]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
f07c2250 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 3 | * All Rights Reserved. |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
93c189c1 | 18 | #include "xfs.h" |
1da177e4 LT |
19 | #include <linux/stddef.h> |
20 | #include <linux/errno.h> | |
5a0e3ad6 | 21 | #include <linux/gfp.h> |
1da177e4 LT |
22 | #include <linux/pagemap.h> |
23 | #include <linux/init.h> | |
24 | #include <linux/vmalloc.h> | |
25 | #include <linux/bio.h> | |
26 | #include <linux/sysctl.h> | |
27 | #include <linux/proc_fs.h> | |
28 | #include <linux/workqueue.h> | |
29 | #include <linux/percpu.h> | |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/hash.h> | |
4df08c52 | 32 | #include <linux/kthread.h> |
b20a3503 | 33 | #include <linux/migrate.h> |
3fcfab16 | 34 | #include <linux/backing-dev.h> |
7dfb7103 | 35 | #include <linux/freezer.h> |
1da177e4 | 36 | |
4fb6e8ad | 37 | #include "xfs_format.h" |
239880ef | 38 | #include "xfs_log_format.h" |
7fd36c44 | 39 | #include "xfs_trans_resv.h" |
239880ef | 40 | #include "xfs_sb.h" |
b7963133 | 41 | #include "xfs_mount.h" |
0b1b213f | 42 | #include "xfs_trace.h" |
239880ef | 43 | #include "xfs_log.h" |
b7963133 | 44 | |
7989cb8e | 45 | static kmem_zone_t *xfs_buf_zone; |
23ea4032 | 46 | |
ce8e922c NS |
47 | #ifdef XFS_BUF_LOCK_TRACKING |
48 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
49 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
50 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 51 | #else |
ce8e922c NS |
52 | # define XB_SET_OWNER(bp) do { } while (0) |
53 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
54 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
55 | #endif |
56 | ||
ce8e922c | 57 | #define xb_to_gfp(flags) \ |
aa5c158e | 58 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN) |
1da177e4 | 59 | |
1da177e4 | 60 | |
73c77e2c JB |
61 | static inline int |
62 | xfs_buf_is_vmapped( | |
63 | struct xfs_buf *bp) | |
64 | { | |
65 | /* | |
66 | * Return true if the buffer is vmapped. | |
67 | * | |
611c9946 DC |
68 | * b_addr is null if the buffer is not mapped, but the code is clever |
69 | * enough to know it doesn't have to map a single page, so the check has | |
70 | * to be both for b_addr and bp->b_page_count > 1. | |
73c77e2c | 71 | */ |
611c9946 | 72 | return bp->b_addr && bp->b_page_count > 1; |
73c77e2c JB |
73 | } |
74 | ||
75 | static inline int | |
76 | xfs_buf_vmap_len( | |
77 | struct xfs_buf *bp) | |
78 | { | |
79 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
80 | } | |
81 | ||
9c7504aa BF |
82 | /* |
83 | * Bump the I/O in flight count on the buftarg if we haven't yet done so for | |
84 | * this buffer. The count is incremented once per buffer (per hold cycle) | |
85 | * because the corresponding decrement is deferred to buffer release. Buffers | |
86 | * can undergo I/O multiple times in a hold-release cycle and per buffer I/O | |
87 | * tracking adds unnecessary overhead. This is used for sychronization purposes | |
88 | * with unmount (see xfs_wait_buftarg()), so all we really need is a count of | |
89 | * in-flight buffers. | |
90 | * | |
91 | * Buffers that are never released (e.g., superblock, iclog buffers) must set | |
92 | * the XBF_NO_IOACCT flag before I/O submission. Otherwise, the buftarg count | |
93 | * never reaches zero and unmount hangs indefinitely. | |
94 | */ | |
95 | static inline void | |
96 | xfs_buf_ioacct_inc( | |
97 | struct xfs_buf *bp) | |
98 | { | |
99 | if (bp->b_flags & (XBF_NO_IOACCT|_XBF_IN_FLIGHT)) | |
100 | return; | |
101 | ||
102 | ASSERT(bp->b_flags & XBF_ASYNC); | |
103 | bp->b_flags |= _XBF_IN_FLIGHT; | |
104 | percpu_counter_inc(&bp->b_target->bt_io_count); | |
105 | } | |
106 | ||
107 | /* | |
108 | * Clear the in-flight state on a buffer about to be released to the LRU or | |
109 | * freed and unaccount from the buftarg. | |
110 | */ | |
111 | static inline void | |
112 | xfs_buf_ioacct_dec( | |
113 | struct xfs_buf *bp) | |
114 | { | |
115 | if (!(bp->b_flags & _XBF_IN_FLIGHT)) | |
116 | return; | |
117 | ||
9c7504aa BF |
118 | bp->b_flags &= ~_XBF_IN_FLIGHT; |
119 | percpu_counter_dec(&bp->b_target->bt_io_count); | |
120 | } | |
121 | ||
430cbeb8 DC |
122 | /* |
123 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
124 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
125 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
126 | * to remove the reference that LRU holds on the buffer. | |
127 | * | |
128 | * This prevents build-up of stale buffers on the LRU. | |
129 | */ | |
130 | void | |
131 | xfs_buf_stale( | |
132 | struct xfs_buf *bp) | |
133 | { | |
43ff2122 CH |
134 | ASSERT(xfs_buf_islocked(bp)); |
135 | ||
430cbeb8 | 136 | bp->b_flags |= XBF_STALE; |
43ff2122 CH |
137 | |
138 | /* | |
139 | * Clear the delwri status so that a delwri queue walker will not | |
140 | * flush this buffer to disk now that it is stale. The delwri queue has | |
141 | * a reference to the buffer, so this is safe to do. | |
142 | */ | |
143 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
144 | ||
9c7504aa BF |
145 | /* |
146 | * Once the buffer is marked stale and unlocked, a subsequent lookup | |
147 | * could reset b_flags. There is no guarantee that the buffer is | |
148 | * unaccounted (released to LRU) before that occurs. Drop in-flight | |
149 | * status now to preserve accounting consistency. | |
150 | */ | |
151 | xfs_buf_ioacct_dec(bp); | |
152 | ||
a4082357 DC |
153 | spin_lock(&bp->b_lock); |
154 | atomic_set(&bp->b_lru_ref, 0); | |
155 | if (!(bp->b_state & XFS_BSTATE_DISPOSE) && | |
e80dfa19 DC |
156 | (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru))) |
157 | atomic_dec(&bp->b_hold); | |
158 | ||
430cbeb8 | 159 | ASSERT(atomic_read(&bp->b_hold) >= 1); |
a4082357 | 160 | spin_unlock(&bp->b_lock); |
430cbeb8 | 161 | } |
1da177e4 | 162 | |
3e85c868 DC |
163 | static int |
164 | xfs_buf_get_maps( | |
165 | struct xfs_buf *bp, | |
166 | int map_count) | |
167 | { | |
168 | ASSERT(bp->b_maps == NULL); | |
169 | bp->b_map_count = map_count; | |
170 | ||
171 | if (map_count == 1) { | |
f4b42421 | 172 | bp->b_maps = &bp->__b_map; |
3e85c868 DC |
173 | return 0; |
174 | } | |
175 | ||
176 | bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map), | |
177 | KM_NOFS); | |
178 | if (!bp->b_maps) | |
2451337d | 179 | return -ENOMEM; |
3e85c868 DC |
180 | return 0; |
181 | } | |
182 | ||
183 | /* | |
184 | * Frees b_pages if it was allocated. | |
185 | */ | |
186 | static void | |
187 | xfs_buf_free_maps( | |
188 | struct xfs_buf *bp) | |
189 | { | |
f4b42421 | 190 | if (bp->b_maps != &bp->__b_map) { |
3e85c868 DC |
191 | kmem_free(bp->b_maps); |
192 | bp->b_maps = NULL; | |
193 | } | |
194 | } | |
195 | ||
4347b9d7 | 196 | struct xfs_buf * |
3e85c868 | 197 | _xfs_buf_alloc( |
4347b9d7 | 198 | struct xfs_buftarg *target, |
3e85c868 DC |
199 | struct xfs_buf_map *map, |
200 | int nmaps, | |
ce8e922c | 201 | xfs_buf_flags_t flags) |
1da177e4 | 202 | { |
4347b9d7 | 203 | struct xfs_buf *bp; |
3e85c868 DC |
204 | int error; |
205 | int i; | |
4347b9d7 | 206 | |
aa5c158e | 207 | bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS); |
4347b9d7 CH |
208 | if (unlikely(!bp)) |
209 | return NULL; | |
210 | ||
1da177e4 | 211 | /* |
12bcb3f7 DC |
212 | * We don't want certain flags to appear in b_flags unless they are |
213 | * specifically set by later operations on the buffer. | |
1da177e4 | 214 | */ |
611c9946 | 215 | flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD); |
ce8e922c | 216 | |
ce8e922c | 217 | atomic_set(&bp->b_hold, 1); |
430cbeb8 | 218 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 219 | init_completion(&bp->b_iowait); |
430cbeb8 | 220 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 221 | INIT_LIST_HEAD(&bp->b_list); |
a731cd11 | 222 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
a4082357 | 223 | spin_lock_init(&bp->b_lock); |
ce8e922c NS |
224 | XB_SET_OWNER(bp); |
225 | bp->b_target = target; | |
3e85c868 | 226 | bp->b_flags = flags; |
de1cbee4 | 227 | |
1da177e4 | 228 | /* |
aa0e8833 DC |
229 | * Set length and io_length to the same value initially. |
230 | * I/O routines should use io_length, which will be the same in | |
1da177e4 LT |
231 | * most cases but may be reset (e.g. XFS recovery). |
232 | */ | |
3e85c868 DC |
233 | error = xfs_buf_get_maps(bp, nmaps); |
234 | if (error) { | |
235 | kmem_zone_free(xfs_buf_zone, bp); | |
236 | return NULL; | |
237 | } | |
238 | ||
239 | bp->b_bn = map[0].bm_bn; | |
240 | bp->b_length = 0; | |
241 | for (i = 0; i < nmaps; i++) { | |
242 | bp->b_maps[i].bm_bn = map[i].bm_bn; | |
243 | bp->b_maps[i].bm_len = map[i].bm_len; | |
244 | bp->b_length += map[i].bm_len; | |
245 | } | |
246 | bp->b_io_length = bp->b_length; | |
247 | ||
ce8e922c NS |
248 | atomic_set(&bp->b_pin_count, 0); |
249 | init_waitqueue_head(&bp->b_waiters); | |
250 | ||
ff6d6af2 | 251 | XFS_STATS_INC(target->bt_mount, xb_create); |
0b1b213f | 252 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 CH |
253 | |
254 | return bp; | |
1da177e4 LT |
255 | } |
256 | ||
257 | /* | |
ce8e922c NS |
258 | * Allocate a page array capable of holding a specified number |
259 | * of pages, and point the page buf at it. | |
1da177e4 LT |
260 | */ |
261 | STATIC int | |
ce8e922c NS |
262 | _xfs_buf_get_pages( |
263 | xfs_buf_t *bp, | |
87937bf8 | 264 | int page_count) |
1da177e4 LT |
265 | { |
266 | /* Make sure that we have a page list */ | |
