | |
390 | bp->b_pages = bp->b_page_array; | |
f8f9ee47 | 391 | bp->b_pages[0] = kmem_to_page(bp->b_addr); |
0e6e847f | 392 | bp->b_page_count = 1; |
611c9946 | 393 | bp->b_flags |= _XBF_KMEM; |
0e6e847f DC |
394 | return 0; |
395 | } | |
396 | ||
397 | use_alloc_page: | |
f4b42421 MT |
398 | start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT; |
399 | end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1) | |
cbb7baab | 400 | >> PAGE_SHIFT; |
795cac72 | 401 | page_count = end - start; |
87937bf8 | 402 | error = _xfs_buf_get_pages(bp, page_count); |
1da177e4 LT |
403 | if (unlikely(error)) |
404 | return error; | |
1da177e4 | 405 | |
ce8e922c | 406 | offset = bp->b_offset; |
0e6e847f | 407 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 408 | |
ce8e922c | 409 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
410 | struct page *page; |
411 | uint retries = 0; | |
0e6e847f DC |
412 | retry: |
413 | page = alloc_page(gfp_mask); | |
1da177e4 | 414 | if (unlikely(page == NULL)) { |
ce8e922c NS |
415 | if (flags & XBF_READ_AHEAD) { |
416 | bp->b_page_count = i; | |
2451337d | 417 | error = -ENOMEM; |
0e6e847f | 418 | goto out_free_pages; |
1da177e4 LT |
419 | } |
420 | ||
421 | /* | |
422 | * This could deadlock. | |
423 | * | |
424 | * But until all the XFS lowlevel code is revamped to | |
425 | * handle buffer allocation failures we can't do much. | |
426 | */ | |
427 | if (!(++retries % 100)) | |
4f10700a | 428 | xfs_err(NULL, |
5bf97b1c TH |
429 | "%s(%u) possible memory allocation deadlock in %s (mode:0x%x)", |
430 | current->comm, current->pid, | |
34a622b2 | 431 | __func__, gfp_mask); |
1da177e4 | 432 | |
dbd329f1 | 433 | XFS_STATS_INC(bp->b_mount, xb_page_retries); |
8aa7e847 | 434 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
435 | goto retry; |
436 | } | |
437 | ||
dbd329f1 | 438 | XFS_STATS_INC(bp->b_mount, xb_page_found); |
1da177e4 | 439 | |
0e6e847f | 440 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 441 | size -= nbytes; |
ce8e922c | 442 | bp->b_pages[i] = page; |
1da177e4 LT |
443 | offset = 0; |
444 | } | |
0e6e847f | 445 | return 0; |
1da177e4 | 446 | |
0e6e847f DC |
447 | out_free_pages: |
448 | for (i = 0; i < bp->b_page_count; i++) | |
449 | __free_page(bp->b_pages[i]); | |
2aa6ba7b | 450 | bp->b_flags &= ~_XBF_PAGES; |
1da177e4 LT |
451 | return error; |
452 | } | |
453 | ||
454 | /* | |
25985edc | 455 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
456 | */ |
457 | STATIC int | |
ce8e922c | 458 | _xfs_buf_map_pages( |
1da177e4 LT |
459 | xfs_buf_t *bp, |
460 | uint flags) | |
461 | { | |
0e6e847f | 462 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 463 | if (bp->b_page_count == 1) { |
0e6e847f | 464 | /* A single page buffer is always mappable */ |
ce8e922c | 465 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
611c9946 DC |
466 | } else if (flags & XBF_UNMAPPED) { |
467 | bp->b_addr = NULL; | |
468 | } else { | |
a19fb380 | 469 | int retried = 0; |
9ba1fb2c | 470 | unsigned nofs_flag; |
ae687e58 DC |
471 | |
472 | /* | |
cf085a1b | 473 | * vm_map_ram() will allocate auxiliary structures (e.g. |
ae687e58 DC |
474 | * pagetables) with GFP_KERNEL, yet we are likely to be under |
475 | * GFP_NOFS context here. Hence we need to tell memory reclaim | |
9ba1fb2c | 476 | * that we are in such a context via PF_MEMALLOC_NOFS to prevent |
ae687e58 DC |
477 | * memory reclaim re-entering the filesystem here and |
478 | * potentially deadlocking. | |
479 | */ | |
9ba1fb2c | 480 | nofs_flag = memalloc_nofs_save(); |
a19fb380 DC |
481 | do { |
482 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
d4efd79a | 483 | -1); |
a19fb380 DC |
484 | if (bp->b_addr) |
485 | break; | |
486 | vm_unmap_aliases(); | |
487 | } while (retried++ <= 1); | |
9ba1fb2c | 488 | memalloc_nofs_restore(nofs_flag); |
a19fb380 DC |
489 | |
490 | if (!bp->b_addr) | |
1da177e4 | 491 | return -ENOMEM; |
ce8e922c | 492 | bp->b_addr += bp->b_offset; |
1da177e4 LT |
493 | } |
494 | ||
495 | return 0; | |
496 | } | |
497 | ||
498 | /* | |
499 | * Finding and Reading Buffers | |
500 | */ | |
6031e73a LS |
501 | static int |
502 | _xfs_buf_obj_cmp( | |
503 | struct rhashtable_compare_arg *arg, | |
504 | const void *obj) | |
505 | { | |
506 | const struct xfs_buf_map *map = arg->key; | |
507 | const struct xfs_buf *bp = obj; | |
508 | ||
509 | /* | |
510 | * The key hashing in the lookup path depends on the key being the | |
511 | * first element of the compare_arg, make sure to assert this. | |
512 | */ | |
513 | BUILD_BUG_ON(offsetof(struct xfs_buf_map, bm_bn) != 0); | |
514 | ||
515 | if (bp->b_bn != map->bm_bn) | |
516 | return 1; | |
517 | ||
518 | if (unlikely(bp->b_length != map->bm_len)) { | |
519 | /* | |
520 | * found a block number match. If the range doesn't | |
521 | * match, the only way this is allowed is if the buffer | |
522 | * in the cache is stale and the transaction that made | |
523 | * it stale has not yet committed. i.e. we are | |
524 | * reallocating a busy extent. Skip this buffer and | |
525 | * continue searching for an exact match. | |
526 | */ | |
527 | ASSERT(bp->b_flags & XBF_STALE); | |
528 | return 1; | |
529 | } | |
530 | return 0; | |
531 | } | |
532 | ||
533 | static const struct rhashtable_params xfs_buf_hash_params = { | |
534 | .min_size = 32, /* empty AGs have minimal footprint */ | |
535 | .nelem_hint = 16, | |
536 | .key_len = sizeof(xfs_daddr_t), | |
537 | .key_offset = offsetof(struct xfs_buf, b_bn), | |
538 | .head_offset = offsetof(struct xfs_buf, b_rhash_head), | |
539 | .automatic_shrinking = true, | |
540 | .obj_cmpfn = _xfs_buf_obj_cmp, | |
541 | }; | |
542 | ||
543 | int | |
544 | xfs_buf_hash_init( | |
545 | struct xfs_perag *pag) | |
546 | { | |
547 | spin_lock_init(&pag->pag_buf_lock); | |
548 | return rhashtable_init(&pag->pag_buf_hash, &xfs_buf_hash_params); | |
549 | } | |
550 | ||
551 | void | |
552 | xfs_buf_hash_destroy( | |
553 | struct xfs_perag *pag) | |
554 | { | |
555 | rhashtable_destroy(&pag->pag_buf_hash); | |
556 | } | |
1da177e4 LT |
557 | |
558 | /* | |
b027d4c9 DC |
559 | * Look up a buffer in the buffer cache and return it referenced and locked |
560 | * in @found_bp. | |
561 | * | |
562 | * If @new_bp is supplied and we have a lookup miss, insert @new_bp into the | |
563 | * cache. | |
564 | * | |
565 | * If XBF_TRYLOCK is set in @flags, only try to lock the buffer and return | |
566 | * -EAGAIN if we fail to lock it. | |
567 | * | |
568 | * Return values are: | |
569 | * -EFSCORRUPTED if have been supplied with an invalid address | |
570 | * -EAGAIN on trylock failure | |
571 | * -ENOENT if we fail to find a match and @new_bp was NULL | |
572 | * 0, with @found_bp: | |
573 | * - @new_bp if we inserted it into the cache | |
574 | * - the buffer we found and locked. | |
1da177e4 | 575 | */ |
b027d4c9 DC |
576 | static int |
577 | xfs_buf_find( | |
e70b73f8 | 578 | struct xfs_buftarg *btp, |
3e85c868 DC |
579 | struct xfs_buf_map *map, |
580 | int nmaps, | |
ce8e922c | 581 | xfs_buf_flags_t flags, |
b027d4c9 DC |
582 | struct xfs_buf *new_bp, |
583 | struct xfs_buf **found_bp) | |
1da177e4 | 584 | { |
74f75a0c | 585 | struct xfs_perag *pag; |
74f75a0c | 586 | xfs_buf_t *bp; |
6031e73a | 587 | struct xfs_buf_map cmap = { .bm_bn = map[0].bm_bn }; |
10616b80 | 588 | xfs_daddr_t eofs; |
3e85c868 | 589 | int i; |
1da177e4 | 590 | |
b027d4c9 DC |
591 | *found_bp = NULL; |
592 | ||
3e85c868 | 593 | for (i = 0; i < nmaps; i++) |
6031e73a | 594 | cmap.bm_len += map[i].bm_len; |
1da177e4 LT |
595 | |
596 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
6031e73a LS |
597 | ASSERT(!(BBTOB(cmap.bm_len) < btp->bt_meta_sectorsize)); |
598 | ASSERT(!(BBTOB(cmap.bm_bn) & (xfs_off_t)btp->bt_meta_sectormask)); | |
1da177e4 | 599 | |
10616b80 DC |
600 | /* |
601 | * Corrupted block numbers can get through to here, unfortunately, so we | |
602 | * have to check that the buffer falls within the filesystem bounds. | |
603 | */ | |
604 | eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks); | |
6031e73a | 605 | if (cmap.bm_bn < 0 || cmap.bm_bn >= eofs) { |
10616b80 | 606 | xfs_alert(btp->bt_mount, |
c219b015 | 607 | "%s: daddr 0x%llx out of range, EOFS 0x%llx", |
6031e73a | 608 | __func__, cmap.bm_bn, eofs); |
7bc0dc27 | 609 | WARN_ON(1); |
b027d4c9 | 610 | return -EFSCORRUPTED; |
10616b80 DC |
611 | } |
612 | ||
74f75a0c | 613 | pag = xfs_perag_get(btp->bt_mount, |
6031e73a | 614 | xfs_daddr_to_agno(btp->bt_mount, cmap.bm_bn)); |
74f75a0c | 615 | |
74f75a0c | 616 | spin_lock(&pag->pag_buf_lock); |
6031e73a LS |
617 | bp = rhashtable_lookup_fast(&pag->pag_buf_hash, &cmap, |
618 | xfs_buf_hash_params); | |
619 | if (bp) { | |
620 | atomic_inc(&bp->b_hold); | |
621 | goto found; | |
1da177e4 LT |
622 | } |
623 | ||
624 | /* No match found */ | |
b027d4c9 | 625 | if (!new_bp) { |
ff6d6af2 | 626 | XFS_STATS_INC(btp->bt_mount, xb_miss_locked); |
74f75a0c DC |
627 | spin_unlock(&pag->pag_buf_lock); |
