]>
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" |
e9e899a2 | 17 | #include "xfs_errortag.h" |
7561d27e | 18 | #include "xfs_error.h" |
b7963133 | 19 | |
7989cb8e | 20 | static kmem_zone_t *xfs_buf_zone; |
23ea4032 | 21 | |
ce8e922c | 22 | #define xb_to_gfp(flags) \ |
aa5c158e | 23 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN) |
1da177e4 | 24 | |
37fd1678 DC |
25 | /* |
26 | * Locking orders | |
27 | * | |
28 | * xfs_buf_ioacct_inc: | |
29 | * xfs_buf_ioacct_dec: | |
30 | * b_sema (caller holds) | |
31 | * b_lock | |
32 | * | |
33 | * xfs_buf_stale: | |
34 | * b_sema (caller holds) | |
35 | * b_lock | |
36 | * lru_lock | |
37 | * | |
38 | * xfs_buf_rele: | |
39 | * b_lock | |
40 | * pag_buf_lock | |
41 | * lru_lock | |
42 | * | |
43 | * xfs_buftarg_wait_rele | |
44 | * lru_lock | |
45 | * b_lock (trylock due to inversion) | |
46 | * | |
47 | * xfs_buftarg_isolate | |
48 | * lru_lock | |
49 | * b_lock (trylock due to inversion) | |
50 | */ | |
1da177e4 | 51 | |
73c77e2c JB |
52 | static inline int |
53 | xfs_buf_is_vmapped( | |
54 | struct xfs_buf *bp) | |
55 | { | |
56 | /* | |
57 | * Return true if the buffer is vmapped. | |
58 | * | |
611c9946 DC |
59 | * b_addr is null if the buffer is not mapped, but the code is clever |
60 | * enough to know it doesn't have to map a single page, so the check has | |
61 | * to be both for b_addr and bp->b_page_count > 1. | |
73c77e2c | 62 | */ |
611c9946 | 63 | return bp->b_addr && bp->b_page_count > 1; |
73c77e2c JB |
64 | } |
65 | ||
66 | static inline int | |
67 | xfs_buf_vmap_len( | |
68 | struct xfs_buf *bp) | |
69 | { | |
70 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
71 | } | |
72 | ||
9c7504aa BF |
73 | /* |
74 | * Bump the I/O in flight count on the buftarg if we haven't yet done so for | |
75 | * this buffer. The count is incremented once per buffer (per hold cycle) | |
76 | * because the corresponding decrement is deferred to buffer release. Buffers | |
77 | * can undergo I/O multiple times in a hold-release cycle and per buffer I/O | |
78 | * tracking adds unnecessary overhead. This is used for sychronization purposes | |
79 | * with unmount (see xfs_wait_buftarg()), so all we really need is a count of | |
80 | * in-flight buffers. | |
81 | * | |
82 | * Buffers that are never released (e.g., superblock, iclog buffers) must set | |
83 | * the XBF_NO_IOACCT flag before I/O submission. Otherwise, the buftarg count | |
84 | * never reaches zero and unmount hangs indefinitely. | |
85 | */ | |
86 | static inline void | |
87 | xfs_buf_ioacct_inc( | |
88 | struct xfs_buf *bp) | |
89 | { | |
63db7c81 | 90 | if (bp->b_flags & XBF_NO_IOACCT) |
9c7504aa BF |
91 | return; |
92 | ||
93 | ASSERT(bp->b_flags & XBF_ASYNC); | |
63db7c81 BF |
94 | spin_lock(&bp->b_lock); |
95 | if (!(bp->b_state & XFS_BSTATE_IN_FLIGHT)) { | |
96 | bp->b_state |= XFS_BSTATE_IN_FLIGHT; | |
97 | percpu_counter_inc(&bp->b_target->bt_io_count); | |
98 | } | |
99 | spin_unlock(&bp->b_lock); | |
9c7504aa BF |
100 | } |
101 | ||
102 | /* | |
103 | * Clear the in-flight state on a buffer about to be released to the LRU or | |
104 | * freed and unaccount from the buftarg. | |
105 | */ | |
106 | static inline void | |
63db7c81 | 107 | __xfs_buf_ioacct_dec( |
9c7504aa BF |
108 | struct xfs_buf *bp) |
109 | { | |
95989c46 | 110 | lockdep_assert_held(&bp->b_lock); |
9c7504aa | 111 | |
63db7c81 BF |
112 | if (bp->b_state & XFS_BSTATE_IN_FLIGHT) { |
113 | bp->b_state &= ~XFS_BSTATE_IN_FLIGHT; | |
114 | percpu_counter_dec(&bp->b_target->bt_io_count); | |
115 | } | |
116 | } | |
117 | ||
118 | static inline void | |
119 | xfs_buf_ioacct_dec( | |
120 | struct xfs_buf *bp) | |
121 | { | |
122 | spin_lock(&bp->b_lock); | |
123 | __xfs_buf_ioacct_dec(bp); | |
124 | spin_unlock(&bp->b_lock); | |
9c7504aa BF |
125 | } |
126 | ||
430cbeb8 DC |
127 | /* |
128 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
129 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
130 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
131 | * to remove the reference that LRU holds on the buffer. | |
132 | * | |
133 | * This prevents build-up of stale buffers on the LRU. | |
134 | */ | |
135 | void | |
136 | xfs_buf_stale( | |
137 | struct xfs_buf *bp) | |
138 | { | |
43ff2122 CH |
139 | ASSERT(xfs_buf_islocked(bp)); |
140 | ||
430cbeb8 | 141 | bp->b_flags |= XBF_STALE; |
43ff2122 CH |
142 | |
143 | /* | |
144 | * Clear the delwri status so that a delwri queue walker will not | |
145 | * flush this buffer to disk now that it is stale. The delwri queue has | |
146 | * a reference to the buffer, so this is safe to do. | |
147 | */ | |
148 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
149 | ||
9c7504aa BF |
150 | /* |
151 | * Once the buffer is marked stale and unlocked, a subsequent lookup | |
152 | * could reset b_flags. There is no guarantee that the buffer is | |
153 | * unaccounted (released to LRU) before that occurs. Drop in-flight | |
154 | * status now to preserve accounting consistency. | |
155 | */ | |
a4082357 | 156 | spin_lock(&bp->b_lock); |
63db7c81 BF |
157 | __xfs_buf_ioacct_dec(bp); |
158 | ||
a4082357 DC |
159 | atomic_set(&bp->b_lru_ref, 0); |
160 | if (!(bp->b_state & XFS_BSTATE_DISPOSE) && | |
e80dfa19 DC |
161 | (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru))) |
162 | atomic_dec(&bp->b_hold); | |
163 | ||
430cbeb8 | 164 | ASSERT(atomic_read(&bp->b_hold) >= 1); |
a4082357 | 165 | spin_unlock(&bp->b_lock); |
430cbeb8 | 166 | } |
1da177e4 | 167 | |
3e85c868 DC |
168 | static int |
169 | xfs_buf_get_maps( | |
170 | struct xfs_buf *bp, | |
171 | int map_count) | |
172 | { | |
173 | ASSERT(bp->b_maps == NULL); | |
174 | bp->b_map_count = map_count; | |
175 | ||
176 | if (map_count == 1) { | |
f4b42421 | 177 | bp->b_maps = &bp->__b_map; |
3e85c868 DC |
178 | return 0; |
179 | } | |
180 | ||
181 | bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map), | |
182 | KM_NOFS); | |
183 | if (!bp->b_maps) | |
2451337d | 184 | return -ENOMEM; |
3e85c868 DC |
185 | return 0; |
186 | } | |
187 | ||
188 | /* | |
189 | * Frees b_pages if it was allocated. | |
190 | */ | |
191 | static void | |
192 | xfs_buf_free_maps( | |
193 | struct xfs_buf *bp) | |
194 | { | |
f4b42421 | 195 | if (bp->b_maps != &bp->__b_map) { |
3e85c868 DC |
196 | kmem_free(bp->b_maps); |
197 | bp->b_maps = NULL; | |
198 | } | |
199 | } | |
200 | ||
32dff5e5 | 201 | static int |
3e85c868 | 202 | _xfs_buf_alloc( |
4347b9d7 | 203 | struct xfs_buftarg *target, |
3e85c868 DC |
204 | struct xfs_buf_map *map, |
205 | int nmaps, | |
32dff5e5 DW |
206 | xfs_buf_flags_t flags, |
207 | struct xfs_buf **bpp) | |
1da177e4 | 208 | { |
4347b9d7 | 209 | struct xfs_buf *bp; |
3e85c868 DC |
210 | int error; |
211 | int i; | |
4347b9d7 | 212 | |
32dff5e5 | 213 | *bpp = NULL; |
aa5c158e | 214 | bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS); |
4347b9d7 | 215 | if (unlikely(!bp)) |
32dff5e5 | 216 | return -ENOMEM; |
4347b9d7 | 217 | |
1da177e4 | 218 | /* |
12bcb3f7 DC |
219 | * We don't want certain flags to appear in b_flags unless they are |
220 | * specifically set by later operations on the buffer. | |
1da177e4 | 221 | */ |
611c9946 | 222 | flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD); |
ce8e922c | 223 | |
ce8e922c | 224 | atomic_set(&bp->b_hold, 1); |
430cbeb8 | 225 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 226 | init_completion(&bp->b_iowait); |
430cbeb8 | 227 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 228 | INIT_LIST_HEAD(&bp->b_list); |
643c8c05 | 229 | INIT_LIST_HEAD(&bp->b_li_list); |
a731cd11 | 230 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
a4082357 | 231 | spin_lock_init(&bp->b_lock); |
ce8e922c | 232 | bp->b_target = target; |
dbd329f1 | 233 | bp->b_mount = target->bt_mount; |
3e85c868 | 234 | bp->b_flags = flags; |
de1cbee4 | 235 | |
1da177e4 | 236 | /* |
aa0e8833 DC |
237 | * Set length and io_length to the same value initially. |
238 | * I/O routines should use io_length, which will be the same in | |
1da177e4 LT |
239 | * most cases but may be reset (e.g. XFS recovery). |
240 | */ | |
3e85c868 DC |
241 | error = xfs_buf_get_maps(bp, nmaps); |
242 | if (error) { | |
377bcd5f | 243 | kmem_cache_free(xfs_buf_zone, bp); |
32dff5e5 | 244 | return error; |
3e85c868 DC |
245 | } |
246 | ||
247 | bp->b_bn = map[0].bm_bn; | |
248 | bp->b_length = 0; | |
249 | for (i = 0; i < nmaps; i++) { | |
250 | bp->b_maps[i].bm_bn = map[i].bm_bn; | |
251 | bp->b_maps[i].bm_len = map[i].bm_len; | |
252 | bp->b_length += map[i].bm_len; | |
253 | } | |
3e85c868 | 254 | |
ce8e922c NS |
255 | atomic_set(&bp->b_pin_count, 0); |
256 | init_waitqueue_head(&bp->b_waiters); | |
257 | ||
dbd329f1 | 258 | XFS_STATS_INC(bp->b_mount, xb_create); |
0b1b213f | 259 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 | 260 | |
32dff5e5 DW |
261 | *bpp = bp; |
262 | return 0; | |
1da177e4 LT |
263 | } |
264 | ||
265 | /* | |
ce8e922c NS |
266 | * Allocate a page array capable of holding a specified number |
267 | * of pages, and point the page buf at it. | |
1da177e4 LT |
268 | */ |
269 | STATIC int | |
ce8e922c NS |
270 | _xfs_buf_get_pages( |
271 | xfs_buf_t *bp, | |
87937bf8 | 272 | int page_count) |
1da177e4 LT |
273 | { |
274 | /* Make sure that we have a page list */ | |
ce8e922c | 275 | if (bp->b_pages == NULL) { |
ce8e922c NS |
276 | bp->b_page_count = page_count; |
277 | if (page_count <= XB_PAGES) { | |
278 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 279 | } else { |
ce8e922c | 280 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
aa5c158e | 281 | page_count, KM_NOFS); |
ce8e922c | 282 | if (bp->b_pages == NULL) |
1da177e4 LT |
283 | return -ENOMEM; |
284 | } | |
ce8e922c | 285 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
286 | } |
287 | return 0; | |
288 | } | |
289 | ||
290 | /* | |
ce8e922c | 291 | * Frees b_pages if it was allocated. |
1da177e4 LT |
292 | */ |
293 | STATIC void | |
ce8e922c | 294 | _xfs_buf_free_pages( |
1da177e4 LT |
295 | xfs_buf_t *bp) |
296 | { | |
ce8e922c | 297 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 298 | kmem_free(bp->b_pages); |
3fc98b1a | 299 | bp->b_pages = NULL; |
1da177e4 LT |
300 | } |
301 | } | |
302 | ||
303 | /* | |
