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 | ||
118 | ASSERT(bp->b_flags & XBF_ASYNC); | |
119 | bp->b_flags &= ~_XBF_IN_FLIGHT; | |
120 | percpu_counter_dec(&bp->b_target->bt_io_count); | |
121 | } | |
122 | ||
430cbeb8 DC |
123 | /* |
124 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
125 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
126 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
127 | * to remove the reference that LRU holds on the buffer. | |
128 | * | |
129 | * This prevents build-up of stale buffers on the LRU. | |
130 | */ | |
131 | void | |
132 | xfs_buf_stale( | |
133 | struct xfs_buf *bp) | |
134 | { | |
43ff2122 CH |
135 | ASSERT(xfs_buf_islocked(bp)); |
136 | ||
430cbeb8 | 137 | bp->b_flags |= XBF_STALE; |
43ff2122 CH |
138 | |
139 | /* | |
140 | * Clear the delwri status so that a delwri queue walker will not | |
141 | * flush this buffer to disk now that it is stale. The delwri queue has | |
142 | * a reference to the buffer, so this is safe to do. | |
143 | */ | |
144 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
145 | ||
9c7504aa BF |
146 | /* |
147 | * Once the buffer is marked stale and unlocked, a subsequent lookup | |
148 | * could reset b_flags. There is no guarantee that the buffer is | |
149 | * unaccounted (released to LRU) before that occurs. Drop in-flight | |
150 | * status now to preserve accounting consistency. | |
151 | */ | |
152 | xfs_buf_ioacct_dec(bp); | |
153 | ||
a4082357 DC |
154 | spin_lock(&bp->b_lock); |
155 | atomic_set(&bp->b_lru_ref, 0); | |
156 | if (!(bp->b_state & XFS_BSTATE_DISPOSE) && | |
e80dfa19 DC |
157 | (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru))) |
158 | atomic_dec(&bp->b_hold); | |
159 | ||
430cbeb8 | 160 | ASSERT(atomic_read(&bp->b_hold) >= 1); |
a4082357 | 161 | spin_unlock(&bp->b_lock); |
430cbeb8 | 162 | } |
1da177e4 | 163 | |
3e85c868 DC |
164 | static int |
165 | xfs_buf_get_maps( | |
166 | struct xfs_buf *bp, | |
167 | int map_count) | |
168 | { | |
169 | ASSERT(bp->b_maps == NULL); | |
170 | bp->b_map_count = map_count; | |
171 | ||
172 | if (map_count == 1) { | |
f4b42421 | 173 | bp->b_maps = &bp->__b_map; |
3e85c868 DC |
174 | return 0; |
175 | } | |
176 | ||
177 | bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map), | |
178 | KM_NOFS); | |
179 | if (!bp->b_maps) | |
2451337d | 180 | return -ENOMEM; |
3e85c868 DC |
181 | return 0; |
182 | } | |
183 | ||
184 | /* | |
185 | * Frees b_pages if it was allocated. | |
186 | */ | |
187 | static void | |
188 | xfs_buf_free_maps( | |
189 | struct xfs_buf *bp) | |
190 | { | |
f4b42421 | 191 | if (bp->b_maps != &bp->__b_map) { |
3e85c868 DC |
192 | kmem_free(bp->b_maps); |
193 | bp->b_maps = NULL; | |
194 | } | |
195 | } | |
196 | ||
4347b9d7 | 197 | struct xfs_buf * |
3e85c868 | 198 | _xfs_buf_alloc( |
4347b9d7 | 199 | struct xfs_buftarg *target, |
3e85c868 DC |
200 | struct xfs_buf_map *map, |
201 | int nmaps, | |
ce8e922c | 202 | xfs_buf_flags_t flags) |
1da177e4 | 203 | { |
4347b9d7 | 204 | struct xfs_buf *bp; |
3e85c868 DC |
205 | int error; |
206 | int i; | |
4347b9d7 | 207 | |
aa5c158e | 208 | bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS); |
4347b9d7 CH |
209 | if (unlikely(!bp)) |
210 | return NULL; | |
211 | ||
1da177e4 | 212 | /* |
12bcb3f7 DC |
213 | * We don't want certain flags to appear in b_flags unless they are |
214 | * specifically set by later operations on the buffer. | |
1da177e4 | 215 | */ |
611c9946 | 216 | flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD); |
ce8e922c | 217 | |
ce8e922c | 218 | atomic_set(&bp->b_hold, 1); |
430cbeb8 | 219 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 220 | init_completion(&bp->b_iowait); |
430cbeb8 | 221 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 222 | INIT_LIST_HEAD(&bp->b_list); |
74f75a0c | 223 | RB_CLEAR_NODE(&bp->b_rbnode); |
a731cd11 | 224 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
a4082357 | 225 | spin_lock_init(&bp->b_lock); |
ce8e922c NS |
226 | XB_SET_OWNER(bp); |
227 | bp->b_target = target; | |
3e85c868 | 228 | bp->b_flags = flags; |
de1cbee4 | 229 | |
1da177e4 | 230 | /* |
aa0e8833 DC |
231 | * Set length and io_length to the same value initially. |
232 | * I/O routines should use io_length, which will be the same in | |
1da177e4 LT |
233 | * most cases but may be reset (e.g. XFS recovery). |
234 | */ | |
3e85c868 DC |
235 | error = xfs_buf_get_maps(bp, nmaps); |
236 | if (error) { | |
237 | kmem_zone_free(xfs_buf_zone, bp); | |
238 | return NULL; | |
239 | } | |
240 | ||
241 | bp->b_bn = map[0].bm_bn; | |
242 | bp->b_length = 0; | |
243 | for (i = 0; i < nmaps; i++) { | |
244 | bp->b_maps[i].bm_bn = map[i].bm_bn; | |
245 | bp->b_maps[i].bm_len = map[i].bm_len; | |
246 | bp->b_length += map[i].bm_len; | |
247 | } | |
248 | bp->b_io_length = bp->b_length; | |
249 | ||
ce8e922c NS |
250 | atomic_set(&bp->b_pin_count, 0); |
251 | init_waitqueue_head(&bp->b_waiters); | |
252 | ||
ff6d6af2 | 253 | XFS_STATS_INC(target->bt_mount, xb_create); |