ce8e922c | 267 | if (bp->b_pages == NULL) { |
ce8e922c NS |
268 | bp->b_page_count = page_count; |
269 | if (page_count <= XB_PAGES) { | |
270 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 271 | } else { |
ce8e922c | 272 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
aa5c158e | 273 | page_count, KM_NOFS); |
ce8e922c | 274 | if (bp->b_pages == NULL) |
1da177e4 LT |
275 | return -ENOMEM; |
276 | } | |
ce8e922c | 277 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
278 | } |
279 | return 0; | |
280 | } | |
281 | ||
282 | /* | |
ce8e922c | 283 | * Frees b_pages if it was allocated. |
1da177e4 LT |
284 | */ |
285 | STATIC void | |
ce8e922c | 286 | _xfs_buf_free_pages( |
1da177e4 LT |
287 | xfs_buf_t *bp) |
288 | { | |
ce8e922c | 289 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 290 | kmem_free(bp->b_pages); |
3fc98b1a | 291 | bp->b_pages = NULL; |
1da177e4 LT |
292 | } |
293 | } | |
294 | ||
295 | /* | |
296 | * Releases the specified buffer. | |
297 | * | |
298 | * The modification state of any associated pages is left unchanged. | |
b46fe825 | 299 | * The buffer must not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
300 | * hashed and refcounted buffers |
301 | */ | |
302 | void | |
ce8e922c | 303 | xfs_buf_free( |
1da177e4 LT |
304 | xfs_buf_t *bp) |
305 | { | |
0b1b213f | 306 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 307 | |
430cbeb8 DC |
308 | ASSERT(list_empty(&bp->b_lru)); |
309 | ||
0e6e847f | 310 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
311 | uint i; |
312 | ||
73c77e2c | 313 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
314 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
315 | bp->b_page_count); | |
1da177e4 | 316 | |
948ecdb4 NS |
317 | for (i = 0; i < bp->b_page_count; i++) { |
318 | struct page *page = bp->b_pages[i]; | |
319 | ||
0e6e847f | 320 | __free_page(page); |
948ecdb4 | 321 | } |
0e6e847f DC |
322 | } else if (bp->b_flags & _XBF_KMEM) |
323 | kmem_free(bp->b_addr); | |
3fc98b1a | 324 | _xfs_buf_free_pages(bp); |
3e85c868 | 325 | xfs_buf_free_maps(bp); |
4347b9d7 | 326 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
327 | } |
328 | ||
329 | /* | |
0e6e847f | 330 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
331 | */ |
332 | STATIC int | |
0e6e847f | 333 | xfs_buf_allocate_memory( |
1da177e4 LT |
334 | xfs_buf_t *bp, |
335 | uint flags) | |
336 | { | |
aa0e8833 | 337 | size_t size; |
1da177e4 | 338 | size_t nbytes, offset; |
ce8e922c | 339 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 340 | unsigned short page_count, i; |
795cac72 | 341 | xfs_off_t start, end; |
1da177e4 LT |
342 | int error; |
343 | ||
0e6e847f DC |
344 | /* |
345 | * for buffers that are contained within a single page, just allocate | |
346 | * the memory from the heap - there's no need for the complexity of | |
347 | * page arrays to keep allocation down to order 0. | |
348 | */ | |
795cac72 DC |
349 | size = BBTOB(bp->b_length); |
350 | if (size < PAGE_SIZE) { | |
aa5c158e | 351 | bp->b_addr = kmem_alloc(size, KM_NOFS); |
0e6e847f DC |
352 | if (!bp->b_addr) { |
353 | /* low memory - use alloc_page loop instead */ | |
354 | goto use_alloc_page; | |
355 | } | |
356 | ||
795cac72 | 357 | if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) != |
0e6e847f DC |
358 | ((unsigned long)bp->b_addr & PAGE_MASK)) { |
359 | /* b_addr spans two pages - use alloc_page instead */ | |
360 | kmem_free(bp->b_addr); | |
361 | bp->b_addr = NULL; | |
362 | goto use_alloc_page; | |
363 | } | |
364 | bp->b_offset = offset_in_page(bp->b_addr); | |
365 | bp->b_pages = bp->b_page_array; | |
366 | bp->b_pages[0] = virt_to_page(bp->b_addr); | |
367 | bp->b_page_count = 1; | |
611c9946 | 368 | bp->b_flags |= _XBF_KMEM; |
0e6e847f DC |
369 | return 0; |
370 | } | |
371 | ||
372 | use_alloc_page: | |
f4b42421 MT |
373 | start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT; |
374 | end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1) | |
cbb7baab | 375 | >> PAGE_SHIFT; |
795cac72 | 376 | page_count = end - start; |
87937bf8 | 377 | error = _xfs_buf_get_pages(bp, page_count); |
1da177e4 LT |
378 | if (unlikely(error)) |
379 | return error; | |
1da177e4 | 380 | |
ce8e922c | 381 | offset = bp->b_offset; |
0e6e847f | 382 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 383 | |
ce8e922c | 384 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
385 | struct page *page; |
386 | uint retries = 0; | |
0e6e847f DC |
387 | retry: |
388 | page = alloc_page(gfp_mask); | |
1da177e4 | 389 | if (unlikely(page == NULL)) { |
ce8e922c NS |
390 | if (flags & XBF_READ_AHEAD) { |
391 | bp->b_page_count = i; | |
2451337d | 392 | error = -ENOMEM; |
0e6e847f | 393 | goto out_free_pages; |
1da177e4 LT |
394 | } |
395 | ||
396 | /* | |
397 | * This could deadlock. | |
398 | * | |
399 | * But until all the XFS lowlevel code is revamped to | |
400 | * handle buffer allocation failures we can't do much. | |
401 | */ | |
402 | if (!(++retries % 100)) | |
4f10700a | 403 | xfs_err(NULL, |
5bf97b1c TH |
404 | "%s(%u) possible memory allocation deadlock in %s (mode:0x%x)", |
405 | current->comm, current->pid, | |
34a622b2 | 406 | __func__, gfp_mask); |
1da177e4 | 407 | |
ff6d6af2 | 408 | XFS_STATS_INC(bp->b_target->bt_mount, xb_page_retries); |
8aa7e847 | 409 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
410 | goto retry; |
411 | } | |
412 | ||
ff6d6af2 | 413 | XFS_STATS_INC(bp->b_target->bt_mount, xb_page_found); |
1da177e4 | 414 | |
0e6e847f | 415 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 416 | size -= nbytes; |
ce8e922c | 417 | bp->b_pages[i] = page; |
1da177e4 LT |
418 | offset = 0; |
419 | } | |
0e6e847f | 420 | return 0; |
1da177e4 | 421 | |
0e6e847f DC |
422 | out_free_pages: |
423 | for (i = 0; i < bp->b_page_count; i++) | |
424 | __free_page(bp->b_pages[i]); | |
2aa6ba7b | 425 | bp->b_flags &= ~_XBF_PAGES; |
1da177e4 LT |
426 | return error; |
427 | } | |
428 | ||
429 | /* | |
25985edc | 430 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
431 | */ |
432 | STATIC int | |
ce8e922c | 433 | _xfs_buf_map_pages( |
1da177e4 LT |
434 | xfs_buf_t *bp, |
435 | uint flags) | |
436 | { | |
0e6e847f | 437 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 438 | if (bp->b_page_count == 1) { |
0e6e847f | 439 | /* A single page buffer is always mappable */ |
ce8e922c | 440 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
611c9946 DC |
441 | } else if (flags & XBF_UNMAPPED) { |
442 | bp->b_addr = NULL; | |
443 | } else { | |
a19fb380 | 444 | int retried = 0; |
ae687e58 DC |
445 | unsigned noio_flag; |
446 | ||
447 | /* | |
448 | * vm_map_ram() will allocate auxillary structures (e.g. | |
449 | * pagetables) with GFP_KERNEL, yet we are likely to be under | |
450 | * GFP_NOFS context here. Hence we need to tell memory reclaim | |
451 | * that we are in such a context via PF_MEMALLOC_NOIO to prevent | |
452 | * memory reclaim re-entering the filesystem here and | |
453 | * potentially deadlocking. | |
454 | */ | |
455 | noio_flag = memalloc_noio_save(); | |
a19fb380 DC |
456 | do { |
457 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
458 | -1, PAGE_KERNEL); | |
459 | if (bp->b_addr) | |
460 | break; | |
461 | vm_unmap_aliases(); | |
462 | } while (retried++ <= 1); | |
ae687e58 | 463 | memalloc_noio_restore(noio_flag); |
a19fb380 DC |
464 | |
465 | if (!bp->b_addr) | |
1da177e4 | 466 | return -ENOMEM; |
ce8e922c | 467 | bp->b_addr += bp->b_offset; |
1da177e4 LT |
468 | } |
469 | ||
470 | return 0; | |
471 | } | |
472 | ||
473 | /* | |
474 | * Finding and Reading Buffers | |
475 | */ | |
6031e73a LS |
476 | static int |
477 | _xfs_buf_obj_cmp( | |
478 | struct rhashtable_compare_arg *arg, | |
479 | const void *obj) | |
480 | { | |
481 | const struct xfs_buf_map *map = arg->key; | |
482 | const struct xfs_buf *bp = obj; | |
483 | ||
484 | /* | |
485 | * The key hashing in the lookup path depends on the key being the | |
486 | * first element of the compare_arg, make sure to assert this. | |
487 | */ | |
488 | BUILD_BUG_ON(offsetof(struct xfs_buf_map, bm_bn) != 0); | |
489 | ||
490 | if (bp->b_bn != map->bm_bn) | |
491 | return 1; | |
492 | ||
493 | if (unlikely(bp->b_length != map->bm_len)) { | |
494 | /* | |
495 | * found a block number match. If the range doesn't | |
496 | * match, the only way this is allowed is if the buffer | |
497 | * in the cache is stale and the transaction that made | |
498 | * it stale has not yet committed. i.e. we are | |
499 | * reallocating a busy extent. Skip this buffer and | |
500 | * continue searching for an exact match. | |
501 | */ | |
502 | ASSERT(bp->b_flags & XBF_STALE); | |
503 | return 1; | |
504 | } | |
505 | return 0; | |
506 | } | |
507 | ||
508 | static const struct rhashtable_params xfs_buf_hash_params = { | |
509 | .min_size = 32, /* empty AGs have minimal footprint */ | |
510 | .nelem_hint = 16, | |
511 | .key_len = sizeof(xfs_daddr_t), | |
512 | .key_offset = offsetof(struct xfs_buf, b_bn), | |
513 | .head_offset = offsetof(struct xfs_buf, b_rhash_head), | |
514 | .automatic_shrinking = true, | |
515 | .obj_cmpfn = _xfs_buf_obj_cmp, | |
516 | }; | |
517 | ||
518 | int | |
519 | xfs_buf_hash_init( | |
520 | struct xfs_perag *pag) | |
521 | { | |
522 | spin_lock_init(&pag->pag_buf_lock); | |
523 | return rhashtable_init(&pag->pag_buf_hash, &xfs_buf_hash_params); | |
524 | } | |
525 | ||
526 | void | |
527 | xfs_buf_hash_destroy( | |
528 | struct xfs_perag *pag) | |
529 | { | |
530 | rhashtable_destroy(&pag->pag_buf_hash); | |