628 | xfs_perag_put(pag); | |
b027d4c9 | 629 | return -ENOENT; |
1da177e4 | 630 | } |
b027d4c9 DC |
631 | |
632 | /* the buffer keeps the perag reference until it is freed */ | |
633 | new_bp->b_pag = pag; | |
634 | rhashtable_insert_fast(&pag->pag_buf_hash, &new_bp->b_rhash_head, | |
635 | xfs_buf_hash_params); | |
636 | spin_unlock(&pag->pag_buf_lock); | |
637 | *found_bp = new_bp; | |
638 | return 0; | |
1da177e4 LT |
639 | |
640 | found: | |
74f75a0c DC |
641 | spin_unlock(&pag->pag_buf_lock); |
642 | xfs_perag_put(pag); | |
1da177e4 | 643 | |
0c842ad4 CH |
644 | if (!xfs_buf_trylock(bp)) { |
645 | if (flags & XBF_TRYLOCK) { | |
ce8e922c | 646 | xfs_buf_rele(bp); |
ff6d6af2 | 647 | XFS_STATS_INC(btp->bt_mount, xb_busy_locked); |
b027d4c9 | 648 | return -EAGAIN; |
1da177e4 | 649 | } |
0c842ad4 | 650 | xfs_buf_lock(bp); |
ff6d6af2 | 651 | XFS_STATS_INC(btp->bt_mount, xb_get_locked_waited); |
1da177e4 LT |
652 | } |
653 | ||
0e6e847f DC |
654 | /* |
655 | * if the buffer is stale, clear all the external state associated with | |
656 | * it. We need to keep flags such as how we allocated the buffer memory | |
657 | * intact here. | |
658 | */ | |
ce8e922c NS |
659 | if (bp->b_flags & XBF_STALE) { |
660 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
611c9946 | 661 | bp->b_flags &= _XBF_KMEM | _XBF_PAGES; |
1813dd64 | 662 | bp->b_ops = NULL; |
2f926587 | 663 | } |
0b1b213f CH |
664 | |
665 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ff6d6af2 | 666 | XFS_STATS_INC(btp->bt_mount, xb_get_locked); |
b027d4c9 DC |
667 | *found_bp = bp; |
668 | return 0; | |
1da177e4 LT |
669 | } |
670 | ||
8925a3dc DC |
671 | struct xfs_buf * |
672 | xfs_buf_incore( | |
673 | struct xfs_buftarg *target, | |
674 | xfs_daddr_t blkno, | |
675 | size_t numblks, | |
676 | xfs_buf_flags_t flags) | |
677 | { | |
b027d4c9 DC |
678 | struct xfs_buf *bp; |
679 | int error; | |
8925a3dc | 680 | DEFINE_SINGLE_BUF_MAP(map, blkno, numblks); |
b027d4c9 DC |
681 | |
682 | error = xfs_buf_find(target, &map, 1, flags, NULL, &bp); | |
683 | if (error) | |
684 | return NULL; | |
685 | return bp; | |
8925a3dc DC |
686 | } |
687 | ||
1da177e4 | 688 | /* |
3815832a DC |
689 | * Assembles a buffer covering the specified range. The code is optimised for |
690 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
691 | * more hits than misses. | |
1da177e4 | 692 | */ |
3848b5f6 | 693 | int |
6dde2707 DC |
694 | xfs_buf_get_map( |
695 | struct xfs_buftarg *target, | |
696 | struct xfs_buf_map *map, | |
697 | int nmaps, | |
3848b5f6 DW |
698 | xfs_buf_flags_t flags, |
699 | struct xfs_buf **bpp) | |
1da177e4 | 700 | { |
3815832a DC |
701 | struct xfs_buf *bp; |
702 | struct xfs_buf *new_bp; | |
0e6e847f | 703 | int error = 0; |
1da177e4 | 704 | |
3848b5f6 | 705 | *bpp = NULL; |
b027d4c9 | 706 | error = xfs_buf_find(target, map, nmaps, flags, NULL, &bp); |
3848b5f6 | 707 | if (!error) |
3815832a | 708 | goto found; |
3848b5f6 DW |
709 | if (error != -ENOENT) |
710 | return error; | |
3815832a | 711 | |
32dff5e5 DW |
712 | error = _xfs_buf_alloc(target, map, nmaps, flags, &new_bp); |
713 | if (error) | |
3848b5f6 | 714 | return error; |
1da177e4 | 715 | |
fe2429b0 DC |
716 | error = xfs_buf_allocate_memory(new_bp, flags); |
717 | if (error) { | |
3e85c868 | 718 | xfs_buf_free(new_bp); |
3848b5f6 | 719 | return error; |
fe2429b0 DC |
720 | } |
721 | ||
b027d4c9 DC |
722 | error = xfs_buf_find(target, map, nmaps, flags, new_bp, &bp); |
723 | if (error) { | |
fe2429b0 | 724 | xfs_buf_free(new_bp); |
3848b5f6 | 725 | return error; |
3815832a DC |
726 | } |
727 | ||
fe2429b0 DC |
728 | if (bp != new_bp) |
729 | xfs_buf_free(new_bp); | |
1da177e4 | 730 | |
3815832a | 731 | found: |
611c9946 | 732 | if (!bp->b_addr) { |
ce8e922c | 733 | error = _xfs_buf_map_pages(bp, flags); |
1da177e4 | 734 | if (unlikely(error)) { |
93baa55a DW |
735 | xfs_warn_ratelimited(target->bt_mount, |
736 | "%s: failed to map %u pages", __func__, | |
737 | bp->b_page_count); | |
a8acad70 | 738 | xfs_buf_relse(bp); |
3848b5f6 | 739 | return error; |
1da177e4 LT |
740 | } |
741 | } | |
742 | ||
b79f4a1c DC |
743 | /* |
744 | * Clear b_error if this is a lookup from a caller that doesn't expect | |
745 | * valid data to be found in the buffer. | |
746 | */ | |
747 | if (!(flags & XBF_READ)) | |
748 | xfs_buf_ioerror(bp, 0); | |
749 | ||
ff6d6af2 | 750 | XFS_STATS_INC(target->bt_mount, xb_get); |
0b1b213f | 751 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
3848b5f6 DW |
752 | *bpp = bp; |
753 | return 0; | |
1da177e4 LT |
754 | } |
755 | ||
5d765b97 CH |
756 | STATIC int |
757 | _xfs_buf_read( | |
758 | xfs_buf_t *bp, | |
759 | xfs_buf_flags_t flags) | |
760 | { | |
43ff2122 | 761 | ASSERT(!(flags & XBF_WRITE)); |
f4b42421 | 762 | ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL); |
5d765b97 | 763 | |
43ff2122 | 764 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD); |
1d5ae5df | 765 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 766 | |
6af88cda | 767 | return xfs_buf_submit(bp); |
5d765b97 CH |
768 | } |
769 | ||
1aff5696 | 770 | /* |
75d02303 | 771 | * Reverify a buffer found in cache without an attached ->b_ops. |
add46b3b | 772 | * |
75d02303 BF |
773 | * If the caller passed an ops structure and the buffer doesn't have ops |
774 | * assigned, set the ops and use it to verify the contents. If verification | |
775 | * fails, clear XBF_DONE. We assume the buffer has no recorded errors and is | |
776 | * already in XBF_DONE state on entry. | |
add46b3b | 777 | * |
75d02303 BF |
778 | * Under normal operations, every in-core buffer is verified on read I/O |
779 | * completion. There are two scenarios that can lead to in-core buffers without | |
780 | * an assigned ->b_ops. The first is during log recovery of buffers on a V4 | |
781 | * filesystem, though these buffers are purged at the end of recovery. The | |
782 | * other is online repair, which intentionally reads with a NULL buffer ops to | |
783 | * run several verifiers across an in-core buffer in order to establish buffer | |
784 | * type. If repair can't establish that, the buffer will be left in memory | |
785 | * with NULL buffer ops. | |
1aff5696 DW |
786 | */ |
787 | int | |
75d02303 | 788 | xfs_buf_reverify( |
1aff5696 DW |
789 | struct xfs_buf *bp, |
790 | const struct xfs_buf_ops *ops) | |
791 | { | |
792 | ASSERT(bp->b_flags & XBF_DONE); | |
793 | ASSERT(bp->b_error == 0); | |
794 | ||
795 | if (!ops || bp->b_ops) | |
796 | return 0; | |
797 | ||
798 | bp->b_ops = ops; | |
799 | bp->b_ops->verify_read(bp); | |
800 | if (bp->b_error) | |
801 | bp->b_flags &= ~XBF_DONE; | |
802 | return bp->b_error; | |
803 | } | |
804 | ||
4ed8e27b | 805 | int |
6dde2707 DC |
806 | xfs_buf_read_map( |
807 | struct xfs_buftarg *target, | |
808 | struct xfs_buf_map *map, | |
809 | int nmaps, | |
c3f8fc73 | 810 | xfs_buf_flags_t flags, |
4ed8e27b | 811 | struct xfs_buf **bpp, |
cdbcf82b DW |
812 | const struct xfs_buf_ops *ops, |
813 | xfs_failaddr_t fa) | |
1da177e4 | 814 | { |
6dde2707 | 815 | struct xfs_buf *bp; |
3848b5f6 | 816 | int error; |
ce8e922c NS |
817 | |
818 | flags |= XBF_READ; | |
4ed8e27b | 819 | *bpp = NULL; |
ce8e922c | 820 | |
3848b5f6 DW |
821 | error = xfs_buf_get_map(target, map, nmaps, flags, &bp); |
822 | if (error) | |
4ed8e27b | 823 | return error; |
0b1b213f | 824 | |
1aff5696 DW |
825 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
826 | ||
827 | if (!(bp->b_flags & XBF_DONE)) { | |
4ed8e27b | 828 | /* Initiate the buffer read and wait. */ |
1aff5696 DW |
829 | XFS_STATS_INC(target->bt_mount, xb_get_read); |
830 | bp->b_ops = ops; | |
4ed8e27b DW |
831 | error = _xfs_buf_read(bp, flags); |
832 | ||
833 | /* Readahead iodone already dropped the buffer, so exit. */ | |
834 | if (flags & XBF_ASYNC) | |
835 | return 0; | |
836 | } else { | |
837 | /* Buffer already read; all we need to do is check it. */ | |
838 | error = xfs_buf_reverify(bp, ops); | |
839 | ||
840 | /* Readahead already finished; drop the buffer and exit. */ | |
841 | if (flags & XBF_ASYNC) { | |
842 | xfs_buf_relse(bp); | |
843 | return 0; | |
844 | } | |
845 | ||
846 | /* We do not want read in the flags */ | |
847 | bp->b_flags &= ~XBF_READ; | |
848 | ASSERT(bp->b_ops != NULL || ops == NULL); | |
1aff5696 DW |
849 | } |
850 | ||
4ed8e27b DW |
851 | /* |
852 | * If we've had a read error, then the contents of the buffer are | |
853 | * invalid and should not be used. To ensure that a followup read tries | |
854 | * to pull the buffer from disk again, we clear the XBF_DONE flag and | |
855 | * mark the buffer stale. This ensures that anyone who has a current | |
856 | * reference to the buffer will interpret it's contents correctly and | |
857 | * future cache lookups will also treat it as an empty, uninitialised | |
858 | * buffer. | |
859 | */ | |
860 | if (error) { | |
861 | if (!XFS_FORCED_SHUTDOWN(target->bt_mount)) | |
