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" |
1da177e4 LT |
7 | #include <linux/stddef.h> |
8 | #include <linux/errno.h> | |
5a0e3ad6 | 9 | #include <linux/gfp.h> |
1da177e4 LT |
10 | #include <linux/pagemap.h> |
11 | #include <linux/init.h> | |
12 | #include <linux/vmalloc.h> | |
13 | #include <linux/bio.h> | |
14 | #include <linux/sysctl.h> | |
15 | #include <linux/proc_fs.h> | |
16 | #include <linux/workqueue.h> | |
17 | #include <linux/percpu.h> | |
18 | #include <linux/blkdev.h> | |
19 | #include <linux/hash.h> | |
4df08c52 | 20 | #include <linux/kthread.h> |
b20a3503 | 21 | #include <linux/migrate.h> |
3fcfab16 | 22 | #include <linux/backing-dev.h> |
7dfb7103 | 23 | #include <linux/freezer.h> |
1da177e4 | 24 | |
4fb6e8ad | 25 | #include "xfs_format.h" |
239880ef | 26 | #include "xfs_log_format.h" |
7fd36c44 | 27 | #include "xfs_trans_resv.h" |
239880ef | 28 | #include "xfs_sb.h" |
b7963133 | 29 | #include "xfs_mount.h" |
0b1b213f | 30 | #include "xfs_trace.h" |
239880ef | 31 | #include "xfs_log.h" |
e9e899a2 | 32 | #include "xfs_errortag.h" |
7561d27e | 33 | #include "xfs_error.h" |
b7963133 | 34 | |
7989cb8e | 35 | static kmem_zone_t *xfs_buf_zone; |
23ea4032 | 36 | |
ce8e922c | 37 | #define xb_to_gfp(flags) \ |
aa5c158e | 38 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN) |
1da177e4 | 39 | |
37fd1678 DC |
40 | /* |
41 | * Locking orders | |
42 | * | |
43 | * xfs_buf_ioacct_inc: | |
44 | * xfs_buf_ioacct_dec: | |
45 | * b_sema (caller holds) | |
46 | * b_lock | |
47 | * | |
48 | * xfs_buf_stale: | |
49 | * b_sema (caller holds) | |
50 | * b_lock | |
51 | * lru_lock | |
52 | * | |
53 | * xfs_buf_rele: | |
54 | * b_lock | |
55 | * pag_buf_lock | |
56 | * lru_lock | |
57 | * | |
58 | * xfs_buftarg_wait_rele | |
59 | * lru_lock | |
60 | * b_lock (trylock due to inversion) | |
61 | * | |
62 | * xfs_buftarg_isolate | |
63 | * lru_lock | |
64 | * b_lock (trylock due to inversion) | |
65 | */ | |
1da177e4 | 66 | |
73c77e2c JB |
67 | static inline int |
68 | xfs_buf_is_vmapped( | |
69 | struct xfs_buf *bp) | |
70 | { | |
71 | /* | |
72 | * Return true if the buffer is vmapped. | |
73 | * | |
611c9946 DC |
74 | * b_addr is null if the buffer is not mapped, but the code is clever |
75 | * enough to know it doesn't have to map a single page, so the check has | |
76 | * to be both for b_addr and bp->b_page_count > 1. | |
73c77e2c | 77 | */ |
611c9946 | 78 | return bp->b_addr && bp->b_page_count > 1; |
73c77e2c JB |
79 | } |
80 | ||
81 | static inline int | |
82 | xfs_buf_vmap_len( | |
83 | struct xfs_buf *bp) | |
84 | { | |
85 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
86 | } | |
87 | ||
9c7504aa BF |
88 | /* |
89 | * Bump the I/O in flight count on the buftarg if we haven't yet done so for | |
90 | * this buffer. The count is incremented once per buffer (per hold cycle) | |
91 | * because the corresponding decrement is deferred to buffer release. Buffers | |
92 | * can undergo I/O multiple times in a hold-release cycle and per buffer I/O | |
93 | * tracking adds unnecessary overhead. This is used for sychronization purposes | |
94 | * with unmount (see xfs_wait_buftarg()), so all we really need is a count of | |
95 | * in-flight buffers. | |
96 | * | |
97 | * Buffers that are never released (e.g., superblock, iclog buffers) must set | |
98 | * the XBF_NO_IOACCT flag before I/O submission. Otherwise, the buftarg count | |
99 | * never reaches zero and unmount hangs indefinitely. | |
100 | */ | |
101 | static inline void | |
102 | xfs_buf_ioacct_inc( | |
103 | struct xfs_buf *bp) | |
104 | { | |
63db7c81 | 105 | if (bp->b_flags & XBF_NO_IOACCT) |
9c7504aa BF |
106 | return; |
107 | ||
108 | ASSERT(bp->b_flags & XBF_ASYNC); | |
63db7c81 BF |
109 | spin_lock(&bp->b_lock); |
110 | if (!(bp->b_state & XFS_BSTATE_IN_FLIGHT)) { | |
111 | bp->b_state |= XFS_BSTATE_IN_FLIGHT; | |
112 | percpu_counter_inc(&bp->b_target->bt_io_count); | |
113 | } | |
114 | spin_unlock(&bp->b_lock); | |
9c7504aa BF |
115 | } |
116 | ||
117 | /* | |
118 | * Clear the in-flight state on a buffer about to be released to the LRU or | |
119 | * freed and unaccount from the buftarg. | |
120 | */ | |
121 | static inline void | |
63db7c81 | 122 | __xfs_buf_ioacct_dec( |
9c7504aa BF |
123 | struct xfs_buf *bp) |
124 | { | |
95989c46 | 125 | lockdep_assert_held(&bp->b_lock); |
9c7504aa | 126 | |
63db7c81 BF |
127 | if (bp->b_state & XFS_BSTATE_IN_FLIGHT) { |
128 | bp->b_state &= ~XFS_BSTATE_IN_FLIGHT; | |
129 | percpu_counter_dec(&bp->b_target->bt_io_count); | |
130 | } | |
131 | } | |
132 | ||
133 | static inline void | |
134 | xfs_buf_ioacct_dec( | |
135 | struct xfs_buf *bp) | |
136 | { | |
137 | spin_lock(&bp->b_lock); | |
138 | __xfs_buf_ioacct_dec(bp); | |
139 | spin_unlock(&bp->b_lock); | |
9c7504aa BF |
140 | } |
141 | ||
430cbeb8 DC |
142 | /* |
143 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
144 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
145 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
146 | * to remove the reference that LRU holds on the buffer. | |
147 | * | |
148 | * This prevents build-up of stale buffers on the LRU. | |
149 | */ | |
150 | void | |
151 | xfs_buf_stale( | |
152 | struct xfs_buf *bp) | |
153 | { | |
43ff2122 CH |
154 | ASSERT(xfs_buf_islocked(bp)); |
155 | ||
430cbeb8 | 156 | bp->b_flags |= XBF_STALE; |
43ff2122 CH |
157 | |
158 | /* | |
159 | * Clear the delwri status so that a delwri queue walker will not | |
160 | * flush this buffer to disk now that it is stale. The delwri queue has | |
161 | * a reference to the buffer, so this is safe to do. | |
162 | */ | |
163 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
164 | ||
9c7504aa BF |
165 | /* |
166 | * Once the buffer is marked stale and unlocked, a subsequent lookup | |
167 | * could reset b_flags. There is no guarantee that the buffer is | |
168 | * unaccounted (released to LRU) before that occurs. Drop in-flight | |
169 | * status now to preserve accounting consistency. | |
170 | */ | |
a4082357 | 171 | spin_lock(&bp->b_lock); |
63db7c81 BF |
172 | __xfs_buf_ioacct_dec(bp); |
173 | ||
a4082357 DC |
174 | atomic_set(&bp->b_lru_ref, 0); |
175 | if (!(bp->b_state & XFS_BSTATE_DISPOSE) && | |
e80dfa19 DC |
176 | (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru))) |
177 | atomic_dec(&bp->b_hold); | |
178 | ||
430cbeb8 | 179 | ASSERT(atomic_read(&bp->b_hold) >= 1); |
a4082357 | 180 | spin_unlock(&bp->b_lock); |
430cbeb8 | 181 | } |
1da177e4 | 182 | |
3e85c868 DC |
183 | static int |
184 | xfs_buf_get_maps( | |
185 | struct xfs_buf *bp, | |
186 | int map_count) | |
187 | { | |
188 | ASSERT(bp->b_maps == NULL); | |
189 | bp->b_map_count = map_count; | |
190 | ||
191 | if (map_count == 1) { | |
f4b42421 | 192 | bp->b_maps = &bp->__b_map; |
3e85c868 DC |
193 | return 0; |
194 | } | |
195 | ||
196 | bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map), | |
197 | KM_NOFS); | |
198 | if (!bp->b_maps) | |
2451337d | 199 | return -ENOMEM; |
3e85c868 DC |
200 | return 0; |
201 | } | |
202 | ||
203 | /* | |
204 | * Frees b_pages if it was allocated. | |
205 | */ | |
206 | static void | |
207 | xfs_buf_free_maps( | |
208 | struct xfs_buf *bp) | |
209 | { | |
f4b42421 | 210 | if (bp->b_maps != &bp->__b_map) { |
3e85c868 DC |
211 | kmem_free(bp->b_maps); |
212 | bp->b_maps = NULL; | |
213 | } | |
214 | } | |
215 | ||
4347b9d7 | 216 | struct xfs_buf * |
3e85c868 | 217 | _xfs_buf_alloc( |
4347b9d7 | 218 | struct xfs_buftarg *target, |
3e85c868 DC |
219 | struct xfs_buf_map *map, |
220 | int nmaps, | |
ce8e922c | 221 | xfs_buf_flags_t flags) |
1da177e4 | 222 | { |
4347b9d7 | 223 | struct xfs_buf *bp; |
3e85c868 DC |
224 | int error; |
225 | int i; | |
4347b9d7 | 226 | |
aa5c158e | 227 | bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS); |
4347b9d7 CH |
228 | if (unlikely(!bp)) |
229 | return NULL; | |
230 | ||
1da177e4 | 231 | /* |
12bcb3f7 DC |
232 | * We don't want certain flags to appear in b_flags unless they are |
233 | * specifically set by later operations on the buffer. | |
1da177e4 | 234 | */ |
611c9946 | 235 | flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD); |
ce8e922c | 236 | |
ce8e922c | 237 | atomic_set(&bp->b_hold, 1); |
430cbeb8 | 238 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 239 | init_completion(&bp->b_iowait); |
430cbeb8 | 240 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 241 | INIT_LIST_HEAD(&bp->b_list); |
643c8c05 | 242 | INIT_LIST_HEAD(&bp->b_li_list); |
a731cd11 | 243 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
a4082357 | 244 | spin_lock_init(&bp->b_lock); |
ce8e922c | 245 | bp->b_target = target; |
3e85c868 | 246 | bp->b_flags = flags; |
de1cbee4 | 247 | |
1da177e4 | 248 | /* |
aa0e8833 DC |
249 | * Set length and io_length to the same value initially. |
250 | * I/O routines should use io_length, which will be the same in | |
1da177e4 LT |
251 | * most cases but may be reset (e.g. XFS recovery). |
252 | */ | |
3e85c868 DC |
253 | error = xfs_buf_get_maps(bp, nmaps); |
254 | if (error) { | |
255 | kmem_zone_free(xfs_buf_zone, bp); | |
256 | return NULL; | |
257 | } | |
258 | ||
259 | bp->b_bn = map[0].bm_bn; | |
260 | bp->b_length = 0; | |
261 | for (i = 0; i < nmaps; i++) { | |
262 | bp->b_maps[i].bm_bn = map[i].bm_bn; | |
263 | bp->b_maps[i].bm_len = map[i].bm_len; | |
264 | bp->b_length += map[i].bm_len; | |
265 | } | |
266 | bp->b_io_length = bp->b_length; | |
267 | ||
ce8e922c NS |
268 | atomic_set(&bp->b_pin_count, 0); |
269 | init_waitqueue_head(&bp->b_waiters); | |
270 | ||
ff6d6af2 | 271 | XFS_STATS_INC(target->bt_mount, xb_create); |
0b1b213f | 272 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 CH |
273 | |
274 | return bp; | |
1da177e4 LT |
275 | } |
276 | ||
277 | /* | |
ce8e922c NS |
278 | * Allocate a page array capable of holding a specified number |
279 | * of pages, and point the page buf at it. | |
1da177e4 LT |
280 | */ |
281 | STATIC int | |
ce8e922c NS |
282 | _xfs_buf_get_pages( |
283 | xfs_buf_t *bp, | |
87937bf8 | 284 | int page_count) |
1da177e4 LT |
285 | { |
286 | /* Make sure that we have a page list */ | |
ce8e922c | 287 | if (bp->b_pages == NULL) { |
ce8e922c NS |
288 | bp->b_page_count = page_count; |
289 | if (page_count <= XB_PAGES) { | |
290 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 291 | } else { |
ce8e922c | 292 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
aa5c158e | 293 | page_count, KM_NOFS); |
ce8e922c | 294 | if (bp->b_pages == NULL) |
1da177e4 LT |
295 | return -ENOMEM; |
296 | } | |
ce8e922c | 297 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