0b1b213f | 254 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 CH |
255 | |
256 | return bp; | |
1da177e4 LT |
257 | } |
258 | ||
259 | /* | |
ce8e922c NS |
260 | * Allocate a page array capable of holding a specified number |
261 | * of pages, and point the page buf at it. | |
1da177e4 LT |
262 | */ |
263 | STATIC int | |
ce8e922c NS |
264 | _xfs_buf_get_pages( |
265 | xfs_buf_t *bp, | |
87937bf8 | 266 | int page_count) |
1da177e4 LT |
267 | { |
268 | /* Make sure that we have a page list */ | |
ce8e922c | 269 | if (bp->b_pages == NULL) { |
ce8e922c NS |
270 | bp->b_page_count = page_count; |
271 | if (page_count <= XB_PAGES) { | |
272 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 273 | } else { |
ce8e922c | 274 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
aa5c158e | 275 | page_count, KM_NOFS); |
ce8e922c | 276 | if (bp->b_pages == NULL) |
1da177e4 LT |
277 | return -ENOMEM; |
278 | } | |
ce8e922c | 279 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
280 | } |
281 | return 0; | |
282 | } | |
283 | ||
284 | /* | |
ce8e922c | 285 | * Frees b_pages if it was allocated. |
1da177e4 LT |
286 | */ |
287 | STATIC void | |
ce8e922c | 288 | _xfs_buf_free_pages( |
1da177e4 LT |
289 | xfs_buf_t *bp) |
290 | { | |
ce8e922c | 291 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 292 | kmem_free(bp->b_pages); |
3fc98b1a | 293 | bp->b_pages = NULL; |
1da177e4 LT |
294 | } |
295 | } | |
296 | ||
297 | /* | |
298 | * Releases the specified buffer. | |
299 | * | |
300 | * The modification state of any associated pages is left unchanged. | |
b46fe825 | 301 | * The buffer must not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
302 | * hashed and refcounted buffers |
303 | */ | |
304 | void | |
ce8e922c | 305 | xfs_buf_free( |
1da177e4 LT |
306 | xfs_buf_t *bp) |
307 | { | |
0b1b213f | 308 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 309 | |
430cbeb8 DC |
310 | ASSERT(list_empty(&bp->b_lru)); |
311 | ||
0e6e847f | 312 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
313 | uint i; |
314 | ||
73c77e2c | 315 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
316 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
317 | bp->b_page_count); | |
1da177e4 | 318 | |
948ecdb4 NS |
319 | for (i = 0; i < bp->b_page_count; i++) { |
320 | struct page *page = bp->b_pages[i]; | |
321 | ||
0e6e847f | 322 | __free_page(page); |
948ecdb4 | 323 | } |
0e6e847f DC |
324 | } else if (bp->b_flags & _XBF_KMEM) |
325 | kmem_free(bp->b_addr); | |
3fc98b1a | 326 | _xfs_buf_free_pages(bp); |
3e85c868 | 327 | xfs_buf_free_maps(bp); |
4347b9d7 | 328 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
329 | } |
330 | ||
331 | /* | |
0e6e847f | 332 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
333 | */ |
334 | STATIC int | |
0e6e847f | 335 | xfs_buf_allocate_memory( |
1da177e4 LT |
336 | xfs_buf_t *bp, |
337 | uint flags) | |
338 | { | |
aa0e8833 | 339 | size_t size; |
1da177e4 | 340 | size_t nbytes, offset; |
ce8e922c | 341 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 342 | unsigned short page_count, i; |
795cac72 | 343 | xfs_off_t start, end; |
1da177e4 LT |
344 | int error; |
345 | ||
0e6e847f DC |
346 | /* |
347 | * for buffers that are contained within a single page, just allocate | |
348 | * the memory from the heap - there's no need for the complexity of | |
349 | * page arrays to keep allocation down to order 0. | |
350 | */ | |
795cac72 DC |
351 | size = BBTOB(bp->b_length); |
352 | if (size < PAGE_SIZE) { | |
aa5c158e | 353 | bp->b_addr = kmem_alloc(size, KM_NOFS); |
0e6e847f DC |
354 | if (!bp->b_addr) { |
355 | /* low memory - use alloc_page loop instead */ | |
356 | goto use_alloc_page; | |
357 | } | |
358 | ||
795cac72 | 359 | if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) != |
0e6e847f DC |
360 | ((unsigned long)bp->b_addr & PAGE_MASK)) { |
361 | /* b_addr spans two pages - use alloc_page instead */ | |
362 | kmem_free(bp->b_addr); | |
363 | bp->b_addr = NULL; | |
364 | goto use_alloc_page; | |
365 | } | |
366 | bp->b_offset = offset_in_page(bp->b_addr); | |
367 | bp->b_pages = bp->b_page_array; | |
368 | bp->b_pages[0] = virt_to_page(bp->b_addr); | |
369 | bp->b_page_count = 1; | |
611c9946 | 370 | bp->b_flags |= _XBF_KMEM; |
0e6e847f DC |
371 | return 0; |
372 | } | |
373 | ||
374 | use_alloc_page: | |
f4b42421 MT |
375 | start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT; |
376 | end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1) | |
cbb7baab | 377 | >> PAGE_SHIFT; |
795cac72 | 378 | page_count = end - start; |
87937bf8 | 379 | error = _xfs_buf_get_pages(bp, page_count); |
1da177e4 LT |
380 | if (unlikely(error)) |
381 | return error; | |
1da177e4 | 382 | |
ce8e922c | 383 | offset = bp->b_offset; |
0e6e847f | 384 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 385 | |
ce8e922c | 386 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
387 | struct page *page; |
388 | uint retries = 0; | |
0e6e847f DC |
389 | retry: |
390 | page = alloc_page(gfp_mask); | |
1da177e4 | 391 | if (unlikely(page == NULL)) { |