cdbcf82b | 862 | xfs_buf_ioerror_alert(bp, fa); |
1aff5696 | 863 | |
4ed8e27b DW |
864 | bp->b_flags &= ~XBF_DONE; |
865 | xfs_buf_stale(bp); | |
1aff5696 | 866 | xfs_buf_relse(bp); |
4ed8e27b DW |
867 | |
868 | /* bad CRC means corrupted metadata */ | |
869 | if (error == -EFSBADCRC) | |
870 | error = -EFSCORRUPTED; | |
871 | return error; | |
1da177e4 LT |
872 | } |
873 | ||
4ed8e27b DW |
874 | *bpp = bp; |
875 | return 0; | |
1da177e4 LT |
876 | } |
877 | ||
1da177e4 | 878 | /* |
ce8e922c NS |
879 | * If we are not low on memory then do the readahead in a deadlock |
880 | * safe manner. | |
1da177e4 LT |
881 | */ |
882 | void | |
6dde2707 DC |
883 | xfs_buf_readahead_map( |
884 | struct xfs_buftarg *target, | |
885 | struct xfs_buf_map *map, | |
c3f8fc73 | 886 | int nmaps, |
1813dd64 | 887 | const struct xfs_buf_ops *ops) |
1da177e4 | 888 | { |
4ed8e27b DW |
889 | struct xfs_buf *bp; |
890 | ||
efa7c9f9 | 891 | if (bdi_read_congested(target->bt_bdev->bd_bdi)) |
1da177e4 LT |
892 | return; |
893 | ||
6dde2707 | 894 | xfs_buf_read_map(target, map, nmaps, |
cdbcf82b DW |
895 | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD, &bp, ops, |
896 | __this_address); | |
1da177e4 LT |
897 | } |
898 | ||
5adc94c2 DC |
899 | /* |
900 | * Read an uncached buffer from disk. Allocates and returns a locked | |
901 | * buffer containing the disk contents or nothing. | |
902 | */ | |
ba372674 | 903 | int |
5adc94c2 | 904 | xfs_buf_read_uncached( |
5adc94c2 DC |
905 | struct xfs_buftarg *target, |
906 | xfs_daddr_t daddr, | |
e70b73f8 | 907 | size_t numblks, |
c3f8fc73 | 908 | int flags, |
ba372674 | 909 | struct xfs_buf **bpp, |
1813dd64 | 910 | const struct xfs_buf_ops *ops) |
5adc94c2 | 911 | { |
eab4e633 | 912 | struct xfs_buf *bp; |
2842b6db | 913 | int error; |
5adc94c2 | 914 | |
ba372674 DC |
915 | *bpp = NULL; |
916 | ||
2842b6db DW |
917 | error = xfs_buf_get_uncached(target, numblks, flags, &bp); |
918 | if (error) | |
919 | return error; | |
5adc94c2 DC |
920 | |
921 | /* set up the buffer for a read IO */ | |
3e85c868 | 922 | ASSERT(bp->b_map_count == 1); |
ba372674 | 923 | bp->b_bn = XFS_BUF_DADDR_NULL; /* always null for uncached buffers */ |
3e85c868 | 924 | bp->b_maps[0].bm_bn = daddr; |
cbb7baab | 925 | bp->b_flags |= XBF_READ; |
1813dd64 | 926 | bp->b_ops = ops; |
5adc94c2 | 927 | |
6af88cda | 928 | xfs_buf_submit(bp); |
ba372674 | 929 | if (bp->b_error) { |
2842b6db | 930 | error = bp->b_error; |
83a0adc3 | 931 | xfs_buf_relse(bp); |
ba372674 | 932 | return error; |
83a0adc3 | 933 | } |
ba372674 DC |
934 | |
935 | *bpp = bp; | |
936 | return 0; | |
1da177e4 LT |
937 | } |
938 | ||
2842b6db | 939 | int |
686865f7 DC |
940 | xfs_buf_get_uncached( |
941 | struct xfs_buftarg *target, | |
e70b73f8 | 942 | size_t numblks, |
2842b6db DW |
943 | int flags, |
944 | struct xfs_buf **bpp) | |
1da177e4 | 945 | { |
e70b73f8 | 946 | unsigned long page_count; |
1fa40b01 | 947 | int error, i; |
3e85c868 DC |
948 | struct xfs_buf *bp; |
949 | DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks); | |
1da177e4 | 950 | |
2842b6db DW |
951 | *bpp = NULL; |
952 | ||
c891c30a | 953 | /* flags might contain irrelevant bits, pass only what we care about */ |
32dff5e5 DW |
954 | error = _xfs_buf_alloc(target, &map, 1, flags & XBF_NO_IOACCT, &bp); |
955 | if (error) | |
1da177e4 | 956 | goto fail; |
1da177e4 | 957 | |
e70b73f8 | 958 | page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT; |
87937bf8 | 959 | error = _xfs_buf_get_pages(bp, page_count); |
1fa40b01 | 960 | if (error) |
1da177e4 LT |
961 | goto fail_free_buf; |
962 | ||
1fa40b01 | 963 | for (i = 0; i < page_count; i++) { |
686865f7 | 964 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
2842b6db DW |
965 | if (!bp->b_pages[i]) { |
966 | error = -ENOMEM; | |
1fa40b01 | 967 | goto fail_free_mem; |
2842b6db | 968 | } |
1da177e4 | 969 | } |
1fa40b01 | 970 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 971 | |
611c9946 | 972 | error = _xfs_buf_map_pages(bp, 0); |
1fa40b01 | 973 | if (unlikely(error)) { |
4f10700a | 974 | xfs_warn(target->bt_mount, |
08e96e1a | 975 | "%s: failed to map pages", __func__); |
1da177e4 | 976 | goto fail_free_mem; |
1fa40b01 | 977 | } |
1da177e4 | 978 | |
686865f7 | 979 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
2842b6db DW |
980 | *bpp = bp; |
981 | return 0; | |
1fa40b01 | 982 | |
1da177e4 | 983 | fail_free_mem: |
1fa40b01 CH |
984 | while (--i >= 0) |
985 | __free_page(bp->b_pages[i]); | |
ca165b88 | 986 | _xfs_buf_free_pages(bp); |
1da177e4 | 987 | fail_free_buf: |
3e85c868 | 988 | xfs_buf_free_maps(bp); |
377bcd5f | 989 | kmem_cache_free(xfs_buf_zone, bp); |
1da177e4 | 990 | fail: |
2842b6db | 991 | return error; |
1da177e4 LT |
992 | } |
993 | ||
994 | /* | |
1da177e4 LT |
995 | * Increment reference count on buffer, to hold the buffer concurrently |
996 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
997 | * Must hold the buffer already to call this function. |
998 | */ | |
999 | void | |
ce8e922c NS |
1000 | xfs_buf_hold( |
1001 | xfs_buf_t *bp) | |
1da177e4 | 1002 | { |
0b1b213f | 1003 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 1004 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
1005 | } |
1006 | ||
1007 | /* | |
9c7504aa BF |
1008 | * Release a hold on the specified buffer. If the hold count is 1, the buffer is |
1009 | * placed on LRU or freed (depending on b_lru_ref). | |
1da177e4 LT |
1010 | */ |
1011 | void | |
ce8e922c NS |
1012 | xfs_buf_rele( |
1013 | xfs_buf_t *bp) | |
1da177e4 | 1014 | { |
74f75a0c | 1015 | struct xfs_perag *pag = bp->b_pag; |
9c7504aa BF |
1016 | bool release; |
1017 | bool freebuf = false; | |
1da177e4 | 1018 | |
0b1b213f | 1019 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 1020 | |
74f75a0c | 1021 | if (!pag) { |
430cbeb8 | 1022 | ASSERT(list_empty(&bp->b_lru)); |
9c7504aa BF |
1023 | if (atomic_dec_and_test(&bp->b_hold)) { |
1024 | xfs_buf_ioacct_dec(bp); | |
fad3aa1e | 1025 | xfs_buf_free(bp); |
9c7504aa | 1026 | } |
fad3aa1e NS |
1027 | return; |
1028 | } | |
1029 | ||
3790689f | 1030 | ASSERT(atomic_read(&bp->b_hold) > 0); |
a4082357 | 1031 | |
37fd1678 DC |
1032 | /* |
1033 | * We grab the b_lock here first to serialise racing xfs_buf_rele() | |
1034 | * calls. The pag_buf_lock being taken on the last reference only | |
1035 | * serialises against racing lookups in xfs_buf_find(). IOWs, the second | |
1036 | * to last reference we drop here is not serialised against the last | |
1037 | * reference until we take bp->b_lock. Hence if we don't grab b_lock | |
1038 | * first, the last "release" reference can win the race to the lock and | |
1039 | * free the buffer before the second-to-last reference is processed, | |
1040 | * leading to a use-after-free scenario. | |
1041 | */ | |
9c7504aa | 1042 | spin_lock(&bp->b_lock); |
37fd1678 | 1043 | release = atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock); |
9c7504aa BF |
1044 | if (!release) { |
1045 | /* | |
1046 | * Drop the in-flight state if the buffer is already on the LRU | |
1047 | * and it holds the only reference. This is racy because we | |
1048 | * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT | |
1049 | * ensures the decrement occurs only once per-buf. | |
1050 | */ | |
1051 | if ((atomic_read(&bp->b_hold) == 1) && !list_empty(&bp->b_lru)) | |
63db7c81 | 1052 | __xfs_buf_ioacct_dec(bp); |
9c7504aa BF |
1053 | goto out_unlock; |
1054 | } | |
1055 | ||
1056 | /* the last reference has been dropped ... */ | |
63db7c81 | 1057 | __xfs_buf_ioacct_dec(bp); |
9c7504aa BF |
1058 | if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) { |
1059 | /* | |
1060 | * If the buffer is added to the LRU take a new reference to the | |
1061 | * buffer for the LRU and clear the (now stale) dispose list | |
1062 | * state flag | |
1063 | */ | |
1064 | if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) { | |
1065 | bp->b_state &= ~XFS_BSTATE_DISPOSE; | |
1066 | atomic_inc(&bp->b_hold); | |
1da177e4 | 1067 | } |
9c7504aa BF |
1068 | spin_unlock(&pag->pag_buf_lock); |
1069 | } else { | |
1070 | /* | |
1071 | * most of the time buffers will already be removed from the | |
1072 | * LRU, so optimise that case by checking for the | |
1073 | * XFS_BSTATE_DISPOSE flag indicating the last list the buffer | |
1074 | * was on was the disposal list | |
1075 | */ | |
1076 | if (!(bp->b_state & XFS_BSTATE_DISPOSE)) { | |
1077 | list_lru_del(&bp->b_target->bt_lru, &bp->b_lru); | |
1078 | } else { | |
1079 | ASSERT(list_empty(&bp->b_lru)); | |
1da177e4 | 1080 | } |
9c7504aa BF |
1081 | |
1082 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); | |
6031e73a LS |
1083 | rhashtable_remove_fast(&pag->pag_buf_hash, &bp->b_rhash_head, |
1084 | xfs_buf_hash_params); | |
9c7504aa BF |
1085 | spin_unlock(&pag->pag_buf_lock); |
1086 | xfs_perag_put(pag); | |
1087 | freebuf = true; | |
1da177e4 | 1088 | } |
9c7504aa BF |
1089 | |
1090 | out_unlock: | |
1091 | spin_unlock(&bp->b_lock); | |