298 | } |
299 | return 0; | |
300 | } | |
301 | ||
302 | /* | |
ce8e922c | 303 | * Frees b_pages if it was allocated. |
1da177e4 LT |
304 | */ |
305 | STATIC void | |
ce8e922c | 306 | _xfs_buf_free_pages( |
1da177e4 LT |
307 | xfs_buf_t *bp) |
308 | { | |
ce8e922c | 309 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 310 | kmem_free(bp->b_pages); |
3fc98b1a | 311 | bp->b_pages = NULL; |
1da177e4 LT |
312 | } |
313 | } | |
314 | ||
315 | /* | |
316 | * Releases the specified buffer. | |
317 | * | |
318 | * The modification state of any associated pages is left unchanged. | |
b46fe825 | 319 | * The buffer must not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
320 | * hashed and refcounted buffers |
321 | */ | |
322 | void | |
ce8e922c | 323 | xfs_buf_free( |
1da177e4 LT |
324 | xfs_buf_t *bp) |
325 | { | |
0b1b213f | 326 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 327 | |
430cbeb8 DC |
328 | ASSERT(list_empty(&bp->b_lru)); |
329 | ||
0e6e847f | 330 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
331 | uint i; |
332 | ||
73c77e2c | 333 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
334 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
335 | bp->b_page_count); | |
1da177e4 | 336 | |
948ecdb4 NS |
337 | for (i = 0; i < bp->b_page_count; i++) { |
338 | struct page *page = bp->b_pages[i]; | |
339 | ||
0e6e847f | 340 | __free_page(page); |
948ecdb4 | 341 | } |
0e6e847f DC |
342 | } else if (bp->b_flags & _XBF_KMEM) |
343 | kmem_free(bp->b_addr); | |
3fc98b1a | 344 | _xfs_buf_free_pages(bp); |
3e85c868 | 345 | xfs_buf_free_maps(bp); |
4347b9d7 | 346 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
347 | } |
348 | ||
349 | /* | |
0e6e847f | 350 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
351 | */ |
352 | STATIC int | |
0e6e847f | 353 | xfs_buf_allocate_memory( |
1da177e4 LT |
354 | xfs_buf_t *bp, |
355 | uint flags) | |
356 | { | |
aa0e8833 | 357 | size_t size; |
1da177e4 | 358 | size_t nbytes, offset; |
ce8e922c | 359 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 360 | unsigned short page_count, i; |
795cac72 | 361 | xfs_off_t start, end; |
1da177e4 LT |
362 | int error; |
363 | ||
0e6e847f DC |
364 | /* |
365 | * for buffers that are contained within a single page, just allocate | |
366 | * the memory from the heap - there's no need for the complexity of | |
367 | * page arrays to keep allocation down to order 0. | |
368 | */ | |
795cac72 DC |
369 | size = BBTOB(bp->b_length); |
370 | if (size < PAGE_SIZE) { | |
aa5c158e | 371 | bp->b_addr = kmem_alloc(size, KM_NOFS); |
0e6e847f DC |
372 | if (!bp->b_addr) { |
373 | /* low memory - use alloc_page loop instead */ | |
374 | goto use_alloc_page; | |
375 | } | |
376 | ||
795cac72 | 377 | if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) != |
0e6e847f DC |
378 | ((unsigned long)bp->b_addr & PAGE_MASK)) { |
379 | /* b_addr spans two pages - use alloc_page instead */ | |
380 | kmem_free(bp->b_addr); | |
381 | bp->b_addr = NULL; | |
382 | goto use_alloc_page; | |
383 | } | |
384 | bp->b_offset = offset_in_page(bp->b_addr); | |
385 | bp->b_pages = bp->b_page_array; | |
386 | bp->b_pages[0] = virt_to_page(bp->b_addr); | |
387 | bp->b_page_count = 1; | |
611c9946 | 388 | bp->b_flags |= _XBF_KMEM; |
0e6e847f DC |
389 | return 0; |
390 | } | |
391 | ||
392 | use_alloc_page: | |
f4b42421 MT |
393 | start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT; |
394 | end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1) | |
cbb7baab | 395 | >> PAGE_SHIFT; |
795cac72 | 396 | page_count = end - start; |
87937bf8 | 397 | error = _xfs_buf_get_pages(bp, page_count); |
1da177e4 LT |
398 | if (unlikely(error)) |
399 | return error; | |
1da177e4 | 400 | |
ce8e922c | 401 | offset = bp->b_offset; |
0e6e847f | 402 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 403 | |
ce8e922c | 404 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
405 | struct page *page; |
406 | uint retries = 0; | |
0e6e847f DC |
407 | retry: |
408 | page = alloc_page(gfp_mask); | |
1da177e4 | 409 | if (unlikely(page == NULL)) { |
ce8e922c NS |
410 | if (flags & XBF_READ_AHEAD) { |
411 | bp->b_page_count = i; | |
2451337d | 412 | error = -ENOMEM; |
0e6e847f | 413 | goto out_free_pages; |
1da177e4 LT |
414 | } |
415 | ||
416 | /* | |
417 | * This could deadlock. | |
418 | * | |
419 | * But until all the XFS lowlevel code is revamped to | |
420 | * handle buffer allocation failures we can't do much. | |
421 | */ | |
422 | if (!(++retries % 100)) | |
4f10700a | 423 | xfs_err(NULL, |
5bf97b1c TH |
424 | "%s(%u) possible memory allocation deadlock in %s (mode:0x%x)", |
425 | current->comm, current->pid, | |
34a622b2 | 426 | __func__, gfp_mask); |
1da177e4 | 427 | |
ff6d6af2 | 428 | XFS_STATS_INC(bp->b_target->bt_mount, xb_page_retries); |
8aa7e847 | 429 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
430 | goto retry; |
431 | } | |
432 | ||
ff6d6af2 | 433 | XFS_STATS_INC(bp->b_target->bt_mount, xb_page_found); |
1da177e4 | 434 | |
0e6e847f | 435 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 436 | size -= nbytes; |
ce8e922c | 437 | bp->b_pages[i] = page; |
1da177e4 LT |
438 | offset = 0; |
439 | } | |
0e6e847f | 440 | return 0; |
1da177e4 | 441 | |
0e6e847f DC |
442 | out_free_pages: |
443 | for (i = 0; i < bp->b_page_count; i++) | |
444 | __free_page(bp->b_pages[i]); | |
2aa6ba7b | 445 | bp->b_flags &= ~_XBF_PAGES; |
1da177e4 LT |
446 | return error; |
447 | } | |
448 | ||
449 | /* | |
25985edc | 450 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
451 | */ |
452 | STATIC int | |
ce8e922c | 453 | _xfs_buf_map_pages( |
1da177e4 LT |
454 | xfs_buf_t *bp, |
455 | uint flags) | |
456 | { | |
0e6e847f | 457 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 458 | if (bp->b_page_count == 1) { |
0e6e847f | 459 | /* A single page buffer is always mappable */ |
ce8e922c | 460 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
611c9946 DC |
461 | } else if (flags & XBF_UNMAPPED) { |
462 | bp->b_addr = NULL; | |
463 | } else { | |
a19fb380 | 464 | int retried = 0; |
9ba1fb2c | 465 | unsigned nofs_flag; |
ae687e58 DC |
466 | |
467 | /* | |
468 | * vm_map_ram() will allocate auxillary structures (e.g. | |
469 | * pagetables) with GFP_KERNEL, yet we are likely to be under | |
470 | * GFP_NOFS context here. Hence we need to tell memory reclaim | |
9ba1fb2c | 471 | * that we are in such a context via PF_MEMALLOC_NOFS to prevent |
ae687e58 DC |
472 | * memory reclaim re-entering the filesystem here and |
473 | * potentially deadlocking. | |
474 | */ | |
9ba1fb2c | 475 | nofs_flag = memalloc_nofs_save(); |
a19fb380 DC |
476 | do { |
477 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
478 | -1, PAGE_KERNEL); | |
479 | if (bp->b_addr) | |
480 | break; | |
481 | vm_unmap_aliases(); | |
482 | } while (retried++ <= 1); | |
9ba1fb2c | 483 | memalloc_nofs_restore(nofs_flag); |
a19fb380 DC |
484 | |
485 | if (!bp->b_addr) | |
1da177e4 | 486 | return -ENOMEM; |
ce8e922c | 487 | bp->b_addr += bp->b_offset; |
1da177e4 LT |
488 | } |
489 | ||
490 | return 0; | |
491 | } | |
492 | ||
493 | /* | |
494 | * Finding and Reading Buffers | |
495 | */ | |
6031e73a LS |
496 | static int |
497 | _xfs_buf_obj_cmp( | |
498 | struct rhashtable_compare_arg *arg, | |
499 | const void *obj) | |
500 | { | |
501 | const struct xfs_buf_map *map = arg->key; | |
502 | const struct xfs_buf *bp = obj; | |
503 | ||
504 | /* | |
505 | * The key hashing in the lookup path depends on the key being the | |
506 | * first element of the compare_arg, make sure to assert this. | |
507 | */ | |
508 | BUILD_BUG_ON(offsetof(struct xfs_buf_map, bm_bn) != 0); | |
509 | ||
510 | if (bp->b_bn != map->bm_bn) | |
511 | return 1; | |
512 | ||
513 | if (unlikely(bp->b_length != map->bm_len)) { | |
514 | /* | |
515 | * found a block number match. If the range doesn't | |
516 | * match, the only way this is allowed is if the buffer | |
517 | * in the cache is stale and the transaction that made | |
518 | * it stale has not yet committed. i.e. we are | |
519 | * reallocating a busy extent. Skip this buffer and | |
520 | * continue searching for an exact match. | |
521 | */ | |
522 | ASSERT(bp->b_flags & XBF_STALE); | |
523 | return 1; | |
524 | } | |
525 | return 0; | |
526 | } | |
527 | ||
528 | static const struct rhashtable_params xfs_buf_hash_params = { | |
529 | .min_size = 32, /* empty AGs have minimal footprint */ | |
530 | .nelem_hint = 16, | |
531 | .key_len = sizeof(xfs_daddr_t), | |
532 | .key_offset = offsetof(struct xfs_buf, b_bn), | |
533 | .head_offset = offsetof(struct xfs_buf, b_rhash_head), | |
534 | .automatic_shrinking = true, | |
535 | .obj_cmpfn = _xfs_buf_obj_cmp, | |
536 | }; | |
537 | ||
538 | int | |
539 | xfs_buf_hash_init( | |
540 | struct xfs_perag *pag) | |
541 | { | |
542 | spin_lock_init(&pag->pag_buf_lock); | |
543 | return rhashtable_init(&pag->pag_buf_hash, &xfs_buf_hash_params); | |
544 | } | |
545 | ||
546 | void | |
547 | xfs_buf_hash_destroy( | |
548 | struct xfs_perag *pag) | |
549 | { | |
550 | rhashtable_destroy(&pag->pag_buf_hash); | |
551 | } | |
1da177e4 LT |
552 | |
553 | /* | |
b027d4c9 DC |
554 | * Look up a buffer in the buffer cache and return it referenced and locked |
555 | * in @found_bp. | |
556 | * | |
557 | * If @new_bp is supplied and we have a lookup miss, insert @new_bp into the | |
558 | * cache. | |
559 | * | |
560 | * If XBF_TRYLOCK is set in @flags, only try to lock the buffer and return | |
561 | * -EAGAIN if we fail to lock it. | |
562 | * | |
563 | * Return values are: | |
564 | * -EFSCORRUPTED if have been supplied with an invalid address | |
565 | * -EAGAIN on trylock failure | |
566 | * -ENOENT if we fail to find a match and @new_bp was NULL | |
567 | * 0, with @found_bp: | |
568 | * - @new_bp if we inserted it into the cache | |
569 | * - the buffer we found and locked. | |
1da177e4 | 570 | */ |
b027d4c9 DC |
571 | static int |
572 | xfs_buf_find( | |
e70b73f8 | 573 | struct xfs_buftarg *btp, |
3e85c868 DC |
574 | struct xfs_buf_map *map, |
575 | int nmaps, | |
ce8e922c | 576 | xfs_buf_flags_t flags, |
b027d4c9 DC |
577 | struct xfs_buf *new_bp, |
578 | struct xfs_buf **found_bp) | |
1da177e4 | 579 | { |
74f75a0c | 580 | struct xfs_perag *pag; |
74f75a0c | 581 | xfs_buf_t *bp; |
6031e73a | 582 | struct xfs_buf_map cmap = { .bm_bn = map[0].bm_bn }; |
10616b80 | 583 | xfs_daddr_t eofs; |
3e85c868 | 584 | int i; |
1da177e4 | 585 | |
b027d4c9 DC |
586 | *found_bp = NULL; |
587 | ||
3e85c868 | 588 | for (i = 0; i < nmaps; i++) |
6031e73a | 589 | cmap.bm_len += map[i].bm_len; |