531 | } | |
1da177e4 LT |
532 | |
533 | /* | |
ce8e922c | 534 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 | 535 | * a given range of an inode. The buffer is returned |
eabbaf11 | 536 | * locked. No I/O is implied by this call. |
1da177e4 LT |
537 | */ |
538 | xfs_buf_t * | |
ce8e922c | 539 | _xfs_buf_find( |
e70b73f8 | 540 | struct xfs_buftarg *btp, |
3e85c868 DC |
541 | struct xfs_buf_map *map, |
542 | int nmaps, | |
ce8e922c NS |
543 | xfs_buf_flags_t flags, |
544 | xfs_buf_t *new_bp) | |
1da177e4 | 545 | { |
74f75a0c | 546 | struct xfs_perag *pag; |
74f75a0c | 547 | xfs_buf_t *bp; |
6031e73a | 548 | struct xfs_buf_map cmap = { .bm_bn = map[0].bm_bn }; |
10616b80 | 549 | xfs_daddr_t eofs; |
3e85c868 | 550 | int i; |
1da177e4 | 551 | |
3e85c868 | 552 | for (i = 0; i < nmaps; i++) |
6031e73a | 553 | cmap.bm_len += map[i].bm_len; |
1da177e4 LT |
554 | |
555 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
6031e73a LS |
556 | ASSERT(!(BBTOB(cmap.bm_len) < btp->bt_meta_sectorsize)); |
557 | ASSERT(!(BBTOB(cmap.bm_bn) & (xfs_off_t)btp->bt_meta_sectormask)); | |
1da177e4 | 558 | |
10616b80 DC |
559 | /* |
560 | * Corrupted block numbers can get through to here, unfortunately, so we | |
561 | * have to check that the buffer falls within the filesystem bounds. | |
562 | */ | |
563 | eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks); | |
6031e73a | 564 | if (cmap.bm_bn < 0 || cmap.bm_bn >= eofs) { |
10616b80 | 565 | /* |
2451337d | 566 | * XXX (dgc): we should really be returning -EFSCORRUPTED here, |
10616b80 DC |
567 | * but none of the higher level infrastructure supports |
568 | * returning a specific error on buffer lookup failures. | |
569 | */ | |
570 | xfs_alert(btp->bt_mount, | |
571 | "%s: Block out of range: block 0x%llx, EOFS 0x%llx ", | |
6031e73a | 572 | __func__, cmap.bm_bn, eofs); |
7bc0dc27 | 573 | WARN_ON(1); |
10616b80 DC |
574 | return NULL; |
575 | } | |
576 | ||
74f75a0c | 577 | pag = xfs_perag_get(btp->bt_mount, |
6031e73a | 578 | xfs_daddr_to_agno(btp->bt_mount, cmap.bm_bn)); |
74f75a0c | 579 | |
74f75a0c | 580 | spin_lock(&pag->pag_buf_lock); |
6031e73a LS |
581 | bp = rhashtable_lookup_fast(&pag->pag_buf_hash, &cmap, |
582 | xfs_buf_hash_params); | |
583 | if (bp) { | |
584 | atomic_inc(&bp->b_hold); | |
585 | goto found; | |
1da177e4 LT |
586 | } |
587 | ||
588 | /* No match found */ | |
ce8e922c | 589 | if (new_bp) { |
74f75a0c DC |
590 | /* the buffer keeps the perag reference until it is freed */ |
591 | new_bp->b_pag = pag; | |
6031e73a LS |
592 | rhashtable_insert_fast(&pag->pag_buf_hash, |
593 | &new_bp->b_rhash_head, | |
594 | xfs_buf_hash_params); | |
74f75a0c | 595 | spin_unlock(&pag->pag_buf_lock); |
1da177e4 | 596 | } else { |
ff6d6af2 | 597 | XFS_STATS_INC(btp->bt_mount, xb_miss_locked); |
74f75a0c DC |
598 | spin_unlock(&pag->pag_buf_lock); |
599 | xfs_perag_put(pag); | |
1da177e4 | 600 | } |
ce8e922c | 601 | return new_bp; |
1da177e4 LT |
602 | |
603 | found: | |
74f75a0c DC |
604 | spin_unlock(&pag->pag_buf_lock); |
605 | xfs_perag_put(pag); | |
1da177e4 | 606 | |
0c842ad4 CH |
607 | if (!xfs_buf_trylock(bp)) { |
608 | if (flags & XBF_TRYLOCK) { | |
ce8e922c | 609 | xfs_buf_rele(bp); |
ff6d6af2 | 610 | XFS_STATS_INC(btp->bt_mount, xb_busy_locked); |
ce8e922c | 611 | return NULL; |
1da177e4 | 612 | } |
0c842ad4 | 613 | xfs_buf_lock(bp); |
ff6d6af2 | 614 | XFS_STATS_INC(btp->bt_mount, xb_get_locked_waited); |
1da177e4 LT |
615 | } |
616 | ||
0e6e847f DC |
617 | /* |
618 | * if the buffer is stale, clear all the external state associated with | |
619 | * it. We need to keep flags such as how we allocated the buffer memory | |
620 | * intact here. | |
621 | */ | |
ce8e922c NS |
622 | if (bp->b_flags & XBF_STALE) { |
623 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
cfb02852 | 624 | ASSERT(bp->b_iodone == NULL); |
611c9946 | 625 | bp->b_flags &= _XBF_KMEM | _XBF_PAGES; |
1813dd64 | 626 | bp->b_ops = NULL; |
2f926587 | 627 | } |
0b1b213f CH |
628 | |
629 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ff6d6af2 | 630 | XFS_STATS_INC(btp->bt_mount, xb_get_locked); |
ce8e922c | 631 | return bp; |
1da177e4 LT |
632 | } |
633 | ||
634 | /* | |
3815832a DC |
635 | * Assembles a buffer covering the specified range. The code is optimised for |
636 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
637 | * more hits than misses. | |
1da177e4 | 638 | */ |
3815832a | 639 | struct xfs_buf * |
6dde2707 DC |
640 | xfs_buf_get_map( |
641 | struct xfs_buftarg *target, | |
642 | struct xfs_buf_map *map, | |
643 | int nmaps, | |
ce8e922c | 644 | xfs_buf_flags_t flags) |
1da177e4 | 645 | { |
3815832a DC |
646 | struct xfs_buf *bp; |
647 | struct xfs_buf *new_bp; | |
0e6e847f | 648 | int error = 0; |
1da177e4 | 649 | |
6dde2707 | 650 | bp = _xfs_buf_find(target, map, nmaps, flags, NULL); |
3815832a DC |
651 | if (likely(bp)) |
652 | goto found; | |
653 | ||
6dde2707 | 654 | new_bp = _xfs_buf_alloc(target, map, nmaps, flags); |
ce8e922c | 655 | if (unlikely(!new_bp)) |
1da177e4 LT |
656 | return NULL; |
657 | ||
fe2429b0 DC |
658 | error = xfs_buf_allocate_memory(new_bp, flags); |
659 | if (error) { | |
3e85c868 | 660 | xfs_buf_free(new_bp); |
fe2429b0 DC |
661 | return NULL; |
662 | } | |
663 | ||
6dde2707 | 664 | bp = _xfs_buf_find(target, map, nmaps, flags, new_bp); |
3815832a | 665 | if (!bp) { |
fe2429b0 | 666 | xfs_buf_free(new_bp); |
3815832a DC |
667 | return NULL; |
668 | } | |
669 | ||
fe2429b0 DC |
670 | if (bp != new_bp) |
671 | xfs_buf_free(new_bp); | |
1da177e4 | 672 | |
3815832a | 673 | found: |
611c9946 | 674 | if (!bp->b_addr) { |
ce8e922c | 675 | error = _xfs_buf_map_pages(bp, flags); |
1da177e4 | 676 | if (unlikely(error)) { |
4f10700a | 677 | xfs_warn(target->bt_mount, |
08e96e1a | 678 | "%s: failed to map pagesn", __func__); |
a8acad70 DC |
679 | xfs_buf_relse(bp); |
680 | return NULL; | |
1da177e4 LT |
681 | } |
682 | } | |
683 | ||
b79f4a1c DC |
684 | /* |
685 | * Clear b_error if this is a lookup from a caller that doesn't expect | |
686 | * valid data to be found in the buffer. | |
687 | */ | |
688 | if (!(flags & XBF_READ)) | |
689 | xfs_buf_ioerror(bp, 0); | |
690 | ||
ff6d6af2 | 691 | XFS_STATS_INC(target->bt_mount, xb_get); |
0b1b213f | 692 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 693 | return bp; |
1da177e4 LT |
694 | } |
695 | ||
5d765b97 CH |
696 | STATIC int |
697 | _xfs_buf_read( | |
698 | xfs_buf_t *bp, | |
699 | xfs_buf_flags_t flags) | |
700 | { | |
43ff2122 | 701 | ASSERT(!(flags & XBF_WRITE)); |
f4b42421 | 702 | ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL); |
5d765b97 | 703 | |
43ff2122 | 704 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD); |
1d5ae5df | 705 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 706 | |
595bff75 DC |
707 | if (flags & XBF_ASYNC) { |
708 | xfs_buf_submit(bp); | |
0e95f19a | 709 | return 0; |
595bff75 DC |
710 | } |
711 | return xfs_buf_submit_wait(bp); | |
5d765b97 CH |
712 | } |
713 | ||
1da177e4 | 714 | xfs_buf_t * |
6dde2707 DC |
715 | xfs_buf_read_map( |
716 | struct xfs_buftarg *target, | |
717 | struct xfs_buf_map *map, | |
718 | int nmaps, | |
c3f8fc73 | 719 | xfs_buf_flags_t flags, |
1813dd64 | 720 | const struct xfs_buf_ops *ops) |
1da177e4 | 721 | { |
6dde2707 | 722 | struct xfs_buf *bp; |
ce8e922c NS |
723 | |
724 | flags |= XBF_READ; | |
725 | ||
6dde2707 | 726 | bp = xfs_buf_get_map(target, map, nmaps, flags); |
ce8e922c | 727 | if (bp) { |
0b1b213f CH |
728 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
729 | ||
b0388bf1 | 730 | if (!(bp->b_flags & XBF_DONE)) { |
ff6d6af2 | 731 | XFS_STATS_INC(target->bt_mount, xb_get_read); |
1813dd64 | 732 | bp->b_ops = ops; |
5d765b97 | 733 | _xfs_buf_read(bp, flags); |
ce8e922c | 734 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
735 | /* |
736 | * Read ahead call which is already satisfied, | |
737 | * drop the buffer | |
738 | */ | |
a8acad70 DC |
739 | xfs_buf_relse(bp); |
740 | return NULL; | |
1da177e4 | 741 | } else { |
1da177e4 | 742 | /* We do not want read in the flags */ |
ce8e922c | 743 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
744 | } |
745 | } | |
746 | ||
ce8e922c | 747 | return bp; |
1da177e4 LT |
748 | } |
749 | ||
1da177e4 | 750 | /* |
ce8e922c NS |
751 | * If we are not low on memory then do the readahead in a deadlock |
752 | * safe manner. | |
1da177e4 LT |
753 | */ |
754 | void | |
6dde2707 DC |
755 | xfs_buf_readahead_map( |
756 | struct xfs_buftarg *target, | |
757 | struct xfs_buf_map *map, | |
c3f8fc73 | 758 | int nmaps, |
1813dd64 | 759 | const struct xfs_buf_ops *ops) |
1da177e4 | 760 | { |
0e6e847f | 761 | if (bdi_read_congested(target->bt_bdi)) |
1da177e4 LT |
762 | return; |
763 | ||
6dde2707 | 764 | xfs_buf_read_map(target, map, nmaps, |
1813dd64 | 765 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD, ops); |
1da177e4 LT |
766 | } |
767 | ||
5adc94c2 DC |
768 | /* |
769 | * Read an uncached buffer from disk. Allocates and returns a locked | |
770 | * buffer containing the disk contents or nothing. | |
771 | */ | |
ba372674 | 772 | int |
5adc94c2 | 773 | xfs_buf_read_uncached( |
5adc94c2 DC |
774 | struct xfs_buftarg *target, |
775 | xfs_daddr_t daddr, | |
e70b73f8 | 776 | size_t numblks, |
c3f8fc73 | 777 | int flags, |
ba372674 | 778 | struct xfs_buf **bpp, |
1813dd64 | 779 | const struct xfs_buf_ops *ops) |
5adc94c2 | 780 | { |
eab4e633 | 781 | struct xfs_buf *bp; |
5adc94c2 | 782 | |
ba372674 DC |
783 | *bpp = NULL; |
784 | ||