1092 | ||
1093 | if (freebuf) | |
1094 | xfs_buf_free(bp); | |
1da177e4 LT |
1095 | } |
1096 | ||
1097 | ||
1098 | /* | |
0e6e847f | 1099 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
1100 | * |
1101 | * If we come across a stale, pinned, locked buffer, we know that we are | |
1102 | * being asked to lock a buffer that has been reallocated. Because it is | |
1103 | * pinned, we know that the log has not been pushed to disk and hence it | |
1104 | * will still be locked. Rather than continuing to have trylock attempts | |
1105 | * fail until someone else pushes the log, push it ourselves before | |
1106 | * returning. This means that the xfsaild will not get stuck trying | |
1107 | * to push on stale inode buffers. | |
1da177e4 LT |
1108 | */ |
1109 | int | |
0c842ad4 CH |
1110 | xfs_buf_trylock( |
1111 | struct xfs_buf *bp) | |
1da177e4 LT |
1112 | { |
1113 | int locked; | |
1114 | ||
ce8e922c | 1115 | locked = down_trylock(&bp->b_sema) == 0; |
fa6c668d | 1116 | if (locked) |
479c6412 | 1117 | trace_xfs_buf_trylock(bp, _RET_IP_); |
fa6c668d | 1118 | else |
479c6412 | 1119 | trace_xfs_buf_trylock_fail(bp, _RET_IP_); |
0c842ad4 | 1120 | return locked; |
1da177e4 | 1121 | } |
1da177e4 LT |
1122 | |
1123 | /* | |
0e6e847f | 1124 | * Lock a buffer object. |
ed3b4d6c DC |
1125 | * |
1126 | * If we come across a stale, pinned, locked buffer, we know that we | |
1127 | * are being asked to lock a buffer that has been reallocated. Because | |
1128 | * it is pinned, we know that the log has not been pushed to disk and | |
1129 | * hence it will still be locked. Rather than sleeping until someone | |
1130 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 1131 | */ |
ce8e922c NS |
1132 | void |
1133 | xfs_buf_lock( | |
0c842ad4 | 1134 | struct xfs_buf *bp) |
1da177e4 | 1135 | { |
0b1b213f CH |
1136 | trace_xfs_buf_lock(bp, _RET_IP_); |
1137 | ||
ed3b4d6c | 1138 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
dbd329f1 | 1139 | xfs_log_force(bp->b_mount, 0); |
ce8e922c | 1140 | down(&bp->b_sema); |
0b1b213f CH |
1141 | |
1142 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
1143 | } |
1144 | ||
1da177e4 | 1145 | void |
ce8e922c | 1146 | xfs_buf_unlock( |
0c842ad4 | 1147 | struct xfs_buf *bp) |
1da177e4 | 1148 | { |
20e8a063 BF |
1149 | ASSERT(xfs_buf_islocked(bp)); |
1150 | ||
ce8e922c | 1151 | up(&bp->b_sema); |
0b1b213f | 1152 | trace_xfs_buf_unlock(bp, _RET_IP_); |
1da177e4 LT |
1153 | } |
1154 | ||
ce8e922c NS |
1155 | STATIC void |
1156 | xfs_buf_wait_unpin( | |
1157 | xfs_buf_t *bp) | |
1da177e4 LT |
1158 | { |
1159 | DECLARE_WAITQUEUE (wait, current); | |
1160 | ||
ce8e922c | 1161 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
1162 | return; |
1163 | ||
ce8e922c | 1164 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1165 | for (;;) { |
1166 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 1167 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 1168 | break; |
7eaceacc | 1169 | io_schedule(); |
1da177e4 | 1170 | } |
ce8e922c | 1171 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1172 | set_current_state(TASK_RUNNING); |
1173 | } | |
1174 | ||
1175 | /* | |
1176 | * Buffer Utility Routines | |
1177 | */ | |
1178 | ||
e8aaba9a DC |
1179 | void |
1180 | xfs_buf_ioend( | |
1181 | struct xfs_buf *bp) | |
1da177e4 | 1182 | { |
e8aaba9a DC |
1183 | bool read = bp->b_flags & XBF_READ; |
1184 | ||
1185 | trace_xfs_buf_iodone(bp, _RET_IP_); | |
1813dd64 DC |
1186 | |
1187 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); | |
d5929de8 | 1188 | |
61be9c52 DC |
1189 | /* |
1190 | * Pull in IO completion errors now. We are guaranteed to be running | |
1191 | * single threaded, so we don't need the lock to read b_io_error. | |
1192 | */ | |
1193 | if (!bp->b_error && bp->b_io_error) | |
1194 | xfs_buf_ioerror(bp, bp->b_io_error); | |
1195 | ||
b01d1461 DC |
1196 | if (read) { |
1197 | if (!bp->b_error && bp->b_ops) | |
1198 | bp->b_ops->verify_read(bp); | |
1199 | if (!bp->b_error) | |
1200 | bp->b_flags |= XBF_DONE; | |
1201 | xfs_buf_ioend_finish(bp); | |
1202 | return; | |
e8aaba9a DC |
1203 | } |
1204 | ||
b6983e80 BF |
1205 | if (!bp->b_error) { |
1206 | bp->b_flags &= ~XBF_WRITE_FAIL; | |
e8aaba9a | 1207 | bp->b_flags |= XBF_DONE; |
b6983e80 | 1208 | } |
1da177e4 | 1209 | |
9fe5c77c DC |
1210 | /* |
1211 | * If this is a log recovery buffer, we aren't doing transactional IO | |
1212 | * yet so we need to let it handle IO completions. | |
1213 | */ | |
1214 | if (bp->b_flags & _XBF_LOGRECOVERY) { | |
1215 | xlog_recover_iodone(bp); | |
1216 | return; | |
1217 | } | |
1218 | ||
f593bf14 DC |
1219 | if (bp->b_flags & _XBF_INODES) { |
1220 | xfs_buf_inode_iodone(bp); | |
1221 | return; | |
1222 | } | |
1223 | ||
0c7e5afb DC |
1224 | if (bp->b_flags & _XBF_DQUOTS) { |
1225 | xfs_buf_dquot_iodone(bp); | |
1226 | return; | |
1227 | } | |
b01d1461 | 1228 | xfs_buf_iodone(bp); |
1da177e4 LT |
1229 | } |
1230 | ||
e8aaba9a DC |
1231 | static void |
1232 | xfs_buf_ioend_work( | |
1233 | struct work_struct *work) | |
1da177e4 | 1234 | { |
e8aaba9a | 1235 | struct xfs_buf *bp = |
b29c70f5 | 1236 | container_of(work, xfs_buf_t, b_ioend_work); |
0b1b213f | 1237 | |
e8aaba9a DC |
1238 | xfs_buf_ioend(bp); |
1239 | } | |
1da177e4 | 1240 | |
211fe1a4 | 1241 | static void |
e8aaba9a DC |
1242 | xfs_buf_ioend_async( |
1243 | struct xfs_buf *bp) | |
1244 | { | |
b29c70f5 | 1245 | INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work); |
dbd329f1 | 1246 | queue_work(bp->b_mount->m_buf_workqueue, &bp->b_ioend_work); |
1da177e4 LT |
1247 | } |
1248 | ||
1da177e4 | 1249 | void |
31ca03c9 | 1250 | __xfs_buf_ioerror( |
ce8e922c | 1251 | xfs_buf_t *bp, |
31ca03c9 DW |
1252 | int error, |
1253 | xfs_failaddr_t failaddr) | |
1da177e4 | 1254 | { |
2451337d DC |
1255 | ASSERT(error <= 0 && error >= -1000); |
1256 | bp->b_error = error; | |
31ca03c9 | 1257 | trace_xfs_buf_ioerror(bp, error, failaddr); |
1da177e4 LT |
1258 | } |
1259 | ||
901796af CH |
1260 | void |
1261 | xfs_buf_ioerror_alert( | |
1262 | struct xfs_buf *bp, | |
cdbcf82b | 1263 | xfs_failaddr_t func) |
901796af | 1264 | { |
f9bccfcc BF |
1265 | xfs_buf_alert_ratelimited(bp, "XFS: metadata IO error", |
1266 | "metadata I/O error in \"%pS\" at daddr 0x%llx len %d error %d", | |
1267 | func, (uint64_t)XFS_BUF_ADDR(bp), | |
1268 | bp->b_length, -bp->b_error); | |
901796af CH |
1269 | } |
1270 | ||
54b3b1f6 BF |
1271 | /* |
1272 | * To simulate an I/O failure, the buffer must be locked and held with at least | |
1273 | * three references. The LRU reference is dropped by the stale call. The buf | |
1274 | * item reference is dropped via ioend processing. The third reference is owned | |
1275 | * by the caller and is dropped on I/O completion if the buffer is XBF_ASYNC. | |
1276 | */ | |
1277 | void | |
1278 | xfs_buf_ioend_fail( | |
1279 | struct xfs_buf *bp) | |
1280 | { | |
1281 | bp->b_flags &= ~XBF_DONE; | |
1282 | xfs_buf_stale(bp); | |
1283 | xfs_buf_ioerror(bp, -EIO); | |
1284 | xfs_buf_ioend(bp); | |
901796af CH |
1285 | } |
1286 | ||
a2dcf5df CH |
1287 | int |
1288 | xfs_bwrite( | |
1289 | struct xfs_buf *bp) | |
1290 | { | |
1291 | int error; | |
1292 | ||
1293 | ASSERT(xfs_buf_islocked(bp)); | |
1294 | ||
1295 | bp->b_flags |= XBF_WRITE; | |
27187754 | 1296 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q | |
b6983e80 | 1297 | XBF_DONE); |
a2dcf5df | 1298 | |
6af88cda | 1299 | error = xfs_buf_submit(bp); |
dbd329f1 CH |
1300 | if (error) |
1301 | xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR); | |
a2dcf5df CH |
1302 | return error; |
1303 | } | |
1304 | ||
9bdd9bd6 | 1305 | static void |
ce8e922c | 1306 | xfs_buf_bio_end_io( |
4246a0b6 | 1307 | struct bio *bio) |
1da177e4 | 1308 | { |
9bdd9bd6 | 1309 | struct xfs_buf *bp = (struct xfs_buf *)bio->bi_private; |
1da177e4 | 1310 | |
7376d745 BF |
1311 | if (!bio->bi_status && |
1312 | (bp->b_flags & XBF_WRITE) && (bp->b_flags & XBF_ASYNC) && | |
43dc0aa8 | 1313 | XFS_TEST_ERROR(false, bp->b_mount, XFS_ERRTAG_BUF_IOERROR)) |
7376d745 | 1314 | bio->bi_status = BLK_STS_IOERR; |
1da177e4 | 1315 | |
37eb17e6 DC |
1316 | /* |
1317 | * don't overwrite existing errors - otherwise we can lose errors on | |
1318 | * buffers that require multiple bios to complete. | |
1319 | */ | |
4e4cbee9 CH |
1320 | if (bio->bi_status) { |
1321 | int error = blk_status_to_errno(bio->bi_status); | |
1322 | ||
1323 | cmpxchg(&bp->b_io_error, 0, error); | |
1324 | } | |
1da177e4 | 1325 | |
37eb17e6 | 1326 | if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
73c77e2c JB |
1327 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); |
1328 | ||
e8aaba9a DC |
1329 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
1330 | xfs_buf_ioend_async(bp); | |
1da177e4 | 1331 | bio_put(bio); |
1da177e4 LT |
1332 | } |
1333 | ||
3e85c868 DC |
1334 | static void |
1335 | xfs_buf_ioapply_map( | |
1336 | struct xfs_buf *bp, | |
1337 | int map, | |