1da177e4 LT |
590 | |
591 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
6031e73a LS |
592 | ASSERT(!(BBTOB(cmap.bm_len) < btp->bt_meta_sectorsize)); |
593 | ASSERT(!(BBTOB(cmap.bm_bn) & (xfs_off_t)btp->bt_meta_sectormask)); | |
1da177e4 | 594 | |
10616b80 DC |
595 | /* |
596 | * Corrupted block numbers can get through to here, unfortunately, so we | |
597 | * have to check that the buffer falls within the filesystem bounds. | |
598 | */ | |
599 | eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks); | |
6031e73a | 600 | if (cmap.bm_bn < 0 || cmap.bm_bn >= eofs) { |
10616b80 | 601 | xfs_alert(btp->bt_mount, |
c219b015 | 602 | "%s: daddr 0x%llx out of range, EOFS 0x%llx", |
6031e73a | 603 | __func__, cmap.bm_bn, eofs); |
7bc0dc27 | 604 | WARN_ON(1); |
b027d4c9 | 605 | return -EFSCORRUPTED; |
10616b80 DC |
606 | } |
607 | ||
74f75a0c | 608 | pag = xfs_perag_get(btp->bt_mount, |
6031e73a | 609 | xfs_daddr_to_agno(btp->bt_mount, cmap.bm_bn)); |
74f75a0c | 610 | |
74f75a0c | 611 | spin_lock(&pag->pag_buf_lock); |
6031e73a LS |
612 | bp = rhashtable_lookup_fast(&pag->pag_buf_hash, &cmap, |
613 | xfs_buf_hash_params); | |
614 | if (bp) { | |
615 | atomic_inc(&bp->b_hold); | |
616 | goto found; | |
1da177e4 LT |
617 | } |
618 | ||
619 | /* No match found */ | |
b027d4c9 | 620 | if (!new_bp) { |
ff6d6af2 | 621 | XFS_STATS_INC(btp->bt_mount, xb_miss_locked); |
74f75a0c DC |
622 | spin_unlock(&pag->pag_buf_lock); |
623 | xfs_perag_put(pag); | |
b027d4c9 | 624 | return -ENOENT; |
1da177e4 | 625 | } |
b027d4c9 DC |
626 | |
627 | /* the buffer keeps the perag reference until it is freed */ | |
628 | new_bp->b_pag = pag; | |
629 | rhashtable_insert_fast(&pag->pag_buf_hash, &new_bp->b_rhash_head, | |
630 | xfs_buf_hash_params); | |
631 | spin_unlock(&pag->pag_buf_lock); | |
632 | *found_bp = new_bp; | |
633 | return 0; | |
1da177e4 LT |
634 | |
635 | found: | |
74f75a0c DC |
636 | spin_unlock(&pag->pag_buf_lock); |
637 | xfs_perag_put(pag); | |
1da177e4 | 638 | |
0c842ad4 CH |
639 | if (!xfs_buf_trylock(bp)) { |
640 | if (flags & XBF_TRYLOCK) { | |
ce8e922c | 641 | xfs_buf_rele(bp); |
ff6d6af2 | 642 | XFS_STATS_INC(btp->bt_mount, xb_busy_locked); |
b027d4c9 | 643 | return -EAGAIN; |
1da177e4 | 644 | } |
0c842ad4 | 645 | xfs_buf_lock(bp); |
ff6d6af2 | 646 | XFS_STATS_INC(btp->bt_mount, xb_get_locked_waited); |
1da177e4 LT |
647 | } |
648 | ||
0e6e847f DC |
649 | /* |
650 | * if the buffer is stale, clear all the external state associated with | |
651 | * it. We need to keep flags such as how we allocated the buffer memory | |
652 | * intact here. | |
653 | */ | |
ce8e922c NS |
654 | if (bp->b_flags & XBF_STALE) { |
655 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
cfb02852 | 656 | ASSERT(bp->b_iodone == NULL); |
611c9946 | 657 | bp->b_flags &= _XBF_KMEM | _XBF_PAGES; |
1813dd64 | 658 | bp->b_ops = NULL; |
2f926587 | 659 | } |
0b1b213f CH |
660 | |
661 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ff6d6af2 | 662 | XFS_STATS_INC(btp->bt_mount, xb_get_locked); |
b027d4c9 DC |
663 | *found_bp = bp; |
664 | return 0; | |
1da177e4 LT |
665 | } |
666 | ||
8925a3dc DC |
667 | struct xfs_buf * |
668 | xfs_buf_incore( | |
669 | struct xfs_buftarg *target, | |
670 | xfs_daddr_t blkno, | |
671 | size_t numblks, | |
672 | xfs_buf_flags_t flags) | |
673 | { | |
b027d4c9 DC |
674 | struct xfs_buf *bp; |
675 | int error; | |
8925a3dc | 676 | DEFINE_SINGLE_BUF_MAP(map, blkno, numblks); |
b027d4c9 DC |
677 | |
678 | error = xfs_buf_find(target, &map, 1, flags, NULL, &bp); | |
679 | if (error) | |
680 | return NULL; | |
681 | return bp; | |
8925a3dc DC |
682 | } |
683 | ||
1da177e4 | 684 | /* |
3815832a DC |
685 | * Assembles a buffer covering the specified range. The code is optimised for |
686 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
687 | * more hits than misses. | |
1da177e4 | 688 | */ |
3815832a | 689 | struct xfs_buf * |
6dde2707 DC |
690 | xfs_buf_get_map( |
691 | struct xfs_buftarg *target, | |
692 | struct xfs_buf_map *map, | |
693 | int nmaps, | |
ce8e922c | 694 | xfs_buf_flags_t flags) |
1da177e4 | 695 | { |
3815832a DC |
696 | struct xfs_buf *bp; |
697 | struct xfs_buf *new_bp; | |
0e6e847f | 698 | int error = 0; |
1da177e4 | 699 | |
b027d4c9 DC |
700 | error = xfs_buf_find(target, map, nmaps, flags, NULL, &bp); |
701 | ||
702 | switch (error) { | |
703 | case 0: | |
704 | /* cache hit */ | |
3815832a | 705 | goto found; |
b027d4c9 DC |
706 | case -EAGAIN: |
707 | /* cache hit, trylock failure, caller handles failure */ | |
708 | ASSERT(flags & XBF_TRYLOCK); | |
709 | return NULL; | |
710 | case -ENOENT: | |
711 | /* cache miss, go for insert */ | |
712 | break; | |
713 | case -EFSCORRUPTED: | |
714 | default: | |
715 | /* | |
716 | * None of the higher layers understand failure types | |
717 | * yet, so return NULL to signal a fatal lookup error. | |
718 | */ | |
719 | return NULL; | |
720 | } | |
3815832a | 721 | |
6dde2707 | 722 | new_bp = _xfs_buf_alloc(target, map, nmaps, flags); |
ce8e922c | 723 | if (unlikely(!new_bp)) |
1da177e4 LT |
724 | return NULL; |
725 | ||
fe2429b0 DC |
726 | error = xfs_buf_allocate_memory(new_bp, flags); |
727 | if (error) { | |
3e85c868 | 728 | xfs_buf_free(new_bp); |
fe2429b0 DC |
729 | return NULL; |
730 | } | |
731 | ||
b027d4c9 DC |
732 | error = xfs_buf_find(target, map, nmaps, flags, new_bp, &bp); |
733 | if (error) { | |
fe2429b0 | 734 | xfs_buf_free(new_bp); |
3815832a DC |
735 | return NULL; |
736 | } | |
737 | ||
fe2429b0 DC |
738 | if (bp != new_bp) |
739 | xfs_buf_free(new_bp); | |
1da177e4 | 740 | |
3815832a | 741 | found: |
611c9946 | 742 | if (!bp->b_addr) { |
ce8e922c | 743 | error = _xfs_buf_map_pages(bp, flags); |
1da177e4 | 744 | if (unlikely(error)) { |
4f10700a | 745 | xfs_warn(target->bt_mount, |
08e96e1a | 746 | "%s: failed to map pagesn", __func__); |
a8acad70 DC |
747 | xfs_buf_relse(bp); |
748 | return NULL; | |
1da177e4 LT |
749 | } |
750 | } | |
751 | ||
b79f4a1c DC |
752 | /* |
753 | * Clear b_error if this is a lookup from a caller that doesn't expect | |
754 | * valid data to be found in the buffer. | |
755 | */ | |
756 | if (!(flags & XBF_READ)) | |
757 | xfs_buf_ioerror(bp, 0); | |
758 | ||
ff6d6af2 | 759 | XFS_STATS_INC(target->bt_mount, xb_get); |
0b1b213f | 760 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 761 | return bp; |
1da177e4 LT |
762 | } |
763 | ||
5d765b97 CH |
764 | STATIC int |
765 | _xfs_buf_read( | |
766 | xfs_buf_t *bp, | |
767 | xfs_buf_flags_t flags) | |
768 | { | |
43ff2122 | 769 | ASSERT(!(flags & XBF_WRITE)); |
f4b42421 | 770 | ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL); |
5d765b97 | 771 | |
43ff2122 | 772 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD); |
1d5ae5df | 773 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 774 | |
6af88cda | 775 | return xfs_buf_submit(bp); |
5d765b97 CH |
776 | } |
777 | ||
1aff5696 | 778 | /* |
75d02303 | 779 | * Reverify a buffer found in cache without an attached ->b_ops. |
add46b3b | 780 | * |
75d02303 BF |
781 | * If the caller passed an ops structure and the buffer doesn't have ops |
782 | * assigned, set the ops and use it to verify the contents. If verification | |
783 | * fails, clear XBF_DONE. We assume the buffer has no recorded errors and is | |
784 | * already in XBF_DONE state on entry. | |
add46b3b | 785 | * |
75d02303 BF |
786 | * Under normal operations, every in-core buffer is verified on read I/O |
787 | * completion. There are two scenarios that can lead to in-core buffers without | |
788 | * an assigned ->b_ops. The first is during log recovery of buffers on a V4 | |
789 | * filesystem, though these buffers are purged at the end of recovery. The | |
790 | * other is online repair, which intentionally reads with a NULL buffer ops to | |
791 | * run several verifiers across an in-core buffer in order to establish buffer | |
792 | * type. If repair can't establish that, the buffer will be left in memory | |
793 | * with NULL buffer ops. | |
1aff5696 DW |
794 | */ |
795 | int | |
75d02303 | 796 | xfs_buf_reverify( |
1aff5696 DW |
797 | struct xfs_buf *bp, |
798 | const struct xfs_buf_ops *ops) | |
799 | { | |
800 | ASSERT(bp->b_flags & XBF_DONE); | |
801 | ASSERT(bp->b_error == 0); | |
802 | ||
803 | if (!ops || bp->b_ops) | |
804 | return 0; | |
805 | ||
806 | bp->b_ops = ops; | |
807 | bp->b_ops->verify_read(bp); | |
808 | if (bp->b_error) | |
809 | bp->b_flags &= ~XBF_DONE; | |
810 | return bp->b_error; | |
811 | } | |
812 | ||
1da177e4 | 813 | xfs_buf_t * |
6dde2707 DC |
814 | xfs_buf_read_map( |
815 | struct xfs_buftarg *target, | |
816 | struct xfs_buf_map *map, | |
817 | int nmaps, | |
c3f8fc73 | 818 | xfs_buf_flags_t flags, |
1813dd64 | 819 | const struct xfs_buf_ops *ops) |
1da177e4 | 820 | { |
6dde2707 | 821 | struct xfs_buf *bp; |
ce8e922c NS |
822 | |
823 | flags |= XBF_READ; | |
824 | ||
6dde2707 | 825 | bp = xfs_buf_get_map(target, map, nmaps, flags); |
1aff5696 DW |
826 | if (!bp) |
827 | return NULL; | |
0b1b213f | 828 | |
1aff5696 DW |
829 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
830 | ||
831 | if (!(bp->b_flags & XBF_DONE)) { | |
832 | XFS_STATS_INC(target->bt_mount, xb_get_read); | |
833 | bp->b_ops = ops; | |
834 | _xfs_buf_read(bp, flags); | |
835 | return bp; | |
836 | } | |
837 | ||
75d02303 | 838 | xfs_buf_reverify(bp, ops); |
1aff5696 DW |
839 | |
840 | if (flags & XBF_ASYNC) { | |
841 | /* | |
842 | * Read ahead call which is already satisfied, | |
843 | * drop the buffer | |
844 | */ | |
845 | xfs_buf_relse(bp); | |
846 | return NULL; | |
1da177e4 LT |
847 | } |
848 | ||
1aff5696 DW |
849 | /* We do not want read in the flags */ |
850 | bp->b_flags &= ~XBF_READ; | |
851 | ASSERT(bp->b_ops != NULL || ops == NULL); | |
ce8e922c | 852 | return bp; |
1da177e4 LT |
853 | } |
854 | ||
1da177e4 | 855 | /* |
ce8e922c NS |
856 | * If we are not low on memory then do the readahead in a deadlock |
857 | * safe manner. | |
1da177e4 LT |
858 | */ |
859 | void | |
6dde2707 DC |
860 | xfs_buf_readahead_map( |
861 | struct xfs_buftarg *target, | |
862 | struct xfs_buf_map *map, | |
c3f8fc73 | 863 | int nmaps, |
1813dd64 | 864 | const struct xfs_buf_ops *ops) |
1da177e4 | 865 | { |
efa7c9f9 | 866 | if (bdi_read_congested(target->bt_bdev->bd_bdi)) |
1da177e4 LT |
867 | return; |
868 | ||
6dde2707 | 869 | xfs_buf_read_map(target, map, nmaps, |
1813dd64 | 870 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD, ops); |
1da177e4 LT |
871 | } |
872 | ||
5adc94c2 DC |
873 | /* |
874 | * Read an uncached buffer from disk. Allocates and returns a locked | |