ce8e922c NS |
392 | if (flags & XBF_READ_AHEAD) { |
393 | bp->b_page_count = i; | |
2451337d | 394 | error = -ENOMEM; |
0e6e847f | 395 | goto out_free_pages; |
1da177e4 LT |
396 | } |
397 | ||
398 | /* | |
399 | * This could deadlock. | |
400 | * | |
401 | * But until all the XFS lowlevel code is revamped to | |
402 | * handle buffer allocation failures we can't do much. | |
403 | */ | |
404 | if (!(++retries % 100)) | |
4f10700a | 405 | xfs_err(NULL, |
5bf97b1c TH |
406 | "%s(%u) possible memory allocation deadlock in %s (mode:0x%x)", |
407 | current->comm, current->pid, | |
34a622b2 | 408 | __func__, gfp_mask); |
1da177e4 | 409 | |
ff6d6af2 | 410 | XFS_STATS_INC(bp->b_target->bt_mount, xb_page_retries); |
8aa7e847 | 411 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
412 | goto retry; |
413 | } | |
414 | ||
ff6d6af2 | 415 | XFS_STATS_INC(bp->b_target->bt_mount, xb_page_found); |
1da177e4 | 416 | |
0e6e847f | 417 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 418 | size -= nbytes; |
ce8e922c | 419 | bp->b_pages[i] = page; |
1da177e4 LT |
420 | offset = 0; |
421 | } | |
0e6e847f | 422 | return 0; |
1da177e4 | 423 | |
0e6e847f DC |
424 | out_free_pages: |
425 | for (i = 0; i < bp->b_page_count; i++) | |
426 | __free_page(bp->b_pages[i]); | |
1da177e4 LT |
427 | return error; |
428 | } | |
429 | ||
430 | /* | |
25985edc | 431 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
432 | */ |
433 | STATIC int | |
ce8e922c | 434 | _xfs_buf_map_pages( |
1da177e4 LT |
435 | xfs_buf_t *bp, |
436 | uint flags) | |
437 | { | |
0e6e847f | 438 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 439 | if (bp->b_page_count == 1) { |
0e6e847f | 440 | /* A single page buffer is always mappable */ |
ce8e922c | 441 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
611c9946 DC |
442 | } else if (flags & XBF_UNMAPPED) { |
443 | bp->b_addr = NULL; | |
444 | } else { | |
a19fb380 | 445 | int retried = 0; |
ae687e58 DC |
446 | unsigned noio_flag; |
447 | ||
448 | /* | |
449 | * vm_map_ram() will allocate auxillary structures (e.g. | |
450 | * pagetables) with GFP_KERNEL, yet we are likely to be under | |
451 | * GFP_NOFS context here. Hence we need to tell memory reclaim | |
452 | * that we are in such a context via PF_MEMALLOC_NOIO to prevent | |
453 | * memory reclaim re-entering the filesystem here and | |
454 | * potentially deadlocking. | |
455 | */ | |
456 | noio_flag = memalloc_noio_save(); | |
a19fb380 DC |
457 | do { |
458 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
459 | -1, PAGE_KERNEL); | |
460 | if (bp->b_addr) | |
461 | break; | |
462 | vm_unmap_aliases(); | |
463 | } while (retried++ <= 1); | |
ae687e58 | 464 | memalloc_noio_restore(noio_flag); |
a19fb380 DC |
465 | |
466 | if (!bp->b_addr) | |
1da177e4 | 467 | return -ENOMEM; |
ce8e922c | 468 | bp->b_addr += bp->b_offset; |
1da177e4 LT |
469 | } |
470 | ||
471 | return 0; | |
472 | } | |
473 | ||
474 | /* | |
475 | * Finding and Reading Buffers | |
476 | */ | |
477 | ||
478 | /* | |
ce8e922c | 479 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 | 480 | * a given range of an inode. The buffer is returned |
eabbaf11 | 481 | * locked. No I/O is implied by this call. |
1da177e4 LT |
482 | */ |
483 | xfs_buf_t * | |
ce8e922c | 484 | _xfs_buf_find( |
e70b73f8 | 485 | struct xfs_buftarg *btp, |
3e85c868 DC |
486 | struct xfs_buf_map *map, |
487 | int nmaps, | |
ce8e922c NS |
488 | xfs_buf_flags_t flags, |
489 | xfs_buf_t *new_bp) | |
1da177e4 | 490 | { |
74f75a0c DC |
491 | struct xfs_perag *pag; |
492 | struct rb_node **rbp; | |
493 | struct rb_node *parent; | |
494 | xfs_buf_t *bp; | |
3e85c868 | 495 | xfs_daddr_t blkno = map[0].bm_bn; |
10616b80 | 496 | xfs_daddr_t eofs; |
3e85c868 DC |
497 | int numblks = 0; |
498 | int i; | |
1da177e4 | 499 | |
3e85c868 DC |
500 | for (i = 0; i < nmaps; i++) |
501 | numblks += map[i].bm_len; | |
1da177e4 LT |
502 | |
503 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
f79af0b9 | 504 | ASSERT(!(BBTOB(numblks) < btp->bt_meta_sectorsize)); |
6da54179 | 505 | ASSERT(!(BBTOB(blkno) & (xfs_off_t)btp->bt_meta_sectormask)); |
1da177e4 | 506 | |
10616b80 DC |
507 | /* |
508 | * Corrupted block numbers can get through to here, unfortunately, so we | |
509 | * have to check that the buffer falls within the filesystem bounds. | |
510 | */ | |
511 | eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks); | |
db52d09e | 512 | if (blkno < 0 || blkno >= eofs) { |
10616b80 | 513 | /* |
2451337d | 514 | * XXX (dgc): we should really be returning -EFSCORRUPTED here, |
10616b80 DC |
515 | * but none of the higher level infrastructure supports |
516 | * returning a specific error on buffer lookup failures. | |
517 | */ | |
518 | xfs_alert(btp->bt_mount, | |
519 | "%s: Block out of range: block 0x%llx, EOFS 0x%llx ", | |
520 | __func__, blkno, eofs); | |
7bc0dc27 | 521 | WARN_ON(1); |
10616b80 DC |
522 | return NULL; |
523 | } | |
524 | ||
74f75a0c DC |
525 | /* get tree root */ |
526 | pag = xfs_perag_get(btp->bt_mount, | |