1338 | int *buf_offset, | |
1339 | int *count, | |
2123ef85 | 1340 | int op) |
1da177e4 | 1341 | { |
3e85c868 DC |
1342 | int page_index; |
1343 | int total_nr_pages = bp->b_page_count; | |
1344 | int nr_pages; | |
1345 | struct bio *bio; | |
1346 | sector_t sector = bp->b_maps[map].bm_bn; | |
1347 | int size; | |
1348 | int offset; | |
1da177e4 | 1349 | |
3e85c868 DC |
1350 | /* skip the pages in the buffer before the start offset */ |
1351 | page_index = 0; | |
1352 | offset = *buf_offset; | |
1353 | while (offset >= PAGE_SIZE) { | |
1354 | page_index++; | |
1355 | offset -= PAGE_SIZE; | |
f538d4da CH |
1356 | } |
1357 | ||
3e85c868 DC |
1358 | /* |
1359 | * Limit the IO size to the length of the current vector, and update the | |
1360 | * remaining IO count for the next time around. | |
1361 | */ | |
1362 | size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count); | |
1363 | *count -= size; | |
1364 | *buf_offset += size; | |
34951f5c | 1365 | |
1da177e4 | 1366 | next_chunk: |
ce8e922c | 1367 | atomic_inc(&bp->b_io_remaining); |
c908e380 | 1368 | nr_pages = min(total_nr_pages, BIO_MAX_PAGES); |
1da177e4 LT |
1369 | |
1370 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
74d46992 | 1371 | bio_set_dev(bio, bp->b_target->bt_bdev); |
4f024f37 | 1372 | bio->bi_iter.bi_sector = sector; |
ce8e922c NS |
1373 | bio->bi_end_io = xfs_buf_bio_end_io; |
1374 | bio->bi_private = bp; | |
2123ef85 | 1375 | bio->bi_opf = op; |
0e6e847f | 1376 | |
3e85c868 | 1377 | for (; size && nr_pages; nr_pages--, page_index++) { |
0e6e847f | 1378 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1379 | |
1380 | if (nbytes > size) | |
1381 | nbytes = size; | |
1382 | ||
3e85c868 DC |
1383 | rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes, |
1384 | offset); | |
ce8e922c | 1385 | if (rbytes < nbytes) |
1da177e4 LT |
1386 | break; |
1387 | ||
1388 | offset = 0; | |
aa0e8833 | 1389 | sector += BTOBB(nbytes); |
1da177e4 LT |
1390 | size -= nbytes; |
1391 | total_nr_pages--; | |
1392 | } | |
1393 | ||
4f024f37 | 1394 | if (likely(bio->bi_iter.bi_size)) { |
73c77e2c JB |
1395 | if (xfs_buf_is_vmapped(bp)) { |
1396 | flush_kernel_vmap_range(bp->b_addr, | |
1397 | xfs_buf_vmap_len(bp)); | |
1398 | } | |
4e49ea4a | 1399 | submit_bio(bio); |
1da177e4 LT |
1400 | if (size) |
1401 | goto next_chunk; | |
1402 | } else { | |
37eb17e6 DC |
1403 | /* |
1404 | * This is guaranteed not to be the last io reference count | |
595bff75 | 1405 | * because the caller (xfs_buf_submit) holds a count itself. |
37eb17e6 DC |
1406 | */ |
1407 | atomic_dec(&bp->b_io_remaining); | |
2451337d | 1408 | xfs_buf_ioerror(bp, -EIO); |
ec53d1db | 1409 | bio_put(bio); |
1da177e4 | 1410 | } |
3e85c868 DC |
1411 | |
1412 | } | |
1413 | ||
1414 | STATIC void | |
1415 | _xfs_buf_ioapply( | |
1416 | struct xfs_buf *bp) | |
1417 | { | |
1418 | struct blk_plug plug; | |
50bfcd0c | 1419 | int op; |
3e85c868 DC |
1420 | int offset; |
1421 | int size; | |
1422 | int i; | |
1423 | ||
c163f9a1 DC |
1424 | /* |
1425 | * Make sure we capture only current IO errors rather than stale errors | |
1426 | * left over from previous use of the buffer (e.g. failed readahead). | |
1427 | */ | |
1428 | bp->b_error = 0; | |
1429 | ||
3e85c868 | 1430 | if (bp->b_flags & XBF_WRITE) { |
50bfcd0c | 1431 | op = REQ_OP_WRITE; |
1813dd64 DC |
1432 | |
1433 | /* | |
1434 | * Run the write verifier callback function if it exists. If | |
1435 | * this function fails it will mark the buffer with an error and | |
1436 | * the IO should not be dispatched. | |
1437 | */ | |
1438 | if (bp->b_ops) { | |
1439 | bp->b_ops->verify_write(bp); | |
1440 | if (bp->b_error) { | |
dbd329f1 | 1441 | xfs_force_shutdown(bp->b_mount, |
1813dd64 DC |
1442 | SHUTDOWN_CORRUPT_INCORE); |
1443 | return; | |
1444 | } | |
400b9d88 | 1445 | } else if (bp->b_bn != XFS_BUF_DADDR_NULL) { |
dbd329f1 | 1446 | struct xfs_mount *mp = bp->b_mount; |
400b9d88 DC |
1447 | |
1448 | /* | |
1449 | * non-crc filesystems don't attach verifiers during | |
1450 | * log recovery, so don't warn for such filesystems. | |
1451 | */ | |
1452 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
1453 | xfs_warn(mp, | |
c219b015 | 1454 | "%s: no buf ops on daddr 0x%llx len %d", |
400b9d88 | 1455 | __func__, bp->b_bn, bp->b_length); |
9c712a13 DW |
1456 | xfs_hex_dump(bp->b_addr, |
1457 | XFS_CORRUPTION_DUMP_LEN); | |
400b9d88 DC |
1458 | dump_stack(); |
1459 | } | |
1813dd64 | 1460 | } |
3e85c868 | 1461 | } else { |
50bfcd0c | 1462 | op = REQ_OP_READ; |
2123ef85 CH |
1463 | if (bp->b_flags & XBF_READ_AHEAD) |
1464 | op |= REQ_RAHEAD; | |
3e85c868 DC |
1465 | } |
1466 | ||
1467 | /* we only use the buffer cache for meta-data */ | |
2123ef85 | 1468 | op |= REQ_META; |
3e85c868 DC |
1469 | |
1470 | /* | |
1471 | * Walk all the vectors issuing IO on them. Set up the initial offset | |
1472 | * into the buffer and the desired IO size before we start - | |
1473 | * _xfs_buf_ioapply_vec() will modify them appropriately for each | |
1474 | * subsequent call. | |
1475 | */ | |
1476 | offset = bp->b_offset; | |
8124b9b6 | 1477 | size = BBTOB(bp->b_length); |
3e85c868 DC |
1478 | blk_start_plug(&plug); |
1479 | for (i = 0; i < bp->b_map_count; i++) { | |
2123ef85 | 1480 | xfs_buf_ioapply_map(bp, i, &offset, &size, op); |
3e85c868 DC |
1481 | if (bp->b_error) |
1482 | break; | |
1483 | if (size <= 0) | |
1484 | break; /* all done */ | |
1485 | } | |
1486 | blk_finish_plug(&plug); | |
1da177e4 LT |
1487 | } |
1488 | ||
595bff75 | 1489 | /* |
bb00b6f1 | 1490 | * Wait for I/O completion of a sync buffer and return the I/O error code. |
595bff75 | 1491 | */ |
eaebb515 | 1492 | static int |
bb00b6f1 | 1493 | xfs_buf_iowait( |
595bff75 | 1494 | struct xfs_buf *bp) |
1da177e4 | 1495 | { |
bb00b6f1 BF |
1496 | ASSERT(!(bp->b_flags & XBF_ASYNC)); |
1497 | ||
1498 | trace_xfs_buf_iowait(bp, _RET_IP_); | |
1499 | wait_for_completion(&bp->b_iowait); | |
1500 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
1501 | ||
1502 | return bp->b_error; | |
1503 | } | |
1504 | ||
1505 | /* | |
1506 | * Buffer I/O submission path, read or write. Asynchronous submission transfers | |
1507 | * the buffer lock ownership and the current reference to the IO. It is not | |
1508 | * safe to reference the buffer after a call to this function unless the caller | |
1509 | * holds an additional reference itself. | |
1510 | */ | |
1511 | int | |
1512 | __xfs_buf_submit( | |
1513 | struct xfs_buf *bp, | |
1514 | bool wait) | |
1515 | { | |
1516 | int error = 0; | |
1517 | ||
595bff75 | 1518 | trace_xfs_buf_submit(bp, _RET_IP_); |
1da177e4 | 1519 | |
43ff2122 | 1520 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
595bff75 DC |
1521 | |
1522 | /* on shutdown we stale and complete the buffer immediately */ | |
dbd329f1 | 1523 | if (XFS_FORCED_SHUTDOWN(bp->b_mount)) { |
54b3b1f6 | 1524 | xfs_buf_ioend_fail(bp); |
eaebb515 | 1525 | return -EIO; |
595bff75 | 1526 | } |
1da177e4 | 1527 | |
bb00b6f1 BF |
1528 | /* |
1529 | * Grab a reference so the buffer does not go away underneath us. For | |
1530 | * async buffers, I/O completion drops the callers reference, which | |
1531 | * could occur before submission returns. | |
1532 | */ | |
1533 | xfs_buf_hold(bp); | |
1534 | ||
375ec69d | 1535 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1536 | xfs_buf_wait_unpin(bp); |
e11bb805 | 1537 | |
61be9c52 DC |
1538 | /* clear the internal error state to avoid spurious errors */ |
1539 | bp->b_io_error = 0; | |
1540 | ||
8d6c1210 | 1541 | /* |
e11bb805 DC |
1542 | * Set the count to 1 initially, this will stop an I/O completion |
1543 | * callout which happens before we have started all the I/O from calling | |
1544 | * xfs_buf_ioend too early. | |
1da177e4 | 1545 | */ |
ce8e922c | 1546 | atomic_set(&bp->b_io_remaining, 1); |
eaebb515 BF |
1547 | if (bp->b_flags & XBF_ASYNC) |
1548 | xfs_buf_ioacct_inc(bp); | |
ce8e922c | 1549 | _xfs_buf_ioapply(bp); |
e11bb805 | 1550 | |
8d6c1210 | 1551 | /* |
595bff75 DC |
1552 | * If _xfs_buf_ioapply failed, we can get back here with only the IO |
1553 | * reference we took above. If we drop it to zero, run completion so | |
1554 | * that we don't return to the caller with completion still pending. | |
8d6c1210 | 1555 | */ |
e8aaba9a | 1556 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
eaebb515 | 1557 | if (bp->b_error || !(bp->b_flags & XBF_ASYNC)) |
e8aaba9a DC |
1558 | xfs_buf_ioend(bp); |
1559 | else | |
1560 | xfs_buf_ioend_async(bp); | |
1561 | } | |
1da177e4 | 1562 | |
6af88cda BF |
1563 | if (wait) |
1564 | error = xfs_buf_iowait(bp); | |
bb00b6f1 | 1565 | |
595bff75 | 1566 | /* |
6af88cda BF |
1567 | * Release the hold that keeps the buffer referenced for the entire |
1568 | * I/O. Note that if the buffer is async, it is not safe to reference | |
1569 | * after this release. | |
595bff75 DC |
1570 | */ |
1571 | xfs_buf_rele(bp); | |
1572 | return error; | |
1da177e4 LT |
1573 | } |