875 | * buffer containing the disk contents or nothing. | |
876 | */ | |
ba372674 | 877 | int |
5adc94c2 | 878 | xfs_buf_read_uncached( |
5adc94c2 DC |
879 | struct xfs_buftarg *target, |
880 | xfs_daddr_t daddr, | |
e70b73f8 | 881 | size_t numblks, |
c3f8fc73 | 882 | int flags, |
ba372674 | 883 | struct xfs_buf **bpp, |
1813dd64 | 884 | const struct xfs_buf_ops *ops) |
5adc94c2 | 885 | { |
eab4e633 | 886 | struct xfs_buf *bp; |
5adc94c2 | 887 | |
ba372674 DC |
888 | *bpp = NULL; |
889 | ||
e70b73f8 | 890 | bp = xfs_buf_get_uncached(target, numblks, flags); |
5adc94c2 | 891 | if (!bp) |
ba372674 | 892 | return -ENOMEM; |
5adc94c2 DC |
893 | |
894 | /* set up the buffer for a read IO */ | |
3e85c868 | 895 | ASSERT(bp->b_map_count == 1); |
ba372674 | 896 | bp->b_bn = XFS_BUF_DADDR_NULL; /* always null for uncached buffers */ |
3e85c868 | 897 | bp->b_maps[0].bm_bn = daddr; |
cbb7baab | 898 | bp->b_flags |= XBF_READ; |
1813dd64 | 899 | bp->b_ops = ops; |
5adc94c2 | 900 | |
6af88cda | 901 | xfs_buf_submit(bp); |
ba372674 DC |
902 | if (bp->b_error) { |
903 | int error = bp->b_error; | |
83a0adc3 | 904 | xfs_buf_relse(bp); |
ba372674 | 905 | return error; |
83a0adc3 | 906 | } |
ba372674 DC |
907 | |
908 | *bpp = bp; | |
909 | return 0; | |
1da177e4 LT |
910 | } |
911 | ||
44396476 DC |
912 | /* |
913 | * Return a buffer allocated as an empty buffer and associated to external | |
914 | * memory via xfs_buf_associate_memory() back to it's empty state. | |
915 | */ | |
916 | void | |
917 | xfs_buf_set_empty( | |
918 | struct xfs_buf *bp, | |
e70b73f8 | 919 | size_t numblks) |
44396476 DC |
920 | { |
921 | if (bp->b_pages) | |
922 | _xfs_buf_free_pages(bp); | |
923 | ||
924 | bp->b_pages = NULL; | |
925 | bp->b_page_count = 0; | |
926 | bp->b_addr = NULL; | |
4e94b71b | 927 | bp->b_length = numblks; |
aa0e8833 | 928 | bp->b_io_length = numblks; |
3e85c868 DC |
929 | |
930 | ASSERT(bp->b_map_count == 1); | |
44396476 | 931 | bp->b_bn = XFS_BUF_DADDR_NULL; |
3e85c868 DC |
932 | bp->b_maps[0].bm_bn = XFS_BUF_DADDR_NULL; |
933 | bp->b_maps[0].bm_len = bp->b_length; | |
44396476 DC |
934 | } |
935 | ||
1da177e4 LT |
936 | static inline struct page * |
937 | mem_to_page( | |
938 | void *addr) | |
939 | { | |
9e2779fa | 940 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
941 | return virt_to_page(addr); |
942 | } else { | |
943 | return vmalloc_to_page(addr); | |
944 | } | |
945 | } | |
946 | ||
947 | int | |
ce8e922c NS |
948 | xfs_buf_associate_memory( |
949 | xfs_buf_t *bp, | |
1da177e4 LT |
950 | void *mem, |
951 | size_t len) | |
952 | { | |
953 | int rval; | |
954 | int i = 0; | |
d1afb678 LM |
955 | unsigned long pageaddr; |
956 | unsigned long offset; | |
957 | size_t buflen; | |
1da177e4 LT |
958 | int page_count; |
959 | ||
0e6e847f | 960 | pageaddr = (unsigned long)mem & PAGE_MASK; |
d1afb678 | 961 | offset = (unsigned long)mem - pageaddr; |
0e6e847f DC |
962 | buflen = PAGE_ALIGN(len + offset); |
963 | page_count = buflen >> PAGE_SHIFT; | |
1da177e4 LT |
964 | |
965 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
966 | if (bp->b_pages) |
967 | _xfs_buf_free_pages(bp); | |
1da177e4 | 968 | |
ce8e922c NS |
969 | bp->b_pages = NULL; |
970 | bp->b_addr = mem; | |
1da177e4 | 971 | |
87937bf8 | 972 | rval = _xfs_buf_get_pages(bp, page_count); |
1da177e4 LT |
973 | if (rval) |
974 | return rval; | |
975 | ||
ce8e922c | 976 | bp->b_offset = offset; |
d1afb678 LM |
977 | |
978 | for (i = 0; i < bp->b_page_count; i++) { | |
979 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
0e6e847f | 980 | pageaddr += PAGE_SIZE; |
1da177e4 | 981 | } |
1da177e4 | 982 | |
aa0e8833 | 983 | bp->b_io_length = BTOBB(len); |
4e94b71b | 984 | bp->b_length = BTOBB(buflen); |
1da177e4 LT |
985 | |
986 | return 0; | |
987 | } | |
988 | ||
989 | xfs_buf_t * | |
686865f7 DC |
990 | xfs_buf_get_uncached( |
991 | struct xfs_buftarg *target, | |
e70b73f8 | 992 | size_t numblks, |
686865f7 | 993 | int flags) |
1da177e4 | 994 | { |
e70b73f8 | 995 | unsigned long page_count; |
1fa40b01 | 996 | int error, i; |
3e85c868 DC |
997 | struct xfs_buf *bp; |
998 | DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks); | |
1da177e4 | 999 | |
c891c30a BF |
1000 | /* flags might contain irrelevant bits, pass only what we care about */ |
1001 | bp = _xfs_buf_alloc(target, &map, 1, flags & XBF_NO_IOACCT); | |
1da177e4 LT |
1002 | if (unlikely(bp == NULL)) |
1003 | goto fail; | |
1da177e4 | 1004 | |
e70b73f8 | 1005 | page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT; |
87937bf8 | 1006 | error = _xfs_buf_get_pages(bp, page_count); |
1fa40b01 | 1007 | if (error) |
1da177e4 LT |
1008 | goto fail_free_buf; |
1009 | ||
1fa40b01 | 1010 | for (i = 0; i < page_count; i++) { |
686865f7 | 1011 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
1012 | if (!bp->b_pages[i]) |
1013 | goto fail_free_mem; | |
1da177e4 | 1014 | } |
1fa40b01 | 1015 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 1016 | |
611c9946 | 1017 | error = _xfs_buf_map_pages(bp, 0); |
1fa40b01 | 1018 | if (unlikely(error)) { |
4f10700a | 1019 | xfs_warn(target->bt_mount, |
08e96e1a | 1020 | "%s: failed to map pages", __func__); |
1da177e4 | 1021 | goto fail_free_mem; |
1fa40b01 | 1022 | } |
1da177e4 | 1023 | |
686865f7 | 1024 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 1025 | return bp; |
1fa40b01 | 1026 | |
1da177e4 | 1027 | fail_free_mem: |
1fa40b01 CH |
1028 | while (--i >= 0) |
1029 | __free_page(bp->b_pages[i]); | |
ca165b88 | 1030 | _xfs_buf_free_pages(bp); |
1da177e4 | 1031 | fail_free_buf: |
3e85c868 | 1032 | xfs_buf_free_maps(bp); |
4347b9d7 | 1033 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
1034 | fail: |
1035 | return NULL; | |
1036 | } | |
1037 | ||
1038 | /* | |
1da177e4 LT |
1039 | * Increment reference count on buffer, to hold the buffer concurrently |
1040 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
1041 | * Must hold the buffer already to call this function. |
1042 | */ | |
1043 | void | |
ce8e922c NS |
1044 | xfs_buf_hold( |
1045 | xfs_buf_t *bp) | |
1da177e4 | 1046 | { |
0b1b213f | 1047 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 1048 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
1049 | } |
1050 | ||
1051 | /* | |
9c7504aa BF |
1052 | * Release a hold on the specified buffer. If the hold count is 1, the buffer is |
1053 | * placed on LRU or freed (depending on b_lru_ref). | |
1da177e4 LT |
1054 | */ |
1055 | void | |
ce8e922c NS |
1056 | xfs_buf_rele( |
1057 | xfs_buf_t *bp) | |
1da177e4 | 1058 | { |
74f75a0c | 1059 | struct xfs_perag *pag = bp->b_pag; |
9c7504aa BF |
1060 | bool release; |
1061 | bool freebuf = false; | |
1da177e4 | 1062 | |
0b1b213f | 1063 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 1064 | |
74f75a0c | 1065 | if (!pag) { |
430cbeb8 | 1066 | ASSERT(list_empty(&bp->b_lru)); |
9c7504aa BF |
1067 | if (atomic_dec_and_test(&bp->b_hold)) { |
1068 | xfs_buf_ioacct_dec(bp); | |
fad3aa1e | 1069 | xfs_buf_free(bp); |
9c7504aa | 1070 | } |
fad3aa1e NS |
1071 | return; |
1072 | } | |
1073 | ||
3790689f | 1074 | ASSERT(atomic_read(&bp->b_hold) > 0); |
a4082357 | 1075 | |
37fd1678 DC |
1076 | /* |
1077 | * We grab the b_lock here first to serialise racing xfs_buf_rele() | |
1078 | * calls. The pag_buf_lock being taken on the last reference only | |
1079 | * serialises against racing lookups in xfs_buf_find(). IOWs, the second | |
1080 | * to last reference we drop here is not serialised against the last | |
1081 | * reference until we take bp->b_lock. Hence if we don't grab b_lock | |
1082 | * first, the last "release" reference can win the race to the lock and | |
1083 | * free the buffer before the second-to-last reference is processed, | |
1084 | * leading to a use-after-free scenario. | |
1085 | */ | |
9c7504aa | 1086 | spin_lock(&bp->b_lock); |
37fd1678 | 1087 | release = atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock); |
9c7504aa BF |
1088 | if (!release) { |
1089 | /* | |
1090 | * Drop the in-flight state if the buffer is already on the LRU | |
1091 | * and it holds the only reference. This is racy because we | |
1092 | * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT | |
1093 | * ensures the decrement occurs only once per-buf. | |
1094 | */ | |
1095 | if ((atomic_read(&bp->b_hold) == 1) && !list_empty(&bp->b_lru)) | |
63db7c81 | 1096 | __xfs_buf_ioacct_dec(bp); |
9c7504aa BF |
1097 | goto out_unlock; |
1098 | } | |
1099 | ||
1100 | /* the last reference has been dropped ... */ | |
63db7c81 | 1101 | __xfs_buf_ioacct_dec(bp); |
9c7504aa BF |
1102 | if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) { |
1103 | /* | |
1104 | * If the buffer is added to the LRU take a new reference to the | |
1105 | * buffer for the LRU and clear the (now stale) dispose list | |
1106 | * state flag | |
1107 | */ | |
1108 | if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) { | |
1109 | bp->b_state &= ~XFS_BSTATE_DISPOSE; | |
1110 | atomic_inc(&bp->b_hold); | |
1da177e4 | 1111 | } |
9c7504aa BF |
1112 | spin_unlock(&pag->pag_buf_lock); |
1113 | } else { | |
1114 | /* | |
1115 | * most of the time buffers will already be removed from the | |
1116 | * LRU, so optimise that case by checking for the | |
1117 | * XFS_BSTATE_DISPOSE flag indicating the last list the buffer | |
1118 | * was on was the disposal list | |
1119 | */ | |
1120 | if (!(bp->b_state & XFS_BSTATE_DISPOSE)) { | |
1121 | list_lru_del(&bp->b_target->bt_lru, &bp->b_lru); | |
1122 | } else { | |
1123 | ASSERT(list_empty(&bp->b_lru)); | |
1da177e4 | 1124 | } |
9c7504aa BF |
1125 | |
1126 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); | |
6031e73a LS |
1127 | rhashtable_remove_fast(&pag->pag_buf_hash, &bp->b_rhash_head, |
1128 | xfs_buf_hash_params); | |
9c7504aa BF |
1129 | spin_unlock(&pag->pag_buf_lock); |
1130 | xfs_perag_put(pag); | |
1131 | freebuf = true; | |
1da177e4 | 1132 | } |
9c7504aa BF |
1133 | |
1134 | out_unlock: | |
1135 | spin_unlock(&bp->b_lock); | |
1136 | ||
1137 | if (freebuf) | |
1138 | xfs_buf_free(bp); | |
1da177e4 LT |
1139 | } |
1140 | ||
1141 | ||
1142 | /* | |
0e6e847f | 1143 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
1144 | * |
1145 | * If we come across a stale, pinned, locked buffer, we know that we are | |
1146 | * being asked to lock a buffer that has been reallocated. Because it is | |
1147 | * pinned, we know that the log has not been pushed to disk and hence it | |
1148 | * will still be locked. Rather than continuing to have trylock attempts | |
1149 | * fail until someone else pushes the log, push it ourselves before | |