e70b73f8 | 785 | bp = xfs_buf_get_uncached(target, numblks, flags); |
5adc94c2 | 786 | if (!bp) |
ba372674 | 787 | return -ENOMEM; |
5adc94c2 DC |
788 | |
789 | /* set up the buffer for a read IO */ | |
3e85c868 | 790 | ASSERT(bp->b_map_count == 1); |
ba372674 | 791 | bp->b_bn = XFS_BUF_DADDR_NULL; /* always null for uncached buffers */ |
3e85c868 | 792 | bp->b_maps[0].bm_bn = daddr; |
cbb7baab | 793 | bp->b_flags |= XBF_READ; |
1813dd64 | 794 | bp->b_ops = ops; |
5adc94c2 | 795 | |
595bff75 | 796 | xfs_buf_submit_wait(bp); |
ba372674 DC |
797 | if (bp->b_error) { |
798 | int error = bp->b_error; | |
83a0adc3 | 799 | xfs_buf_relse(bp); |
ba372674 | 800 | return error; |
83a0adc3 | 801 | } |
ba372674 DC |
802 | |
803 | *bpp = bp; | |
804 | return 0; | |
1da177e4 LT |
805 | } |
806 | ||
44396476 DC |
807 | /* |
808 | * Return a buffer allocated as an empty buffer and associated to external | |
809 | * memory via xfs_buf_associate_memory() back to it's empty state. | |
810 | */ | |
811 | void | |
812 | xfs_buf_set_empty( | |
813 | struct xfs_buf *bp, | |
e70b73f8 | 814 | size_t numblks) |
44396476 DC |
815 | { |
816 | if (bp->b_pages) | |
817 | _xfs_buf_free_pages(bp); | |
818 | ||
819 | bp->b_pages = NULL; | |
820 | bp->b_page_count = 0; | |
821 | bp->b_addr = NULL; | |
4e94b71b | 822 | bp->b_length = numblks; |
aa0e8833 | 823 | bp->b_io_length = numblks; |
3e85c868 DC |
824 | |
825 | ASSERT(bp->b_map_count == 1); | |
44396476 | 826 | bp->b_bn = XFS_BUF_DADDR_NULL; |
3e85c868 DC |
827 | bp->b_maps[0].bm_bn = XFS_BUF_DADDR_NULL; |
828 | bp->b_maps[0].bm_len = bp->b_length; | |
44396476 DC |
829 | } |
830 | ||
1da177e4 LT |
831 | static inline struct page * |
832 | mem_to_page( | |
833 | void *addr) | |
834 | { | |
9e2779fa | 835 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
836 | return virt_to_page(addr); |
837 | } else { | |
838 | return vmalloc_to_page(addr); | |
839 | } | |
840 | } | |
841 | ||
842 | int | |
ce8e922c NS |
843 | xfs_buf_associate_memory( |
844 | xfs_buf_t *bp, | |
1da177e4 LT |
845 | void *mem, |
846 | size_t len) | |
847 | { | |
848 | int rval; | |
849 | int i = 0; | |
d1afb678 LM |
850 | unsigned long pageaddr; |
851 | unsigned long offset; | |
852 | size_t buflen; | |
1da177e4 LT |
853 | int page_count; |
854 | ||
0e6e847f | 855 | pageaddr = (unsigned long)mem & PAGE_MASK; |
d1afb678 | 856 | offset = (unsigned long)mem - pageaddr; |
0e6e847f DC |
857 | buflen = PAGE_ALIGN(len + offset); |
858 | page_count = buflen >> PAGE_SHIFT; | |
1da177e4 LT |
859 | |
860 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
861 | if (bp->b_pages) |
862 | _xfs_buf_free_pages(bp); | |
1da177e4 | 863 | |
ce8e922c NS |
864 | bp->b_pages = NULL; |
865 | bp->b_addr = mem; | |
1da177e4 | 866 | |
87937bf8 | 867 | rval = _xfs_buf_get_pages(bp, page_count); |
1da177e4 LT |
868 | if (rval) |
869 | return rval; | |
870 | ||
ce8e922c | 871 | bp->b_offset = offset; |
d1afb678 LM |
872 | |
873 | for (i = 0; i < bp->b_page_count; i++) { | |
874 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
0e6e847f | 875 | pageaddr += PAGE_SIZE; |
1da177e4 | 876 | } |
1da177e4 | 877 | |
aa0e8833 | 878 | bp->b_io_length = BTOBB(len); |
4e94b71b | 879 | bp->b_length = BTOBB(buflen); |
1da177e4 LT |
880 | |
881 | return 0; | |
882 | } | |
883 | ||
884 | xfs_buf_t * | |
686865f7 DC |
885 | xfs_buf_get_uncached( |
886 | struct xfs_buftarg *target, | |
e70b73f8 | 887 | size_t numblks, |
686865f7 | 888 | int flags) |
1da177e4 | 889 | { |
e70b73f8 | 890 | unsigned long page_count; |
1fa40b01 | 891 | int error, i; |
3e85c868 DC |
892 | struct xfs_buf *bp; |
893 | DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks); | |
1da177e4 | 894 | |
c891c30a BF |
895 | /* flags might contain irrelevant bits, pass only what we care about */ |
896 | bp = _xfs_buf_alloc(target, &map, 1, flags & XBF_NO_IOACCT); | |
1da177e4 LT |
897 | if (unlikely(bp == NULL)) |
898 | goto fail; | |
1da177e4 | 899 | |
e70b73f8 | 900 | page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT; |
87937bf8 | 901 | error = _xfs_buf_get_pages(bp, page_count); |
1fa40b01 | 902 | if (error) |
1da177e4 LT |
903 | goto fail_free_buf; |
904 | ||
1fa40b01 | 905 | for (i = 0; i < page_count; i++) { |
686865f7 | 906 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
907 | if (!bp->b_pages[i]) |
908 | goto fail_free_mem; | |
1da177e4 | 909 | } |
1fa40b01 | 910 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 911 | |
611c9946 | 912 | error = _xfs_buf_map_pages(bp, 0); |
1fa40b01 | 913 | if (unlikely(error)) { |
4f10700a | 914 | xfs_warn(target->bt_mount, |
08e96e1a | 915 | "%s: failed to map pages", __func__); |
1da177e4 | 916 | goto fail_free_mem; |
1fa40b01 | 917 | } |
1da177e4 | 918 | |
686865f7 | 919 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 920 | return bp; |
1fa40b01 | 921 | |
1da177e4 | 922 | fail_free_mem: |
1fa40b01 CH |
923 | while (--i >= 0) |
924 | __free_page(bp->b_pages[i]); | |
ca165b88 | 925 | _xfs_buf_free_pages(bp); |
1da177e4 | 926 | fail_free_buf: |
3e85c868 | 927 | xfs_buf_free_maps(bp); |
4347b9d7 | 928 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
929 | fail: |
930 | return NULL; | |
931 | } | |
932 | ||
933 | /* | |
1da177e4 LT |
934 | * Increment reference count on buffer, to hold the buffer concurrently |
935 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
936 | * Must hold the buffer already to call this function. |
937 | */ | |
938 | void | |
ce8e922c NS |
939 | xfs_buf_hold( |
940 | xfs_buf_t *bp) | |
1da177e4 | 941 | { |
0b1b213f | 942 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 943 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
944 | } |
945 | ||
946 | /* | |
9c7504aa BF |
947 | * Release a hold on the specified buffer. If the hold count is 1, the buffer is |
948 | * placed on LRU or freed (depending on b_lru_ref). | |
1da177e4 LT |
949 | */ |
950 | void | |
ce8e922c NS |
951 | xfs_buf_rele( |
952 | xfs_buf_t *bp) | |
1da177e4 | 953 | { |
74f75a0c | 954 | struct xfs_perag *pag = bp->b_pag; |
9c7504aa BF |
955 | bool release; |
956 | bool freebuf = false; | |
1da177e4 | 957 | |
0b1b213f | 958 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 959 | |
74f75a0c | 960 | if (!pag) { |
430cbeb8 | 961 | ASSERT(list_empty(&bp->b_lru)); |
9c7504aa BF |
962 | if (atomic_dec_and_test(&bp->b_hold)) { |
963 | xfs_buf_ioacct_dec(bp); | |
fad3aa1e | 964 | xfs_buf_free(bp); |
9c7504aa | 965 | } |
fad3aa1e NS |
966 | return; |
967 | } | |
968 | ||
3790689f | 969 | ASSERT(atomic_read(&bp->b_hold) > 0); |
a4082357 | 970 | |
9c7504aa BF |
971 | release = atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock); |
972 | spin_lock(&bp->b_lock); | |
973 | if (!release) { | |
974 | /* | |
975 | * Drop the in-flight state if the buffer is already on the LRU | |
976 | * and it holds the only reference. This is racy because we | |
977 | * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT | |
978 | * ensures the decrement occurs only once per-buf. | |
979 | */ | |
980 | if ((atomic_read(&bp->b_hold) == 1) && !list_empty(&bp->b_lru)) | |
981 | xfs_buf_ioacct_dec(bp); | |
982 | goto out_unlock; | |
983 | } | |
984 | ||
985 | /* the last reference has been dropped ... */ | |
986 | xfs_buf_ioacct_dec(bp); | |
987 | if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) { | |
988 | /* | |
989 | * If the buffer is added to the LRU take a new reference to the | |
990 | * buffer for the LRU and clear the (now stale) dispose list | |
991 | * state flag | |
992 | */ | |
993 | if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) { | |
994 | bp->b_state &= ~XFS_BSTATE_DISPOSE; | |
995 | atomic_inc(&bp->b_hold); | |
1da177e4 | 996 | } |
9c7504aa BF |
997 | spin_unlock(&pag->pag_buf_lock); |
998 | } else { | |
999 | /* | |
1000 | * most of the time buffers will already be removed from the | |
1001 | * LRU, so optimise that case by checking for the | |
1002 | * XFS_BSTATE_DISPOSE flag indicating the last list the buffer | |
1003 | * was on was the disposal list | |
1004 | */ | |
1005 | if (!(bp->b_state & XFS_BSTATE_DISPOSE)) { | |
1006 | list_lru_del(&bp->b_target->bt_lru, &bp->b_lru); | |
1007 | } else { | |
1008 | ASSERT(list_empty(&bp->b_lru)); | |
1da177e4 | 1009 | } |
9c7504aa BF |
1010 | |
1011 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); | |
6031e73a LS |
1012 | rhashtable_remove_fast(&pag->pag_buf_hash, &bp->b_rhash_head, |
1013 | xfs_buf_hash_params); | |
9c7504aa BF |
1014 | spin_unlock(&pag->pag_buf_lock); |
1015 | xfs_perag_put(pag); | |
1016 | freebuf = true; | |
1da177e4 | 1017 | } |
9c7504aa BF |
1018 | |
1019 | out_unlock: | |
1020 | spin_unlock(&bp->b_lock); | |
1021 | ||
1022 | if (freebuf) | |
1023 | xfs_buf_free(bp); | |
1da177e4 LT |
1024 | } |
1025 | ||
1026 | ||
1027 | /* | |
0e6e847f | 1028 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
1029 | * |
1030 | * If we come across a stale, pinned, locked buffer, we know that we are | |
1031 | * being asked to lock a buffer that has been reallocated. Because it is | |
1032 | * pinned, we know that the log has not been pushed to disk and hence it | |
1033 | * will still be locked. Rather than continuing to have trylock attempts | |
1034 | * fail until someone else pushes the log, push it ourselves before | |
1035 | * returning. This means that the xfsaild will not get stuck trying | |
1036 | * to push on stale inode buffers. | |
1da177e4 LT |
1037 | */ |
1038 | int | |
0c842ad4 CH |
1039 | xfs_buf_trylock( |
1040 | struct xfs_buf *bp) | |