1574 | ||
88ee2df7 | 1575 | void * |
ce8e922c | 1576 | xfs_buf_offset( |
88ee2df7 | 1577 | struct xfs_buf *bp, |
1da177e4 LT |
1578 | size_t offset) |
1579 | { | |
1580 | struct page *page; | |
1581 | ||
611c9946 | 1582 | if (bp->b_addr) |
62926044 | 1583 | return bp->b_addr + offset; |
1da177e4 | 1584 | |
ce8e922c | 1585 | offset += bp->b_offset; |
0e6e847f | 1586 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
88ee2df7 | 1587 | return page_address(page) + (offset & (PAGE_SIZE-1)); |
1da177e4 LT |
1588 | } |
1589 | ||
1da177e4 | 1590 | void |
f9a196ee CH |
1591 | xfs_buf_zero( |
1592 | struct xfs_buf *bp, | |
1593 | size_t boff, | |
1594 | size_t bsize) | |
1da177e4 | 1595 | { |
795cac72 | 1596 | size_t bend; |
1da177e4 LT |
1597 | |
1598 | bend = boff + bsize; | |
1599 | while (boff < bend) { | |
795cac72 DC |
1600 | struct page *page; |
1601 | int page_index, page_offset, csize; | |
1602 | ||
1603 | page_index = (boff + bp->b_offset) >> PAGE_SHIFT; | |
1604 | page_offset = (boff + bp->b_offset) & ~PAGE_MASK; | |
1605 | page = bp->b_pages[page_index]; | |
1606 | csize = min_t(size_t, PAGE_SIZE - page_offset, | |
8124b9b6 | 1607 | BBTOB(bp->b_length) - boff); |
1da177e4 | 1608 | |
795cac72 | 1609 | ASSERT((csize + page_offset) <= PAGE_SIZE); |
1da177e4 | 1610 | |
f9a196ee | 1611 | memset(page_address(page) + page_offset, 0, csize); |
1da177e4 LT |
1612 | |
1613 | boff += csize; | |
1da177e4 LT |
1614 | } |
1615 | } | |
1616 | ||
8d57c216 DW |
1617 | /* |
1618 | * Log a message about and stale a buffer that a caller has decided is corrupt. | |
1619 | * | |
1620 | * This function should be called for the kinds of metadata corruption that | |
1621 | * cannot be detect from a verifier, such as incorrect inter-block relationship | |
1622 | * data. Do /not/ call this function from a verifier function. | |
1623 | * | |
1624 | * The buffer must be XBF_DONE prior to the call. Afterwards, the buffer will | |
1625 | * be marked stale, but b_error will not be set. The caller is responsible for | |
1626 | * releasing the buffer or fixing it. | |
1627 | */ | |
1628 | void | |
1629 | __xfs_buf_mark_corrupt( | |
1630 | struct xfs_buf *bp, | |
1631 | xfs_failaddr_t fa) | |
1632 | { | |
1633 | ASSERT(bp->b_flags & XBF_DONE); | |
1634 | ||
e83cf875 | 1635 | xfs_buf_corruption_error(bp, fa); |
8d57c216 DW |
1636 | xfs_buf_stale(bp); |
1637 | } | |
1638 | ||
1da177e4 | 1639 | /* |
ce8e922c | 1640 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1641 | */ |
1642 | ||
1643 | /* | |
430cbeb8 DC |
1644 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1645 | * returned. These buffers will have an elevated hold count, so wait on those | |
1646 | * while freeing all the buffers only held by the LRU. | |
1da177e4 | 1647 | */ |
e80dfa19 DC |
1648 | static enum lru_status |
1649 | xfs_buftarg_wait_rele( | |
1650 | struct list_head *item, | |
3f97b163 | 1651 | struct list_lru_one *lru, |
e80dfa19 DC |
1652 | spinlock_t *lru_lock, |
1653 | void *arg) | |
1654 | ||
1da177e4 | 1655 | { |
e80dfa19 | 1656 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); |
a4082357 | 1657 | struct list_head *dispose = arg; |
430cbeb8 | 1658 | |
e80dfa19 | 1659 | if (atomic_read(&bp->b_hold) > 1) { |
a4082357 | 1660 | /* need to wait, so skip it this pass */ |
e80dfa19 | 1661 | trace_xfs_buf_wait_buftarg(bp, _RET_IP_); |
a4082357 | 1662 | return LRU_SKIP; |
1da177e4 | 1663 | } |
a4082357 DC |
1664 | if (!spin_trylock(&bp->b_lock)) |
1665 | return LRU_SKIP; | |
e80dfa19 | 1666 | |
a4082357 DC |
1667 | /* |
1668 | * clear the LRU reference count so the buffer doesn't get | |
1669 | * ignored in xfs_buf_rele(). | |
1670 | */ | |
1671 | atomic_set(&bp->b_lru_ref, 0); | |
1672 | bp->b_state |= XFS_BSTATE_DISPOSE; | |
3f97b163 | 1673 | list_lru_isolate_move(lru, item, dispose); |
a4082357 DC |
1674 | spin_unlock(&bp->b_lock); |
1675 | return LRU_REMOVED; | |
1da177e4 LT |
1676 | } |
1677 | ||
e80dfa19 DC |
1678 | void |
1679 | xfs_wait_buftarg( | |
1680 | struct xfs_buftarg *btp) | |
1681 | { | |
a4082357 | 1682 | LIST_HEAD(dispose); |
61948b6f BF |
1683 | int loop = 0; |
1684 | bool write_fail = false; | |
a4082357 | 1685 | |
85bec546 | 1686 | /* |
9c7504aa BF |
1687 | * First wait on the buftarg I/O count for all in-flight buffers to be |
1688 | * released. This is critical as new buffers do not make the LRU until | |
1689 | * they are released. | |
1690 | * | |
1691 | * Next, flush the buffer workqueue to ensure all completion processing | |
1692 | * has finished. Just waiting on buffer locks is not sufficient for | |
1693 | * async IO as the reference count held over IO is not released until | |
1694 | * after the buffer lock is dropped. Hence we need to ensure here that | |
1695 | * all reference counts have been dropped before we start walking the | |
1696 | * LRU list. | |
85bec546 | 1697 | */ |
9c7504aa BF |
1698 | while (percpu_counter_sum(&btp->bt_io_count)) |
1699 | delay(100); | |
800b2694 | 1700 | flush_workqueue(btp->bt_mount->m_buf_workqueue); |
85bec546 | 1701 | |
a4082357 DC |
1702 | /* loop until there is nothing left on the lru list. */ |
1703 | while (list_lru_count(&btp->bt_lru)) { | |
e80dfa19 | 1704 | list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele, |
a4082357 DC |
1705 | &dispose, LONG_MAX); |
1706 | ||
1707 | while (!list_empty(&dispose)) { | |
1708 | struct xfs_buf *bp; | |
1709 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1710 | list_del_init(&bp->b_lru); | |
ac8809f9 | 1711 | if (bp->b_flags & XBF_WRITE_FAIL) { |
61948b6f BF |
1712 | write_fail = true; |
1713 | xfs_buf_alert_ratelimited(bp, | |
1714 | "XFS: Corruption Alert", | |
c219b015 | 1715 | "Corruption Alert: Buffer at daddr 0x%llx had permanent write failures!", |
ac8809f9 DC |
1716 | (long long)bp->b_bn); |
1717 | } | |
a4082357 DC |
1718 | xfs_buf_rele(bp); |
1719 | } | |
1720 | if (loop++ != 0) | |
1721 | delay(100); | |
1722 | } | |
61948b6f BF |
1723 | |
1724 | /* | |
1725 | * If one or more failed buffers were freed, that means dirty metadata | |
1726 | * was thrown away. This should only ever happen after I/O completion | |
1727 | * handling has elevated I/O error(s) to permanent failures and shuts | |
1728 | * down the fs. | |
1729 | */ | |
1730 | if (write_fail) { | |
1731 | ASSERT(XFS_FORCED_SHUTDOWN(btp->bt_mount)); | |
1732 | xfs_alert(btp->bt_mount, | |
1733 | "Please run xfs_repair to determine the extent of the problem."); | |
1734 | } | |
e80dfa19 DC |
1735 | } |
1736 | ||
1737 | static enum lru_status | |
1738 | xfs_buftarg_isolate( | |
1739 | struct list_head *item, | |
3f97b163 | 1740 | struct list_lru_one *lru, |
e80dfa19 DC |
1741 | spinlock_t *lru_lock, |
1742 | void *arg) | |
1743 | { | |
1744 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); | |
1745 | struct list_head *dispose = arg; | |
1746 | ||
a4082357 DC |
1747 | /* |
1748 | * we are inverting the lru lock/bp->b_lock here, so use a trylock. | |
1749 | * If we fail to get the lock, just skip it. | |
1750 | */ | |
1751 | if (!spin_trylock(&bp->b_lock)) | |
1752 | return LRU_SKIP; | |
e80dfa19 DC |
1753 | /* |
1754 | * Decrement the b_lru_ref count unless the value is already | |
1755 | * zero. If the value is already zero, we need to reclaim the | |
1756 | * buffer, otherwise it gets another trip through the LRU. | |
1757 | */ | |
19957a18 | 1758 | if (atomic_add_unless(&bp->b_lru_ref, -1, 0)) { |
a4082357 | 1759 | spin_unlock(&bp->b_lock); |
e80dfa19 | 1760 | return LRU_ROTATE; |
a4082357 | 1761 | } |
e80dfa19 | 1762 | |
a4082357 | 1763 | bp->b_state |= XFS_BSTATE_DISPOSE; |
3f97b163 | 1764 | list_lru_isolate_move(lru, item, dispose); |
a4082357 | 1765 | spin_unlock(&bp->b_lock); |
e80dfa19 DC |
1766 | return LRU_REMOVED; |
1767 | } | |
1768 | ||
addbda40 | 1769 | static unsigned long |
e80dfa19 | 1770 | xfs_buftarg_shrink_scan( |
ff57ab21 | 1771 | struct shrinker *shrink, |
1495f230 | 1772 | struct shrink_control *sc) |
a6867a68 | 1773 | { |
ff57ab21 DC |
1774 | struct xfs_buftarg *btp = container_of(shrink, |
1775 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1776 | LIST_HEAD(dispose); |
addbda40 | 1777 | unsigned long freed; |
430cbeb8 | 1778 | |
503c358c VD |
1779 | freed = list_lru_shrink_walk(&btp->bt_lru, sc, |
1780 | xfs_buftarg_isolate, &dispose); | |
430cbeb8 DC |
1781 | |
1782 | while (!list_empty(&dispose)) { | |
e80dfa19 | 1783 | struct xfs_buf *bp; |
430cbeb8 DC |
1784 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); |
1785 | list_del_init(&bp->b_lru); | |
1786 | xfs_buf_rele(bp); | |
1787 | } | |
1788 | ||
e80dfa19 DC |
1789 | return freed; |
1790 | } | |
1791 | ||
addbda40 | 1792 | static unsigned long |
e80dfa19 DC |
1793 | xfs_buftarg_shrink_count( |
1794 | struct shrinker *shrink, | |
1795 | struct shrink_control *sc) | |
1796 | { | |
1797 | struct xfs_buftarg *btp = container_of(shrink, | |
1798 | struct xfs_buftarg, bt_shrinker); | |
503c358c | 1799 | return list_lru_shrink_count(&btp->bt_lru, sc); |