1150 | * returning. This means that the xfsaild will not get stuck trying | |
1151 | * to push on stale inode buffers. | |
1da177e4 LT |
1152 | */ |
1153 | int | |
0c842ad4 CH |
1154 | xfs_buf_trylock( |
1155 | struct xfs_buf *bp) | |
1da177e4 LT |
1156 | { |
1157 | int locked; | |
1158 | ||
ce8e922c | 1159 | locked = down_trylock(&bp->b_sema) == 0; |
fa6c668d | 1160 | if (locked) |
479c6412 | 1161 | trace_xfs_buf_trylock(bp, _RET_IP_); |
fa6c668d | 1162 | else |
479c6412 | 1163 | trace_xfs_buf_trylock_fail(bp, _RET_IP_); |
0c842ad4 | 1164 | return locked; |
1da177e4 | 1165 | } |
1da177e4 LT |
1166 | |
1167 | /* | |
0e6e847f | 1168 | * Lock a buffer object. |
ed3b4d6c DC |
1169 | * |
1170 | * If we come across a stale, pinned, locked buffer, we know that we | |
1171 | * are being asked to lock a buffer that has been reallocated. Because | |
1172 | * it is pinned, we know that the log has not been pushed to disk and | |
1173 | * hence it will still be locked. Rather than sleeping until someone | |
1174 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 1175 | */ |
ce8e922c NS |
1176 | void |
1177 | xfs_buf_lock( | |
0c842ad4 | 1178 | struct xfs_buf *bp) |
1da177e4 | 1179 | { |
0b1b213f CH |
1180 | trace_xfs_buf_lock(bp, _RET_IP_); |
1181 | ||
ed3b4d6c | 1182 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 1183 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c | 1184 | down(&bp->b_sema); |
0b1b213f CH |
1185 | |
1186 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
1187 | } |
1188 | ||
1da177e4 | 1189 | void |
ce8e922c | 1190 | xfs_buf_unlock( |
0c842ad4 | 1191 | struct xfs_buf *bp) |
1da177e4 | 1192 | { |
20e8a063 BF |
1193 | ASSERT(xfs_buf_islocked(bp)); |
1194 | ||
ce8e922c | 1195 | up(&bp->b_sema); |
0b1b213f | 1196 | trace_xfs_buf_unlock(bp, _RET_IP_); |
1da177e4 LT |
1197 | } |
1198 | ||
ce8e922c NS |
1199 | STATIC void |
1200 | xfs_buf_wait_unpin( | |
1201 | xfs_buf_t *bp) | |
1da177e4 LT |
1202 | { |
1203 | DECLARE_WAITQUEUE (wait, current); | |
1204 | ||
ce8e922c | 1205 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
1206 | return; |
1207 | ||
ce8e922c | 1208 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1209 | for (;;) { |
1210 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 1211 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 1212 | break; |
7eaceacc | 1213 | io_schedule(); |
1da177e4 | 1214 | } |
ce8e922c | 1215 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1216 | set_current_state(TASK_RUNNING); |
1217 | } | |
1218 | ||
1219 | /* | |
1220 | * Buffer Utility Routines | |
1221 | */ | |
1222 | ||
e8aaba9a DC |
1223 | void |
1224 | xfs_buf_ioend( | |
1225 | struct xfs_buf *bp) | |
1da177e4 | 1226 | { |
e8aaba9a DC |
1227 | bool read = bp->b_flags & XBF_READ; |
1228 | ||
1229 | trace_xfs_buf_iodone(bp, _RET_IP_); | |
1813dd64 DC |
1230 | |
1231 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); | |
d5929de8 | 1232 | |
61be9c52 DC |
1233 | /* |
1234 | * Pull in IO completion errors now. We are guaranteed to be running | |
1235 | * single threaded, so we don't need the lock to read b_io_error. | |
1236 | */ | |
1237 | if (!bp->b_error && bp->b_io_error) | |
1238 | xfs_buf_ioerror(bp, bp->b_io_error); | |
1239 | ||
e8aaba9a DC |
1240 | /* Only validate buffers that were read without errors */ |
1241 | if (read && !bp->b_error && bp->b_ops) { | |
1242 | ASSERT(!bp->b_iodone); | |
1813dd64 | 1243 | bp->b_ops->verify_read(bp); |
e8aaba9a DC |
1244 | } |
1245 | ||
1246 | if (!bp->b_error) | |
1247 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1248 | |
80f6c29d | 1249 | if (bp->b_iodone) |
ce8e922c NS |
1250 | (*(bp->b_iodone))(bp); |
1251 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 | 1252 | xfs_buf_relse(bp); |
595bff75 | 1253 | else |
1813dd64 | 1254 | complete(&bp->b_iowait); |
1da177e4 LT |
1255 | } |
1256 | ||
e8aaba9a DC |
1257 | static void |
1258 | xfs_buf_ioend_work( | |
1259 | struct work_struct *work) | |
1da177e4 | 1260 | { |
e8aaba9a | 1261 | struct xfs_buf *bp = |
b29c70f5 | 1262 | container_of(work, xfs_buf_t, b_ioend_work); |
0b1b213f | 1263 | |
e8aaba9a DC |
1264 | xfs_buf_ioend(bp); |
1265 | } | |
1da177e4 | 1266 | |
211fe1a4 | 1267 | static void |
e8aaba9a DC |
1268 | xfs_buf_ioend_async( |
1269 | struct xfs_buf *bp) | |
1270 | { | |
b29c70f5 BF |
1271 | INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work); |
1272 | queue_work(bp->b_ioend_wq, &bp->b_ioend_work); | |
1da177e4 LT |
1273 | } |
1274 | ||
1da177e4 | 1275 | void |
31ca03c9 | 1276 | __xfs_buf_ioerror( |
ce8e922c | 1277 | xfs_buf_t *bp, |
31ca03c9 DW |
1278 | int error, |
1279 | xfs_failaddr_t failaddr) | |
1da177e4 | 1280 | { |
2451337d DC |
1281 | ASSERT(error <= 0 && error >= -1000); |
1282 | bp->b_error = error; | |
31ca03c9 | 1283 | trace_xfs_buf_ioerror(bp, error, failaddr); |
1da177e4 LT |
1284 | } |
1285 | ||
901796af CH |
1286 | void |
1287 | xfs_buf_ioerror_alert( | |
1288 | struct xfs_buf *bp, | |
1289 | const char *func) | |
1290 | { | |
1291 | xfs_alert(bp->b_target->bt_mount, | |
c219b015 DW |
1292 | "metadata I/O error in \"%s\" at daddr 0x%llx len %d error %d", |
1293 | func, (uint64_t)XFS_BUF_ADDR(bp), bp->b_length, | |
1294 | -bp->b_error); | |
901796af CH |
1295 | } |
1296 | ||
a2dcf5df CH |
1297 | int |
1298 | xfs_bwrite( | |
1299 | struct xfs_buf *bp) | |
1300 | { | |
1301 | int error; | |
1302 | ||
1303 | ASSERT(xfs_buf_islocked(bp)); | |
1304 | ||
1305 | bp->b_flags |= XBF_WRITE; | |
27187754 DC |
1306 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q | |
1307 | XBF_WRITE_FAIL | XBF_DONE); | |
a2dcf5df | 1308 | |
6af88cda | 1309 | error = xfs_buf_submit(bp); |
a2dcf5df CH |
1310 | if (error) { |
1311 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1312 | SHUTDOWN_META_IO_ERROR); | |
1313 | } | |
1314 | return error; | |
1315 | } | |
1316 | ||
9bdd9bd6 | 1317 | static void |
ce8e922c | 1318 | xfs_buf_bio_end_io( |
4246a0b6 | 1319 | struct bio *bio) |
1da177e4 | 1320 | { |
9bdd9bd6 | 1321 | struct xfs_buf *bp = (struct xfs_buf *)bio->bi_private; |
1da177e4 | 1322 | |
37eb17e6 DC |
1323 | /* |
1324 | * don't overwrite existing errors - otherwise we can lose errors on | |
1325 | * buffers that require multiple bios to complete. | |
1326 | */ | |
4e4cbee9 CH |
1327 | if (bio->bi_status) { |
1328 | int error = blk_status_to_errno(bio->bi_status); | |
1329 | ||
1330 | cmpxchg(&bp->b_io_error, 0, error); | |
1331 | } | |
1da177e4 | 1332 | |
37eb17e6 | 1333 | if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
73c77e2c JB |
1334 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); |
1335 | ||
e8aaba9a DC |
1336 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
1337 | xfs_buf_ioend_async(bp); | |
1da177e4 | 1338 | bio_put(bio); |
1da177e4 LT |
1339 | } |
1340 | ||
3e85c868 DC |
1341 | static void |
1342 | xfs_buf_ioapply_map( | |
1343 | struct xfs_buf *bp, | |
1344 | int map, | |
1345 | int *buf_offset, | |
1346 | int *count, | |
50bfcd0c MC |
1347 | int op, |
1348 | int op_flags) | |
1da177e4 | 1349 | { |
3e85c868 DC |
1350 | int page_index; |
1351 | int total_nr_pages = bp->b_page_count; | |
1352 | int nr_pages; | |
1353 | struct bio *bio; | |
1354 | sector_t sector = bp->b_maps[map].bm_bn; | |
1355 | int size; | |
1356 | int offset; | |
1da177e4 | 1357 | |
3e85c868 DC |
1358 | /* skip the pages in the buffer before the start offset */ |
1359 | page_index = 0; | |
1360 | offset = *buf_offset; | |
1361 | while (offset >= PAGE_SIZE) { | |
1362 | page_index++; | |
1363 | offset -= PAGE_SIZE; | |
f538d4da CH |
1364 | } |
1365 | ||
3e85c868 DC |
1366 | /* |
1367 | * Limit the IO size to the length of the current vector, and update the | |
1368 | * remaining IO count for the next time around. | |
1369 | */ | |
1370 | size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count); | |
1371 | *count -= size; | |
1372 | *buf_offset += size; | |
34951f5c | 1373 | |
1da177e4 | 1374 | next_chunk: |
ce8e922c | 1375 | atomic_inc(&bp->b_io_remaining); |
c908e380 | 1376 | nr_pages = min(total_nr_pages, BIO_MAX_PAGES); |
1da177e4 LT |
1377 | |
1378 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
74d46992 | 1379 | bio_set_dev(bio, bp->b_target->bt_bdev); |
4f024f37 | 1380 | bio->bi_iter.bi_sector = sector; |
ce8e922c NS |
1381 | bio->bi_end_io = xfs_buf_bio_end_io; |
1382 | bio->bi_private = bp; | |
50bfcd0c | 1383 | bio_set_op_attrs(bio, op, op_flags); |
0e6e847f | 1384 | |
3e85c868 | 1385 | for (; size && nr_pages; nr_pages--, page_index++) { |
0e6e847f | 1386 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1387 | |
1388 | if (nbytes > size) | |
1389 | nbytes = size; | |
1390 | ||
3e85c868 DC |
1391 | rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes, |
1392 | offset); | |
ce8e922c | 1393 | if (rbytes < nbytes) |
1da177e4 LT |
1394 | break; |
1395 | ||
1396 | offset = 0; | |
aa0e8833 | 1397 | sector += BTOBB(nbytes); |
1da177e4 LT |
1398 | size -= nbytes; |
1399 | total_nr_pages--; | |
1400 | } | |
1401 | ||
4f024f37 | 1402 | if (likely(bio->bi_iter.bi_size)) { |
73c77e2c JB |
1403 | if (xfs_buf_is_vmapped(bp)) { |
1404 | flush_kernel_vmap_range(bp->b_addr, | |
1405 | xfs_buf_vmap_len(bp)); | |
1406 | } | |
4e49ea4a | 1407 | submit_bio(bio); |
1da177e4 LT |
1408 | if (size) |
1409 | goto next_chunk; | |
1410 | } else { | |
37eb17e6 DC |
1411 | /* |
1412 | * This is guaranteed not to be the last io reference count | |
595bff75 | 1413 | * because the caller (xfs_buf_submit) holds a count itself. |
37eb17e6 DC |
1414 | */ |
1415 | atomic_dec(&bp->b_io_remaining); | |
2451337d | 1416 | xfs_buf_ioerror(bp, -EIO); |
ec53d1db | 1417 | bio_put(bio); |
1da177e4 | 1418 | } |
3e85c868 DC |
1419 | |
1420 | } | |
1421 | ||
1422 | STATIC void | |
1423 | _xfs_buf_ioapply( | |
1424 | struct xfs_buf *bp) | |
1425 | { | |
1426 | struct blk_plug plug; | |
50bfcd0c MC |
1427 | int op; |
1428 | int op_flags = 0; | |
3e85c868 DC |
1429 | int offset; |
1430 | int size; | |
1431 | int i; | |
1432 | ||
c163f9a1 DC |
1433 | /* |
1434 | * Make sure we capture only current IO errors rather than stale errors | |
1435 | * left over from previous use of the buffer (e.g. failed readahead). | |
1436 | */ | |
1437 | bp->b_error = 0; | |
1438 | ||
b29c70f5 BF |
1439 | /* |
1440 | * Initialize the I/O completion workqueue if we haven't yet or the | |
1441 | * submitter has not opted to specify a custom one. | |
1442 | */ | |
1443 | if (!bp->b_ioend_wq) | |
1444 | bp->b_ioend_wq = bp->b_target->bt_mount->m_buf_workqueue; | |
1445 | ||
3e85c868 | 1446 | if (bp->b_flags & XBF_WRITE) { |