1da177e4 LT |
1041 | { |
1042 | int locked; | |
1043 | ||
ce8e922c | 1044 | locked = down_trylock(&bp->b_sema) == 0; |
479c6412 | 1045 | if (locked) { |
ce8e922c | 1046 | XB_SET_OWNER(bp); |
479c6412 DW |
1047 | trace_xfs_buf_trylock(bp, _RET_IP_); |
1048 | } else { | |
1049 | trace_xfs_buf_trylock_fail(bp, _RET_IP_); | |
1050 | } | |
0c842ad4 | 1051 | return locked; |
1da177e4 | 1052 | } |
1da177e4 LT |
1053 | |
1054 | /* | |
0e6e847f | 1055 | * Lock a buffer object. |
ed3b4d6c DC |
1056 | * |
1057 | * If we come across a stale, pinned, locked buffer, we know that we | |
1058 | * are being asked to lock a buffer that has been reallocated. Because | |
1059 | * it is pinned, we know that the log has not been pushed to disk and | |
1060 | * hence it will still be locked. Rather than sleeping until someone | |
1061 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 1062 | */ |
ce8e922c NS |
1063 | void |
1064 | xfs_buf_lock( | |
0c842ad4 | 1065 | struct xfs_buf *bp) |
1da177e4 | 1066 | { |
0b1b213f CH |
1067 | trace_xfs_buf_lock(bp, _RET_IP_); |
1068 | ||
ed3b4d6c | 1069 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 1070 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
1071 | down(&bp->b_sema); |
1072 | XB_SET_OWNER(bp); | |
0b1b213f CH |
1073 | |
1074 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
1075 | } |
1076 | ||
1da177e4 | 1077 | void |
ce8e922c | 1078 | xfs_buf_unlock( |
0c842ad4 | 1079 | struct xfs_buf *bp) |
1da177e4 | 1080 | { |
ce8e922c NS |
1081 | XB_CLEAR_OWNER(bp); |
1082 | up(&bp->b_sema); | |
0b1b213f CH |
1083 | |
1084 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
1085 | } |
1086 | ||
ce8e922c NS |
1087 | STATIC void |
1088 | xfs_buf_wait_unpin( | |
1089 | xfs_buf_t *bp) | |
1da177e4 LT |
1090 | { |
1091 | DECLARE_WAITQUEUE (wait, current); | |
1092 | ||
ce8e922c | 1093 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
1094 | return; |
1095 | ||
ce8e922c | 1096 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1097 | for (;;) { |
1098 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 1099 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 1100 | break; |
7eaceacc | 1101 | io_schedule(); |
1da177e4 | 1102 | } |
ce8e922c | 1103 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1104 | set_current_state(TASK_RUNNING); |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * Buffer Utility Routines | |
1109 | */ | |
1110 | ||
e8aaba9a DC |
1111 | void |
1112 | xfs_buf_ioend( | |
1113 | struct xfs_buf *bp) | |
1da177e4 | 1114 | { |
e8aaba9a DC |
1115 | bool read = bp->b_flags & XBF_READ; |
1116 | ||
1117 | trace_xfs_buf_iodone(bp, _RET_IP_); | |
1813dd64 DC |
1118 | |
1119 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); | |
d5929de8 | 1120 | |
61be9c52 DC |
1121 | /* |
1122 | * Pull in IO completion errors now. We are guaranteed to be running | |
1123 | * single threaded, so we don't need the lock to read b_io_error. | |
1124 | */ | |
1125 | if (!bp->b_error && bp->b_io_error) | |
1126 | xfs_buf_ioerror(bp, bp->b_io_error); | |
1127 | ||
e8aaba9a DC |
1128 | /* Only validate buffers that were read without errors */ |
1129 | if (read && !bp->b_error && bp->b_ops) { | |
1130 | ASSERT(!bp->b_iodone); | |
1813dd64 | 1131 | bp->b_ops->verify_read(bp); |
e8aaba9a DC |
1132 | } |
1133 | ||
1134 | if (!bp->b_error) | |
1135 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1136 | |
80f6c29d | 1137 | if (bp->b_iodone) |
ce8e922c NS |
1138 | (*(bp->b_iodone))(bp); |
1139 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 | 1140 | xfs_buf_relse(bp); |
595bff75 | 1141 | else |
1813dd64 | 1142 | complete(&bp->b_iowait); |
1da177e4 LT |
1143 | } |
1144 | ||
e8aaba9a DC |
1145 | static void |
1146 | xfs_buf_ioend_work( | |
1147 | struct work_struct *work) | |
1da177e4 | 1148 | { |
e8aaba9a | 1149 | struct xfs_buf *bp = |
b29c70f5 | 1150 | container_of(work, xfs_buf_t, b_ioend_work); |
0b1b213f | 1151 | |
e8aaba9a DC |
1152 | xfs_buf_ioend(bp); |
1153 | } | |
1da177e4 | 1154 | |
211fe1a4 | 1155 | static void |
e8aaba9a DC |
1156 | xfs_buf_ioend_async( |
1157 | struct xfs_buf *bp) | |
1158 | { | |
b29c70f5 BF |
1159 | INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work); |
1160 | queue_work(bp->b_ioend_wq, &bp->b_ioend_work); | |
1da177e4 LT |
1161 | } |
1162 | ||
1da177e4 | 1163 | void |
ce8e922c NS |
1164 | xfs_buf_ioerror( |
1165 | xfs_buf_t *bp, | |
1166 | int error) | |
1da177e4 | 1167 | { |
2451337d DC |
1168 | ASSERT(error <= 0 && error >= -1000); |
1169 | bp->b_error = error; | |
0b1b213f | 1170 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1171 | } |
1172 | ||
901796af CH |
1173 | void |
1174 | xfs_buf_ioerror_alert( | |
1175 | struct xfs_buf *bp, | |
1176 | const char *func) | |
1177 | { | |
1178 | xfs_alert(bp->b_target->bt_mount, | |
aa0e8833 | 1179 | "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d", |
2451337d | 1180 | (__uint64_t)XFS_BUF_ADDR(bp), func, -bp->b_error, bp->b_length); |
901796af CH |
1181 | } |
1182 | ||
a2dcf5df CH |
1183 | int |
1184 | xfs_bwrite( | |
1185 | struct xfs_buf *bp) | |
1186 | { | |
1187 | int error; | |
1188 | ||
1189 | ASSERT(xfs_buf_islocked(bp)); | |
1190 | ||
1191 | bp->b_flags |= XBF_WRITE; | |
27187754 DC |
1192 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q | |
1193 | XBF_WRITE_FAIL | XBF_DONE); | |
a2dcf5df | 1194 | |
595bff75 | 1195 | error = xfs_buf_submit_wait(bp); |
a2dcf5df CH |
1196 | if (error) { |
1197 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1198 | SHUTDOWN_META_IO_ERROR); | |
1199 | } | |
1200 | return error; | |
1201 | } | |
1202 | ||
9bdd9bd6 | 1203 | static void |
ce8e922c | 1204 | xfs_buf_bio_end_io( |
4246a0b6 | 1205 | struct bio *bio) |
1da177e4 | 1206 | { |
9bdd9bd6 | 1207 | struct xfs_buf *bp = (struct xfs_buf *)bio->bi_private; |
1da177e4 | 1208 | |
37eb17e6 DC |
1209 | /* |
1210 | * don't overwrite existing errors - otherwise we can lose errors on | |
1211 | * buffers that require multiple bios to complete. | |
1212 | */ | |
9bdd9bd6 BF |
1213 | if (bio->bi_error) |
1214 | cmpxchg(&bp->b_io_error, 0, bio->bi_error); | |
1da177e4 | 1215 | |
37eb17e6 | 1216 | if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
73c77e2c JB |
1217 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); |
1218 | ||
e8aaba9a DC |
1219 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
1220 | xfs_buf_ioend_async(bp); | |
1da177e4 | 1221 | bio_put(bio); |
1da177e4 LT |
1222 | } |
1223 | ||
3e85c868 DC |
1224 | static void |
1225 | xfs_buf_ioapply_map( | |
1226 | struct xfs_buf *bp, | |
1227 | int map, | |
1228 | int *buf_offset, | |
1229 | int *count, | |
50bfcd0c MC |
1230 | int op, |
1231 | int op_flags) | |
1da177e4 | 1232 | { |
3e85c868 DC |
1233 | int page_index; |
1234 | int total_nr_pages = bp->b_page_count; | |
1235 | int nr_pages; | |
1236 | struct bio *bio; | |
1237 | sector_t sector = bp->b_maps[map].bm_bn; | |
1238 | int size; | |
1239 | int offset; | |
1da177e4 | 1240 | |
ce8e922c | 1241 | total_nr_pages = bp->b_page_count; |
1da177e4 | 1242 | |
3e85c868 DC |
1243 | /* skip the pages in the buffer before the start offset */ |
1244 | page_index = 0; | |
1245 | offset = *buf_offset; | |
1246 | while (offset >= PAGE_SIZE) { | |
1247 | page_index++; | |
1248 | offset -= PAGE_SIZE; | |
f538d4da CH |
1249 | } |
1250 | ||
3e85c868 DC |
1251 | /* |
1252 | * Limit the IO size to the length of the current vector, and update the | |
1253 | * remaining IO count for the next time around. | |
1254 | */ | |
1255 | size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count); | |
1256 | *count -= size; | |
1257 | *buf_offset += size; | |
34951f5c | 1258 | |
1da177e4 | 1259 | next_chunk: |
ce8e922c | 1260 | atomic_inc(&bp->b_io_remaining); |
c908e380 | 1261 | nr_pages = min(total_nr_pages, BIO_MAX_PAGES); |
1da177e4 LT |
1262 | |
1263 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1264 | bio->bi_bdev = bp->b_target->bt_bdev; |
4f024f37 | 1265 | bio->bi_iter.bi_sector = sector; |
ce8e922c NS |
1266 | bio->bi_end_io = xfs_buf_bio_end_io; |
1267 | bio->bi_private = bp; | |
50bfcd0c | 1268 | bio_set_op_attrs(bio, op, op_flags); |
0e6e847f | 1269 | |
3e85c868 | 1270 | for (; size && nr_pages; nr_pages--, page_index++) { |
0e6e847f | 1271 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1272 | |
1273 | if (nbytes > size) | |
1274 | nbytes = size; | |
1275 | ||
3e85c868 DC |
1276 | rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes, |
1277 | offset); | |
ce8e922c | 1278 | if (rbytes < nbytes) |
1da177e4 LT |
1279 | break; |
1280 | ||
1281 | offset = 0; | |
aa0e8833 | 1282 | sector += BTOBB(nbytes); |
1da177e4 LT |
1283 | size -= nbytes; |
1284 | total_nr_pages--; | |
1285 | } | |
1286 | ||
4f024f37 | 1287 | if (likely(bio->bi_iter.bi_size)) { |
73c77e2c JB |
1288 | if (xfs_buf_is_vmapped(bp)) { |
1289 | flush_kernel_vmap_range(bp->b_addr, | |
1290 | xfs_buf_vmap_len(bp)); | |
1291 | } | |
4e49ea4a | 1292 | submit_bio(bio); |
1da177e4 LT |
1293 | if (size) |
1294 | goto next_chunk; | |
1295 | } else { | |
37eb17e6 DC |
1296 | /* |
1297 | * This is guaranteed not to be the last io reference count | |
595bff75 | 1298 | * because the caller (xfs_buf_submit) holds a count itself. |
37eb17e6 DC |
1299 | */ |
1300 | atomic_dec(&bp->b_io_remaining); | |
2451337d | 1301 | xfs_buf_ioerror(bp, -EIO); |
ec53d1db | 1302 | bio_put(bio); |
1da177e4 | 1303 | } |
3e85c868 DC |
1304 | |
1305 | } | |
1306 | ||
1307 | STATIC void | |
1308 | _xfs_buf_ioapply( | |