304 | * Releases the specified buffer. | |
305 | * | |
306 | * The modification state of any associated pages is left unchanged. | |
b46fe825 | 307 | * The buffer must not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
308 | * hashed and refcounted buffers |
309 | */ | |
25a40957 | 310 | static void |
ce8e922c | 311 | xfs_buf_free( |
1da177e4 LT |
312 | xfs_buf_t *bp) |
313 | { | |
0b1b213f | 314 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 315 | |
430cbeb8 DC |
316 | ASSERT(list_empty(&bp->b_lru)); |
317 | ||
0e6e847f | 318 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
319 | uint i; |
320 | ||
73c77e2c | 321 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
322 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
323 | bp->b_page_count); | |
1da177e4 | 324 | |
948ecdb4 NS |
325 | for (i = 0; i < bp->b_page_count; i++) { |
326 | struct page *page = bp->b_pages[i]; | |
327 | ||
0e6e847f | 328 | __free_page(page); |
948ecdb4 | 329 | } |
0e6e847f DC |
330 | } else if (bp->b_flags & _XBF_KMEM) |
331 | kmem_free(bp->b_addr); | |
3fc98b1a | 332 | _xfs_buf_free_pages(bp); |
3e85c868 | 333 | xfs_buf_free_maps(bp); |
377bcd5f | 334 | kmem_cache_free(xfs_buf_zone, bp); |
1da177e4 LT |
335 | } |
336 | ||
337 | /* | |
0e6e847f | 338 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
339 | */ |
340 | STATIC int | |
0e6e847f | 341 | xfs_buf_allocate_memory( |
1da177e4 LT |
342 | xfs_buf_t *bp, |
343 | uint flags) | |
344 | { | |
aa0e8833 | 345 | size_t size; |
1da177e4 | 346 | size_t nbytes, offset; |
ce8e922c | 347 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 348 | unsigned short page_count, i; |
795cac72 | 349 | xfs_off_t start, end; |
1da177e4 | 350 | int error; |
3219e8cf BD |
351 | xfs_km_flags_t kmflag_mask = 0; |
352 | ||
353 | /* | |
354 | * assure zeroed buffer for non-read cases. | |
355 | */ | |
356 | if (!(flags & XBF_READ)) { | |
357 | kmflag_mask |= KM_ZERO; | |
358 | gfp_mask |= __GFP_ZERO; | |
359 | } | |
1da177e4 | 360 | |
0e6e847f DC |
361 | /* |
362 | * for buffers that are contained within a single page, just allocate | |
363 | * the memory from the heap - there's no need for the complexity of | |
364 | * page arrays to keep allocation down to order 0. | |
365 | */ | |
795cac72 DC |
366 | size = BBTOB(bp->b_length); |
367 | if (size < PAGE_SIZE) { | |
f8f9ee47 | 368 | int align_mask = xfs_buftarg_dma_alignment(bp->b_target); |
3219e8cf BD |
369 | bp->b_addr = kmem_alloc_io(size, align_mask, |
370 | KM_NOFS | kmflag_mask); | |
0e6e847f DC |
371 | if (!bp->b_addr) { |
372 | /* low memory - use alloc_page loop instead */ | |
373 | goto use_alloc_page; | |
374 | } | |
375 | ||
795cac72 | 376 | if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) != |
0e6e847f DC |
377 | ((unsigned long)bp->b_addr & PAGE_MASK)) { |
378 | /* b_addr spans two pages - use alloc_page instead */ | |
379 | kmem_free(bp->b_addr); | |
380 | bp->b_addr = NULL; | |
381 | goto use_alloc_page; | |
382 | } | |
383 | bp->b_offset = offset_in_page(bp->b_addr); | |
384 | bp->b_pages = bp->b_page_array; | |
f8f9ee47 | 385 | bp->b_pages[0] = kmem_to_page(bp->b_addr); |
0e6e847f | 386 | bp->b_page_count = 1; |
611c9946 | 387 | bp->b_flags |= _XBF_KMEM; |
0e6e847f DC |
388 | return 0; |
389 | } | |
390 | ||
391 | use_alloc_page: | |
f4b42421 MT |
392 | start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT; |
393 | end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1) | |
cbb7baab | 394 | >> PAGE_SHIFT; |
795cac72 | 395 | page_count = end - start; |
87937bf8 | 396 | error = _xfs_buf_get_pages(bp, page_count); |
1da177e4 LT |
397 | if (unlikely(error)) |
398 | return error; | |
1da177e4 | 399 | |
ce8e922c | 400 | offset = bp->b_offset; |
0e6e847f | 401 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 402 | |
ce8e922c | 403 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
404 | struct page *page; |
405 | uint retries = 0; | |
0e6e847f DC |
406 | retry: |
407 | page = alloc_page(gfp_mask); | |
1da177e4 | 408 | if (unlikely(page == NULL)) { |
ce8e922c NS |
409 | if (flags & XBF_READ_AHEAD) { |
410 | bp->b_page_count = i; | |
2451337d | 411 | error = -ENOMEM; |
0e6e847f | 412 | goto out_free_pages; |
1da177e4 LT |
413 | } |
414 | ||
415 | /* | |
416 | * This could deadlock. | |
417 | * | |
418 | * But until all the XFS lowlevel code is revamped to | |
419 | * handle buffer allocation failures we can't do much. | |
420 | */ | |
421 | if (!(++retries % 100)) | |
4f10700a | 422 | xfs_err(NULL, |
5bf97b1c TH |
423 | "%s(%u) possible memory allocation deadlock in %s (mode:0x%x)", |
424 | current->comm, current->pid, | |
34a622b2 | 425 | __func__, gfp_mask); |
1da177e4 | 426 | |
dbd329f1 | 427 | XFS_STATS_INC(bp->b_mount, xb_page_retries); |
8aa7e847 | 428 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
429 | goto retry; |
430 | } | |
431 | ||
dbd329f1 | 432 | XFS_STATS_INC(bp->b_mount, xb_page_found); |
1da177e4 | 433 | |
0e6e847f | 434 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 435 | size -= nbytes; |
ce8e922c | 436 | bp->b_pages[i] = page; |
1da177e4 LT |
437 | offset = 0; |
438 | } | |
0e6e847f | 439 | return 0; |
1da177e4 | 440 | |
0e6e847f DC |
441 | out_free_pages: |
442 | for (i = 0; i < bp->b_page_count; i++) | |
443 | __free_page(bp->b_pages[i]); | |
2aa6ba7b | 444 | bp->b_flags &= ~_XBF_PAGES; |
1da177e4 LT |
445 | return error; |
446 | } | |
447 | ||
448 | /* | |
25985edc | 449 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
450 | */ |
451 | STATIC int | |
ce8e922c | 452 | _xfs_buf_map_pages( |
1da177e4 LT |
453 | xfs_buf_t *bp, |
454 | uint flags) | |
455 | { | |
0e6e847f | 456 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 457 | if (bp->b_page_count == 1) { |
0e6e847f | 458 | /* A single page buffer is always mappable */ |
ce8e922c | 459 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
611c9946 DC |
460 | } else if (flags & XBF_UNMAPPED) { |
461 | bp->b_addr = NULL; | |
462 | } else { | |
a19fb380 | 463 | int retried = 0; |
9ba1fb2c | 464 | unsigned nofs_flag; |
ae687e58 DC |
465 | |
466 | /* | |
cf085a1b | 467 | * vm_map_ram() will allocate auxiliary structures (e.g. |
ae687e58 DC |
468 | * pagetables) with GFP_KERNEL, yet we are likely to be under |
469 | * GFP_NOFS context here. Hence we need to tell memory reclaim | |
9ba1fb2c | 470 | * that we are in such a context via PF_MEMALLOC_NOFS to prevent |
ae687e58 DC |
471 | * memory reclaim re-entering the filesystem here and |
472 | * potentially deadlocking. | |
473 | */ | |
9ba1fb2c | 474 | nofs_flag = memalloc_nofs_save(); |
a19fb380 DC |
475 | do { |
476 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
477 | -1, PAGE_KERNEL); | |
478 | if (bp->b_addr) | |
479 | break; | |
480 | vm_unmap_aliases(); | |
481 | } while (retried++ <= 1); | |
9ba1fb2c | 482 | memalloc_nofs_restore(nofs_flag); |
a19fb380 DC |
483 | |
484 | if (!bp->b_addr) | |
1da177e4 | 485 | return -ENOMEM; |
ce8e922c | 486 | bp->b_addr += bp->b_offset; |
1da177e4 LT |
487 | } |
488 | ||
489 | return 0; | |
490 | } | |
491 | ||
492 | /* | |
493 | * Finding and Reading Buffers | |
494 | */ | |
6031e73a LS |
495 | static int |
496 | _xfs_buf_obj_cmp( | |
497 | struct rhashtable_compare_arg *arg, | |
498 | const void *obj) | |
499 | { | |
500 | const struct xfs_buf_map *map = arg->key; | |
501 | const struct xfs_buf *bp = obj; | |
502 | ||
503 | /* | |
504 | * The key hashing in the lookup path depends on the key being the | |
505 | * first element of the compare_arg, make sure to assert this. | |
506 | */ | |
507 | BUILD_BUG_ON(offsetof(struct xfs_buf_map, bm_bn) != 0); | |
508 | ||
509 | if (bp->b_bn != map->bm_bn) | |
510 | return 1; | |
511 | ||
512 | if (unlikely(bp->b_length != map->bm_len)) { | |
513 | /* | |
514 | * found a block number match. If the range doesn't | |
515 | * match, the only way this is allowed is if the buffer | |
516 | * in the cache is stale and the transaction that made | |
517 | * it stale has not yet committed. i.e. we are | |
518 | * reallocating a busy extent. Skip this buffer and | |
519 | * continue searching for an exact match. | |
520 | */ | |
521 | ASSERT(bp->b_flags & XBF_STALE); | |
522 | return 1; | |
523 | } | |
524 | return 0; | |
525 | } | |
526 | ||
527 | static const struct rhashtable_params xfs_buf_hash_params = { | |
528 | .min_size = 32, /* empty AGs have minimal footprint */ | |
529 | .nelem_hint = 16, | |
530 | .key_len = sizeof(xfs_daddr_t), | |
531 | .key_offset = offsetof(struct xfs_buf, b_bn), | |
532 | .head_offset = offsetof(struct xfs_buf, b_rhash_head), | |
533 | .automatic_shrinking = true, | |
534 | .obj_cmpfn = _xfs_buf_obj_cmp, | |
535 | }; | |
536 | ||
537 | int | |
538 | xfs_buf_hash_init( | |
539 | struct xfs_perag *pag) | |
540 | { | |
541 | spin_lock_init(&pag->pag_buf_lock); | |
542 | return rhashtable_init(&pag->pag_buf_hash, &xfs_buf_hash_params); | |
543 | } | |
544 | ||
545 | void | |
546 | xfs_buf_hash_destroy( | |
547 | struct xfs_perag *pag) | |
548 | { | |
549 | rhashtable_destroy(&pag->pag_buf_hash); | |
550 | } | |
1da177e4 LT |
551 | |
552 | /* | |
b027d4c9 DC |
553 | * Look up a buffer in the buffer cache and return it referenced and locked |
554 | * in @found_bp. | |
555 | * | |
556 | * If @new_bp is supplied and we have a lookup miss, insert @new_bp into the | |
557 | * cache. | |
558 | * | |
559 | * If XBF_TRYLOCK is set in @flags, only try to lock the buffer and return | |
560 | * -EAGAIN if we fail to lock it. | |
561 | * | |
562 | * Return values are: | |
563 | * -EFSCORRUPTED if have been supplied with an invalid address | |
564 | * -EAGAIN on trylock failure | |
565 | * -ENOENT if we fail to find a match and @new_bp was NULL | |
566 | * 0, with @found_bp: | |
567 | * - @new_bp if we inserted it into the cache | |
568 | * - the buffer we found and locked. | |
1da177e4 | 569 | */ |
b027d4c9 DC |
570 | static int |
571 | xfs_buf_find( | |
e70b73f8 | 572 | struct xfs_buftarg *btp, |
3e85c868 DC |
573 | struct xfs_buf_map *map, |
574 | int nmaps, | |
ce8e922c | 575 | xfs_buf_flags_t flags, |
b027d4c9 DC |
576 | struct xfs_buf *new_bp, |
577 | struct xfs_buf **found_bp) | |
1da177e4 | 578 | { |
74f75a0c | 579 | struct xfs_perag *pag; |