e70b73f8 | 527 | xfs_daddr_to_agno(btp->bt_mount, blkno)); |
74f75a0c DC |
528 | |
529 | /* walk tree */ | |
530 | spin_lock(&pag->pag_buf_lock); | |
531 | rbp = &pag->pag_buf_tree.rb_node; | |
532 | parent = NULL; | |
533 | bp = NULL; | |
534 | while (*rbp) { | |
535 | parent = *rbp; | |
536 | bp = rb_entry(parent, struct xfs_buf, b_rbnode); | |
537 | ||
de1cbee4 | 538 | if (blkno < bp->b_bn) |
74f75a0c | 539 | rbp = &(*rbp)->rb_left; |
de1cbee4 | 540 | else if (blkno > bp->b_bn) |
74f75a0c DC |
541 | rbp = &(*rbp)->rb_right; |
542 | else { | |
543 | /* | |
de1cbee4 | 544 | * found a block number match. If the range doesn't |
74f75a0c DC |
545 | * match, the only way this is allowed is if the buffer |
546 | * in the cache is stale and the transaction that made | |
547 | * it stale has not yet committed. i.e. we are | |
548 | * reallocating a busy extent. Skip this buffer and | |
549 | * continue searching to the right for an exact match. | |
550 | */ | |
4e94b71b | 551 | if (bp->b_length != numblks) { |
74f75a0c DC |
552 | ASSERT(bp->b_flags & XBF_STALE); |
553 | rbp = &(*rbp)->rb_right; | |
554 | continue; | |
555 | } | |
ce8e922c | 556 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
557 | goto found; |
558 | } | |
559 | } | |
560 | ||
561 | /* No match found */ | |
ce8e922c | 562 | if (new_bp) { |
74f75a0c DC |
563 | rb_link_node(&new_bp->b_rbnode, parent, rbp); |
564 | rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree); | |
565 | /* the buffer keeps the perag reference until it is freed */ | |
566 | new_bp->b_pag = pag; | |
567 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 568 | } else { |
ff6d6af2 | 569 | XFS_STATS_INC(btp->bt_mount, xb_miss_locked); |
74f75a0c DC |
570 | spin_unlock(&pag->pag_buf_lock); |
571 | xfs_perag_put(pag); | |
1da177e4 | 572 | } |
ce8e922c | 573 | return new_bp; |
1da177e4 LT |
574 | |
575 | found: | |
74f75a0c DC |
576 | spin_unlock(&pag->pag_buf_lock); |
577 | xfs_perag_put(pag); | |
1da177e4 | 578 | |
0c842ad4 CH |
579 | if (!xfs_buf_trylock(bp)) { |
580 | if (flags & XBF_TRYLOCK) { | |
ce8e922c | 581 | xfs_buf_rele(bp); |
ff6d6af2 | 582 | XFS_STATS_INC(btp->bt_mount, xb_busy_locked); |
ce8e922c | 583 | return NULL; |
1da177e4 | 584 | } |
0c842ad4 | 585 | xfs_buf_lock(bp); |
ff6d6af2 | 586 | XFS_STATS_INC(btp->bt_mount, xb_get_locked_waited); |
1da177e4 LT |
587 | } |
588 | ||
0e6e847f DC |
589 | /* |
590 | * if the buffer is stale, clear all the external state associated with | |
591 | * it. We need to keep flags such as how we allocated the buffer memory | |
592 | * intact here. | |
593 | */ | |
ce8e922c NS |
594 | if (bp->b_flags & XBF_STALE) { |
595 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
cfb02852 | 596 | ASSERT(bp->b_iodone == NULL); |
611c9946 | 597 | bp->b_flags &= _XBF_KMEM | _XBF_PAGES; |
1813dd64 | 598 | bp->b_ops = NULL; |
2f926587 | 599 | } |
0b1b213f CH |
600 | |
601 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ff6d6af2 | 602 | XFS_STATS_INC(btp->bt_mount, xb_get_locked); |
ce8e922c | 603 | return bp; |
1da177e4 LT |
604 | } |
605 | ||
606 | /* | |
3815832a DC |
607 | * Assembles a buffer covering the specified range. The code is optimised for |
608 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
609 | * more hits than misses. | |
1da177e4 | 610 | */ |
3815832a | 611 | struct xfs_buf * |
6dde2707 DC |
612 | xfs_buf_get_map( |
613 | struct xfs_buftarg *target, | |
614 | struct xfs_buf_map *map, | |
615 | int nmaps, | |
ce8e922c | 616 | xfs_buf_flags_t flags) |
1da177e4 | 617 | { |
3815832a DC |
618 | struct xfs_buf *bp; |
619 | struct xfs_buf *new_bp; | |
0e6e847f | 620 | int error = 0; |
1da177e4 | 621 | |
6dde2707 | 622 | bp = _xfs_buf_find(target, map, nmaps, flags, NULL); |
3815832a DC |
623 | if (likely(bp)) |
624 | goto found; | |
625 | ||
6dde2707 | 626 | new_bp = _xfs_buf_alloc(target, map, nmaps, flags); |
ce8e922c | 627 | if (unlikely(!new_bp)) |
1da177e4 LT |
628 | return NULL; |
629 | ||
fe2429b0 DC |
630 | error = xfs_buf_allocate_memory(new_bp, flags); |
631 | if (error) { | |
3e85c868 | 632 | xfs_buf_free(new_bp); |
fe2429b0 DC |
633 | return NULL; |
634 | } | |
635 | ||
6dde2707 | 636 | bp = _xfs_buf_find(target, map, nmaps, flags, new_bp); |
3815832a | 637 | if (!bp) { |
fe2429b0 | 638 | xfs_buf_free(new_bp); |
3815832a DC |
639 | return NULL; |
640 | } | |
641 | ||
fe2429b0 DC |
642 | if (bp != new_bp) |
643 | xfs_buf_free(new_bp); | |
1da177e4 | 644 | |
3815832a | 645 | found: |
611c9946 | 646 | if (!bp->b_addr) { |
ce8e922c | 647 | error = _xfs_buf_map_pages(bp, flags); |
1da177e4 | 648 | if (unlikely(error)) { |
4f10700a | 649 | xfs_warn(target->bt_mount, |
08e96e1a | 650 | "%s: failed to map pagesn", __func__); |
a8acad70 DC |
651 | xfs_buf_relse(bp); |
652 | return NULL; | |
1da177e4 LT |
653 | } |
654 | } | |
655 | ||
b79f4a1c DC |
656 | /* |
657 | * Clear b_error if this is a lookup from a caller that doesn't expect | |
658 | * valid data to be found in the buffer. | |
659 | */ | |
660 | if (!(flags & XBF_READ)) | |