1309 | struct xfs_buf *bp) | |
1310 | { | |
1311 | struct blk_plug plug; | |
50bfcd0c MC |
1312 | int op; |
1313 | int op_flags = 0; | |
3e85c868 DC |
1314 | int offset; |
1315 | int size; | |
1316 | int i; | |
1317 | ||
c163f9a1 DC |
1318 | /* |
1319 | * Make sure we capture only current IO errors rather than stale errors | |
1320 | * left over from previous use of the buffer (e.g. failed readahead). | |
1321 | */ | |
1322 | bp->b_error = 0; | |
1323 | ||
b29c70f5 BF |
1324 | /* |
1325 | * Initialize the I/O completion workqueue if we haven't yet or the | |
1326 | * submitter has not opted to specify a custom one. | |
1327 | */ | |
1328 | if (!bp->b_ioend_wq) | |
1329 | bp->b_ioend_wq = bp->b_target->bt_mount->m_buf_workqueue; | |
1330 | ||
3e85c868 | 1331 | if (bp->b_flags & XBF_WRITE) { |
50bfcd0c | 1332 | op = REQ_OP_WRITE; |
3e85c868 | 1333 | if (bp->b_flags & XBF_SYNCIO) |
70fd7614 | 1334 | op_flags = REQ_SYNC; |
3e85c868 | 1335 | if (bp->b_flags & XBF_FUA) |
50bfcd0c | 1336 | op_flags |= REQ_FUA; |
3e85c868 | 1337 | if (bp->b_flags & XBF_FLUSH) |
28a8f0d3 | 1338 | op_flags |= REQ_PREFLUSH; |
1813dd64 DC |
1339 | |
1340 | /* | |
1341 | * Run the write verifier callback function if it exists. If | |
1342 | * this function fails it will mark the buffer with an error and | |
1343 | * the IO should not be dispatched. | |
1344 | */ | |
1345 | if (bp->b_ops) { | |
1346 | bp->b_ops->verify_write(bp); | |
1347 | if (bp->b_error) { | |
1348 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1349 | SHUTDOWN_CORRUPT_INCORE); | |
1350 | return; | |
1351 | } | |
400b9d88 DC |
1352 | } else if (bp->b_bn != XFS_BUF_DADDR_NULL) { |
1353 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
1354 | ||
1355 | /* | |
1356 | * non-crc filesystems don't attach verifiers during | |
1357 | * log recovery, so don't warn for such filesystems. | |
1358 | */ | |
1359 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
1360 | xfs_warn(mp, | |
1361 | "%s: no ops on block 0x%llx/0x%x", | |
1362 | __func__, bp->b_bn, bp->b_length); | |
1363 | xfs_hex_dump(bp->b_addr, 64); | |
1364 | dump_stack(); | |
1365 | } | |
1813dd64 | 1366 | } |
3e85c868 | 1367 | } else if (bp->b_flags & XBF_READ_AHEAD) { |
50bfcd0c MC |
1368 | op = REQ_OP_READ; |
1369 | op_flags = REQ_RAHEAD; | |
3e85c868 | 1370 | } else { |
50bfcd0c | 1371 | op = REQ_OP_READ; |
3e85c868 DC |
1372 | } |
1373 | ||
1374 | /* we only use the buffer cache for meta-data */ | |
50bfcd0c | 1375 | op_flags |= REQ_META; |
3e85c868 DC |
1376 | |
1377 | /* | |
1378 | * Walk all the vectors issuing IO on them. Set up the initial offset | |
1379 | * into the buffer and the desired IO size before we start - | |
1380 | * _xfs_buf_ioapply_vec() will modify them appropriately for each | |
1381 | * subsequent call. | |
1382 | */ | |
1383 | offset = bp->b_offset; | |
1384 | size = BBTOB(bp->b_io_length); | |
1385 | blk_start_plug(&plug); | |
1386 | for (i = 0; i < bp->b_map_count; i++) { | |
50bfcd0c | 1387 | xfs_buf_ioapply_map(bp, i, &offset, &size, op, op_flags); |
3e85c868 DC |
1388 | if (bp->b_error) |
1389 | break; | |
1390 | if (size <= 0) | |
1391 | break; /* all done */ | |
1392 | } | |
1393 | blk_finish_plug(&plug); | |
1da177e4 LT |
1394 | } |
1395 | ||
595bff75 DC |
1396 | /* |
1397 | * Asynchronous IO submission path. This transfers the buffer lock ownership and | |
1398 | * the current reference to the IO. It is not safe to reference the buffer after | |
1399 | * a call to this function unless the caller holds an additional reference | |
1400 | * itself. | |
1401 | */ | |
0e95f19a | 1402 | void |
595bff75 DC |
1403 | xfs_buf_submit( |
1404 | struct xfs_buf *bp) | |
1da177e4 | 1405 | { |
595bff75 | 1406 | trace_xfs_buf_submit(bp, _RET_IP_); |
1da177e4 | 1407 | |
43ff2122 | 1408 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
595bff75 DC |
1409 | ASSERT(bp->b_flags & XBF_ASYNC); |
1410 | ||
1411 | /* on shutdown we stale and complete the buffer immediately */ | |
1412 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { | |
1413 | xfs_buf_ioerror(bp, -EIO); | |
1414 | bp->b_flags &= ~XBF_DONE; | |
1415 | xfs_buf_stale(bp); | |
1416 | xfs_buf_ioend(bp); | |
1417 | return; | |
1418 | } | |
1da177e4 | 1419 | |
375ec69d | 1420 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1421 | xfs_buf_wait_unpin(bp); |
e11bb805 | 1422 | |
61be9c52 DC |
1423 | /* clear the internal error state to avoid spurious errors */ |
1424 | bp->b_io_error = 0; | |
1425 | ||
e11bb805 | 1426 | /* |
595bff75 DC |
1427 | * The caller's reference is released during I/O completion. |
1428 | * This occurs some time after the last b_io_remaining reference is | |
1429 | * released, so after we drop our Io reference we have to have some | |
1430 | * other reference to ensure the buffer doesn't go away from underneath | |
1431 | * us. Take a direct reference to ensure we have safe access to the | |
1432 | * buffer until we are finished with it. | |
e11bb805 | 1433 | */ |
ce8e922c | 1434 | xfs_buf_hold(bp); |
1da177e4 | 1435 | |
8d6c1210 | 1436 | /* |
e11bb805 DC |
1437 | * Set the count to 1 initially, this will stop an I/O completion |
1438 | * callout which happens before we have started all the I/O from calling | |
1439 | * xfs_buf_ioend too early. | |
1da177e4 | 1440 | */ |
ce8e922c | 1441 | atomic_set(&bp->b_io_remaining, 1); |
9c7504aa | 1442 | xfs_buf_ioacct_inc(bp); |
ce8e922c | 1443 | _xfs_buf_ioapply(bp); |
e11bb805 | 1444 | |
8d6c1210 | 1445 | /* |
595bff75 DC |
1446 | * If _xfs_buf_ioapply failed, we can get back here with only the IO |
1447 | * reference we took above. If we drop it to zero, run completion so | |
1448 | * that we don't return to the caller with completion still pending. | |
8d6c1210 | 1449 | */ |
e8aaba9a | 1450 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
595bff75 | 1451 | if (bp->b_error) |
e8aaba9a DC |
1452 | xfs_buf_ioend(bp); |
1453 | else | |
1454 | xfs_buf_ioend_async(bp); | |
1455 | } | |
1da177e4 | 1456 | |
ce8e922c | 1457 | xfs_buf_rele(bp); |
595bff75 | 1458 | /* Note: it is not safe to reference bp now we've dropped our ref */ |
1da177e4 LT |
1459 | } |
1460 | ||
1461 | /* | |
595bff75 | 1462 | * Synchronous buffer IO submission path, read or write. |
1da177e4 LT |
1463 | */ |
1464 | int | |
595bff75 DC |
1465 | xfs_buf_submit_wait( |
1466 | struct xfs_buf *bp) | |
1da177e4 | 1467 | { |
595bff75 | 1468 | int error; |
0b1b213f | 1469 | |
595bff75 DC |
1470 | trace_xfs_buf_submit_wait(bp, _RET_IP_); |
1471 | ||
1472 | ASSERT(!(bp->b_flags & (_XBF_DELWRI_Q | XBF_ASYNC))); | |
0b1b213f | 1473 | |
595bff75 DC |
1474 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
1475 | xfs_buf_ioerror(bp, -EIO); | |
1476 | xfs_buf_stale(bp); | |
1477 | bp->b_flags &= ~XBF_DONE; | |
1478 | return -EIO; | |
1479 | } | |
1480 | ||
1481 | if (bp->b_flags & XBF_WRITE) | |
1482 | xfs_buf_wait_unpin(bp); | |
1483 | ||
1484 | /* clear the internal error state to avoid spurious errors */ | |
1485 | bp->b_io_error = 0; | |
1486 | ||
1487 | /* | |
1488 | * For synchronous IO, the IO does not inherit the submitters reference | |
1489 | * count, nor the buffer lock. Hence we cannot release the reference we | |
1490 | * are about to take until we've waited for all IO completion to occur, | |
1491 | * including any xfs_buf_ioend_async() work that may be pending. | |
1492 | */ | |
1493 | xfs_buf_hold(bp); | |
1494 | ||
1495 | /* | |
1496 | * Set the count to 1 initially, this will stop an I/O completion | |
1497 | * callout which happens before we have started all the I/O from calling | |
1498 | * xfs_buf_ioend too early. | |
1499 | */ | |
1500 | atomic_set(&bp->b_io_remaining, 1); | |
1501 | _xfs_buf_ioapply(bp); | |
1502 | ||
1503 | /* | |
1504 | * make sure we run completion synchronously if it raced with us and is | |
1505 | * already complete. | |
1506 | */ | |
1507 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) | |
1508 | xfs_buf_ioend(bp); | |
0b1b213f | 1509 | |
595bff75 DC |
1510 | /* wait for completion before gathering the error from the buffer */ |
1511 | trace_xfs_buf_iowait(bp, _RET_IP_); | |
1512 | wait_for_completion(&bp->b_iowait); | |
0b1b213f | 1513 | trace_xfs_buf_iowait_done(bp, _RET_IP_); |
595bff75 DC |
1514 | error = bp->b_error; |
1515 | ||
1516 | /* | |
1517 | * all done now, we can release the hold that keeps the buffer | |
1518 | * referenced for the entire IO. | |
1519 | */ | |
1520 | xfs_buf_rele(bp); | |
1521 | return error; | |
1da177e4 LT |
1522 | } |
1523 | ||
88ee2df7 | 1524 | void * |
ce8e922c | 1525 | xfs_buf_offset( |
88ee2df7 | 1526 | struct xfs_buf *bp, |
1da177e4 LT |
1527 | size_t offset) |
1528 | { | |
1529 | struct page *page; | |
1530 | ||
611c9946 | 1531 | if (bp->b_addr) |
62926044 | 1532 | return bp->b_addr + offset; |
1da177e4 | 1533 | |
ce8e922c | 1534 | offset += bp->b_offset; |
0e6e847f | 1535 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
88ee2df7 | 1536 | return page_address(page) + (offset & (PAGE_SIZE-1)); |
1da177e4 LT |
1537 | } |
1538 | ||
1539 | /* | |
1da177e4 LT |
1540 | * Move data into or out of a buffer. |
1541 | */ | |
1542 | void | |
ce8e922c NS |
1543 | xfs_buf_iomove( |
1544 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1545 | size_t boff, /* starting buffer offset */ |
1546 | size_t bsize, /* length to copy */ | |
b9c48649 | 1547 | void *data, /* data address */ |
ce8e922c | 1548 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 | 1549 | { |
795cac72 | 1550 | size_t bend; |
1da177e4 LT |
1551 | |