74f75a0c | 580 | xfs_buf_t *bp; |
6031e73a | 581 | struct xfs_buf_map cmap = { .bm_bn = map[0].bm_bn }; |
10616b80 | 582 | xfs_daddr_t eofs; |
3e85c868 | 583 | int i; |
1da177e4 | 584 | |
b027d4c9 DC |
585 | *found_bp = NULL; |
586 | ||
3e85c868 | 587 | for (i = 0; i < nmaps; i++) |
6031e73a | 588 | cmap.bm_len += map[i].bm_len; |
1da177e4 LT |
589 | |
590 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
6031e73a LS |
591 | ASSERT(!(BBTOB(cmap.bm_len) < btp->bt_meta_sectorsize)); |
592 | ASSERT(!(BBTOB(cmap.bm_bn) & (xfs_off_t)btp->bt_meta_sectormask)); | |
1da177e4 | 593 | |
10616b80 DC |
594 | /* |
595 | * Corrupted block numbers can get through to here, unfortunately, so we | |
596 | * have to check that the buffer falls within the filesystem bounds. | |
597 | */ | |
598 | eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks); | |
6031e73a | 599 | if (cmap.bm_bn < 0 || cmap.bm_bn >= eofs) { |
10616b80 | 600 | xfs_alert(btp->bt_mount, |
c219b015 | 601 | "%s: daddr 0x%llx out of range, EOFS 0x%llx", |
6031e73a | 602 | __func__, cmap.bm_bn, eofs); |
7bc0dc27 | 603 | WARN_ON(1); |
b027d4c9 | 604 | return -EFSCORRUPTED; |
10616b80 DC |
605 | } |
606 | ||
74f75a0c | 607 | pag = xfs_perag_get(btp->bt_mount, |
6031e73a | 608 | xfs_daddr_to_agno(btp->bt_mount, cmap.bm_bn)); |
74f75a0c | 609 | |
74f75a0c | 610 | spin_lock(&pag->pag_buf_lock); |
6031e73a LS |
611 | bp = rhashtable_lookup_fast(&pag->pag_buf_hash, &cmap, |
612 | xfs_buf_hash_params); | |
613 | if (bp) { | |
614 | atomic_inc(&bp->b_hold); | |
615 | goto found; | |
1da177e4 LT |
616 | } |
617 | ||
618 | /* No match found */ | |
b027d4c9 | 619 | if (!new_bp) { |
ff6d6af2 | 620 | XFS_STATS_INC(btp->bt_mount, xb_miss_locked); |
74f75a0c DC |
621 | spin_unlock(&pag->pag_buf_lock); |
622 | xfs_perag_put(pag); | |
b027d4c9 | 623 | return -ENOENT; |
1da177e4 | 624 | } |
b027d4c9 DC |
625 | |
626 | /* the buffer keeps the perag reference until it is freed */ | |
627 | new_bp->b_pag = pag; | |
628 | rhashtable_insert_fast(&pag->pag_buf_hash, &new_bp->b_rhash_head, | |
629 | xfs_buf_hash_params); | |
630 | spin_unlock(&pag->pag_buf_lock); | |
631 | *found_bp = new_bp; | |
632 | return 0; | |
1da177e4 LT |
633 | |
634 | found: | |
74f75a0c DC |
635 | spin_unlock(&pag->pag_buf_lock); |
636 | xfs_perag_put(pag); | |
1da177e4 | 637 | |
0c842ad4 CH |
638 | if (!xfs_buf_trylock(bp)) { |
639 | if (flags & XBF_TRYLOCK) { | |
ce8e922c | 640 | xfs_buf_rele(bp); |
ff6d6af2 | 641 | XFS_STATS_INC(btp->bt_mount, xb_busy_locked); |
b027d4c9 | 642 | return -EAGAIN; |
1da177e4 | 643 | } |
0c842ad4 | 644 | xfs_buf_lock(bp); |
ff6d6af2 | 645 | XFS_STATS_INC(btp->bt_mount, xb_get_locked_waited); |
1da177e4 LT |
646 | } |
647 | ||
0e6e847f DC |
648 | /* |
649 | * if the buffer is stale, clear all the external state associated with | |
650 | * it. We need to keep flags such as how we allocated the buffer memory | |
651 | * intact here. | |
652 | */ | |
ce8e922c NS |
653 | if (bp->b_flags & XBF_STALE) { |
654 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
cfb02852 | 655 | ASSERT(bp->b_iodone == NULL); |
611c9946 | 656 | bp->b_flags &= _XBF_KMEM | _XBF_PAGES; |
1813dd64 | 657 | bp->b_ops = NULL; |
2f926587 | 658 | } |
0b1b213f CH |
659 | |
660 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ff6d6af2 | 661 | XFS_STATS_INC(btp->bt_mount, xb_get_locked); |
b027d4c9 DC |
662 | *found_bp = bp; |
663 | return 0; | |
1da177e4 LT |
664 | } |
665 | ||
8925a3dc DC |
666 | struct xfs_buf * |
667 | xfs_buf_incore( | |
668 | struct xfs_buftarg *target, | |
669 | xfs_daddr_t blkno, | |
670 | size_t numblks, | |
671 | xfs_buf_flags_t flags) | |
672 | { | |
b027d4c9 DC |
673 | struct xfs_buf *bp; |
674 | int error; | |
8925a3dc | 675 | DEFINE_SINGLE_BUF_MAP(map, blkno, numblks); |
b027d4c9 DC |
676 | |
677 | error = xfs_buf_find(target, &map, 1, flags, NULL, &bp); | |
678 | if (error) | |
679 | return NULL; | |
680 | return bp; | |
8925a3dc DC |
681 | } |
682 | ||
1da177e4 | 683 | /* |
3815832a DC |
684 | * Assembles a buffer covering the specified range. The code is optimised for |
685 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
686 | * more hits than misses. | |
1da177e4 | 687 | */ |
3848b5f6 | 688 | int |
6dde2707 DC |
689 | xfs_buf_get_map( |
690 | struct xfs_buftarg *target, | |
691 | struct xfs_buf_map *map, | |
692 | int nmaps, | |
3848b5f6 DW |
693 | xfs_buf_flags_t flags, |
694 | struct xfs_buf **bpp) | |
1da177e4 | 695 | { |
3815832a DC |
696 | struct xfs_buf *bp; |
697 | struct xfs_buf *new_bp; | |
0e6e847f | 698 | int error = 0; |
1da177e4 | 699 | |
3848b5f6 | 700 | *bpp = NULL; |
b027d4c9 | 701 | error = xfs_buf_find(target, map, nmaps, flags, NULL, &bp); |
3848b5f6 | 702 | if (!error) |
3815832a | 703 | goto found; |
3848b5f6 DW |
704 | if (error != -ENOENT) |
705 | return error; | |
3815832a | 706 | |
32dff5e5 DW |
707 | error = _xfs_buf_alloc(target, map, nmaps, flags, &new_bp); |
708 | if (error) | |
3848b5f6 | 709 | return error; |
1da177e4 | 710 | |
fe2429b0 DC |
711 | error = xfs_buf_allocate_memory(new_bp, flags); |
712 | if (error) { | |
3e85c868 | 713 | xfs_buf_free(new_bp); |
3848b5f6 | 714 | return error; |
fe2429b0 DC |
715 | } |
716 | ||
b027d4c9 DC |
717 | error = xfs_buf_find(target, map, nmaps, flags, new_bp, &bp); |
718 | if (error) { | |
fe2429b0 | 719 | xfs_buf_free(new_bp); |
3848b5f6 | 720 | return error; |
3815832a DC |
721 | } |
722 | ||
fe2429b0 DC |
723 | if (bp != new_bp) |
724 | xfs_buf_free(new_bp); | |
1da177e4 | 725 | |
3815832a | 726 | found: |
611c9946 | 727 | if (!bp->b_addr) { |
ce8e922c | 728 | error = _xfs_buf_map_pages(bp, flags); |
1da177e4 | 729 | if (unlikely(error)) { |
93baa55a DW |
730 | xfs_warn_ratelimited(target->bt_mount, |
731 | "%s: failed to map %u pages", __func__, | |
732 | bp->b_page_count); | |
a8acad70 | 733 | xfs_buf_relse(bp); |
3848b5f6 | 734 | return error; |
1da177e4 LT |
735 | } |
736 | } | |
737 | ||
b79f4a1c DC |
738 | /* |
739 | * Clear b_error if this is a lookup from a caller that doesn't expect | |
740 | * valid data to be found in the buffer. | |
741 | */ | |
742 | if (!(flags & XBF_READ)) | |
743 | xfs_buf_ioerror(bp, 0); | |
744 | ||
ff6d6af2 | 745 | XFS_STATS_INC(target->bt_mount, xb_get); |
0b1b213f | 746 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
3848b5f6 DW |
747 | *bpp = bp; |
748 | return 0; | |
1da177e4 LT |
749 | } |
750 | ||
5d765b97 CH |
751 | STATIC int |
752 | _xfs_buf_read( | |
753 | xfs_buf_t *bp, | |
754 | xfs_buf_flags_t flags) | |
755 | { | |
43ff2122 | 756 | ASSERT(!(flags & XBF_WRITE)); |
f4b42421 | 757 | ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL); |
5d765b97 | 758 | |
43ff2122 | 759 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD); |
1d5ae5df | 760 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 761 | |
6af88cda | 762 | return xfs_buf_submit(bp); |
5d765b97 CH |
763 | } |
764 | ||
1aff5696 | 765 | /* |
75d02303 | 766 | * Reverify a buffer found in cache without an attached ->b_ops. |
add46b3b | 767 | * |
75d02303 BF |
768 | * If the caller passed an ops structure and the buffer doesn't have ops |
769 | * assigned, set the ops and use it to verify the contents. If verification | |
770 | * fails, clear XBF_DONE. We assume the buffer has no recorded errors and is | |
771 | * already in XBF_DONE state on entry. | |
add46b3b | 772 | * |
75d02303 BF |
773 | * Under normal operations, every in-core buffer is verified on read I/O |
774 | * completion. There are two scenarios that can lead to in-core buffers without | |
775 | * an assigned ->b_ops. The first is during log recovery of buffers on a V4 | |
776 | * filesystem, though these buffers are purged at the end of recovery. The | |
777 | * other is online repair, which intentionally reads with a NULL buffer ops to | |
778 | * run several verifiers across an in-core buffer in order to establish buffer | |
779 | * type. If repair can't establish that, the buffer will be left in memory | |
780 | * with NULL buffer ops. | |
1aff5696 DW |
781 | */ |
782 | int | |
75d02303 | 783 | xfs_buf_reverify( |
1aff5696 DW |
784 | struct xfs_buf *bp, |
785 | const struct xfs_buf_ops *ops) | |
786 | { | |
787 | ASSERT(bp->b_flags & XBF_DONE); | |
788 | ASSERT(bp->b_error == 0); | |
789 | ||
790 | if (!ops || bp->b_ops) | |
791 | return 0; | |
792 | ||
793 | bp->b_ops = ops; | |
794 | bp->b_ops->verify_read(bp); | |
795 | if (bp->b_error) | |
796 | bp->b_flags &= ~XBF_DONE; | |
797 | return bp->b_error; | |
798 | } | |
799 | ||
4ed8e27b | 800 | int |
6dde2707 DC |
801 | xfs_buf_read_map( |
802 | struct xfs_buftarg *target, | |
803 | struct xfs_buf_map *map, | |
804 | int nmaps, | |
c3f8fc73 | 805 | xfs_buf_flags_t flags, |
4ed8e27b | 806 | struct xfs_buf **bpp, |
cdbcf82b DW |
807 | const struct xfs_buf_ops *ops, |
808 | xfs_failaddr_t fa) | |
1da177e4 | 809 | { |
6dde2707 | 810 | struct xfs_buf *bp; |
3848b5f6 | 811 | int error; |
ce8e922c NS |
812 | |
813 | flags |= XBF_READ; | |
4ed8e27b | 814 | *bpp = NULL; |
ce8e922c | 815 | |
3848b5f6 DW |
816 | error = xfs_buf_get_map(target, map, nmaps, flags, &bp); |
817 | if (error) | |
4ed8e27b | 818 | return error; |
0b1b213f | 819 | |
1aff5696 DW |
820 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
821 | ||
822 | if (!(bp->b_flags & XBF_DONE)) { | |
4ed8e27b | 823 | /* Initiate the buffer read and wait. */ |
1aff5696 DW |
824 | XFS_STATS_INC(target->bt_mount, xb_get_read); |
825 | bp->b_ops = ops; | |
4ed8e27b DW |
826 | error = _xfs_buf_read(bp, flags); |
827 | ||
828 | /* Readahead iodone already dropped the buffer, so exit. */ | |
829 | if (flags & XBF_ASYNC) | |
830 | return 0; | |
831 | } else { | |
832 | /* Buffer already read; all we need to do is check it. */ | |
833 | error = xfs_buf_reverify(bp, ops); | |
834 | ||
835 | /* Readahead already finished; drop the buffer and exit. */ | |
836 | if (flags & XBF_ASYNC) { | |
837 | xfs_buf_relse(bp); | |
838 | return 0; | |
839 | } | |
840 | ||
841 | /* We do not want read in the flags */ | |
842 | bp->b_flags &= ~XBF_READ; | |
843 | ASSERT(bp->b_ops != NULL || ops == NULL); | |
1aff5696 DW |
844 | } |
845 | ||
4ed8e27b DW |
846 | /* |
847 | * If we've had a read error, then the contents of the buffer are | |