50bfcd0c | 1447 | op = REQ_OP_WRITE; |
3e85c868 | 1448 | if (bp->b_flags & XBF_SYNCIO) |
70fd7614 | 1449 | op_flags = REQ_SYNC; |
3e85c868 | 1450 | if (bp->b_flags & XBF_FUA) |
50bfcd0c | 1451 | op_flags |= REQ_FUA; |
3e85c868 | 1452 | if (bp->b_flags & XBF_FLUSH) |
28a8f0d3 | 1453 | op_flags |= REQ_PREFLUSH; |
1813dd64 DC |
1454 | |
1455 | /* | |
1456 | * Run the write verifier callback function if it exists. If | |
1457 | * this function fails it will mark the buffer with an error and | |
1458 | * the IO should not be dispatched. | |
1459 | */ | |
1460 | if (bp->b_ops) { | |
1461 | bp->b_ops->verify_write(bp); | |
1462 | if (bp->b_error) { | |
1463 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1464 | SHUTDOWN_CORRUPT_INCORE); | |
1465 | return; | |
1466 | } | |
400b9d88 DC |
1467 | } else if (bp->b_bn != XFS_BUF_DADDR_NULL) { |
1468 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
1469 | ||
1470 | /* | |
1471 | * non-crc filesystems don't attach verifiers during | |
1472 | * log recovery, so don't warn for such filesystems. | |
1473 | */ | |
1474 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
1475 | xfs_warn(mp, | |
c219b015 | 1476 | "%s: no buf ops on daddr 0x%llx len %d", |
400b9d88 | 1477 | __func__, bp->b_bn, bp->b_length); |
9c712a13 DW |
1478 | xfs_hex_dump(bp->b_addr, |
1479 | XFS_CORRUPTION_DUMP_LEN); | |
400b9d88 DC |
1480 | dump_stack(); |
1481 | } | |
1813dd64 | 1482 | } |
3e85c868 | 1483 | } else if (bp->b_flags & XBF_READ_AHEAD) { |
50bfcd0c MC |
1484 | op = REQ_OP_READ; |
1485 | op_flags = REQ_RAHEAD; | |
3e85c868 | 1486 | } else { |
50bfcd0c | 1487 | op = REQ_OP_READ; |
3e85c868 DC |
1488 | } |
1489 | ||
1490 | /* we only use the buffer cache for meta-data */ | |
50bfcd0c | 1491 | op_flags |= REQ_META; |
3e85c868 DC |
1492 | |
1493 | /* | |
1494 | * Walk all the vectors issuing IO on them. Set up the initial offset | |
1495 | * into the buffer and the desired IO size before we start - | |
1496 | * _xfs_buf_ioapply_vec() will modify them appropriately for each | |
1497 | * subsequent call. | |
1498 | */ | |
1499 | offset = bp->b_offset; | |
1500 | size = BBTOB(bp->b_io_length); | |
1501 | blk_start_plug(&plug); | |
1502 | for (i = 0; i < bp->b_map_count; i++) { | |
50bfcd0c | 1503 | xfs_buf_ioapply_map(bp, i, &offset, &size, op, op_flags); |
3e85c868 DC |
1504 | if (bp->b_error) |
1505 | break; | |
1506 | if (size <= 0) | |
1507 | break; /* all done */ | |
1508 | } | |
1509 | blk_finish_plug(&plug); | |
1da177e4 LT |
1510 | } |
1511 | ||
595bff75 | 1512 | /* |
bb00b6f1 | 1513 | * Wait for I/O completion of a sync buffer and return the I/O error code. |
595bff75 | 1514 | */ |
eaebb515 | 1515 | static int |
bb00b6f1 | 1516 | xfs_buf_iowait( |
595bff75 | 1517 | struct xfs_buf *bp) |
1da177e4 | 1518 | { |
bb00b6f1 BF |
1519 | ASSERT(!(bp->b_flags & XBF_ASYNC)); |
1520 | ||
1521 | trace_xfs_buf_iowait(bp, _RET_IP_); | |
1522 | wait_for_completion(&bp->b_iowait); | |
1523 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
1524 | ||
1525 | return bp->b_error; | |
1526 | } | |
1527 | ||
1528 | /* | |
1529 | * Buffer I/O submission path, read or write. Asynchronous submission transfers | |
1530 | * the buffer lock ownership and the current reference to the IO. It is not | |
1531 | * safe to reference the buffer after a call to this function unless the caller | |
1532 | * holds an additional reference itself. | |
1533 | */ | |
1534 | int | |
1535 | __xfs_buf_submit( | |
1536 | struct xfs_buf *bp, | |
1537 | bool wait) | |
1538 | { | |
1539 | int error = 0; | |
1540 | ||
595bff75 | 1541 | trace_xfs_buf_submit(bp, _RET_IP_); |
1da177e4 | 1542 | |
43ff2122 | 1543 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
595bff75 DC |
1544 | |
1545 | /* on shutdown we stale and complete the buffer immediately */ | |
1546 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { | |
1547 | xfs_buf_ioerror(bp, -EIO); | |
1548 | bp->b_flags &= ~XBF_DONE; | |
1549 | xfs_buf_stale(bp); | |
465fa17f | 1550 | xfs_buf_ioend(bp); |
eaebb515 | 1551 | return -EIO; |
595bff75 | 1552 | } |
1da177e4 | 1553 | |
bb00b6f1 BF |
1554 | /* |
1555 | * Grab a reference so the buffer does not go away underneath us. For | |
1556 | * async buffers, I/O completion drops the callers reference, which | |
1557 | * could occur before submission returns. | |
1558 | */ | |
1559 | xfs_buf_hold(bp); | |
1560 | ||
375ec69d | 1561 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1562 | xfs_buf_wait_unpin(bp); |
e11bb805 | 1563 | |
61be9c52 DC |
1564 | /* clear the internal error state to avoid spurious errors */ |
1565 | bp->b_io_error = 0; | |
1566 | ||
8d6c1210 | 1567 | /* |
e11bb805 DC |
1568 | * Set the count to 1 initially, this will stop an I/O completion |
1569 | * callout which happens before we have started all the I/O from calling | |
1570 | * xfs_buf_ioend too early. | |
1da177e4 | 1571 | */ |
ce8e922c | 1572 | atomic_set(&bp->b_io_remaining, 1); |
eaebb515 BF |
1573 | if (bp->b_flags & XBF_ASYNC) |
1574 | xfs_buf_ioacct_inc(bp); | |
ce8e922c | 1575 | _xfs_buf_ioapply(bp); |
e11bb805 | 1576 | |
8d6c1210 | 1577 | /* |
595bff75 DC |
1578 | * If _xfs_buf_ioapply failed, we can get back here with only the IO |
1579 | * reference we took above. If we drop it to zero, run completion so | |
1580 | * that we don't return to the caller with completion still pending. | |
8d6c1210 | 1581 | */ |
e8aaba9a | 1582 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
eaebb515 | 1583 | if (bp->b_error || !(bp->b_flags & XBF_ASYNC)) |
e8aaba9a DC |
1584 | xfs_buf_ioend(bp); |
1585 | else | |
1586 | xfs_buf_ioend_async(bp); | |
1587 | } | |
1da177e4 | 1588 | |
6af88cda BF |
1589 | if (wait) |
1590 | error = xfs_buf_iowait(bp); | |
bb00b6f1 | 1591 | |
595bff75 | 1592 | /* |
6af88cda BF |
1593 | * Release the hold that keeps the buffer referenced for the entire |
1594 | * I/O. Note that if the buffer is async, it is not safe to reference | |
1595 | * after this release. | |
595bff75 DC |
1596 | */ |
1597 | xfs_buf_rele(bp); | |
1598 | return error; | |
1da177e4 LT |
1599 | } |
1600 | ||
88ee2df7 | 1601 | void * |
ce8e922c | 1602 | xfs_buf_offset( |
88ee2df7 | 1603 | struct xfs_buf *bp, |
1da177e4 LT |
1604 | size_t offset) |
1605 | { | |
1606 | struct page *page; | |
1607 | ||
611c9946 | 1608 | if (bp->b_addr) |
62926044 | 1609 | return bp->b_addr + offset; |
1da177e4 | 1610 | |
ce8e922c | 1611 | offset += bp->b_offset; |
0e6e847f | 1612 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
88ee2df7 | 1613 | return page_address(page) + (offset & (PAGE_SIZE-1)); |
1da177e4 LT |
1614 | } |
1615 | ||
1616 | /* | |
1da177e4 LT |
1617 | * Move data into or out of a buffer. |
1618 | */ | |
1619 | void | |
ce8e922c NS |
1620 | xfs_buf_iomove( |
1621 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1622 | size_t boff, /* starting buffer offset */ |
1623 | size_t bsize, /* length to copy */ | |
b9c48649 | 1624 | void *data, /* data address */ |
ce8e922c | 1625 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 | 1626 | { |
795cac72 | 1627 | size_t bend; |
1da177e4 LT |
1628 | |
1629 | bend = boff + bsize; | |
1630 | while (boff < bend) { | |
795cac72 DC |
1631 | struct page *page; |
1632 | int page_index, page_offset, csize; | |
1633 | ||
1634 | page_index = (boff + bp->b_offset) >> PAGE_SHIFT; | |
1635 | page_offset = (boff + bp->b_offset) & ~PAGE_MASK; | |
1636 | page = bp->b_pages[page_index]; | |
1637 | csize = min_t(size_t, PAGE_SIZE - page_offset, | |
1638 | BBTOB(bp->b_io_length) - boff); | |
1da177e4 | 1639 | |
795cac72 | 1640 | ASSERT((csize + page_offset) <= PAGE_SIZE); |
1da177e4 LT |
1641 | |
1642 | switch (mode) { | |
ce8e922c | 1643 | case XBRW_ZERO: |
795cac72 | 1644 | memset(page_address(page) + page_offset, 0, csize); |
1da177e4 | 1645 | break; |
ce8e922c | 1646 | case XBRW_READ: |
795cac72 | 1647 | memcpy(data, page_address(page) + page_offset, csize); |
1da177e4 | 1648 | break; |
ce8e922c | 1649 | case XBRW_WRITE: |
795cac72 | 1650 | memcpy(page_address(page) + page_offset, data, csize); |
1da177e4 LT |
1651 | } |
1652 | ||
1653 | boff += csize; | |
1654 | data += csize; | |
1655 | } | |
1656 | } | |
1657 | ||
1658 | /* | |
ce8e922c | 1659 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1660 | */ |
1661 | ||
1662 | /* | |
430cbeb8 DC |
1663 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1664 | * returned. These buffers will have an elevated hold count, so wait on those | |
1665 | * while freeing all the buffers only held by the LRU. | |
1da177e4 | 1666 | */ |
e80dfa19 DC |
1667 | static enum lru_status |
1668 | xfs_buftarg_wait_rele( | |
1669 | struct list_head *item, | |
3f97b163 | 1670 | struct list_lru_one *lru, |
e80dfa19 DC |
1671 | spinlock_t *lru_lock, |
1672 | void *arg) | |
1673 | ||
1da177e4 | 1674 | { |
e80dfa19 | 1675 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); |
a4082357 | 1676 | struct list_head *dispose = arg; |
430cbeb8 | 1677 | |
e80dfa19 | 1678 | if (atomic_read(&bp->b_hold) > 1) { |
a4082357 | 1679 | /* need to wait, so skip it this pass */ |
e80dfa19 | 1680 | trace_xfs_buf_wait_buftarg(bp, _RET_IP_); |
a4082357 | 1681 | return LRU_SKIP; |
1da177e4 | 1682 | } |
a4082357 DC |
1683 | if (!spin_trylock(&bp->b_lock)) |
1684 | return LRU_SKIP; | |
e80dfa19 | 1685 | |
a4082357 DC |
1686 | /* |
1687 | * clear the LRU reference count so the buffer doesn't get | |
1688 | * ignored in xfs_buf_rele(). | |
1689 | */ | |
1690 | atomic_set(&bp->b_lru_ref, 0); | |
1691 | bp->b_state |= XFS_BSTATE_DISPOSE; | |
3f97b163 | 1692 | list_lru_isolate_move(lru, item, dispose); |
a4082357 DC |
1693 | spin_unlock(&bp->b_lock); |
1694 | return LRU_REMOVED; | |
1da177e4 LT |
1695 | } |
1696 | ||
e80dfa19 DC |
1697 | void |
1698 | xfs_wait_buftarg( | |
1699 | struct xfs_buftarg *btp) | |
1700 | { | |
a4082357 DC |
1701 | LIST_HEAD(dispose); |
1702 | int loop = 0; | |
1703 | ||
85bec546 | 1704 | /* |
9c7504aa BF |
1705 | * First wait on the buftarg I/O count for all in-flight buffers to be |
1706 | * released. This is critical as new buffers do not make the LRU until | |
1707 | * they are released. | |
1708 | * | |
1709 | * Next, flush the buffer workqueue to ensure all completion processing | |
1710 | * has finished. Just waiting on buffer locks is not sufficient for | |
1711 | * async IO as the reference count held over IO is not released until | |
1712 | * after the buffer lock is dropped. Hence we need to ensure here that | |
1713 | * all reference counts have been dropped before we start walking the | |
1714 | * LRU list. | |
85bec546 | 1715 | */ |