1552 | bend = boff + bsize; | |
1553 | while (boff < bend) { | |
795cac72 DC |
1554 | struct page *page; |
1555 | int page_index, page_offset, csize; | |
1556 | ||
1557 | page_index = (boff + bp->b_offset) >> PAGE_SHIFT; | |
1558 | page_offset = (boff + bp->b_offset) & ~PAGE_MASK; | |
1559 | page = bp->b_pages[page_index]; | |
1560 | csize = min_t(size_t, PAGE_SIZE - page_offset, | |
1561 | BBTOB(bp->b_io_length) - boff); | |
1da177e4 | 1562 | |
795cac72 | 1563 | ASSERT((csize + page_offset) <= PAGE_SIZE); |
1da177e4 LT |
1564 | |
1565 | switch (mode) { | |
ce8e922c | 1566 | case XBRW_ZERO: |
795cac72 | 1567 | memset(page_address(page) + page_offset, 0, csize); |
1da177e4 | 1568 | break; |
ce8e922c | 1569 | case XBRW_READ: |
795cac72 | 1570 | memcpy(data, page_address(page) + page_offset, csize); |
1da177e4 | 1571 | break; |
ce8e922c | 1572 | case XBRW_WRITE: |
795cac72 | 1573 | memcpy(page_address(page) + page_offset, data, csize); |
1da177e4 LT |
1574 | } |
1575 | ||
1576 | boff += csize; | |
1577 | data += csize; | |
1578 | } | |
1579 | } | |
1580 | ||
1581 | /* | |
ce8e922c | 1582 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1583 | */ |
1584 | ||
1585 | /* | |
430cbeb8 DC |
1586 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1587 | * returned. These buffers will have an elevated hold count, so wait on those | |
1588 | * while freeing all the buffers only held by the LRU. | |
1da177e4 | 1589 | */ |
e80dfa19 DC |
1590 | static enum lru_status |
1591 | xfs_buftarg_wait_rele( | |
1592 | struct list_head *item, | |
3f97b163 | 1593 | struct list_lru_one *lru, |
e80dfa19 DC |
1594 | spinlock_t *lru_lock, |
1595 | void *arg) | |
1596 | ||
1da177e4 | 1597 | { |
e80dfa19 | 1598 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); |
a4082357 | 1599 | struct list_head *dispose = arg; |
430cbeb8 | 1600 | |
e80dfa19 | 1601 | if (atomic_read(&bp->b_hold) > 1) { |
a4082357 | 1602 | /* need to wait, so skip it this pass */ |
e80dfa19 | 1603 | trace_xfs_buf_wait_buftarg(bp, _RET_IP_); |
a4082357 | 1604 | return LRU_SKIP; |
1da177e4 | 1605 | } |
a4082357 DC |
1606 | if (!spin_trylock(&bp->b_lock)) |
1607 | return LRU_SKIP; | |
e80dfa19 | 1608 | |
a4082357 DC |
1609 | /* |
1610 | * clear the LRU reference count so the buffer doesn't get | |
1611 | * ignored in xfs_buf_rele(). | |
1612 | */ | |
1613 | atomic_set(&bp->b_lru_ref, 0); | |
1614 | bp->b_state |= XFS_BSTATE_DISPOSE; | |
3f97b163 | 1615 | list_lru_isolate_move(lru, item, dispose); |
a4082357 DC |
1616 | spin_unlock(&bp->b_lock); |
1617 | return LRU_REMOVED; | |
1da177e4 LT |
1618 | } |
1619 | ||
e80dfa19 DC |
1620 | void |
1621 | xfs_wait_buftarg( | |
1622 | struct xfs_buftarg *btp) | |
1623 | { | |
a4082357 DC |
1624 | LIST_HEAD(dispose); |
1625 | int loop = 0; | |
1626 | ||
85bec546 | 1627 | /* |
9c7504aa BF |
1628 | * First wait on the buftarg I/O count for all in-flight buffers to be |
1629 | * released. This is critical as new buffers do not make the LRU until | |
1630 | * they are released. | |
1631 | * | |
1632 | * Next, flush the buffer workqueue to ensure all completion processing | |
1633 | * has finished. Just waiting on buffer locks is not sufficient for | |
1634 | * async IO as the reference count held over IO is not released until | |
1635 | * after the buffer lock is dropped. Hence we need to ensure here that | |
1636 | * all reference counts have been dropped before we start walking the | |
1637 | * LRU list. | |
85bec546 | 1638 | */ |
9c7504aa BF |
1639 | while (percpu_counter_sum(&btp->bt_io_count)) |
1640 | delay(100); | |
800b2694 | 1641 | flush_workqueue(btp->bt_mount->m_buf_workqueue); |
85bec546 | 1642 | |
a4082357 DC |
1643 | /* loop until there is nothing left on the lru list. */ |
1644 | while (list_lru_count(&btp->bt_lru)) { | |
e80dfa19 | 1645 | list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele, |
a4082357 DC |
1646 | &dispose, LONG_MAX); |
1647 | ||
1648 | while (!list_empty(&dispose)) { | |
1649 | struct xfs_buf *bp; | |
1650 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1651 | list_del_init(&bp->b_lru); | |
ac8809f9 DC |
1652 | if (bp->b_flags & XBF_WRITE_FAIL) { |
1653 | xfs_alert(btp->bt_mount, | |
f41febd2 | 1654 | "Corruption Alert: Buffer at block 0x%llx had permanent write failures!", |
ac8809f9 | 1655 | (long long)bp->b_bn); |
f41febd2 JP |
1656 | xfs_alert(btp->bt_mount, |
1657 | "Please run xfs_repair to determine the extent of the problem."); | |
ac8809f9 | 1658 | } |
a4082357 DC |
1659 | xfs_buf_rele(bp); |
1660 | } | |
1661 | if (loop++ != 0) | |
1662 | delay(100); | |
1663 | } | |
e80dfa19 DC |
1664 | } |
1665 | ||
1666 | static enum lru_status | |
1667 | xfs_buftarg_isolate( | |
1668 | struct list_head *item, | |
3f97b163 | 1669 | struct list_lru_one *lru, |
e80dfa19 DC |
1670 | spinlock_t *lru_lock, |
1671 | void *arg) | |
1672 | { | |
1673 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); | |
1674 | struct list_head *dispose = arg; | |
1675 | ||
a4082357 DC |
1676 | /* |
1677 | * we are inverting the lru lock/bp->b_lock here, so use a trylock. | |
1678 | * If we fail to get the lock, just skip it. | |
1679 | */ | |
1680 | if (!spin_trylock(&bp->b_lock)) | |
1681 | return LRU_SKIP; | |
e80dfa19 DC |
1682 | /* |
1683 | * Decrement the b_lru_ref count unless the value is already | |
1684 | * zero. If the value is already zero, we need to reclaim the | |
1685 | * buffer, otherwise it gets another trip through the LRU. | |
1686 | */ | |
a4082357 DC |
1687 | if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) { |
1688 | spin_unlock(&bp->b_lock); | |
e80dfa19 | 1689 | return LRU_ROTATE; |
a4082357 | 1690 | } |
e80dfa19 | 1691 | |
a4082357 | 1692 | bp->b_state |= XFS_BSTATE_DISPOSE; |
3f97b163 | 1693 | list_lru_isolate_move(lru, item, dispose); |
a4082357 | 1694 | spin_unlock(&bp->b_lock); |
e80dfa19 DC |
1695 | return LRU_REMOVED; |
1696 | } | |
1697 | ||
addbda40 | 1698 | static unsigned long |
e80dfa19 | 1699 | xfs_buftarg_shrink_scan( |
ff57ab21 | 1700 | struct shrinker *shrink, |
1495f230 | 1701 | struct shrink_control *sc) |
a6867a68 | 1702 | { |
ff57ab21 DC |
1703 | struct xfs_buftarg *btp = container_of(shrink, |
1704 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1705 | LIST_HEAD(dispose); |
addbda40 | 1706 | unsigned long freed; |
430cbeb8 | 1707 | |
503c358c VD |
1708 | freed = list_lru_shrink_walk(&btp->bt_lru, sc, |
1709 | xfs_buftarg_isolate, &dispose); | |
430cbeb8 DC |
1710 | |
1711 | while (!list_empty(&dispose)) { | |
e80dfa19 | 1712 | struct xfs_buf *bp; |
430cbeb8 DC |
1713 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); |
1714 | list_del_init(&bp->b_lru); | |
1715 | xfs_buf_rele(bp); | |
1716 | } | |
1717 | ||
e80dfa19 DC |
1718 | return freed; |
1719 | } | |
1720 | ||
addbda40 | 1721 | static unsigned long |
e80dfa19 DC |
1722 | xfs_buftarg_shrink_count( |
1723 | struct shrinker *shrink, | |
1724 | struct shrink_control *sc) | |
1725 | { | |
1726 | struct xfs_buftarg *btp = container_of(shrink, | |
1727 | struct xfs_buftarg, bt_shrinker); | |
503c358c | 1728 | return list_lru_shrink_count(&btp->bt_lru, sc); |
a6867a68 DC |
1729 | } |
1730 | ||
1da177e4 LT |
1731 | void |
1732 | xfs_free_buftarg( | |
b7963133 CH |
1733 | struct xfs_mount *mp, |
1734 | struct xfs_buftarg *btp) | |
1da177e4 | 1735 | { |
ff57ab21 | 1736 | unregister_shrinker(&btp->bt_shrinker); |
9c7504aa BF |
1737 | ASSERT(percpu_counter_sum(&btp->bt_io_count) == 0); |
1738 | percpu_counter_destroy(&btp->bt_io_count); | |
f5e1dd34 | 1739 | list_lru_destroy(&btp->bt_lru); |
ff57ab21 | 1740 | |
2291dab2 | 1741 | xfs_blkdev_issue_flush(btp); |
a6867a68 | 1742 | |
f0e2d93c | 1743 | kmem_free(btp); |
1da177e4 LT |
1744 | } |
1745 | ||
3fefdeee ES |
1746 | int |
1747 | xfs_setsize_buftarg( | |
1da177e4 | 1748 | xfs_buftarg_t *btp, |
3fefdeee | 1749 | unsigned int sectorsize) |
1da177e4 | 1750 | { |
7c71ee78 | 1751 | /* Set up metadata sector size info */ |
6da54179 ES |
1752 | btp->bt_meta_sectorsize = sectorsize; |
1753 | btp->bt_meta_sectormask = sectorsize - 1; | |
1da177e4 | 1754 | |
ce8e922c | 1755 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
4f10700a | 1756 | xfs_warn(btp->bt_mount, |
a1c6f057 DM |
1757 | "Cannot set_blocksize to %u on device %pg", |
1758 | sectorsize, btp->bt_bdev); | |
2451337d | 1759 | return -EINVAL; |
1da177e4 LT |
1760 | } |
1761 | ||
7c71ee78 ES |
1762 | /* Set up device logical sector size mask */ |
1763 | btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev); | |
1764 | btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1; | |
1765 | ||
1da177e4 LT |
1766 | return 0; |
1767 | } | |
1768 | ||
1769 | /* | |
3fefdeee ES |
1770 | * When allocating the initial buffer target we have not yet |
1771 | * read in the superblock, so don't know what sized sectors | |
1772 | * are being used at this early stage. Play safe. | |
ce8e922c | 1773 | */ |
1da177e4 LT |
1774 | STATIC int |
1775 | xfs_setsize_buftarg_early( | |
1776 | xfs_buftarg_t *btp, | |
1777 | struct block_device *bdev) | |
1778 | { | |
a96c4151 | 1779 | return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev)); |
1da177e4 LT |
1780 | } |
1781 | ||
1da177e4 LT |
1782 | xfs_buftarg_t * |
1783 | xfs_alloc_buftarg( | |
ebad861b | 1784 | struct xfs_mount *mp, |
34dcefd7 | 1785 | struct block_device *bdev) |
1da177e4 LT |
1786 | { |
1787 | xfs_buftarg_t *btp; | |
1788 | ||
b17cb364 | 1789 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP | KM_NOFS); |
1da177e4 | 1790 | |