a6867a68 DC |
1800 | } |
1801 | ||
1da177e4 LT |
1802 | void |
1803 | xfs_free_buftarg( | |
b7963133 | 1804 | struct xfs_buftarg *btp) |
1da177e4 | 1805 | { |
ff57ab21 | 1806 | unregister_shrinker(&btp->bt_shrinker); |
9c7504aa BF |
1807 | ASSERT(percpu_counter_sum(&btp->bt_io_count) == 0); |
1808 | percpu_counter_destroy(&btp->bt_io_count); | |
f5e1dd34 | 1809 | list_lru_destroy(&btp->bt_lru); |
ff57ab21 | 1810 | |
2291dab2 | 1811 | xfs_blkdev_issue_flush(btp); |
a6867a68 | 1812 | |
f0e2d93c | 1813 | kmem_free(btp); |
1da177e4 LT |
1814 | } |
1815 | ||
3fefdeee ES |
1816 | int |
1817 | xfs_setsize_buftarg( | |
1da177e4 | 1818 | xfs_buftarg_t *btp, |
3fefdeee | 1819 | unsigned int sectorsize) |
1da177e4 | 1820 | { |
7c71ee78 | 1821 | /* Set up metadata sector size info */ |
6da54179 ES |
1822 | btp->bt_meta_sectorsize = sectorsize; |
1823 | btp->bt_meta_sectormask = sectorsize - 1; | |
1da177e4 | 1824 | |
ce8e922c | 1825 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
4f10700a | 1826 | xfs_warn(btp->bt_mount, |
a1c6f057 DM |
1827 | "Cannot set_blocksize to %u on device %pg", |
1828 | sectorsize, btp->bt_bdev); | |
2451337d | 1829 | return -EINVAL; |
1da177e4 LT |
1830 | } |
1831 | ||
7c71ee78 ES |
1832 | /* Set up device logical sector size mask */ |
1833 | btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev); | |
1834 | btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1; | |
1835 | ||
1da177e4 LT |
1836 | return 0; |
1837 | } | |
1838 | ||
1839 | /* | |
3fefdeee ES |
1840 | * When allocating the initial buffer target we have not yet |
1841 | * read in the superblock, so don't know what sized sectors | |
1842 | * are being used at this early stage. Play safe. | |
ce8e922c | 1843 | */ |
1da177e4 LT |
1844 | STATIC int |
1845 | xfs_setsize_buftarg_early( | |
1846 | xfs_buftarg_t *btp, | |
1847 | struct block_device *bdev) | |
1848 | { | |
a96c4151 | 1849 | return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev)); |
1da177e4 LT |
1850 | } |
1851 | ||
1da177e4 LT |
1852 | xfs_buftarg_t * |
1853 | xfs_alloc_buftarg( | |
ebad861b | 1854 | struct xfs_mount *mp, |
486aff5e DW |
1855 | struct block_device *bdev, |
1856 | struct dax_device *dax_dev) | |
1da177e4 LT |
1857 | { |
1858 | xfs_buftarg_t *btp; | |
1859 | ||
707e0dda | 1860 | btp = kmem_zalloc(sizeof(*btp), KM_NOFS); |
1da177e4 | 1861 | |
ebad861b | 1862 | btp->bt_mount = mp; |
ce8e922c NS |
1863 | btp->bt_dev = bdev->bd_dev; |
1864 | btp->bt_bdev = bdev; | |
486aff5e | 1865 | btp->bt_daxdev = dax_dev; |
0e6e847f | 1866 | |
f9bccfcc BF |
1867 | /* |
1868 | * Buffer IO error rate limiting. Limit it to no more than 10 messages | |
1869 | * per 30 seconds so as to not spam logs too much on repeated errors. | |
1870 | */ | |
1871 | ratelimit_state_init(&btp->bt_ioerror_rl, 30 * HZ, | |
1872 | DEFAULT_RATELIMIT_BURST); | |
1873 | ||
1da177e4 | 1874 | if (xfs_setsize_buftarg_early(btp, bdev)) |
d210a987 | 1875 | goto error_free; |
5ca302c8 GC |
1876 | |
1877 | if (list_lru_init(&btp->bt_lru)) | |
d210a987 | 1878 | goto error_free; |
5ca302c8 | 1879 | |
9c7504aa | 1880 | if (percpu_counter_init(&btp->bt_io_count, 0, GFP_KERNEL)) |
d210a987 | 1881 | goto error_lru; |
9c7504aa | 1882 | |
e80dfa19 DC |
1883 | btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count; |
1884 | btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan; | |
ff57ab21 | 1885 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; |
e80dfa19 | 1886 | btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE; |
d210a987 MH |
1887 | if (register_shrinker(&btp->bt_shrinker)) |
1888 | goto error_pcpu; | |
1da177e4 LT |
1889 | return btp; |
1890 | ||
d210a987 MH |
1891 | error_pcpu: |
1892 | percpu_counter_destroy(&btp->bt_io_count); | |
1893 | error_lru: | |
1894 | list_lru_destroy(&btp->bt_lru); | |
1895 | error_free: | |
f0e2d93c | 1896 | kmem_free(btp); |
1da177e4 LT |
1897 | return NULL; |
1898 | } | |
1899 | ||
20e8a063 BF |
1900 | /* |
1901 | * Cancel a delayed write list. | |
1902 | * | |
1903 | * Remove each buffer from the list, clear the delwri queue flag and drop the | |
1904 | * associated buffer reference. | |
1905 | */ | |
1906 | void | |
1907 | xfs_buf_delwri_cancel( | |
1908 | struct list_head *list) | |
1909 | { | |
1910 | struct xfs_buf *bp; | |
1911 | ||
1912 | while (!list_empty(list)) { | |
1913 | bp = list_first_entry(list, struct xfs_buf, b_list); | |
1914 | ||
1915 | xfs_buf_lock(bp); | |
1916 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
1917 | list_del_init(&bp->b_list); | |
1918 | xfs_buf_relse(bp); | |
1919 | } | |
1920 | } | |
1921 | ||
1da177e4 | 1922 | /* |
43ff2122 CH |
1923 | * Add a buffer to the delayed write list. |
1924 | * | |
1925 | * This queues a buffer for writeout if it hasn't already been. Note that | |
1926 | * neither this routine nor the buffer list submission functions perform | |
1927 | * any internal synchronization. It is expected that the lists are thread-local | |
1928 | * to the callers. | |
1929 | * | |
1930 | * Returns true if we queued up the buffer, or false if it already had | |
1931 | * been on the buffer list. | |
1da177e4 | 1932 | */ |
43ff2122 | 1933 | bool |
ce8e922c | 1934 | xfs_buf_delwri_queue( |
43ff2122 CH |
1935 | struct xfs_buf *bp, |
1936 | struct list_head *list) | |
1da177e4 | 1937 | { |
43ff2122 | 1938 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 1939 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1940 | |
43ff2122 CH |
1941 | /* |
1942 | * If the buffer is already marked delwri it already is queued up | |
1943 | * by someone else for imediate writeout. Just ignore it in that | |
1944 | * case. | |
1945 | */ | |
1946 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
1947 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
1948 | return false; | |
1da177e4 | 1949 | } |
1da177e4 | 1950 | |
43ff2122 | 1951 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
1952 | |
1953 | /* | |
43ff2122 CH |
1954 | * If a buffer gets written out synchronously or marked stale while it |
1955 | * is on a delwri list we lazily remove it. To do this, the other party | |
1956 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
1957 | * It remains referenced and on the list. In a rare corner case it | |
1958 | * might get readded to a delwri list after the synchronous writeout, in | |
1959 | * which case we need just need to re-add the flag here. | |
d808f617 | 1960 | */ |
43ff2122 CH |
1961 | bp->b_flags |= _XBF_DELWRI_Q; |
1962 | if (list_empty(&bp->b_list)) { | |
1963 | atomic_inc(&bp->b_hold); | |
1964 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 1965 | } |
585e6d88 | 1966 | |
43ff2122 | 1967 | return true; |
585e6d88 DC |
1968 | } |
1969 | ||
089716aa DC |
1970 | /* |
1971 | * Compare function is more complex than it needs to be because | |
1972 | * the return value is only 32 bits and we are doing comparisons | |
1973 | * on 64 bit values | |
1974 | */ | |
1975 | static int | |
1976 | xfs_buf_cmp( | |
1977 | void *priv, | |
1978 | struct list_head *a, | |
1979 | struct list_head *b) | |
1980 | { | |
1981 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1982 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1983 | xfs_daddr_t diff; | |
1984 | ||
f4b42421 | 1985 | diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn; |
089716aa DC |
1986 | if (diff < 0) |
1987 | return -1; | |
1988 | if (diff > 0) | |
1989 | return 1; | |
1990 | return 0; | |
1991 | } | |
1992 | ||
26f1fe85 | 1993 | /* |
e339dd8d BF |
1994 | * Submit buffers for write. If wait_list is specified, the buffers are |
1995 | * submitted using sync I/O and placed on the wait list such that the caller can | |
1996 | * iowait each buffer. Otherwise async I/O is used and the buffers are released | |
1997 | * at I/O completion time. In either case, buffers remain locked until I/O | |
1998 | * completes and the buffer is released from the queue. | |
26f1fe85 | 1999 | */ |
43ff2122 | 2000 | static int |
26f1fe85 | 2001 | xfs_buf_delwri_submit_buffers( |
43ff2122 | 2002 | struct list_head *buffer_list, |
26f1fe85 | 2003 | struct list_head *wait_list) |
1da177e4 | 2004 | { |
43ff2122 CH |
2005 | struct xfs_buf *bp, *n; |
2006 | int pinned = 0; | |
26f1fe85 | 2007 | struct blk_plug plug; |
43ff2122 | 2008 | |
26f1fe85 | 2009 | list_sort(NULL, buffer_list, xfs_buf_cmp); |
43ff2122 | 2010 | |
26f1fe85 | 2011 | blk_start_plug(&plug); |
43ff2122 | 2012 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { |
26f1fe85 | 2013 | if (!wait_list) { |
43ff2122 CH |
2014 | if (xfs_buf_ispinned(bp)) { |
2015 | pinned++; | |
2016 | continue; | |
2017 | } | |
2018 | if (!xfs_buf_trylock(bp)) | |
2019 | continue; | |
2020 | } else { | |
2021 | xfs_buf_lock(bp); | |
2022 | } | |
978c7b2f | 2023 | |
43ff2122 CH |
2024 | /* |
2025 | * Someone else might have written the buffer synchronously or | |
2026 | * marked it stale in the meantime. In that case only the | |