9c7504aa BF |
1716 | while (percpu_counter_sum(&btp->bt_io_count)) |
1717 | delay(100); | |
800b2694 | 1718 | flush_workqueue(btp->bt_mount->m_buf_workqueue); |
85bec546 | 1719 | |
a4082357 DC |
1720 | /* loop until there is nothing left on the lru list. */ |
1721 | while (list_lru_count(&btp->bt_lru)) { | |
e80dfa19 | 1722 | list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele, |
a4082357 DC |
1723 | &dispose, LONG_MAX); |
1724 | ||
1725 | while (!list_empty(&dispose)) { | |
1726 | struct xfs_buf *bp; | |
1727 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1728 | list_del_init(&bp->b_lru); | |
ac8809f9 DC |
1729 | if (bp->b_flags & XBF_WRITE_FAIL) { |
1730 | xfs_alert(btp->bt_mount, | |
c219b015 | 1731 | "Corruption Alert: Buffer at daddr 0x%llx had permanent write failures!", |
ac8809f9 | 1732 | (long long)bp->b_bn); |
f41febd2 JP |
1733 | xfs_alert(btp->bt_mount, |
1734 | "Please run xfs_repair to determine the extent of the problem."); | |
ac8809f9 | 1735 | } |
a4082357 DC |
1736 | xfs_buf_rele(bp); |
1737 | } | |
1738 | if (loop++ != 0) | |
1739 | delay(100); | |
1740 | } | |
e80dfa19 DC |
1741 | } |
1742 | ||
1743 | static enum lru_status | |
1744 | xfs_buftarg_isolate( | |
1745 | struct list_head *item, | |
3f97b163 | 1746 | struct list_lru_one *lru, |
e80dfa19 DC |
1747 | spinlock_t *lru_lock, |
1748 | void *arg) | |
1749 | { | |
1750 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); | |
1751 | struct list_head *dispose = arg; | |
1752 | ||
a4082357 DC |
1753 | /* |
1754 | * we are inverting the lru lock/bp->b_lock here, so use a trylock. | |
1755 | * If we fail to get the lock, just skip it. | |
1756 | */ | |
1757 | if (!spin_trylock(&bp->b_lock)) | |
1758 | return LRU_SKIP; | |
e80dfa19 DC |
1759 | /* |
1760 | * Decrement the b_lru_ref count unless the value is already | |
1761 | * zero. If the value is already zero, we need to reclaim the | |
1762 | * buffer, otherwise it gets another trip through the LRU. | |
1763 | */ | |
19957a18 | 1764 | if (atomic_add_unless(&bp->b_lru_ref, -1, 0)) { |
a4082357 | 1765 | spin_unlock(&bp->b_lock); |
e80dfa19 | 1766 | return LRU_ROTATE; |
a4082357 | 1767 | } |
e80dfa19 | 1768 | |
a4082357 | 1769 | bp->b_state |= XFS_BSTATE_DISPOSE; |
3f97b163 | 1770 | list_lru_isolate_move(lru, item, dispose); |
a4082357 | 1771 | spin_unlock(&bp->b_lock); |
e80dfa19 DC |
1772 | return LRU_REMOVED; |
1773 | } | |
1774 | ||
addbda40 | 1775 | static unsigned long |
e80dfa19 | 1776 | xfs_buftarg_shrink_scan( |
ff57ab21 | 1777 | struct shrinker *shrink, |
1495f230 | 1778 | struct shrink_control *sc) |
a6867a68 | 1779 | { |
ff57ab21 DC |
1780 | struct xfs_buftarg *btp = container_of(shrink, |
1781 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1782 | LIST_HEAD(dispose); |
addbda40 | 1783 | unsigned long freed; |
430cbeb8 | 1784 | |
503c358c VD |
1785 | freed = list_lru_shrink_walk(&btp->bt_lru, sc, |
1786 | xfs_buftarg_isolate, &dispose); | |
430cbeb8 DC |
1787 | |
1788 | while (!list_empty(&dispose)) { | |
e80dfa19 | 1789 | struct xfs_buf *bp; |
430cbeb8 DC |
1790 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); |
1791 | list_del_init(&bp->b_lru); | |
1792 | xfs_buf_rele(bp); | |
1793 | } | |
1794 | ||
e80dfa19 DC |
1795 | return freed; |
1796 | } | |
1797 | ||
addbda40 | 1798 | static unsigned long |
e80dfa19 DC |
1799 | xfs_buftarg_shrink_count( |
1800 | struct shrinker *shrink, | |
1801 | struct shrink_control *sc) | |
1802 | { | |
1803 | struct xfs_buftarg *btp = container_of(shrink, | |
1804 | struct xfs_buftarg, bt_shrinker); | |
503c358c | 1805 | return list_lru_shrink_count(&btp->bt_lru, sc); |
a6867a68 DC |
1806 | } |
1807 | ||
1da177e4 LT |
1808 | void |
1809 | xfs_free_buftarg( | |
b7963133 | 1810 | struct xfs_buftarg *btp) |
1da177e4 | 1811 | { |
ff57ab21 | 1812 | unregister_shrinker(&btp->bt_shrinker); |
9c7504aa BF |
1813 | ASSERT(percpu_counter_sum(&btp->bt_io_count) == 0); |
1814 | percpu_counter_destroy(&btp->bt_io_count); | |
f5e1dd34 | 1815 | list_lru_destroy(&btp->bt_lru); |
ff57ab21 | 1816 | |
2291dab2 | 1817 | xfs_blkdev_issue_flush(btp); |
a6867a68 | 1818 | |
f0e2d93c | 1819 | kmem_free(btp); |
1da177e4 LT |
1820 | } |
1821 | ||
3fefdeee ES |
1822 | int |
1823 | xfs_setsize_buftarg( | |
1da177e4 | 1824 | xfs_buftarg_t *btp, |
3fefdeee | 1825 | unsigned int sectorsize) |
1da177e4 | 1826 | { |
7c71ee78 | 1827 | /* Set up metadata sector size info */ |
6da54179 ES |
1828 | btp->bt_meta_sectorsize = sectorsize; |
1829 | btp->bt_meta_sectormask = sectorsize - 1; | |
1da177e4 | 1830 | |
ce8e922c | 1831 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
4f10700a | 1832 | xfs_warn(btp->bt_mount, |
a1c6f057 DM |
1833 | "Cannot set_blocksize to %u on device %pg", |
1834 | sectorsize, btp->bt_bdev); | |
2451337d | 1835 | return -EINVAL; |
1da177e4 LT |
1836 | } |
1837 | ||
7c71ee78 ES |
1838 | /* Set up device logical sector size mask */ |
1839 | btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev); | |
1840 | btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1; | |
1841 | ||
1da177e4 LT |
1842 | return 0; |
1843 | } | |
1844 | ||
1845 | /* | |
3fefdeee ES |
1846 | * When allocating the initial buffer target we have not yet |
1847 | * read in the superblock, so don't know what sized sectors | |
1848 | * are being used at this early stage. Play safe. | |
ce8e922c | 1849 | */ |
1da177e4 LT |
1850 | STATIC int |
1851 | xfs_setsize_buftarg_early( | |
1852 | xfs_buftarg_t *btp, | |
1853 | struct block_device *bdev) | |
1854 | { | |
a96c4151 | 1855 | return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev)); |
1da177e4 LT |
1856 | } |
1857 | ||
1da177e4 LT |
1858 | xfs_buftarg_t * |
1859 | xfs_alloc_buftarg( | |
ebad861b | 1860 | struct xfs_mount *mp, |
486aff5e DW |
1861 | struct block_device *bdev, |
1862 | struct dax_device *dax_dev) | |
1da177e4 LT |
1863 | { |
1864 | xfs_buftarg_t *btp; | |
1865 | ||
b17cb364 | 1866 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP | KM_NOFS); |
1da177e4 | 1867 | |
ebad861b | 1868 | btp->bt_mount = mp; |
ce8e922c NS |
1869 | btp->bt_dev = bdev->bd_dev; |
1870 | btp->bt_bdev = bdev; | |
486aff5e | 1871 | btp->bt_daxdev = dax_dev; |
0e6e847f | 1872 | |
1da177e4 | 1873 | if (xfs_setsize_buftarg_early(btp, bdev)) |
d210a987 | 1874 | goto error_free; |
5ca302c8 GC |
1875 | |
1876 | if (list_lru_init(&btp->bt_lru)) | |
d210a987 | 1877 | goto error_free; |
5ca302c8 | 1878 | |
9c7504aa | 1879 | if (percpu_counter_init(&btp->bt_io_count, 0, GFP_KERNEL)) |
d210a987 | 1880 | goto error_lru; |
9c7504aa | 1881 | |
e80dfa19 DC |
1882 | btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count; |
1883 | btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan; | |
ff57ab21 | 1884 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; |
e80dfa19 | 1885 | btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE; |
d210a987 MH |
1886 | if (register_shrinker(&btp->bt_shrinker)) |
1887 | goto error_pcpu; | |
1da177e4 LT |
1888 | return btp; |
1889 | ||
d210a987 MH |
1890 | error_pcpu: |
1891 | percpu_counter_destroy(&btp->bt_io_count); | |
1892 | error_lru: | |
1893 | list_lru_destroy(&btp->bt_lru); | |
1894 | error_free: | |
f0e2d93c | 1895 | kmem_free(btp); |
1da177e4 LT |
1896 | return NULL; |
1897 | } | |
1898 | ||
20e8a063 BF |
1899 | /* |
1900 | * Cancel a delayed write list. | |
1901 | * | |
1902 | * Remove each buffer from the list, clear the delwri queue flag and drop the | |
1903 | * associated buffer reference. | |
1904 | */ | |
1905 | void | |
1906 | xfs_buf_delwri_cancel( | |
1907 | struct list_head *list) | |
1908 | { | |
1909 | struct xfs_buf *bp; | |
1910 | ||
1911 | while (!list_empty(list)) { | |
1912 | bp = list_first_entry(list, struct xfs_buf, b_list); | |
1913 | ||
1914 | xfs_buf_lock(bp); | |
1915 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
1916 | list_del_init(&bp->b_list); | |
1917 | xfs_buf_relse(bp); | |
1918 | } | |
1919 | } | |
1920 | ||
1da177e4 | 1921 | /* |
43ff2122 CH |
1922 | * Add a buffer to the delayed write list. |
1923 | * | |
1924 | * This queues a buffer for writeout if it hasn't already been. Note that | |
1925 | * neither this routine nor the buffer list submission functions perform | |
1926 | * any internal synchronization. It is expected that the lists are thread-local | |
1927 | * to the callers. | |
1928 | * | |
1929 | * Returns true if we queued up the buffer, or false if it already had | |
1930 | * been on the buffer list. | |
1da177e4 | 1931 | */ |
43ff2122 | 1932 | bool |
ce8e922c | 1933 | xfs_buf_delwri_queue( |
43ff2122 CH |
1934 | struct xfs_buf *bp, |
1935 | struct list_head *list) | |
1da177e4 | 1936 | { |
43ff2122 | 1937 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 1938 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1939 | |
43ff2122 CH |
1940 | /* |
1941 | * If the buffer is already marked delwri it already is queued up | |
1942 | * by someone else for imediate writeout. Just ignore it in that | |
1943 | * case. | |
1944 | */ | |
1945 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
1946 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
1947 | return false; | |
1da177e4 | 1948 | } |
1da177e4 | 1949 | |
43ff2122 | 1950 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
1951 | |
1952 | /* | |
43ff2122 CH |
1953 | * If a buffer gets written out synchronously or marked stale while it |
1954 | * is on a delwri list we lazily remove it. To do this, the other party | |
1955 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
1956 | * It remains referenced and on the list. In a rare corner case it | |
1957 | * might get readded to a delwri list after the synchronous writeout, in | |
1958 | * which case we need just need to re-add the flag here. | |
d808f617 | 1959 | */ |
43ff2122 CH |
1960 | bp->b_flags |= _XBF_DELWRI_Q; |
1961 | if (list_empty(&bp->b_list)) { | |
1962 | atomic_inc(&bp->b_hold); | |
1963 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 1964 | } |
585e6d88 | 1965 | |
43ff2122 | 1966 | return true; |
585e6d88 DC |
1967 | } |
1968 | ||
089716aa DC |
1969 | /* |
1970 | * Compare function is more complex than it needs to be because | |
1971 | * the return value is only 32 bits and we are doing comparisons | |
1972 | * on 64 bit values | |
1973 | */ | |
1974 | static int | |
1975 | xfs_buf_cmp( | |
1976 | void *priv, | |
1977 | struct list_head *a, | |
1978 | struct list_head *b) | |
1979 | { | |
1980 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1981 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1982 | xfs_daddr_t diff; | |
1983 | ||