661 | xfs_buf_ioerror(bp, 0); | |
662 | ||
ff6d6af2 | 663 | XFS_STATS_INC(target->bt_mount, xb_get); |
0b1b213f | 664 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 665 | return bp; |
1da177e4 LT |
666 | } |
667 | ||
5d765b97 CH |
668 | STATIC int |
669 | _xfs_buf_read( | |
670 | xfs_buf_t *bp, | |
671 | xfs_buf_flags_t flags) | |
672 | { | |
43ff2122 | 673 | ASSERT(!(flags & XBF_WRITE)); |
f4b42421 | 674 | ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL); |
5d765b97 | 675 | |
43ff2122 | 676 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD); |
1d5ae5df | 677 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 678 | |
595bff75 DC |
679 | if (flags & XBF_ASYNC) { |
680 | xfs_buf_submit(bp); | |
0e95f19a | 681 | return 0; |
595bff75 DC |
682 | } |
683 | return xfs_buf_submit_wait(bp); | |
5d765b97 CH |
684 | } |
685 | ||
1da177e4 | 686 | xfs_buf_t * |
6dde2707 DC |
687 | xfs_buf_read_map( |
688 | struct xfs_buftarg *target, | |
689 | struct xfs_buf_map *map, | |
690 | int nmaps, | |
c3f8fc73 | 691 | xfs_buf_flags_t flags, |
1813dd64 | 692 | const struct xfs_buf_ops *ops) |
1da177e4 | 693 | { |
6dde2707 | 694 | struct xfs_buf *bp; |
ce8e922c NS |
695 | |
696 | flags |= XBF_READ; | |
697 | ||
6dde2707 | 698 | bp = xfs_buf_get_map(target, map, nmaps, flags); |
ce8e922c | 699 | if (bp) { |
0b1b213f CH |
700 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
701 | ||
b0388bf1 | 702 | if (!(bp->b_flags & XBF_DONE)) { |
ff6d6af2 | 703 | XFS_STATS_INC(target->bt_mount, xb_get_read); |
1813dd64 | 704 | bp->b_ops = ops; |
5d765b97 | 705 | _xfs_buf_read(bp, flags); |
ce8e922c | 706 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
707 | /* |
708 | * Read ahead call which is already satisfied, | |
709 | * drop the buffer | |
710 | */ | |
a8acad70 DC |
711 | xfs_buf_relse(bp); |
712 | return NULL; | |
1da177e4 | 713 | } else { |
1da177e4 | 714 | /* We do not want read in the flags */ |
ce8e922c | 715 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
716 | } |
717 | } | |
718 | ||
ce8e922c | 719 | return bp; |
1da177e4 LT |
720 | } |
721 | ||
1da177e4 | 722 | /* |
ce8e922c NS |
723 | * If we are not low on memory then do the readahead in a deadlock |
724 | * safe manner. | |
1da177e4 LT |
725 | */ |
726 | void | |
6dde2707 DC |
727 | xfs_buf_readahead_map( |
728 | struct xfs_buftarg *target, | |
729 | struct xfs_buf_map *map, | |
c3f8fc73 | 730 | int nmaps, |
1813dd64 | 731 | const struct xfs_buf_ops *ops) |
1da177e4 | 732 | { |
0e6e847f | 733 | if (bdi_read_congested(target->bt_bdi)) |
1da177e4 LT |
734 | return; |
735 | ||
6dde2707 | 736 | xfs_buf_read_map(target, map, nmaps, |
1813dd64 | 737 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD, ops); |
1da177e4 LT |
738 | } |
739 | ||
5adc94c2 DC |
740 | /* |
741 | * Read an uncached buffer from disk. Allocates and returns a locked | |
742 | * buffer containing the disk contents or nothing. | |
743 | */ | |
ba372674 | 744 | int |
5adc94c2 | 745 | xfs_buf_read_uncached( |
5adc94c2 DC |
746 | struct xfs_buftarg *target, |
747 | xfs_daddr_t daddr, | |
e70b73f8 | 748 | size_t numblks, |
c3f8fc73 | 749 | int flags, |
ba372674 | 750 | struct xfs_buf **bpp, |
1813dd64 | 751 | const struct xfs_buf_ops *ops) |
5adc94c2 | 752 | { |
eab4e633 | 753 | struct xfs_buf *bp; |
5adc94c2 | 754 | |
ba372674 DC |
755 | *bpp = NULL; |
756 | ||
e70b73f8 | 757 | bp = xfs_buf_get_uncached(target, numblks, flags); |
5adc94c2 | 758 | if (!bp) |
ba372674 | 759 | return -ENOMEM; |
5adc94c2 DC |
760 | |
761 | /* set up the buffer for a read IO */ | |
3e85c868 | 762 | ASSERT(bp->b_map_count == 1); |
ba372674 | 763 | bp->b_bn = XFS_BUF_DADDR_NULL; /* always null for uncached buffers */ |
3e85c868 | 764 | bp->b_maps[0].bm_bn = daddr; |
cbb7baab | 765 | bp->b_flags |= XBF_READ; |
1813dd64 | 766 | bp->b_ops = ops; |
5adc94c2 | 767 | |
595bff75 | 768 | xfs_buf_submit_wait(bp); |
ba372674 DC |
769 | if (bp->b_error) { |
770 | int error = bp->b_error; | |
83a0adc3 | 771 | xfs_buf_relse(bp); |
ba372674 | 772 | return error; |
83a0adc3 | 773 | } |
ba372674 DC |
774 | |
775 | *bpp = bp; | |
776 | return 0; | |
1da177e4 LT |
777 | } |
778 | ||
44396476 DC |
779 | /* |
780 | * Return a buffer allocated as an empty buffer and associated to external | |
781 | * memory via xfs_buf_associate_memory() back to it's empty state. | |
782 | */ | |
783 | void | |
784 | xfs_buf_set_empty( | |
785 | struct xfs_buf *bp, | |
e70b73f8 | 786 | size_t numblks) |
44396476 DC |
787 | { |
788 | if (bp->b_pages) | |
789 | _xfs_buf_free_pages(bp); | |
790 | ||
791 | bp->b_pages = NULL; | |
792 | bp->b_page_count = 0; | |
793 | bp->b_addr = NULL; | |
4e94b71b | 794 | bp->b_length = numblks; |
aa0e8833 | 795 | bp->b_io_length = numblks; |
3e85c868 DC |
796 | |
797 | ASSERT(bp->b_map_count == 1); | |
44396476 | 798 | bp->b_bn = XFS_BUF_DADDR_NULL; |
3e85c868 DC |
799 | bp->b_maps[0].bm_bn = XFS_BUF_DADDR_NULL; |
800 | bp->b_maps[0].bm_len = bp->b_length; | |
44396476 DC |
801 | } |
802 | ||
1da177e4 LT |
803 | static inline struct page * |
804 | mem_to_page( | |