848 | * invalid and should not be used. To ensure that a followup read tries | |
849 | * to pull the buffer from disk again, we clear the XBF_DONE flag and | |
850 | * mark the buffer stale. This ensures that anyone who has a current | |
851 | * reference to the buffer will interpret it's contents correctly and | |
852 | * future cache lookups will also treat it as an empty, uninitialised | |
853 | * buffer. | |
854 | */ | |
855 | if (error) { | |
856 | if (!XFS_FORCED_SHUTDOWN(target->bt_mount)) | |
cdbcf82b | 857 | xfs_buf_ioerror_alert(bp, fa); |
1aff5696 | 858 | |
4ed8e27b DW |
859 | bp->b_flags &= ~XBF_DONE; |
860 | xfs_buf_stale(bp); | |
1aff5696 | 861 | xfs_buf_relse(bp); |
4ed8e27b DW |
862 | |
863 | /* bad CRC means corrupted metadata */ | |
864 | if (error == -EFSBADCRC) | |
865 | error = -EFSCORRUPTED; | |
866 | return error; | |
1da177e4 LT |
867 | } |
868 | ||
4ed8e27b DW |
869 | *bpp = bp; |
870 | return 0; | |
1da177e4 LT |
871 | } |
872 | ||
1da177e4 | 873 | /* |
ce8e922c NS |
874 | * If we are not low on memory then do the readahead in a deadlock |
875 | * safe manner. | |
1da177e4 LT |
876 | */ |
877 | void | |
6dde2707 DC |
878 | xfs_buf_readahead_map( |
879 | struct xfs_buftarg *target, | |
880 | struct xfs_buf_map *map, | |
c3f8fc73 | 881 | int nmaps, |
1813dd64 | 882 | const struct xfs_buf_ops *ops) |
1da177e4 | 883 | { |
4ed8e27b DW |
884 | struct xfs_buf *bp; |
885 | ||
efa7c9f9 | 886 | if (bdi_read_congested(target->bt_bdev->bd_bdi)) |
1da177e4 LT |
887 | return; |
888 | ||
6dde2707 | 889 | xfs_buf_read_map(target, map, nmaps, |
cdbcf82b DW |
890 | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD, &bp, ops, |
891 | __this_address); | |
1da177e4 LT |
892 | } |
893 | ||
5adc94c2 DC |
894 | /* |
895 | * Read an uncached buffer from disk. Allocates and returns a locked | |
896 | * buffer containing the disk contents or nothing. | |
897 | */ | |
ba372674 | 898 | int |
5adc94c2 | 899 | xfs_buf_read_uncached( |
5adc94c2 DC |
900 | struct xfs_buftarg *target, |
901 | xfs_daddr_t daddr, | |
e70b73f8 | 902 | size_t numblks, |
c3f8fc73 | 903 | int flags, |
ba372674 | 904 | struct xfs_buf **bpp, |
1813dd64 | 905 | const struct xfs_buf_ops *ops) |
5adc94c2 | 906 | { |
eab4e633 | 907 | struct xfs_buf *bp; |
2842b6db | 908 | int error; |
5adc94c2 | 909 | |
ba372674 DC |
910 | *bpp = NULL; |
911 | ||
2842b6db DW |
912 | error = xfs_buf_get_uncached(target, numblks, flags, &bp); |
913 | if (error) | |
914 | return error; | |
5adc94c2 DC |
915 | |
916 | /* set up the buffer for a read IO */ | |
3e85c868 | 917 | ASSERT(bp->b_map_count == 1); |
ba372674 | 918 | bp->b_bn = XFS_BUF_DADDR_NULL; /* always null for uncached buffers */ |
3e85c868 | 919 | bp->b_maps[0].bm_bn = daddr; |
cbb7baab | 920 | bp->b_flags |= XBF_READ; |
1813dd64 | 921 | bp->b_ops = ops; |
5adc94c2 | 922 | |
6af88cda | 923 | xfs_buf_submit(bp); |
ba372674 | 924 | if (bp->b_error) { |
2842b6db | 925 | error = bp->b_error; |
83a0adc3 | 926 | xfs_buf_relse(bp); |
ba372674 | 927 | return error; |
83a0adc3 | 928 | } |
ba372674 DC |
929 | |
930 | *bpp = bp; | |
931 | return 0; | |
1da177e4 LT |
932 | } |
933 | ||
2842b6db | 934 | int |
686865f7 DC |
935 | xfs_buf_get_uncached( |
936 | struct xfs_buftarg *target, | |
e70b73f8 | 937 | size_t numblks, |
2842b6db DW |
938 | int flags, |
939 | struct xfs_buf **bpp) | |
1da177e4 | 940 | { |
e70b73f8 | 941 | unsigned long page_count; |
1fa40b01 | 942 | int error, i; |
3e85c868 DC |
943 | struct xfs_buf *bp; |
944 | DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks); | |
1da177e4 | 945 | |
2842b6db DW |
946 | *bpp = NULL; |
947 | ||
c891c30a | 948 | /* flags might contain irrelevant bits, pass only what we care about */ |
32dff5e5 DW |
949 | error = _xfs_buf_alloc(target, &map, 1, flags & XBF_NO_IOACCT, &bp); |
950 | if (error) | |
1da177e4 | 951 | goto fail; |
1da177e4 | 952 | |
e70b73f8 | 953 | page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT; |
87937bf8 | 954 | error = _xfs_buf_get_pages(bp, page_count); |
1fa40b01 | 955 | if (error) |
1da177e4 LT |
956 | goto fail_free_buf; |
957 | ||
1fa40b01 | 958 | for (i = 0; i < page_count; i++) { |
686865f7 | 959 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
2842b6db DW |
960 | if (!bp->b_pages[i]) { |
961 | error = -ENOMEM; | |
1fa40b01 | 962 | goto fail_free_mem; |
2842b6db | 963 | } |
1da177e4 | 964 | } |
1fa40b01 | 965 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 966 | |
611c9946 | 967 | error = _xfs_buf_map_pages(bp, 0); |
1fa40b01 | 968 | if (unlikely(error)) { |
4f10700a | 969 | xfs_warn(target->bt_mount, |
08e96e1a | 970 | "%s: failed to map pages", __func__); |
1da177e4 | 971 | goto fail_free_mem; |
1fa40b01 | 972 | } |
1da177e4 | 973 | |
686865f7 | 974 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
2842b6db DW |
975 | *bpp = bp; |
976 | return 0; | |
1fa40b01 | 977 | |
1da177e4 | 978 | fail_free_mem: |
1fa40b01 CH |
979 | while (--i >= 0) |
980 | __free_page(bp->b_pages[i]); | |
ca165b88 | 981 | _xfs_buf_free_pages(bp); |
1da177e4 | 982 | fail_free_buf: |
3e85c868 | 983 | xfs_buf_free_maps(bp); |
377bcd5f | 984 | kmem_cache_free(xfs_buf_zone, bp); |
1da177e4 | 985 | fail: |
2842b6db | 986 | return error; |
1da177e4 LT |
987 | } |
988 | ||
989 | /* | |
1da177e4 LT |
990 | * Increment reference count on buffer, to hold the buffer concurrently |
991 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
992 | * Must hold the buffer already to call this function. |
993 | */ | |
994 | void | |
ce8e922c NS |
995 | xfs_buf_hold( |
996 | xfs_buf_t *bp) | |
1da177e4 | 997 | { |
0b1b213f | 998 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 999 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
1000 | } |
1001 | ||
1002 | /* | |
9c7504aa BF |
1003 | * Release a hold on the specified buffer. If the hold count is 1, the buffer is |
1004 | * placed on LRU or freed (depending on b_lru_ref). | |
1da177e4 LT |
1005 | */ |
1006 | void | |
ce8e922c NS |
1007 | xfs_buf_rele( |
1008 | xfs_buf_t *bp) | |
1da177e4 | 1009 | { |
74f75a0c | 1010 | struct xfs_perag *pag = bp->b_pag; |
9c7504aa BF |
1011 | bool release; |
1012 | bool freebuf = false; | |
1da177e4 | 1013 | |
0b1b213f | 1014 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 1015 | |
74f75a0c | 1016 | if (!pag) { |
430cbeb8 | 1017 | ASSERT(list_empty(&bp->b_lru)); |
9c7504aa BF |
1018 | if (atomic_dec_and_test(&bp->b_hold)) { |
1019 | xfs_buf_ioacct_dec(bp); | |
fad3aa1e | 1020 | xfs_buf_free(bp); |
9c7504aa | 1021 | } |
fad3aa1e NS |
1022 | return; |
1023 | } | |
1024 | ||
3790689f | 1025 | ASSERT(atomic_read(&bp->b_hold) > 0); |
a4082357 | 1026 | |
37fd1678 DC |
1027 | /* |
1028 | * We grab the b_lock here first to serialise racing xfs_buf_rele() | |
1029 | * calls. The pag_buf_lock being taken on the last reference only | |
1030 | * serialises against racing lookups in xfs_buf_find(). IOWs, the second | |
1031 | * to last reference we drop here is not serialised against the last | |
1032 | * reference until we take bp->b_lock. Hence if we don't grab b_lock | |
1033 | * first, the last "release" reference can win the race to the lock and | |
1034 | * free the buffer before the second-to-last reference is processed, | |
1035 | * leading to a use-after-free scenario. | |
1036 | */ | |
9c7504aa | 1037 | spin_lock(&bp->b_lock); |
37fd1678 | 1038 | release = atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock); |
9c7504aa BF |
1039 | if (!release) { |
1040 | /* | |
1041 | * Drop the in-flight state if the buffer is already on the LRU | |
1042 | * and it holds the only reference. This is racy because we | |
1043 | * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT | |
1044 | * ensures the decrement occurs only once per-buf. | |
1045 | */ | |
1046 | if ((atomic_read(&bp->b_hold) == 1) && !list_empty(&bp->b_lru)) | |
63db7c81 | 1047 | __xfs_buf_ioacct_dec(bp); |
9c7504aa BF |
1048 | goto out_unlock; |
1049 | } | |
1050 | ||
1051 | /* the last reference has been dropped ... */ | |
63db7c81 | 1052 | __xfs_buf_ioacct_dec(bp); |
9c7504aa BF |
1053 | if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) { |
1054 | /* | |
1055 | * If the buffer is added to the LRU take a new reference to the | |
1056 | * buffer for the LRU and clear the (now stale) dispose list | |
1057 | * state flag | |
1058 | */ | |
1059 | if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) { | |
1060 | bp->b_state &= ~XFS_BSTATE_DISPOSE; | |
1061 | atomic_inc(&bp->b_hold); | |
1da177e4 | 1062 | } |
9c7504aa BF |
1063 | spin_unlock(&pag->pag_buf_lock); |
1064 | } else { | |
1065 | /* | |
1066 | * most of the time buffers will already be removed from the | |
1067 | * LRU, so optimise that case by checking for the | |
1068 | * XFS_BSTATE_DISPOSE flag indicating the last list the buffer | |
1069 | * was on was the disposal list | |
1070 | */ | |
1071 | if (!(bp->b_state & XFS_BSTATE_DISPOSE)) { | |
1072 | list_lru_del(&bp->b_target->bt_lru, &bp->b_lru); | |
1073 | } else { | |
1074 | ASSERT(list_empty(&bp->b_lru)); | |
1da177e4 | 1075 | } |
9c7504aa BF |
1076 | |
1077 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); | |
6031e73a LS |
1078 | rhashtable_remove_fast(&pag->pag_buf_hash, &bp->b_rhash_head, |
1079 | xfs_buf_hash_params); | |
9c7504aa BF |
1080 | spin_unlock(&pag->pag_buf_lock); |
1081 | xfs_perag_put(pag); | |
1082 | freebuf = true; | |
1da177e4 | 1083 | } |
9c7504aa BF |
1084 | |
1085 | out_unlock: | |
1086 | spin_unlock(&bp->b_lock); | |
1087 | ||
1088 | if (freebuf) | |
1089 | xfs_buf_free(bp); | |
1da177e4 LT |
1090 | } |
1091 | ||
1092 | ||
1093 | /* | |
0e6e847f | 1094 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
1095 | * |
1096 | * If we come across a stale, pinned, locked buffer, we know that we are | |
1097 | * being asked to lock a buffer that has been reallocated. Because it is | |
1098 | * pinned, we know that the log has not been pushed to disk and hence it | |
1099 | * will still be locked. Rather than continuing to have trylock attempts | |
1100 | * fail until someone else pushes the log, push it ourselves before | |
1101 | * returning. This means that the xfsaild will not get stuck trying | |
1102 | * to push on stale inode buffers. | |
1da177e4 LT |