f4b42421 | 1984 | diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn; |
089716aa DC |
1985 | if (diff < 0) |
1986 | return -1; | |
1987 | if (diff > 0) | |
1988 | return 1; | |
1989 | return 0; | |
1990 | } | |
1991 | ||
26f1fe85 | 1992 | /* |
e339dd8d BF |
1993 | * Submit buffers for write. If wait_list is specified, the buffers are |
1994 | * submitted using sync I/O and placed on the wait list such that the caller can | |
1995 | * iowait each buffer. Otherwise async I/O is used and the buffers are released | |
1996 | * at I/O completion time. In either case, buffers remain locked until I/O | |
1997 | * completes and the buffer is released from the queue. | |
26f1fe85 | 1998 | */ |
43ff2122 | 1999 | static int |
26f1fe85 | 2000 | xfs_buf_delwri_submit_buffers( |
43ff2122 | 2001 | struct list_head *buffer_list, |
26f1fe85 | 2002 | struct list_head *wait_list) |
1da177e4 | 2003 | { |
43ff2122 CH |
2004 | struct xfs_buf *bp, *n; |
2005 | int pinned = 0; | |
26f1fe85 | 2006 | struct blk_plug plug; |
43ff2122 | 2007 | |
26f1fe85 | 2008 | list_sort(NULL, buffer_list, xfs_buf_cmp); |
43ff2122 | 2009 | |
26f1fe85 | 2010 | blk_start_plug(&plug); |
43ff2122 | 2011 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { |
26f1fe85 | 2012 | if (!wait_list) { |
43ff2122 CH |
2013 | if (xfs_buf_ispinned(bp)) { |
2014 | pinned++; | |
2015 | continue; | |
2016 | } | |
2017 | if (!xfs_buf_trylock(bp)) | |
2018 | continue; | |
2019 | } else { | |
2020 | xfs_buf_lock(bp); | |
2021 | } | |
978c7b2f | 2022 | |
43ff2122 CH |
2023 | /* |
2024 | * Someone else might have written the buffer synchronously or | |
2025 | * marked it stale in the meantime. In that case only the | |
2026 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
2027 | * reference and remove it from the list here. | |
2028 | */ | |
2029 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
2030 | list_del_init(&bp->b_list); | |
2031 | xfs_buf_relse(bp); | |
2032 | continue; | |
2033 | } | |
c9c12971 | 2034 | |
43ff2122 | 2035 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
a1b7ea5d | 2036 | |
cf53e99d | 2037 | /* |
e339dd8d BF |
2038 | * If we have a wait list, each buffer (and associated delwri |
2039 | * queue reference) transfers to it and is submitted | |
2040 | * synchronously. Otherwise, drop the buffer from the delwri | |
2041 | * queue and submit async. | |
cf53e99d | 2042 | */ |
bbfeb614 | 2043 | bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_WRITE_FAIL); |
e339dd8d | 2044 | bp->b_flags |= XBF_WRITE; |
26f1fe85 | 2045 | if (wait_list) { |
e339dd8d | 2046 | bp->b_flags &= ~XBF_ASYNC; |
26f1fe85 | 2047 | list_move_tail(&bp->b_list, wait_list); |
e339dd8d BF |
2048 | } else { |
2049 | bp->b_flags |= XBF_ASYNC; | |
ce8e922c | 2050 | list_del_init(&bp->b_list); |
e339dd8d | 2051 | } |
6af88cda | 2052 | __xfs_buf_submit(bp, false); |
43ff2122 CH |
2053 | } |
2054 | blk_finish_plug(&plug); | |
1da177e4 | 2055 | |
43ff2122 | 2056 | return pinned; |
1da177e4 LT |
2057 | } |
2058 | ||
2059 | /* | |
43ff2122 CH |
2060 | * Write out a buffer list asynchronously. |
2061 | * | |
2062 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
2063 | * out and not wait for I/O completion on any of the buffers. This interface | |
2064 | * is only safely useable for callers that can track I/O completion by higher | |
2065 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
2066 | * function. | |
efc3289c BF |
2067 | * |
2068 | * Note: this function will skip buffers it would block on, and in doing so | |
2069 | * leaves them on @buffer_list so they can be retried on a later pass. As such, | |
2070 | * it is up to the caller to ensure that the buffer list is fully submitted or | |
2071 | * cancelled appropriately when they are finished with the list. Failure to | |
2072 | * cancel or resubmit the list until it is empty will result in leaked buffers | |
2073 | * at unmount time. | |
1da177e4 LT |
2074 | */ |
2075 | int | |
43ff2122 CH |
2076 | xfs_buf_delwri_submit_nowait( |
2077 | struct list_head *buffer_list) | |
1da177e4 | 2078 | { |
26f1fe85 | 2079 | return xfs_buf_delwri_submit_buffers(buffer_list, NULL); |
43ff2122 | 2080 | } |
1da177e4 | 2081 | |
43ff2122 CH |
2082 | /* |
2083 | * Write out a buffer list synchronously. | |
2084 | * | |
2085 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
2086 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
2087 | * so callers must have some other way of tracking buffers if they require such | |
2088 | * functionality. | |
2089 | */ | |
2090 | int | |
2091 | xfs_buf_delwri_submit( | |
2092 | struct list_head *buffer_list) | |
2093 | { | |
26f1fe85 | 2094 | LIST_HEAD (wait_list); |
43ff2122 CH |
2095 | int error = 0, error2; |
2096 | struct xfs_buf *bp; | |
1da177e4 | 2097 | |
26f1fe85 | 2098 | xfs_buf_delwri_submit_buffers(buffer_list, &wait_list); |
1da177e4 | 2099 | |
43ff2122 | 2100 | /* Wait for IO to complete. */ |
26f1fe85 DC |
2101 | while (!list_empty(&wait_list)) { |
2102 | bp = list_first_entry(&wait_list, struct xfs_buf, b_list); | |
a1b7ea5d | 2103 | |
089716aa | 2104 | list_del_init(&bp->b_list); |
cf53e99d | 2105 | |
e339dd8d BF |
2106 | /* |
2107 | * Wait on the locked buffer, check for errors and unlock and | |
2108 | * release the delwri queue reference. | |
2109 | */ | |
2110 | error2 = xfs_buf_iowait(bp); | |
43ff2122 CH |
2111 | xfs_buf_relse(bp); |
2112 | if (!error) | |
2113 | error = error2; | |
1da177e4 LT |
2114 | } |
2115 | ||
43ff2122 | 2116 | return error; |
1da177e4 LT |
2117 | } |
2118 | ||
7912e7fe BF |
2119 | /* |
2120 | * Push a single buffer on a delwri queue. | |
2121 | * | |
2122 | * The purpose of this function is to submit a single buffer of a delwri queue | |
2123 | * and return with the buffer still on the original queue. The waiting delwri | |
2124 | * buffer submission infrastructure guarantees transfer of the delwri queue | |
2125 | * buffer reference to a temporary wait list. We reuse this infrastructure to | |
2126 | * transfer the buffer back to the original queue. | |
2127 | * | |
2128 | * Note the buffer transitions from the queued state, to the submitted and wait | |
2129 | * listed state and back to the queued state during this call. The buffer | |
2130 | * locking and queue management logic between _delwri_pushbuf() and | |
2131 | * _delwri_queue() guarantee that the buffer cannot be queued to another list | |
2132 | * before returning. | |
2133 | */ | |
2134 | int | |
2135 | xfs_buf_delwri_pushbuf( | |
2136 | struct xfs_buf *bp, | |
2137 | struct list_head *buffer_list) | |
2138 | { | |
2139 | LIST_HEAD (submit_list); | |
2140 | int error; | |
2141 | ||
2142 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
2143 | ||
2144 | trace_xfs_buf_delwri_pushbuf(bp, _RET_IP_); | |
2145 | ||
2146 | /* | |
2147 | * Isolate the buffer to a new local list so we can submit it for I/O | |
2148 | * independently from the rest of the original list. | |
2149 | */ | |
2150 | xfs_buf_lock(bp); | |
2151 | list_move(&bp->b_list, &submit_list); | |
2152 | xfs_buf_unlock(bp); | |
2153 | ||
2154 | /* | |
2155 | * Delwri submission clears the DELWRI_Q buffer flag and returns with | |
e339dd8d | 2156 | * the buffer on the wait list with the original reference. Rather than |
7912e7fe BF |
2157 | * bounce the buffer from a local wait list back to the original list |
2158 | * after I/O completion, reuse the original list as the wait list. | |
2159 | */ | |
2160 | xfs_buf_delwri_submit_buffers(&submit_list, buffer_list); | |
2161 | ||
2162 | /* | |
e339dd8d BF |
2163 | * The buffer is now locked, under I/O and wait listed on the original |
2164 | * delwri queue. Wait for I/O completion, restore the DELWRI_Q flag and | |
2165 | * return with the buffer unlocked and on the original queue. | |
7912e7fe | 2166 | */ |
e339dd8d | 2167 | error = xfs_buf_iowait(bp); |
7912e7fe BF |
2168 | bp->b_flags |= _XBF_DELWRI_Q; |
2169 | xfs_buf_unlock(bp); | |
2170 | ||
2171 | return error; | |
2172 | } | |
2173 | ||
04d8b284 | 2174 | int __init |
ce8e922c | 2175 | xfs_buf_init(void) |
1da177e4 | 2176 | { |
8758280f NS |
2177 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
2178 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 2179 | if (!xfs_buf_zone) |
0b1b213f | 2180 | goto out; |
04d8b284 | 2181 | |
23ea4032 | 2182 | return 0; |
1da177e4 | 2183 | |
0b1b213f | 2184 | out: |
8758280f | 2185 | return -ENOMEM; |
1da177e4 LT |
2186 | } |
2187 | ||
1da177e4 | 2188 | void |
ce8e922c | 2189 | xfs_buf_terminate(void) |
1da177e4 | 2190 | { |
ce8e922c | 2191 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 2192 | } |
7561d27e BF |
2193 | |
2194 | void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref) | |
2195 | { | |
7561d27e BF |
2196 | /* |
2197 | * Set the lru reference count to 0 based on the error injection tag. | |
2198 | * This allows userspace to disrupt buffer caching for debug/testing | |
2199 | * purposes. | |
2200 | */ | |
4eadcf9a BF |
2201 | if (XFS_TEST_ERROR(false, bp->b_target->bt_mount, |
2202 | XFS_ERRTAG_BUF_LRU_REF)) | |
7561d27e BF |
2203 | lru_ref = 0; |
2204 | ||
2205 | atomic_set(&bp->b_lru_ref, lru_ref); | |
2206 | } | |
8473fee3 BF |
2207 | |
2208 | /* | |
2209 | * Verify an on-disk magic value against the magic value specified in the | |
2210 | * verifier structure. The verifier magic is in disk byte order so the caller is | |
2211 | * expected to pass the value directly from disk. | |
2212 | */ | |
2213 | bool | |
2214 | xfs_verify_magic( | |
2215 | struct xfs_buf *bp, | |
15baadf7 | 2216 | __be32 dmagic) |
8473fee3 BF |
2217 | { |
2218 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
2219 | int idx; | |
2220 | ||
2221 | idx = xfs_sb_version_hascrc(&mp->m_sb); | |
2222 | if (unlikely(WARN_ON(!bp->b_ops || !bp->b_ops->magic[idx]))) | |
2223 | return false; | |
2224 | return dmagic == bp->b_ops->magic[idx]; | |
2225 | } | |
15baadf7 DW |
2226 | /* |
2227 | * Verify an on-disk magic value against the magic value specified in the | |
2228 | * verifier structure. The verifier magic is in disk byte order so the caller is | |
2229 | * expected to pass the value directly from disk. | |
2230 | */ | |
2231 | bool | |
2232 | xfs_verify_magic16( | |
2233 | struct xfs_buf *bp, | |
2234 | __be16 dmagic) | |
2235 | { | |
2236 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
2237 | int idx; | |
2238 | ||
2239 | idx = xfs_sb_version_hascrc(&mp->m_sb); | |
2240 | if (unlikely(WARN_ON(!bp->b_ops || !bp->b_ops->magic16[idx]))) | |
2241 | return false; | |
2242 | return dmagic == bp->b_ops->magic16[idx]; | |
2243 | } |