805 | void *addr) | |
806 | { | |
9e2779fa | 807 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
808 | return virt_to_page(addr); |
809 | } else { | |
810 | return vmalloc_to_page(addr); | |
811 | } | |
812 | } | |
813 | ||
814 | int | |
ce8e922c NS |
815 | xfs_buf_associate_memory( |
816 | xfs_buf_t *bp, | |
1da177e4 LT |
817 | void *mem, |
818 | size_t len) | |
819 | { | |
820 | int rval; | |
821 | int i = 0; | |
d1afb678 LM |
822 | unsigned long pageaddr; |
823 | unsigned long offset; | |
824 | size_t buflen; | |
1da177e4 LT |
825 | int page_count; |
826 | ||
0e6e847f | 827 | pageaddr = (unsigned long)mem & PAGE_MASK; |
d1afb678 | 828 | offset = (unsigned long)mem - pageaddr; |
0e6e847f DC |
829 | buflen = PAGE_ALIGN(len + offset); |
830 | page_count = buflen >> PAGE_SHIFT; | |
1da177e4 LT |
831 | |
832 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
833 | if (bp->b_pages) |
834 | _xfs_buf_free_pages(bp); | |
1da177e4 | 835 | |
ce8e922c NS |
836 | bp->b_pages = NULL; |
837 | bp->b_addr = mem; | |
1da177e4 | 838 | |
87937bf8 | 839 | rval = _xfs_buf_get_pages(bp, page_count); |
1da177e4 LT |
840 | if (rval) |
841 | return rval; | |
842 | ||
ce8e922c | 843 | bp->b_offset = offset; |
d1afb678 LM |
844 | |
845 | for (i = 0; i < bp->b_page_count; i++) { | |
846 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
0e6e847f | 847 | pageaddr += PAGE_SIZE; |
1da177e4 | 848 | } |
1da177e4 | 849 | |
aa0e8833 | 850 | bp->b_io_length = BTOBB(len); |
4e94b71b | 851 | bp->b_length = BTOBB(buflen); |
1da177e4 LT |
852 | |
853 | return 0; | |
854 | } | |
855 | ||
856 | xfs_buf_t * | |
686865f7 DC |
857 | xfs_buf_get_uncached( |
858 | struct xfs_buftarg *target, | |
e70b73f8 | 859 | size_t numblks, |
686865f7 | 860 | int flags) |
1da177e4 | 861 | { |
e70b73f8 | 862 | unsigned long page_count; |
1fa40b01 | 863 | int error, i; |
3e85c868 DC |
864 | struct xfs_buf *bp; |
865 | DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks); | |
1da177e4 | 866 | |
c891c30a BF |
867 | /* flags might contain irrelevant bits, pass only what we care about */ |
868 | bp = _xfs_buf_alloc(target, &map, 1, flags & XBF_NO_IOACCT); | |
1da177e4 LT |
869 | if (unlikely(bp == NULL)) |
870 | goto fail; | |
1da177e4 | 871 | |
e70b73f8 | 872 | page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT; |
87937bf8 | 873 | error = _xfs_buf_get_pages(bp, page_count); |
1fa40b01 | 874 | if (error) |
1da177e4 LT |
875 | goto fail_free_buf; |
876 | ||
1fa40b01 | 877 | for (i = 0; i < page_count; i++) { |
686865f7 | 878 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
879 | if (!bp->b_pages[i]) |
880 | goto fail_free_mem; | |
1da177e4 | 881 | } |
1fa40b01 | 882 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 883 | |
611c9946 | 884 | error = _xfs_buf_map_pages(bp, 0); |
1fa40b01 | 885 | if (unlikely(error)) { |
4f10700a | 886 | xfs_warn(target->bt_mount, |
08e96e1a | 887 | "%s: failed to map pages", __func__); |
1da177e4 | 888 | goto fail_free_mem; |
1fa40b01 | 889 | } |
1da177e4 | 890 | |
686865f7 | 891 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 892 | return bp; |
1fa40b01 | 893 | |
1da177e4 | 894 | fail_free_mem: |
1fa40b01 CH |
895 | while (--i >= 0) |
896 | __free_page(bp->b_pages[i]); | |
ca165b88 | 897 | _xfs_buf_free_pages(bp); |
1da177e4 | 898 | fail_free_buf: |
3e85c868 | 899 | xfs_buf_free_maps(bp); |
4347b9d7 | 900 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
901 | fail: |
902 | return NULL; | |
903 | } | |
904 | ||
905 | /* | |
1da177e4 LT |
906 | * Increment reference count on buffer, to hold the buffer concurrently |
907 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
908 | * Must hold the buffer already to call this function. |
909 | */ | |
910 | void | |
ce8e922c NS |
911 | xfs_buf_hold( |
912 | xfs_buf_t *bp) | |
1da177e4 | 913 | { |
0b1b213f | 914 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 915 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
916 | } |
917 | ||
918 | /* | |
9c7504aa BF |
919 | * Release a hold on the specified buffer. If the hold count is 1, the buffer is |
920 | * placed on LRU or freed (depending on b_lru_ref). | |
1da177e4 LT |
921 | */ |
922 | void | |
ce8e922c NS |
923 | xfs_buf_rele( |
924 | xfs_buf_t *bp) | |
1da177e4 | 925 | { |
74f75a0c | 926 | struct xfs_perag *pag = bp->b_pag; |
9c7504aa BF |
927 | bool release; |
928 | bool freebuf = false; | |
1da177e4 | 929 | |
0b1b213f | 930 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 931 | |
74f75a0c | 932 | if (!pag) { |
430cbeb8 | 933 | ASSERT(list_empty(&bp->b_lru)); |
74f75a0c | 934 | ASSERT(RB_EMPTY_NODE(&bp->b_rbnode)); |
9c7504aa BF |
935 | if (atomic_dec_and_test(&bp->b_hold)) { |
936 | xfs_buf_ioacct_dec(bp); | |
fad3aa1e | 937 | xfs_buf_free(bp); |
9c7504aa | 938 | } |
fad3aa1e NS |
939 | return; |
940 | } | |
941 | ||
74f75a0c | 942 | ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode)); |
430cbeb8 | 943 | |
3790689f | 944 | ASSERT(atomic_read(&bp->b_hold) > 0); |
a4082357 | 945 | |
9c7504aa BF |