1103 | */ |
1104 | int | |
0c842ad4 CH |
1105 | xfs_buf_trylock( |
1106 | struct xfs_buf *bp) | |
1da177e4 LT |
1107 | { |
1108 | int locked; | |
1109 | ||
ce8e922c | 1110 | locked = down_trylock(&bp->b_sema) == 0; |
fa6c668d | 1111 | if (locked) |
479c6412 | 1112 | trace_xfs_buf_trylock(bp, _RET_IP_); |
fa6c668d | 1113 | else |
479c6412 | 1114 | trace_xfs_buf_trylock_fail(bp, _RET_IP_); |
0c842ad4 | 1115 | return locked; |
1da177e4 | 1116 | } |
1da177e4 LT |
1117 | |
1118 | /* | |
0e6e847f | 1119 | * Lock a buffer object. |
ed3b4d6c DC |
1120 | * |
1121 | * If we come across a stale, pinned, locked buffer, we know that we | |
1122 | * are being asked to lock a buffer that has been reallocated. Because | |
1123 | * it is pinned, we know that the log has not been pushed to disk and | |
1124 | * hence it will still be locked. Rather than sleeping until someone | |
1125 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 1126 | */ |
ce8e922c NS |
1127 | void |
1128 | xfs_buf_lock( | |
0c842ad4 | 1129 | struct xfs_buf *bp) |
1da177e4 | 1130 | { |
0b1b213f CH |
1131 | trace_xfs_buf_lock(bp, _RET_IP_); |
1132 | ||
ed3b4d6c | 1133 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
dbd329f1 | 1134 | xfs_log_force(bp->b_mount, 0); |
ce8e922c | 1135 | down(&bp->b_sema); |
0b1b213f CH |
1136 | |
1137 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
1138 | } |
1139 | ||
1da177e4 | 1140 | void |
ce8e922c | 1141 | xfs_buf_unlock( |
0c842ad4 | 1142 | struct xfs_buf *bp) |
1da177e4 | 1143 | { |
20e8a063 BF |
1144 | ASSERT(xfs_buf_islocked(bp)); |
1145 | ||
ce8e922c | 1146 | up(&bp->b_sema); |
0b1b213f | 1147 | trace_xfs_buf_unlock(bp, _RET_IP_); |
1da177e4 LT |
1148 | } |
1149 | ||
ce8e922c NS |
1150 | STATIC void |
1151 | xfs_buf_wait_unpin( | |
1152 | xfs_buf_t *bp) | |
1da177e4 LT |
1153 | { |
1154 | DECLARE_WAITQUEUE (wait, current); | |
1155 | ||
ce8e922c | 1156 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
1157 | return; |
1158 | ||
ce8e922c | 1159 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1160 | for (;;) { |
1161 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 1162 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 1163 | break; |
7eaceacc | 1164 | io_schedule(); |
1da177e4 | 1165 | } |
ce8e922c | 1166 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1167 | set_current_state(TASK_RUNNING); |
1168 | } | |
1169 | ||
1170 | /* | |
1171 | * Buffer Utility Routines | |
1172 | */ | |
1173 | ||
e8aaba9a DC |
1174 | void |
1175 | xfs_buf_ioend( | |
1176 | struct xfs_buf *bp) | |
1da177e4 | 1177 | { |
e8aaba9a DC |
1178 | bool read = bp->b_flags & XBF_READ; |
1179 | ||
1180 | trace_xfs_buf_iodone(bp, _RET_IP_); | |
1813dd64 DC |
1181 | |
1182 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); | |
d5929de8 | 1183 | |
61be9c52 DC |
1184 | /* |
1185 | * Pull in IO completion errors now. We are guaranteed to be running | |
1186 | * single threaded, so we don't need the lock to read b_io_error. | |
1187 | */ | |
1188 | if (!bp->b_error && bp->b_io_error) | |
1189 | xfs_buf_ioerror(bp, bp->b_io_error); | |
1190 | ||
e8aaba9a DC |
1191 | /* Only validate buffers that were read without errors */ |
1192 | if (read && !bp->b_error && bp->b_ops) { | |
1193 | ASSERT(!bp->b_iodone); | |
1813dd64 | 1194 | bp->b_ops->verify_read(bp); |
e8aaba9a DC |
1195 | } |
1196 | ||
1197 | if (!bp->b_error) | |
1198 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1199 | |
80f6c29d | 1200 | if (bp->b_iodone) |
ce8e922c NS |
1201 | (*(bp->b_iodone))(bp); |
1202 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 | 1203 | xfs_buf_relse(bp); |
595bff75 | 1204 | else |
1813dd64 | 1205 | complete(&bp->b_iowait); |
1da177e4 LT |
1206 | } |
1207 | ||
e8aaba9a DC |
1208 | static void |
1209 | xfs_buf_ioend_work( | |
1210 | struct work_struct *work) | |
1da177e4 | 1211 | { |
e8aaba9a | 1212 | struct xfs_buf *bp = |
b29c70f5 | 1213 | container_of(work, xfs_buf_t, b_ioend_work); |
0b1b213f | 1214 | |
e8aaba9a DC |
1215 | xfs_buf_ioend(bp); |
1216 | } | |
1da177e4 | 1217 | |
211fe1a4 | 1218 | static void |
e8aaba9a DC |
1219 | xfs_buf_ioend_async( |
1220 | struct xfs_buf *bp) | |
1221 | { | |
b29c70f5 | 1222 | INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work); |
dbd329f1 | 1223 | queue_work(bp->b_mount->m_buf_workqueue, &bp->b_ioend_work); |
1da177e4 LT |
1224 | } |
1225 | ||
1da177e4 | 1226 | void |
31ca03c9 | 1227 | __xfs_buf_ioerror( |
ce8e922c | 1228 | xfs_buf_t *bp, |
31ca03c9 DW |
1229 | int error, |
1230 | xfs_failaddr_t failaddr) | |
1da177e4 | 1231 | { |
2451337d DC |
1232 | ASSERT(error <= 0 && error >= -1000); |
1233 | bp->b_error = error; | |
31ca03c9 | 1234 | trace_xfs_buf_ioerror(bp, error, failaddr); |
1da177e4 LT |
1235 | } |
1236 | ||
901796af CH |
1237 | void |
1238 | xfs_buf_ioerror_alert( | |
1239 | struct xfs_buf *bp, | |
cdbcf82b | 1240 | xfs_failaddr_t func) |
901796af | 1241 | { |
dbd329f1 | 1242 | xfs_alert(bp->b_mount, |
cdbcf82b | 1243 | "metadata I/O error in \"%pS\" at daddr 0x%llx len %d error %d", |
c219b015 DW |
1244 | func, (uint64_t)XFS_BUF_ADDR(bp), bp->b_length, |
1245 | -bp->b_error); | |
901796af CH |
1246 | } |
1247 | ||
a2dcf5df CH |
1248 | int |
1249 | xfs_bwrite( | |
1250 | struct xfs_buf *bp) | |
1251 | { | |
1252 | int error; | |
1253 | ||
1254 | ASSERT(xfs_buf_islocked(bp)); | |
1255 | ||
1256 | bp->b_flags |= XBF_WRITE; | |
27187754 DC |
1257 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q | |
1258 | XBF_WRITE_FAIL | XBF_DONE); | |
a2dcf5df | 1259 | |
6af88cda | 1260 | error = xfs_buf_submit(bp); |
dbd329f1 CH |
1261 | if (error) |
1262 | xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR); | |
a2dcf5df CH |
1263 | return error; |
1264 | } | |
1265 | ||
9bdd9bd6 | 1266 | static void |
ce8e922c | 1267 | xfs_buf_bio_end_io( |
4246a0b6 | 1268 | struct bio *bio) |
1da177e4 | 1269 | { |
9bdd9bd6 | 1270 | struct xfs_buf *bp = (struct xfs_buf *)bio->bi_private; |
1da177e4 | 1271 | |
37eb17e6 DC |
1272 | /* |
1273 | * don't overwrite existing errors - otherwise we can lose errors on | |
1274 | * buffers that require multiple bios to complete. | |
1275 | */ | |
4e4cbee9 CH |
1276 | if (bio->bi_status) { |
1277 | int error = blk_status_to_errno(bio->bi_status); | |
1278 | ||
1279 | cmpxchg(&bp->b_io_error, 0, error); | |
1280 | } | |
1da177e4 | 1281 | |
37eb17e6 | 1282 | if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
73c77e2c JB |
1283 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); |
1284 | ||
e8aaba9a DC |
1285 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
1286 | xfs_buf_ioend_async(bp); | |
1da177e4 | 1287 | bio_put(bio); |
1da177e4 LT |
1288 | } |
1289 | ||
3e85c868 DC |
1290 | static void |
1291 | xfs_buf_ioapply_map( | |
1292 | struct xfs_buf *bp, | |
1293 | int map, | |
1294 | int *buf_offset, | |
1295 | int *count, | |
2123ef85 | 1296 | int op) |
1da177e4 | 1297 | { |
3e85c868 DC |
1298 | int page_index; |
1299 | int total_nr_pages = bp->b_page_count; | |
1300 | int nr_pages; | |
1301 | struct bio *bio; | |
1302 | sector_t sector = bp->b_maps[map].bm_bn; | |
1303 | int size; | |
1304 | int offset; | |
1da177e4 | 1305 | |
3e85c868 DC |
1306 | /* skip the pages in the buffer before the start offset */ |
1307 | page_index = 0; | |
1308 | offset = *buf_offset; | |
1309 | while (offset >= PAGE_SIZE) { | |
1310 | page_index++; | |
1311 | offset -= PAGE_SIZE; | |
f538d4da CH |
1312 | } |
1313 | ||
3e85c868 DC |
1314 | /* |
1315 | * Limit the IO size to the length of the current vector, and update the | |
1316 | * remaining IO count for the next time around. | |
1317 | */ | |
1318 | size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count); | |
1319 | *count -= size; | |
1320 | *buf_offset += size; | |
34951f5c | 1321 | |
1da177e4 | 1322 | next_chunk: |
ce8e922c | 1323 | atomic_inc(&bp->b_io_remaining); |
c908e380 | 1324 | nr_pages = min(total_nr_pages, BIO_MAX_PAGES); |
1da177e4 LT |
1325 | |
1326 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
74d46992 | 1327 | bio_set_dev(bio, bp->b_target->bt_bdev); |
4f024f37 | 1328 | bio->bi_iter.bi_sector = sector; |
ce8e922c NS |
1329 | bio->bi_end_io = xfs_buf_bio_end_io; |
1330 | bio->bi_private = bp; | |
2123ef85 | 1331 | bio->bi_opf = op; |
0e6e847f | 1332 | |
3e85c868 | 1333 | for (; size && nr_pages; nr_pages--, page_index++) { |
0e6e847f | 1334 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1335 | |
1336 | if (nbytes > size) | |
1337 | nbytes = size; | |
1338 | ||
3e85c868 DC |
1339 | rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes, |
1340 | offset); | |
ce8e922c | 1341 | if (rbytes < nbytes) |
1da177e4 LT |
1342 | break; |
1343 | ||
1344 | offset = 0; | |
aa0e8833 | 1345 | sector += BTOBB(nbytes); |
1da177e4 LT |
1346 | size -= nbytes; |
1347 | total_nr_pages--; | |
1348 | } | |
1349 | ||
4f024f37 | 1350 | if (likely(bio->bi_iter.bi_size)) { |
73c77e2c JB |
1351 | if (xfs_buf_is_vmapped(bp)) { |
1352 | flush_kernel_vmap_range(bp->b_addr, | |
1353 | xfs_buf_vmap_len(bp)); | |
1354 | } | |
4e49ea4a | 1355 | submit_bio(bio); |
1da177e4 LT |
1356 | if (size) |
1357 | goto next_chunk; | |
1358 | } else { | |
37eb17e6 DC |
1359 | /* |
1360 | * This is guaranteed not to be the last io reference count | |
595bff75 | 1361 | * because the caller (xfs_buf_submit) holds a count itself. |
37eb17e6 DC |
1362 | */ |
1363 | atomic_dec(&bp->b_io_remaining); | |
2451337d | 1364 | xfs_buf_ioerror(bp, -EIO); |
ec53d1db | 1365 | bio_put(bio); |
1da177e4 | 1366 | } |
3e85c868 DC |
1367 | |
1368 | } | |
1369 | ||
1370 | STATIC void | |
1371 | _xfs_buf_ioapply( | |
1372 | struct xfs_buf *bp) | |
1373 | { | |
1374 | struct blk_plug plug; | |
50bfcd0c | 1375 | int op; |
3e85c868 DC |
1376 | int offset; |
1377 | int size; | |
1378 | int i; | |
1379 | ||
c163f9a1 DC |
1380 | /* |
1381 | * Make sure we capture only current IO errors rather than stale errors | |
1382 | * left over from previous use of the buffer (e.g. failed readahead). | |
1383 | */ | |
1384 | bp->b_error = 0; | |
1385 | ||
3e85c868 | 1386 | if (bp->b_flags & XBF_WRITE) { |
50bfcd0c | 1387 | op = REQ_OP_WRITE; |
1813dd64 DC |
1388 | |
1389 | /* | |
1390 | * Run the write verifier callback function if it exists. If | |