2027 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
2028 | * reference and remove it from the list here. | |
2029 | */ | |
2030 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
2031 | list_del_init(&bp->b_list); | |
2032 | xfs_buf_relse(bp); | |
2033 | continue; | |
2034 | } | |
c9c12971 | 2035 | |
43ff2122 | 2036 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
a1b7ea5d | 2037 | |
cf53e99d | 2038 | /* |
e339dd8d BF |
2039 | * If we have a wait list, each buffer (and associated delwri |
2040 | * queue reference) transfers to it and is submitted | |
2041 | * synchronously. Otherwise, drop the buffer from the delwri | |
2042 | * queue and submit async. | |
cf53e99d | 2043 | */ |
b6983e80 | 2044 | bp->b_flags &= ~_XBF_DELWRI_Q; |
e339dd8d | 2045 | bp->b_flags |= XBF_WRITE; |
26f1fe85 | 2046 | if (wait_list) { |
e339dd8d | 2047 | bp->b_flags &= ~XBF_ASYNC; |
26f1fe85 | 2048 | list_move_tail(&bp->b_list, wait_list); |
e339dd8d BF |
2049 | } else { |
2050 | bp->b_flags |= XBF_ASYNC; | |
ce8e922c | 2051 | list_del_init(&bp->b_list); |
e339dd8d | 2052 | } |
6af88cda | 2053 | __xfs_buf_submit(bp, false); |
43ff2122 CH |
2054 | } |
2055 | blk_finish_plug(&plug); | |
1da177e4 | 2056 | |
43ff2122 | 2057 | return pinned; |
1da177e4 LT |
2058 | } |
2059 | ||
2060 | /* | |
43ff2122 CH |
2061 | * Write out a buffer list asynchronously. |
2062 | * | |
2063 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
2064 | * out and not wait for I/O completion on any of the buffers. This interface | |
2065 | * is only safely useable for callers that can track I/O completion by higher | |
2066 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
2067 | * function. | |
efc3289c BF |
2068 | * |
2069 | * Note: this function will skip buffers it would block on, and in doing so | |
2070 | * leaves them on @buffer_list so they can be retried on a later pass. As such, | |
2071 | * it is up to the caller to ensure that the buffer list is fully submitted or | |
2072 | * cancelled appropriately when they are finished with the list. Failure to | |
2073 | * cancel or resubmit the list until it is empty will result in leaked buffers | |
2074 | * at unmount time. | |
1da177e4 LT |
2075 | */ |
2076 | int | |
43ff2122 CH |
2077 | xfs_buf_delwri_submit_nowait( |
2078 | struct list_head *buffer_list) | |
1da177e4 | 2079 | { |
26f1fe85 | 2080 | return xfs_buf_delwri_submit_buffers(buffer_list, NULL); |
43ff2122 | 2081 | } |
1da177e4 | 2082 | |
43ff2122 CH |
2083 | /* |
2084 | * Write out a buffer list synchronously. | |
2085 | * | |
2086 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
2087 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
2088 | * so callers must have some other way of tracking buffers if they require such | |
2089 | * functionality. | |
2090 | */ | |
2091 | int | |
2092 | xfs_buf_delwri_submit( | |
2093 | struct list_head *buffer_list) | |
2094 | { | |
26f1fe85 | 2095 | LIST_HEAD (wait_list); |
43ff2122 CH |
2096 | int error = 0, error2; |
2097 | struct xfs_buf *bp; | |
1da177e4 | 2098 | |
26f1fe85 | 2099 | xfs_buf_delwri_submit_buffers(buffer_list, &wait_list); |
1da177e4 | 2100 | |
43ff2122 | 2101 | /* Wait for IO to complete. */ |
26f1fe85 DC |
2102 | while (!list_empty(&wait_list)) { |
2103 | bp = list_first_entry(&wait_list, struct xfs_buf, b_list); | |
a1b7ea5d | 2104 | |
089716aa | 2105 | list_del_init(&bp->b_list); |
cf53e99d | 2106 | |
e339dd8d BF |
2107 | /* |
2108 | * Wait on the locked buffer, check for errors and unlock and | |
2109 | * release the delwri queue reference. | |
2110 | */ | |
2111 | error2 = xfs_buf_iowait(bp); | |
43ff2122 CH |
2112 | xfs_buf_relse(bp); |
2113 | if (!error) | |
2114 | error = error2; | |
1da177e4 LT |
2115 | } |
2116 | ||
43ff2122 | 2117 | return error; |
1da177e4 LT |
2118 | } |
2119 | ||
7912e7fe BF |
2120 | /* |
2121 | * Push a single buffer on a delwri queue. | |
2122 | * | |
2123 | * The purpose of this function is to submit a single buffer of a delwri queue | |
2124 | * and return with the buffer still on the original queue. The waiting delwri | |
2125 | * buffer submission infrastructure guarantees transfer of the delwri queue | |
2126 | * buffer reference to a temporary wait list. We reuse this infrastructure to | |
2127 | * transfer the buffer back to the original queue. | |
2128 | * | |
2129 | * Note the buffer transitions from the queued state, to the submitted and wait | |
2130 | * listed state and back to the queued state during this call. The buffer | |
2131 | * locking and queue management logic between _delwri_pushbuf() and | |
2132 | * _delwri_queue() guarantee that the buffer cannot be queued to another list | |
2133 | * before returning. | |
2134 | */ | |
2135 | int | |
2136 | xfs_buf_delwri_pushbuf( | |
2137 | struct xfs_buf *bp, | |
2138 | struct list_head *buffer_list) | |
2139 | { | |
2140 | LIST_HEAD (submit_list); | |
2141 | int error; | |
2142 | ||
2143 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
2144 | ||
2145 | trace_xfs_buf_delwri_pushbuf(bp, _RET_IP_); | |
2146 | ||
2147 | /* | |
2148 | * Isolate the buffer to a new local list so we can submit it for I/O | |
2149 | * independently from the rest of the original list. | |
2150 | */ | |
2151 | xfs_buf_lock(bp); | |
2152 | list_move(&bp->b_list, &submit_list); | |
2153 | xfs_buf_unlock(bp); | |
2154 | ||
2155 | /* | |
2156 | * Delwri submission clears the DELWRI_Q buffer flag and returns with | |
e339dd8d | 2157 | * the buffer on the wait list with the original reference. Rather than |
7912e7fe BF |
2158 | * bounce the buffer from a local wait list back to the original list |
2159 | * after I/O completion, reuse the original list as the wait list. | |
2160 | */ | |
2161 | xfs_buf_delwri_submit_buffers(&submit_list, buffer_list); | |
2162 | ||
2163 | /* | |
e339dd8d BF |
2164 | * The buffer is now locked, under I/O and wait listed on the original |
2165 | * delwri queue. Wait for I/O completion, restore the DELWRI_Q flag and | |
2166 | * return with the buffer unlocked and on the original queue. | |
7912e7fe | 2167 | */ |
e339dd8d | 2168 | error = xfs_buf_iowait(bp); |
7912e7fe BF |
2169 | bp->b_flags |= _XBF_DELWRI_Q; |
2170 | xfs_buf_unlock(bp); | |
2171 | ||
2172 | return error; | |
2173 | } | |
2174 | ||
04d8b284 | 2175 | int __init |
ce8e922c | 2176 | xfs_buf_init(void) |
1da177e4 | 2177 | { |
12eba65b DC |
2178 | xfs_buf_zone = kmem_cache_create("xfs_buf", sizeof(struct xfs_buf), 0, |
2179 | SLAB_HWCACHE_ALIGN | | |
2180 | SLAB_RECLAIM_ACCOUNT | | |
2181 | SLAB_MEM_SPREAD, | |
2182 | NULL); | |
ce8e922c | 2183 | if (!xfs_buf_zone) |
0b1b213f | 2184 | goto out; |
04d8b284 | 2185 | |
23ea4032 | 2186 | return 0; |
1da177e4 | 2187 | |
0b1b213f | 2188 | out: |
8758280f | 2189 | return -ENOMEM; |
1da177e4 LT |
2190 | } |
2191 | ||
1da177e4 | 2192 | void |
ce8e922c | 2193 | xfs_buf_terminate(void) |
1da177e4 | 2194 | { |
aaf54eb8 | 2195 | kmem_cache_destroy(xfs_buf_zone); |
1da177e4 | 2196 | } |
7561d27e BF |
2197 | |
2198 | void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref) | |
2199 | { | |
7561d27e BF |
2200 | /* |
2201 | * Set the lru reference count to 0 based on the error injection tag. | |
2202 | * This allows userspace to disrupt buffer caching for debug/testing | |
2203 | * purposes. | |
2204 | */ | |
dbd329f1 | 2205 | if (XFS_TEST_ERROR(false, bp->b_mount, XFS_ERRTAG_BUF_LRU_REF)) |
7561d27e BF |
2206 | lru_ref = 0; |
2207 | ||
2208 | atomic_set(&bp->b_lru_ref, lru_ref); | |
2209 | } | |
8473fee3 BF |
2210 | |
2211 | /* | |
2212 | * Verify an on-disk magic value against the magic value specified in the | |
2213 | * verifier structure. The verifier magic is in disk byte order so the caller is | |
2214 | * expected to pass the value directly from disk. | |
2215 | */ | |
2216 | bool | |
2217 | xfs_verify_magic( | |
2218 | struct xfs_buf *bp, | |
15baadf7 | 2219 | __be32 dmagic) |
8473fee3 | 2220 | { |
dbd329f1 | 2221 | struct xfs_mount *mp = bp->b_mount; |
8473fee3 BF |
2222 | int idx; |
2223 | ||
2224 | idx = xfs_sb_version_hascrc(&mp->m_sb); | |
14ed8688 | 2225 | if (WARN_ON(!bp->b_ops || !bp->b_ops->magic[idx])) |
8473fee3 BF |
2226 | return false; |
2227 | return dmagic == bp->b_ops->magic[idx]; | |
2228 | } | |
15baadf7 DW |
2229 | /* |
2230 | * Verify an on-disk magic value against the magic value specified in the | |
2231 | * verifier structure. The verifier magic is in disk byte order so the caller is | |
2232 | * expected to pass the value directly from disk. | |
2233 | */ | |
2234 | bool | |
2235 | xfs_verify_magic16( | |
2236 | struct xfs_buf *bp, | |
2237 | __be16 dmagic) | |
2238 | { | |
dbd329f1 | 2239 | struct xfs_mount *mp = bp->b_mount; |
15baadf7 DW |
2240 | int idx; |
2241 | ||
2242 | idx = xfs_sb_version_hascrc(&mp->m_sb); | |
14ed8688 | 2243 | if (WARN_ON(!bp->b_ops || !bp->b_ops->magic16[idx])) |
15baadf7 DW |
2244 | return false; |
2245 | return dmagic == bp->b_ops->magic16[idx]; | |
2246 | } |