946 | release = atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock); |
947 | spin_lock(&bp->b_lock); | |
948 | if (!release) { | |
949 | /* | |
950 | * Drop the in-flight state if the buffer is already on the LRU | |
951 | * and it holds the only reference. This is racy because we | |
952 | * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT | |
953 | * ensures the decrement occurs only once per-buf. | |
954 | */ | |
955 | if ((atomic_read(&bp->b_hold) == 1) && !list_empty(&bp->b_lru)) | |
956 | xfs_buf_ioacct_dec(bp); | |
957 | goto out_unlock; | |
958 | } | |
959 | ||
960 | /* the last reference has been dropped ... */ | |
961 | xfs_buf_ioacct_dec(bp); | |
962 | if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) { | |
963 | /* | |
964 | * If the buffer is added to the LRU take a new reference to the | |
965 | * buffer for the LRU and clear the (now stale) dispose list | |
966 | * state flag | |
967 | */ | |
968 | if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) { | |
969 | bp->b_state &= ~XFS_BSTATE_DISPOSE; | |
970 | atomic_inc(&bp->b_hold); | |
1da177e4 | 971 | } |
9c7504aa BF |
972 | spin_unlock(&pag->pag_buf_lock); |
973 | } else { | |
974 | /* | |
975 | * most of the time buffers will already be removed from the | |
976 | * LRU, so optimise that case by checking for the | |
977 | * XFS_BSTATE_DISPOSE flag indicating the last list the buffer | |
978 | * was on was the disposal list | |
979 | */ | |
980 | if (!(bp->b_state & XFS_BSTATE_DISPOSE)) { | |
981 | list_lru_del(&bp->b_target->bt_lru, &bp->b_lru); | |
982 | } else { | |
983 | ASSERT(list_empty(&bp->b_lru)); | |
1da177e4 | 984 | } |
9c7504aa BF |
985 | |
986 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); | |
987 | rb_erase(&bp->b_rbnode, &pag->pag_buf_tree); | |
988 | spin_unlock(&pag->pag_buf_lock); | |
989 | xfs_perag_put(pag); | |
990 | freebuf = true; | |
1da177e4 | 991 | } |
9c7504aa BF |
992 | |
993 | out_unlock: | |
994 | spin_unlock(&bp->b_lock); | |
995 | ||
996 | if (freebuf) | |
997 | xfs_buf_free(bp); | |
1da177e4 LT |
998 | } |
999 | ||
1000 | ||
1001 | /* | |
0e6e847f | 1002 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
1003 | * |
1004 | * If we come across a stale, pinned, locked buffer, we know that we are | |
1005 | * being asked to lock a buffer that has been reallocated. Because it is | |
1006 | * pinned, we know that the log has not been pushed to disk and hence it | |
1007 | * will still be locked. Rather than continuing to have trylock attempts | |
1008 | * fail until someone else pushes the log, push it ourselves before | |
1009 | * returning. This means that the xfsaild will not get stuck trying | |
1010 | * to push on stale inode buffers. | |
1da177e4 LT |
1011 | */ |
1012 | int | |
0c842ad4 CH |
1013 | xfs_buf_trylock( |
1014 | struct xfs_buf *bp) | |
1da177e4 LT |
1015 | { |
1016 | int locked; | |
1017 | ||
ce8e922c | 1018 | locked = down_trylock(&bp->b_sema) == 0; |
479c6412 | 1019 | if (locked) { |
ce8e922c | 1020 | XB_SET_OWNER(bp); |
479c6412 DW |
1021 | trace_xfs_buf_trylock(bp, _RET_IP_); |
1022 | } else { | |
1023 | trace_xfs_buf_trylock_fail(bp, _RET_IP_); | |
1024 | } | |
0c842ad4 | 1025 | return locked; |
1da177e4 | 1026 | } |
1da177e4 LT |
1027 | |
1028 | /* | |
0e6e847f | 1029 | * Lock a buffer object. |
ed3b4d6c DC |
1030 | * |
1031 | * If we come across a stale, pinned, locked buffer, we know that we | |
1032 | * are being asked to lock a buffer that has been reallocated. Because | |
1033 | * it is pinned, we know that the log has not been pushed to disk and | |
1034 | * hence it will still be locked. Rather than sleeping until someone | |
1035 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 1036 | */ |
ce8e922c NS |
1037 | void |
1038 | xfs_buf_lock( | |
0c842ad4 | 1039 | struct xfs_buf *bp) |
1da177e4 | 1040 | { |
0b1b213f CH |
1041 | trace_xfs_buf_lock(bp, _RET_IP_); |
1042 | ||
ed3b4d6c | 1043 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 1044 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
1045 | down(&bp->b_sema); |
1046 | XB_SET_OWNER(bp); | |
0b1b213f CH |
1047 | |
1048 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
1049 | } |
1050 | ||
1da177e4 | 1051 | void |
ce8e922c | 1052 | xfs_buf_unlock( |
0c842ad4 | 1053 | struct xfs_buf *bp) |
1da177e4 | 1054 | { |
ce8e922c NS |
1055 | XB_CLEAR_OWNER(bp); |
1056 | up(&bp->b_sema); | |
0b1b213f CH |
1057 | |
1058 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
1059 | } |
1060 | ||
ce8e922c NS |
1061 | STATIC void |
1062 | xfs_buf_wait_unpin( | |
1063 | xfs_buf_t *bp) | |
1da177e4 LT |
1064 | { |
1065 | DECLARE_WAITQUEUE (wait, current); | |
1066 | ||
ce8e922c | 1067 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
1068 | return; |
1069 | ||
ce8e922c | 1070 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1071 | for (;;) { |
1072 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 1073 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 1074 | break; |
7eaceacc | 1075 | io_schedule(); |
1da177e4 | 1076 | } |