1391 | * this function fails it will mark the buffer with an error and | |
1392 | * the IO should not be dispatched. | |
1393 | */ | |
1394 | if (bp->b_ops) { | |
1395 | bp->b_ops->verify_write(bp); | |
1396 | if (bp->b_error) { | |
dbd329f1 | 1397 | xfs_force_shutdown(bp->b_mount, |
1813dd64 DC |
1398 | SHUTDOWN_CORRUPT_INCORE); |
1399 | return; | |
1400 | } | |
400b9d88 | 1401 | } else if (bp->b_bn != XFS_BUF_DADDR_NULL) { |
dbd329f1 | 1402 | struct xfs_mount *mp = bp->b_mount; |
400b9d88 DC |
1403 | |
1404 | /* | |
1405 | * non-crc filesystems don't attach verifiers during | |
1406 | * log recovery, so don't warn for such filesystems. | |
1407 | */ | |
1408 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
1409 | xfs_warn(mp, | |
c219b015 | 1410 | "%s: no buf ops on daddr 0x%llx len %d", |
400b9d88 | 1411 | __func__, bp->b_bn, bp->b_length); |
9c712a13 DW |
1412 | xfs_hex_dump(bp->b_addr, |
1413 | XFS_CORRUPTION_DUMP_LEN); | |
400b9d88 DC |
1414 | dump_stack(); |
1415 | } | |
1813dd64 | 1416 | } |
3e85c868 | 1417 | } else { |
50bfcd0c | 1418 | op = REQ_OP_READ; |
2123ef85 CH |
1419 | if (bp->b_flags & XBF_READ_AHEAD) |
1420 | op |= REQ_RAHEAD; | |
3e85c868 DC |
1421 | } |
1422 | ||
1423 | /* we only use the buffer cache for meta-data */ | |
2123ef85 | 1424 | op |= REQ_META; |
3e85c868 DC |
1425 | |
1426 | /* | |
1427 | * Walk all the vectors issuing IO on them. Set up the initial offset | |
1428 | * into the buffer and the desired IO size before we start - | |
1429 | * _xfs_buf_ioapply_vec() will modify them appropriately for each | |
1430 | * subsequent call. | |
1431 | */ | |
1432 | offset = bp->b_offset; | |
8124b9b6 | 1433 | size = BBTOB(bp->b_length); |
3e85c868 DC |
1434 | blk_start_plug(&plug); |
1435 | for (i = 0; i < bp->b_map_count; i++) { | |
2123ef85 | 1436 | xfs_buf_ioapply_map(bp, i, &offset, &size, op); |
3e85c868 DC |
1437 | if (bp->b_error) |
1438 | break; | |
1439 | if (size <= 0) | |
1440 | break; /* all done */ | |
1441 | } | |
1442 | blk_finish_plug(&plug); | |
1da177e4 LT |
1443 | } |
1444 | ||
595bff75 | 1445 | /* |
bb00b6f1 | 1446 | * Wait for I/O completion of a sync buffer and return the I/O error code. |
595bff75 | 1447 | */ |
eaebb515 | 1448 | static int |
bb00b6f1 | 1449 | xfs_buf_iowait( |
595bff75 | 1450 | struct xfs_buf *bp) |
1da177e4 | 1451 | { |
bb00b6f1 BF |
1452 | ASSERT(!(bp->b_flags & XBF_ASYNC)); |
1453 | ||
1454 | trace_xfs_buf_iowait(bp, _RET_IP_); | |
1455 | wait_for_completion(&bp->b_iowait); | |
1456 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
1457 | ||
1458 | return bp->b_error; | |
1459 | } | |
1460 | ||
1461 | /* | |
1462 | * Buffer I/O submission path, read or write. Asynchronous submission transfers | |
1463 | * the buffer lock ownership and the current reference to the IO. It is not | |
1464 | * safe to reference the buffer after a call to this function unless the caller | |
1465 | * holds an additional reference itself. | |
1466 | */ | |
1467 | int | |
1468 | __xfs_buf_submit( | |
1469 | struct xfs_buf *bp, | |
1470 | bool wait) | |
1471 | { | |
1472 | int error = 0; | |
1473 | ||
595bff75 | 1474 | trace_xfs_buf_submit(bp, _RET_IP_); |
1da177e4 | 1475 | |
43ff2122 | 1476 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
595bff75 DC |
1477 | |
1478 | /* on shutdown we stale and complete the buffer immediately */ | |
dbd329f1 | 1479 | if (XFS_FORCED_SHUTDOWN(bp->b_mount)) { |
595bff75 DC |
1480 | xfs_buf_ioerror(bp, -EIO); |
1481 | bp->b_flags &= ~XBF_DONE; | |
1482 | xfs_buf_stale(bp); | |
465fa17f | 1483 | xfs_buf_ioend(bp); |
eaebb515 | 1484 | return -EIO; |
595bff75 | 1485 | } |
1da177e4 | 1486 | |
bb00b6f1 BF |
1487 | /* |
1488 | * Grab a reference so the buffer does not go away underneath us. For | |
1489 | * async buffers, I/O completion drops the callers reference, which | |
1490 | * could occur before submission returns. | |
1491 | */ | |
1492 | xfs_buf_hold(bp); | |
1493 | ||
375ec69d | 1494 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1495 | xfs_buf_wait_unpin(bp); |
e11bb805 | 1496 | |
61be9c52 DC |
1497 | /* clear the internal error state to avoid spurious errors */ |
1498 | bp->b_io_error = 0; | |
1499 | ||
8d6c1210 | 1500 | /* |
e11bb805 DC |
1501 | * Set the count to 1 initially, this will stop an I/O completion |
1502 | * callout which happens before we have started all the I/O from calling | |
1503 | * xfs_buf_ioend too early. | |
1da177e4 | 1504 | */ |
ce8e922c | 1505 | atomic_set(&bp->b_io_remaining, 1); |
eaebb515 BF |
1506 | if (bp->b_flags & XBF_ASYNC) |
1507 | xfs_buf_ioacct_inc(bp); | |
ce8e922c | 1508 | _xfs_buf_ioapply(bp); |
e11bb805 | 1509 | |
8d6c1210 | 1510 | /* |
595bff75 DC |
1511 | * If _xfs_buf_ioapply failed, we can get back here with only the IO |
1512 | * reference we took above. If we drop it to zero, run completion so | |
1513 | * that we don't return to the caller with completion still pending. | |
8d6c1210 | 1514 | */ |
e8aaba9a | 1515 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
eaebb515 | 1516 | if (bp->b_error || !(bp->b_flags & XBF_ASYNC)) |
e8aaba9a DC |
1517 | xfs_buf_ioend(bp); |
1518 | else | |
1519 | xfs_buf_ioend_async(bp); | |
1520 | } | |
1da177e4 | 1521 | |
6af88cda BF |
1522 | if (wait) |
1523 | error = xfs_buf_iowait(bp); | |
bb00b6f1 | 1524 | |
595bff75 | 1525 | /* |
6af88cda BF |
1526 | * Release the hold that keeps the buffer referenced for the entire |
1527 | * I/O. Note that if the buffer is async, it is not safe to reference | |
1528 | * after this release. | |
595bff75 DC |
1529 | */ |
1530 | xfs_buf_rele(bp); | |
1531 | return error; | |
1da177e4 LT |
1532 | } |
1533 | ||
88ee2df7 | 1534 | void * |
ce8e922c | 1535 | xfs_buf_offset( |
88ee2df7 | 1536 | struct xfs_buf *bp, |
1da177e4 LT |
1537 | size_t offset) |
1538 | { | |
1539 | struct page *page; | |
1540 | ||
611c9946 | 1541 | if (bp->b_addr) |
62926044 | 1542 | return bp->b_addr + offset; |
1da177e4 | 1543 | |
ce8e922c | 1544 | offset += bp->b_offset; |
0e6e847f | 1545 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
88ee2df7 | 1546 | return page_address(page) + (offset & (PAGE_SIZE-1)); |
1da177e4 LT |
1547 | } |
1548 | ||
1da177e4 | 1549 | void |
f9a196ee CH |
1550 | xfs_buf_zero( |
1551 | struct xfs_buf *bp, | |
1552 | size_t boff, | |
1553 | size_t bsize) | |
1da177e4 | 1554 | { |
795cac72 | 1555 | size_t bend; |
1da177e4 LT |
1556 | |
1557 | bend = boff + bsize; | |
1558 | while (boff < bend) { | |
795cac72 DC |
1559 | struct page *page; |
1560 | int page_index, page_offset, csize; | |
1561 | ||
1562 | page_index = (boff + bp->b_offset) >> PAGE_SHIFT; | |
1563 | page_offset = (boff + bp->b_offset) & ~PAGE_MASK; | |
1564 | page = bp->b_pages[page_index]; | |
1565 | csize = min_t(size_t, PAGE_SIZE - page_offset, | |
8124b9b6 | 1566 | BBTOB(bp->b_length) - boff); |
1da177e4 | 1567 | |
795cac72 | 1568 | ASSERT((csize + page_offset) <= PAGE_SIZE); |
1da177e4 | 1569 | |
f9a196ee | 1570 | memset(page_address(page) + page_offset, 0, csize); |
1da177e4 LT |
1571 | |
1572 | boff += csize; | |
1da177e4 LT |
1573 | } |
1574 | } | |
1575 | ||
1576 | /* | |
ce8e922c | 1577 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1578 | */ |
1579 | ||
1580 | /* | |
430cbeb8 DC |
1581 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1582 | * returned. These buffers will have an elevated hold count, so wait on those | |
1583 | * while freeing all the buffers only held by the LRU. | |
1da177e4 | 1584 | */ |
e80dfa19 DC |
1585 | static enum lru_status |
1586 | xfs_buftarg_wait_rele( | |
1587 | struct list_head *item, | |
3f97b163 | 1588 | struct list_lru_one *lru, |
e80dfa19 DC |
1589 | spinlock_t *lru_lock, |
1590 | void *arg) | |
1591 | ||
1da177e4 | 1592 | { |
e80dfa19 | 1593 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); |
a4082357 | 1594 | struct list_head *dispose = arg; |
430cbeb8 | 1595 | |
e80dfa19 | 1596 | if (atomic_read(&bp->b_hold) > 1) { |
a4082357 | 1597 | /* need to wait, so skip it this pass */ |
e80dfa19 | 1598 | trace_xfs_buf_wait_buftarg(bp, _RET_IP_); |
a4082357 | 1599 | return LRU_SKIP; |
1da177e4 | 1600 | } |
a4082357 DC |
1601 | if (!spin_trylock(&bp->b_lock)) |
1602 | return LRU_SKIP; | |
e80dfa19 | 1603 | |
a4082357 DC |
1604 | /* |
1605 | * clear the LRU reference count so the buffer doesn't get | |
1606 | * ignored in xfs_buf_rele(). | |
1607 | */ | |
1608 | atomic_set(&bp->b_lru_ref, 0); | |
1609 | bp->b_state |= XFS_BSTATE_DISPOSE; | |
3f97b163 | 1610 | list_lru_isolate_move(lru, item, dispose); |
a4082357 DC |
1611 | spin_unlock(&bp->b_lock); |
1612 | return LRU_REMOVED; | |
1da177e4 LT |
1613 | } |
1614 | ||
e80dfa19 DC |
1615 | void |
1616 | xfs_wait_buftarg( | |
1617 | struct xfs_buftarg *btp) | |
1618 | { | |
a4082357 DC |
1619 | LIST_HEAD(dispose); |
1620 | int loop = 0; | |
1621 | ||
85bec546 | 1622 | /* |
9c7504aa BF |
1623 | * First wait on the buftarg I/O count for all in-flight buffers to be |
1624 | * released. This is critical as new buffers do not make the LRU until | |
1625 | * they are released. | |
1626 | * | |
1627 | * Next, flush the buffer workqueue to ensure all completion processing | |
1628 | * has finished. Just waiting on buffer locks is not sufficient for | |
1629 | * async IO as the reference count held over IO is not released until | |
1630 | * after the buffer lock is dropped. Hence we need to ensure here that | |
1631 | * all reference counts have been dropped before we start walking the | |
1632 | * LRU list. | |
85bec546 | 1633 | */ |
9c7504aa BF |
1634 | while (percpu_counter_sum(&btp->bt_io_count)) |
1635 | delay(100); | |
800b2694 | 1636 | flush_workqueue(btp->bt_mount->m_buf_workqueue); |
85bec546 | 1637 | |
a4082357 DC |
1638 | /* loop until there is nothing left on the lru list. */ |
1639 | while (list_lru_count(&btp->bt_lru)) { | |
e80dfa19 | 1640 | list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele, |
a4082357 DC |
1641 | &dispose, LONG_MAX); |
1642 | ||
1643 | while (!list_empty(&dispose)) { | |
1644 | struct xfs_buf *bp; | |
1645 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1646 | list_del_init(&bp->b_lru); | |
ac8809f9 DC |
1647 | if (bp->b_flags & XBF_WRITE_FAIL) { |
1648 | xfs_alert(btp->bt_mount, | |