ebad861b | 1791 | btp->bt_mount = mp; |
ce8e922c NS |
1792 | btp->bt_dev = bdev->bd_dev; |
1793 | btp->bt_bdev = bdev; | |
0e6e847f | 1794 | btp->bt_bdi = blk_get_backing_dev_info(bdev); |
0e6e847f | 1795 | |
1da177e4 LT |
1796 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1797 | goto error; | |
5ca302c8 GC |
1798 | |
1799 | if (list_lru_init(&btp->bt_lru)) | |
1800 | goto error; | |
1801 | ||
9c7504aa BF |
1802 | if (percpu_counter_init(&btp->bt_io_count, 0, GFP_KERNEL)) |
1803 | goto error; | |
1804 | ||
e80dfa19 DC |
1805 | btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count; |
1806 | btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan; | |
ff57ab21 | 1807 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; |
e80dfa19 | 1808 | btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE; |
ff57ab21 | 1809 | register_shrinker(&btp->bt_shrinker); |
1da177e4 LT |
1810 | return btp; |
1811 | ||
1812 | error: | |
f0e2d93c | 1813 | kmem_free(btp); |
1da177e4 LT |
1814 | return NULL; |
1815 | } | |
1816 | ||
1da177e4 | 1817 | /* |
43ff2122 CH |
1818 | * Add a buffer to the delayed write list. |
1819 | * | |
1820 | * This queues a buffer for writeout if it hasn't already been. Note that | |
1821 | * neither this routine nor the buffer list submission functions perform | |
1822 | * any internal synchronization. It is expected that the lists are thread-local | |
1823 | * to the callers. | |
1824 | * | |
1825 | * Returns true if we queued up the buffer, or false if it already had | |
1826 | * been on the buffer list. | |
1da177e4 | 1827 | */ |
43ff2122 | 1828 | bool |
ce8e922c | 1829 | xfs_buf_delwri_queue( |
43ff2122 CH |
1830 | struct xfs_buf *bp, |
1831 | struct list_head *list) | |
1da177e4 | 1832 | { |
43ff2122 | 1833 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 1834 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1835 | |
43ff2122 CH |
1836 | /* |
1837 | * If the buffer is already marked delwri it already is queued up | |
1838 | * by someone else for imediate writeout. Just ignore it in that | |
1839 | * case. | |
1840 | */ | |
1841 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
1842 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
1843 | return false; | |
1da177e4 | 1844 | } |
1da177e4 | 1845 | |
43ff2122 | 1846 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
1847 | |
1848 | /* | |
43ff2122 CH |
1849 | * If a buffer gets written out synchronously or marked stale while it |
1850 | * is on a delwri list we lazily remove it. To do this, the other party | |
1851 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
1852 | * It remains referenced and on the list. In a rare corner case it | |
1853 | * might get readded to a delwri list after the synchronous writeout, in | |
1854 | * which case we need just need to re-add the flag here. | |
d808f617 | 1855 | */ |
43ff2122 CH |
1856 | bp->b_flags |= _XBF_DELWRI_Q; |
1857 | if (list_empty(&bp->b_list)) { | |
1858 | atomic_inc(&bp->b_hold); | |
1859 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 1860 | } |
585e6d88 | 1861 | |
43ff2122 | 1862 | return true; |
585e6d88 DC |
1863 | } |
1864 | ||
089716aa DC |
1865 | /* |
1866 | * Compare function is more complex than it needs to be because | |
1867 | * the return value is only 32 bits and we are doing comparisons | |
1868 | * on 64 bit values | |
1869 | */ | |
1870 | static int | |
1871 | xfs_buf_cmp( | |
1872 | void *priv, | |
1873 | struct list_head *a, | |
1874 | struct list_head *b) | |
1875 | { | |
1876 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1877 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1878 | xfs_daddr_t diff; | |
1879 | ||
f4b42421 | 1880 | diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn; |
089716aa DC |
1881 | if (diff < 0) |
1882 | return -1; | |
1883 | if (diff > 0) | |
1884 | return 1; | |
1885 | return 0; | |
1886 | } | |
1887 | ||
26f1fe85 DC |
1888 | /* |
1889 | * submit buffers for write. | |
1890 | * | |
1891 | * When we have a large buffer list, we do not want to hold all the buffers | |
1892 | * locked while we block on the request queue waiting for IO dispatch. To avoid | |
1893 | * this problem, we lock and submit buffers in groups of 50, thereby minimising | |
1894 | * the lock hold times for lists which may contain thousands of objects. | |
1895 | * | |
1896 | * To do this, we sort the buffer list before we walk the list to lock and | |
1897 | * submit buffers, and we plug and unplug around each group of buffers we | |
1898 | * submit. | |
1899 | */ | |
43ff2122 | 1900 | static int |
26f1fe85 | 1901 | xfs_buf_delwri_submit_buffers( |
43ff2122 | 1902 | struct list_head *buffer_list, |
26f1fe85 | 1903 | struct list_head *wait_list) |
1da177e4 | 1904 | { |
43ff2122 | 1905 | struct xfs_buf *bp, *n; |
26f1fe85 | 1906 | LIST_HEAD (submit_list); |
43ff2122 | 1907 | int pinned = 0; |
26f1fe85 | 1908 | struct blk_plug plug; |
43ff2122 | 1909 | |
26f1fe85 | 1910 | list_sort(NULL, buffer_list, xfs_buf_cmp); |
43ff2122 | 1911 | |
26f1fe85 | 1912 | blk_start_plug(&plug); |
43ff2122 | 1913 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { |
26f1fe85 | 1914 | if (!wait_list) { |
43ff2122 CH |
1915 | if (xfs_buf_ispinned(bp)) { |
1916 | pinned++; | |
1917 | continue; | |
1918 | } | |
1919 | if (!xfs_buf_trylock(bp)) | |
1920 | continue; | |
1921 | } else { | |
1922 | xfs_buf_lock(bp); | |
1923 | } | |
978c7b2f | 1924 | |
43ff2122 CH |
1925 | /* |
1926 | * Someone else might have written the buffer synchronously or | |
1927 | * marked it stale in the meantime. In that case only the | |
1928 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
1929 | * reference and remove it from the list here. | |
1930 | */ | |
1931 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
1932 | list_del_init(&bp->b_list); | |
1933 | xfs_buf_relse(bp); | |
1934 | continue; | |
1935 | } | |
c9c12971 | 1936 | |
43ff2122 | 1937 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
a1b7ea5d | 1938 | |
cf53e99d | 1939 | /* |
26f1fe85 DC |
1940 | * We do all IO submission async. This means if we need |
1941 | * to wait for IO completion we need to take an extra | |
1942 | * reference so the buffer is still valid on the other | |
1943 | * side. We need to move the buffer onto the io_list | |
1944 | * at this point so the caller can still access it. | |
cf53e99d | 1945 | */ |
bbfeb614 | 1946 | bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_WRITE_FAIL); |
26f1fe85 DC |
1947 | bp->b_flags |= XBF_WRITE | XBF_ASYNC; |
1948 | if (wait_list) { | |
cf53e99d | 1949 | xfs_buf_hold(bp); |
26f1fe85 DC |
1950 | list_move_tail(&bp->b_list, wait_list); |
1951 | } else | |
ce8e922c | 1952 | list_del_init(&bp->b_list); |
8dac3921 | 1953 | |
595bff75 | 1954 | xfs_buf_submit(bp); |
43ff2122 CH |
1955 | } |
1956 | blk_finish_plug(&plug); | |
1da177e4 | 1957 | |
43ff2122 | 1958 | return pinned; |
1da177e4 LT |
1959 | } |
1960 | ||
1961 | /* | |
43ff2122 CH |
1962 | * Write out a buffer list asynchronously. |
1963 | * | |
1964 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
1965 | * out and not wait for I/O completion on any of the buffers. This interface | |
1966 | * is only safely useable for callers that can track I/O completion by higher | |
1967 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
1968 | * function. | |
1da177e4 LT |
1969 | */ |
1970 | int | |
43ff2122 CH |
1971 | xfs_buf_delwri_submit_nowait( |
1972 | struct list_head *buffer_list) | |
1da177e4 | 1973 | { |
26f1fe85 | 1974 | return xfs_buf_delwri_submit_buffers(buffer_list, NULL); |
43ff2122 | 1975 | } |
1da177e4 | 1976 | |
43ff2122 CH |
1977 | /* |
1978 | * Write out a buffer list synchronously. | |
1979 | * | |
1980 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
1981 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
1982 | * so callers must have some other way of tracking buffers if they require such | |
1983 | * functionality. | |
1984 | */ | |
1985 | int | |
1986 | xfs_buf_delwri_submit( | |
1987 | struct list_head *buffer_list) | |
1988 | { | |
26f1fe85 | 1989 | LIST_HEAD (wait_list); |
43ff2122 CH |
1990 | int error = 0, error2; |
1991 | struct xfs_buf *bp; | |
1da177e4 | 1992 | |
26f1fe85 | 1993 | xfs_buf_delwri_submit_buffers(buffer_list, &wait_list); |
1da177e4 | 1994 | |
43ff2122 | 1995 | /* Wait for IO to complete. */ |
26f1fe85 DC |
1996 | while (!list_empty(&wait_list)) { |
1997 | bp = list_first_entry(&wait_list, struct xfs_buf, b_list); | |
a1b7ea5d | 1998 | |
089716aa | 1999 | list_del_init(&bp->b_list); |
cf53e99d DC |
2000 | |
2001 | /* locking the buffer will wait for async IO completion. */ | |
2002 | xfs_buf_lock(bp); | |
2003 | error2 = bp->b_error; | |
43ff2122 CH |
2004 | xfs_buf_relse(bp); |
2005 | if (!error) | |
2006 | error = error2; | |
1da177e4 LT |
2007 | } |
2008 | ||
43ff2122 | 2009 | return error; |
1da177e4 LT |
2010 | } |
2011 | ||
04d8b284 | 2012 | int __init |
ce8e922c | 2013 | xfs_buf_init(void) |
1da177e4 | 2014 | { |
8758280f NS |
2015 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
2016 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 2017 | if (!xfs_buf_zone) |
0b1b213f | 2018 | goto out; |
04d8b284 | 2019 | |
23ea4032 | 2020 | return 0; |
1da177e4 | 2021 | |
0b1b213f | 2022 | out: |
8758280f | 2023 | return -ENOMEM; |
1da177e4 LT |
2024 | } |
2025 | ||
1da177e4 | 2026 | void |
ce8e922c | 2027 | xfs_buf_terminate(void) |
1da177e4 | 2028 | { |
ce8e922c | 2029 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 2030 | } |