c219b015 | 1649 | "Corruption Alert: Buffer at daddr 0x%llx had permanent write failures!", |
ac8809f9 | 1650 | (long long)bp->b_bn); |
f41febd2 JP |
1651 | xfs_alert(btp->bt_mount, |
1652 | "Please run xfs_repair to determine the extent of the problem."); | |
ac8809f9 | 1653 | } |
a4082357 DC |
1654 | xfs_buf_rele(bp); |
1655 | } | |
1656 | if (loop++ != 0) | |
1657 | delay(100); | |
1658 | } | |
e80dfa19 DC |
1659 | } |
1660 | ||
1661 | static enum lru_status | |
1662 | xfs_buftarg_isolate( | |
1663 | struct list_head *item, | |
3f97b163 | 1664 | struct list_lru_one *lru, |
e80dfa19 DC |
1665 | spinlock_t *lru_lock, |
1666 | void *arg) | |
1667 | { | |
1668 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); | |
1669 | struct list_head *dispose = arg; | |
1670 | ||
a4082357 DC |
1671 | /* |
1672 | * we are inverting the lru lock/bp->b_lock here, so use a trylock. | |
1673 | * If we fail to get the lock, just skip it. | |
1674 | */ | |
1675 | if (!spin_trylock(&bp->b_lock)) | |
1676 | return LRU_SKIP; | |
e80dfa19 DC |
1677 | /* |
1678 | * Decrement the b_lru_ref count unless the value is already | |
1679 | * zero. If the value is already zero, we need to reclaim the | |
1680 | * buffer, otherwise it gets another trip through the LRU. | |
1681 | */ | |
19957a18 | 1682 | if (atomic_add_unless(&bp->b_lru_ref, -1, 0)) { |
a4082357 | 1683 | spin_unlock(&bp->b_lock); |
e80dfa19 | 1684 | return LRU_ROTATE; |
a4082357 | 1685 | } |
e80dfa19 | 1686 | |
a4082357 | 1687 | bp->b_state |= XFS_BSTATE_DISPOSE; |
3f97b163 | 1688 | list_lru_isolate_move(lru, item, dispose); |
a4082357 | 1689 | spin_unlock(&bp->b_lock); |
e80dfa19 DC |
1690 | return LRU_REMOVED; |
1691 | } | |
1692 | ||
addbda40 | 1693 | static unsigned long |
e80dfa19 | 1694 | xfs_buftarg_shrink_scan( |
ff57ab21 | 1695 | struct shrinker *shrink, |
1495f230 | 1696 | struct shrink_control *sc) |
a6867a68 | 1697 | { |
ff57ab21 DC |
1698 | struct xfs_buftarg *btp = container_of(shrink, |
1699 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1700 | LIST_HEAD(dispose); |
addbda40 | 1701 | unsigned long freed; |
430cbeb8 | 1702 | |
503c358c VD |
1703 | freed = list_lru_shrink_walk(&btp->bt_lru, sc, |
1704 | xfs_buftarg_isolate, &dispose); | |
430cbeb8 DC |
1705 | |
1706 | while (!list_empty(&dispose)) { | |
e80dfa19 | 1707 | struct xfs_buf *bp; |
430cbeb8 DC |
1708 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); |
1709 | list_del_init(&bp->b_lru); | |
1710 | xfs_buf_rele(bp); | |
1711 | } | |
1712 | ||
e80dfa19 DC |
1713 | return freed; |
1714 | } | |
1715 | ||
addbda40 | 1716 | static unsigned long |
e80dfa19 DC |
1717 | xfs_buftarg_shrink_count( |
1718 | struct shrinker *shrink, | |
1719 | struct shrink_control *sc) | |
1720 | { | |
1721 | struct xfs_buftarg *btp = container_of(shrink, | |
1722 | struct xfs_buftarg, bt_shrinker); | |
503c358c | 1723 | return list_lru_shrink_count(&btp->bt_lru, sc); |
a6867a68 DC |
1724 | } |
1725 | ||
1da177e4 LT |
1726 | void |
1727 | xfs_free_buftarg( | |
b7963133 | 1728 | struct xfs_buftarg *btp) |
1da177e4 | 1729 | { |
ff57ab21 | 1730 | unregister_shrinker(&btp->bt_shrinker); |
9c7504aa BF |
1731 | ASSERT(percpu_counter_sum(&btp->bt_io_count) == 0); |
1732 | percpu_counter_destroy(&btp->bt_io_count); | |
f5e1dd34 | 1733 | list_lru_destroy(&btp->bt_lru); |
ff57ab21 | 1734 | |
2291dab2 | 1735 | xfs_blkdev_issue_flush(btp); |
a6867a68 | 1736 | |
f0e2d93c | 1737 | kmem_free(btp); |
1da177e4 LT |
1738 | } |
1739 | ||
3fefdeee ES |
1740 | int |
1741 | xfs_setsize_buftarg( | |
1da177e4 | 1742 | xfs_buftarg_t *btp, |
3fefdeee | 1743 | unsigned int sectorsize) |
1da177e4 | 1744 | { |
7c71ee78 | 1745 | /* Set up metadata sector size info */ |
6da54179 ES |
1746 | btp->bt_meta_sectorsize = sectorsize; |
1747 | btp->bt_meta_sectormask = sectorsize - 1; | |
1da177e4 | 1748 | |
ce8e922c | 1749 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
4f10700a | 1750 | xfs_warn(btp->bt_mount, |
a1c6f057 DM |
1751 | "Cannot set_blocksize to %u on device %pg", |
1752 | sectorsize, btp->bt_bdev); | |
2451337d | 1753 | return -EINVAL; |
1da177e4 LT |
1754 | } |
1755 | ||
7c71ee78 ES |
1756 | /* Set up device logical sector size mask */ |
1757 | btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev); | |
1758 | btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1; | |
1759 | ||
1da177e4 LT |
1760 | return 0; |
1761 | } | |
1762 | ||
1763 | /* | |
3fefdeee ES |
1764 | * When allocating the initial buffer target we have not yet |
1765 | * read in the superblock, so don't know what sized sectors | |
1766 | * are being used at this early stage. Play safe. | |
ce8e922c | 1767 | */ |
1da177e4 LT |
1768 | STATIC int |
1769 | xfs_setsize_buftarg_early( | |
1770 | xfs_buftarg_t *btp, | |
1771 | struct block_device *bdev) | |
1772 | { | |
a96c4151 | 1773 | return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev)); |
1da177e4 LT |
1774 | } |
1775 | ||
1da177e4 LT |
1776 | xfs_buftarg_t * |
1777 | xfs_alloc_buftarg( | |
ebad861b | 1778 | struct xfs_mount *mp, |
486aff5e DW |
1779 | struct block_device *bdev, |
1780 | struct dax_device *dax_dev) | |
1da177e4 LT |
1781 | { |
1782 | xfs_buftarg_t *btp; | |
1783 | ||
707e0dda | 1784 | btp = kmem_zalloc(sizeof(*btp), KM_NOFS); |
1da177e4 | 1785 | |
ebad861b | 1786 | btp->bt_mount = mp; |
ce8e922c NS |
1787 | btp->bt_dev = bdev->bd_dev; |
1788 | btp->bt_bdev = bdev; | |
486aff5e | 1789 | btp->bt_daxdev = dax_dev; |
0e6e847f | 1790 | |
1da177e4 | 1791 | if (xfs_setsize_buftarg_early(btp, bdev)) |
d210a987 | 1792 | goto error_free; |
5ca302c8 GC |
1793 | |
1794 | if (list_lru_init(&btp->bt_lru)) | |
d210a987 | 1795 | goto error_free; |
5ca302c8 | 1796 | |
9c7504aa | 1797 | if (percpu_counter_init(&btp->bt_io_count, 0, GFP_KERNEL)) |
d210a987 | 1798 | goto error_lru; |
9c7504aa | 1799 | |
e80dfa19 DC |
1800 | btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count; |
1801 | btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan; | |
ff57ab21 | 1802 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; |
e80dfa19 | 1803 | btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE; |
d210a987 MH |
1804 | if (register_shrinker(&btp->bt_shrinker)) |
1805 | goto error_pcpu; | |
1da177e4 LT |
1806 | return btp; |
1807 | ||
d210a987 MH |
1808 | error_pcpu: |
1809 | percpu_counter_destroy(&btp->bt_io_count); | |
1810 | error_lru: | |
1811 | list_lru_destroy(&btp->bt_lru); | |
1812 | error_free: | |
f0e2d93c | 1813 | kmem_free(btp); |
1da177e4 LT |
1814 | return NULL; |
1815 | } | |
1816 | ||
20e8a063 BF |
1817 | /* |
1818 | * Cancel a delayed write list. | |
1819 | * | |
1820 | * Remove each buffer from the list, clear the delwri queue flag and drop the | |
1821 | * associated buffer reference. | |
1822 | */ | |
1823 | void | |
1824 | xfs_buf_delwri_cancel( | |
1825 | struct list_head *list) | |
1826 | { | |
1827 | struct xfs_buf *bp; | |
1828 | ||
1829 | while (!list_empty(list)) { | |
1830 | bp = list_first_entry(list, struct xfs_buf, b_list); | |
1831 | ||
1832 | xfs_buf_lock(bp); | |
1833 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
1834 | list_del_init(&bp->b_list); | |
1835 | xfs_buf_relse(bp); | |
1836 | } | |
1837 | } | |
1838 | ||
1da177e4 | 1839 | /* |
43ff2122 CH |
1840 | * Add a buffer to the delayed write list. |
1841 | * | |
1842 | * This queues a buffer for writeout if it hasn't already been. Note that | |
1843 | * neither this routine nor the buffer list submission functions perform | |
1844 | * any internal synchronization. It is expected that the lists are thread-local | |
1845 | * to the callers. | |
1846 | * | |
1847 | * Returns true if we queued up the buffer, or false if it already had | |
1848 | * been on the buffer list. | |
1da177e4 | 1849 | */ |
43ff2122 | 1850 | bool |
ce8e922c | 1851 | xfs_buf_delwri_queue( |
43ff2122 CH |
1852 | struct xfs_buf *bp, |
1853 | struct list_head *list) | |
1da177e4 | 1854 | { |
43ff2122 | 1855 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 1856 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1857 | |
43ff2122 CH |
1858 | /* |
1859 | * If the buffer is already marked delwri it already is queued up | |
1860 | * by someone else for imediate writeout. Just ignore it in that | |
1861 | * case. | |
1862 | */ | |
1863 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
1864 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
1865 | return false; | |
1da177e4 | 1866 | } |
1da177e4 | 1867 | |
43ff2122 | 1868 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
1869 | |
1870 | /* | |
43ff2122 CH |
1871 | * If a buffer gets written out synchronously or marked stale while it |
1872 | * is on a delwri list we lazily remove it. To do this, the other party | |
1873 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
1874 | * It remains referenced and on the list. In a rare corner case it | |
1875 | * might get readded to a delwri list after the synchronous writeout, in | |
1876 | * which case we need just need to re-add the flag here. | |
d808f617 | 1877 | */ |
43ff2122 CH |
1878 | bp->b_flags |= _XBF_DELWRI_Q; |
1879 | if (list_empty(&bp->b_list)) { | |
1880 | atomic_inc(&bp->b_hold); | |
1881 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 1882 | } |
585e6d88 | 1883 | |
43ff2122 | 1884 | return true; |
585e6d88 DC |
1885 | } |
1886 | ||
089716aa DC |
1887 | /* |
1888 | * Compare function is more complex than it needs to be because | |
1889 | * the return value is only 32 bits and we are doing comparisons | |
1890 | * on 64 bit values | |
1891 | */ | |
1892 | static int | |
1893 | xfs_buf_cmp( | |
1894 | void *priv, | |
1895 | struct list_head *a, | |
1896 | struct list_head *b) | |
1897 | { | |
1898 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1899 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1900 | xfs_daddr_t diff; | |
1901 | ||
f4b42421 | 1902 | diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn; |
089716aa DC |
1903 | if (diff < 0) |
1904 | return -1; | |
1905 | if (diff > 0) | |
1906 | return 1; | |
1907 | return 0; | |
1908 | } | |
1909 | ||
26f1fe85 | 1910 | /* |
e339dd8d BF |
1911 | * Submit buffers for write. If wait_list is specified, the buffers are |
1912 | * submitted using sync I/O and placed on the wait list such that the caller can | |
1913 | * iowait each buffer. Otherwise async I/O is used and the buffers are released | |
1914 | * at I/O completion time. In either case, buffers remain locked until I/O | |
1915 | * completes and the buffer is released from the queue. | |
26f1fe85 | 1916 | */ |
43ff2122 | 1917 | static int |
26f1fe85 | 1918 | xfs_buf_delwri_submit_buffers( |
43ff2122 | 1919 | struct list_head *buffer_list, |
26f1fe85 | 1920 | struct list_head *wait_list) |
1da177e4 | 1921 | { |
43ff2122 CH |
1922 | struct xfs_buf *bp, *n; |
1923 | int pinned = 0; | |
26f1fe85 | 1924 | struct blk_plug plug; |
43ff2122 | 1925 | |
26f1fe85 | 1926 | list_sort(NULL, buffer_list, xfs_buf_cmp); |
43ff2122 | 1927 | |
26f1fe85 | 1928 | blk_start_plug(&plug); |
43ff2122 | 1929 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { |
26f1fe85 | 1930 | if (!wait_list) { |
43ff2122 CH |
1931 | if (xfs_buf_ispinned(bp)) { |
1932 | pinned++; | |
1933 | continue; | |
1934 | } | |
1935 | if (!xfs_buf_trylock(bp)) | |
1936 | continue; | |
1937 | } else { | |
1938 | xfs_buf_lock(bp); | |
1939 | } | |
978c7b2f | 1940 | |
43ff2122 CH |
1941 | /* |
1942 | * Someone else might have written the buffer synchronously or | |
1943 | * marked it stale in the meantime. In that case only the | |
1944 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
1945 | * reference and remove it from the list here. | |
1946 | */ | |
1947 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
1948 | list_del_init(&bp->b_list); | |
1949 | xfs_buf_relse(bp); | |
1950 | continue; | |
1951 | } | |
c9c12971 | 1952 | |
43ff2122 | 1953 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
a1b7ea5d | 1954 | |
cf53e99d | 1955 | /* |
e339dd8d BF |
1956 | * If we have a wait list, each buffer (and associated delwri |
1957 | * queue reference) transfers to it and is submitted | |
1958 | * synchronously. Otherwise, drop the buffer from the delwri | |
1959 | * queue and submit async. | |
cf53e99d | 1960 | */ |
bbfeb614 | 1961 | bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_WRITE_FAIL); |
e339dd8d | 1962 | bp->b_flags |= XBF_WRITE; |
26f1fe85 | 1963 | if (wait_list) { |
e339dd8d | 1964 | bp->b_flags &= ~XBF_ASYNC; |
26f1fe85 | 1965 | list_move_tail(&bp->b_list, wait_list); |
e339dd8d BF |
1966 | } else { |
1967 | bp->b_flags |= XBF_ASYNC; | |
ce8e922c | 1968 | list_del_init(&bp->b_list); |
e339dd8d | 1969 | } |
6af88cda | 1970 | __xfs_buf_submit(bp, false); |
43ff2122 CH |
1971 | } |
1972 | blk_finish_plug(&plug); | |
1da177e4 | 1973 | |
43ff2122 | 1974 | return pinned; |
1da177e4 LT |
1975 | } |
1976 | ||
1977 | /* | |
43ff2122 CH |
1978 | * Write out a buffer list asynchronously. |
1979 | * | |
1980 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
1981 | * out and not wait for I/O completion on any of the buffers. This interface | |
1982 | * is only safely useable for callers that can track I/O completion by higher | |
1983 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
1984 | * function. | |
efc3289c BF |
1985 | * |
1986 | * Note: this function will skip buffers it would block on, and in doing so | |
1987 | * leaves them on @buffer_list so they can be retried on a later pass. As such, | |
1988 | * it is up to the caller to ensure that the buffer list is fully submitted or | |
1989 | * cancelled appropriately when they are finished with the list. Failure to | |
1990 | * cancel or resubmit the list until it is empty will result in leaked buffers | |
1991 | * at unmount time. | |
1da177e4 LT |
1992 | */ |
1993 | int | |
43ff2122 CH |
1994 | xfs_buf_delwri_submit_nowait( |
1995 | struct list_head *buffer_list) | |
1da177e4 | 1996 | { |
26f1fe85 | 1997 | return xfs_buf_delwri_submit_buffers(buffer_list, NULL); |
43ff2122 | 1998 | } |
1da177e4 | 1999 | |
43ff2122 CH |
2000 | /* |
2001 | * Write out a buffer list synchronously. | |
2002 | * | |
2003 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
2004 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
2005 | * so callers must have some other way of tracking buffers if they require such | |
2006 | * functionality. | |
2007 | */ | |
2008 | int | |
2009 | xfs_buf_delwri_submit( | |
2010 | struct list_head *buffer_list) | |
2011 | { | |
26f1fe85 | 2012 | LIST_HEAD (wait_list); |
43ff2122 CH |
2013 | int error = 0, error2; |
2014 | struct xfs_buf *bp; | |
1da177e4 | 2015 | |
26f1fe85 | 2016 | xfs_buf_delwri_submit_buffers(buffer_list, &wait_list); |
1da177e4 | 2017 | |
43ff2122 | 2018 | /* Wait for IO to complete. */ |
26f1fe85 DC |
2019 | while (!list_empty(&wait_list)) { |
2020 | bp = list_first_entry(&wait_list, struct xfs_buf, b_list); | |
a1b7ea5d | 2021 | |
089716aa | 2022 | list_del_init(&bp->b_list); |
cf53e99d | 2023 | |
e339dd8d BF |
2024 | /* |
2025 | * Wait on the locked buffer, check for errors and unlock and | |
2026 | * release the delwri queue reference. | |
2027 | */ | |
2028 | error2 = xfs_buf_iowait(bp); | |
43ff2122 CH |
2029 | xfs_buf_relse(bp); |
2030 | if (!error) | |
2031 | error = error2; | |
1da177e4 LT |
2032 | } |
2033 | ||
43ff2122 | 2034 | return error; |
1da177e4 LT |
2035 | } |
2036 | ||
7912e7fe BF |
2037 | /* |
2038 | * Push a single buffer on a delwri queue. | |
2039 | * | |
2040 | * The purpose of this function is to submit a single buffer of a delwri queue | |
2041 | * and return with the buffer still on the original queue. The waiting delwri | |
2042 | * buffer submission infrastructure guarantees transfer of the delwri queue | |
2043 | * buffer reference to a temporary wait list. We reuse this infrastructure to | |
2044 | * transfer the buffer back to the original queue. | |
2045 | * | |
2046 | * Note the buffer transitions from the queued state, to the submitted and wait | |
2047 | * listed state and back to the queued state during this call. The buffer | |
2048 | * locking and queue management logic between _delwri_pushbuf() and | |
2049 | * _delwri_queue() guarantee that the buffer cannot be queued to another list | |
2050 | * before returning. | |
2051 | */ | |
2052 | int | |
2053 | xfs_buf_delwri_pushbuf( | |
2054 | struct xfs_buf *bp, | |
2055 | struct list_head *buffer_list) | |
2056 | { | |
2057 | LIST_HEAD (submit_list); | |
2058 | int error; | |
2059 | ||
2060 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
2061 | ||
2062 | trace_xfs_buf_delwri_pushbuf(bp, _RET_IP_); | |
2063 | ||
2064 | /* | |
2065 | * Isolate the buffer to a new local list so we can submit it for I/O | |
2066 | * independently from the rest of the original list. | |
2067 | */ | |
2068 | xfs_buf_lock(bp); | |
2069 | list_move(&bp->b_list, &submit_list); | |
2070 | xfs_buf_unlock(bp); | |
2071 | ||
2072 | /* | |
2073 | * Delwri submission clears the DELWRI_Q buffer flag and returns with | |
e339dd8d | 2074 | * the buffer on the wait list with the original reference. Rather than |
7912e7fe BF |
2075 | * bounce the buffer from a local wait list back to the original list |
2076 | * after I/O completion, reuse the original list as the wait list. | |
2077 | */ | |
2078 | xfs_buf_delwri_submit_buffers(&submit_list, buffer_list); | |
2079 | ||
2080 | /* | |
e339dd8d BF |
2081 | * The buffer is now locked, under I/O and wait listed on the original |
2082 | * delwri queue. Wait for I/O completion, restore the DELWRI_Q flag and | |
2083 | * return with the buffer unlocked and on the original queue. | |
7912e7fe | 2084 | */ |
e339dd8d | 2085 | error = xfs_buf_iowait(bp); |
7912e7fe BF |
2086 | bp->b_flags |= _XBF_DELWRI_Q; |
2087 | xfs_buf_unlock(bp); | |
2088 | ||
2089 | return error; | |
2090 | } | |
2091 | ||
04d8b284 | 2092 | int __init |
ce8e922c | 2093 | xfs_buf_init(void) |
1da177e4 | 2094 | { |
b1231760 CM |
2095 | xfs_buf_zone = kmem_cache_create("xfs_buf", |
2096 | sizeof(struct xfs_buf), 0, | |
2097 | SLAB_HWCACHE_ALIGN, NULL); | |
ce8e922c | 2098 | if (!xfs_buf_zone) |
0b1b213f | 2099 | goto out; |
04d8b284 | 2100 | |
23ea4032 | 2101 | return 0; |
1da177e4 | 2102 | |
0b1b213f | 2103 | out: |
8758280f | 2104 | return -ENOMEM; |
1da177e4 LT |
2105 | } |
2106 | ||
1da177e4 | 2107 | void |
ce8e922c | 2108 | xfs_buf_terminate(void) |
1da177e4 | 2109 | { |
aaf54eb8 | 2110 | kmem_cache_destroy(xfs_buf_zone); |
1da177e4 | 2111 | } |
7561d27e BF |
2112 | |
2113 | void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref) | |
2114 | { | |
7561d27e BF |
2115 | /* |
2116 | * Set the lru reference count to 0 based on the error injection tag. | |
2117 | * This allows userspace to disrupt buffer caching for debug/testing | |
2118 | * purposes. | |
2119 | */ | |
dbd329f1 | 2120 | if (XFS_TEST_ERROR(false, bp->b_mount, XFS_ERRTAG_BUF_LRU_REF)) |
7561d27e BF |
2121 | lru_ref = 0; |
2122 | ||
2123 | atomic_set(&bp->b_lru_ref, lru_ref); | |
2124 | } | |
8473fee3 BF |
2125 | |
2126 | /* | |
2127 | * Verify an on-disk magic value against the magic value specified in the | |
2128 | * verifier structure. The verifier magic is in disk byte order so the caller is | |
2129 | * expected to pass the value directly from disk. | |
2130 | */ | |
2131 | bool | |
2132 | xfs_verify_magic( | |
2133 | struct xfs_buf *bp, | |
15baadf7 | 2134 | __be32 dmagic) |
8473fee3 | 2135 | { |
dbd329f1 | 2136 | struct xfs_mount *mp = bp->b_mount; |
8473fee3 BF |
2137 | int idx; |
2138 | ||
2139 | idx = xfs_sb_version_hascrc(&mp->m_sb); | |
14ed8688 | 2140 | if (WARN_ON(!bp->b_ops || !bp->b_ops->magic[idx])) |
8473fee3 BF |
2141 | return false; |
2142 | return dmagic == bp->b_ops->magic[idx]; | |
2143 | } | |
15baadf7 DW |
2144 | /* |
2145 | * Verify an on-disk magic value against the magic value specified in the | |
2146 | * verifier structure. The verifier magic is in disk byte order so the caller is | |
2147 | * expected to pass the value directly from disk. | |
2148 | */ | |
2149 | bool | |
2150 | xfs_verify_magic16( | |
2151 | struct xfs_buf *bp, | |
2152 | __be16 dmagic) | |
2153 | { | |
dbd329f1 | 2154 | struct xfs_mount *mp = bp->b_mount; |
15baadf7 DW |
2155 | int idx; |
2156 | ||
2157 | idx = xfs_sb_version_hascrc(&mp->m_sb); | |
14ed8688 | 2158 | if (WARN_ON(!bp->b_ops || !bp->b_ops->magic16[idx])) |
15baadf7 DW |
2159 | return false; |
2160 | return dmagic == bp->b_ops->magic16[idx]; | |
2161 | } |