ce8e922c | 1077 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1078 | set_current_state(TASK_RUNNING); |
1079 | } | |
1080 | ||
1081 | /* | |
1082 | * Buffer Utility Routines | |
1083 | */ | |
1084 | ||
e8aaba9a DC |
1085 | void |
1086 | xfs_buf_ioend( | |
1087 | struct xfs_buf *bp) | |
1da177e4 | 1088 | { |
e8aaba9a DC |
1089 | bool read = bp->b_flags & XBF_READ; |
1090 | ||
1091 | trace_xfs_buf_iodone(bp, _RET_IP_); | |
1813dd64 DC |
1092 | |
1093 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); | |
d5929de8 | 1094 | |
61be9c52 DC |
1095 | /* |
1096 | * Pull in IO completion errors now. We are guaranteed to be running | |
1097 | * single threaded, so we don't need the lock to read b_io_error. | |
1098 | */ | |
1099 | if (!bp->b_error && bp->b_io_error) | |
1100 | xfs_buf_ioerror(bp, bp->b_io_error); | |
1101 | ||
e8aaba9a DC |
1102 | /* Only validate buffers that were read without errors */ |
1103 | if (read && !bp->b_error && bp->b_ops) { | |
1104 | ASSERT(!bp->b_iodone); | |
1813dd64 | 1105 | bp->b_ops->verify_read(bp); |
e8aaba9a DC |
1106 | } |
1107 | ||
1108 | if (!bp->b_error) | |
1109 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1110 | |
80f6c29d | 1111 | if (bp->b_iodone) |
ce8e922c NS |
1112 | (*(bp->b_iodone))(bp); |
1113 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 | 1114 | xfs_buf_relse(bp); |
595bff75 | 1115 | else |
1813dd64 | 1116 | complete(&bp->b_iowait); |
1da177e4 LT |
1117 | } |
1118 | ||
e8aaba9a DC |
1119 | static void |
1120 | xfs_buf_ioend_work( | |
1121 | struct work_struct *work) | |
1da177e4 | 1122 | { |
e8aaba9a | 1123 | struct xfs_buf *bp = |
b29c70f5 | 1124 | container_of(work, xfs_buf_t, b_ioend_work); |
0b1b213f | 1125 | |
e8aaba9a DC |
1126 | xfs_buf_ioend(bp); |
1127 | } | |
1da177e4 | 1128 | |
211fe1a4 | 1129 | static void |
e8aaba9a DC |
1130 | xfs_buf_ioend_async( |
1131 | struct xfs_buf *bp) | |
1132 | { | |
b29c70f5 BF |
1133 | INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work); |
1134 | queue_work(bp->b_ioend_wq, &bp->b_ioend_work); | |
1da177e4 LT |
1135 | } |
1136 | ||
1da177e4 | 1137 | void |
ce8e922c NS |
1138 | xfs_buf_ioerror( |
1139 | xfs_buf_t *bp, | |
1140 | int error) | |
1da177e4 | 1141 | { |
2451337d DC |
1142 | ASSERT(error <= 0 && error >= -1000); |
1143 | bp->b_error = error; | |
0b1b213f | 1144 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1145 | } |
1146 | ||
901796af CH |
1147 | void |
1148 | xfs_buf_ioerror_alert( | |
1149 | struct xfs_buf *bp, | |
1150 | const char *func) | |
1151 | { | |
1152 | xfs_alert(bp->b_target->bt_mount, | |
aa0e8833 | 1153 | "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d", |
2451337d | 1154 | (__uint64_t)XFS_BUF_ADDR(bp), func, -bp->b_error, bp->b_length); |
901796af CH |
1155 | } |
1156 | ||
a2dcf5df CH |
1157 | int |
1158 | xfs_bwrite( | |
1159 | struct xfs_buf *bp) | |
1160 | { | |
1161 | int error; | |
1162 | ||
1163 | ASSERT(xfs_buf_islocked(bp)); | |
1164 | ||
1165 | bp->b_flags |= XBF_WRITE; | |
27187754 DC |
1166 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q | |
1167 | XBF_WRITE_FAIL | XBF_DONE); | |
a2dcf5df | 1168 | |
595bff75 | 1169 | error = xfs_buf_submit_wait(bp); |
a2dcf5df CH |
1170 | if (error) { |
1171 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1172 | SHUTDOWN_META_IO_ERROR); | |
1173 | } | |
1174 | return error; | |
1175 | } | |
1176 | ||
9bdd9bd6 | 1177 | static void |
ce8e922c | 1178 | xfs_buf_bio_end_io( |
4246a0b6 | 1179 | struct bio *bio) |
1da177e4 | 1180 | { |
9bdd9bd6 | 1181 | struct xfs_buf *bp = (struct xfs_buf *)bio->bi_private; |
1da177e4 | 1182 | |
37eb17e6 DC |
1183 | /* |
1184 | * don't overwrite existing errors - otherwise we can lose errors on | |
1185 | * buffers that require multiple bios to complete. | |
1186 | */ | |
9bdd9bd6 BF |
1187 | if (bio->bi_error) |
1188 | cmpxchg(&bp->b_io_error, 0, bio->bi_error); | |
1da177e4 | 1189 | |
37eb17e6 | 1190 | if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
73c77e2c JB |
1191 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); |
1192 | ||
e8aaba9a DC |
1193 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
1194 | xfs_buf_ioend_async(bp); | |
1da177e4 | 1195 | bio_put(bio); |
1da177e4 LT |
1196 | } |
1197 | ||
3e85c868 DC |
1198 | static void |
1199 | xfs_buf_ioapply_map( | |
1200 | struct xfs_buf *bp, | |
1201 | int map, | |
1202 | int *buf_offset, | |
1203 | int *count, | |
50bfcd0c MC |
1204 | int op, |
1205 | int op_flags) | |
1da177e4 | 1206 | { |
3e85c868 DC |
1207 | int page_index; |
1208 | int total_nr_pages = bp->b_page_count; | |
1209 | int nr_pages; | |
1210 | struct bio *bio; | |
1211 | sector_t sector = bp->b_maps[map].bm_bn; | |
1212 | int size; | |
1213 | int offset; | |
1da177e4 | 1214 | |
ce8e922c | 1215 | total_nr_pages = bp->b_page_count; |
1da177e4 | 1216 | |
3e85c868 DC |
1217 | /* skip the pages in the buffer before the start offset */ |
1218 | page_index = 0; | |
1219 | offset = *buf_offset; | |
1220 | while (offset >= PAGE_SIZE) { | |
1221 | page_index++; | |
1222 | offset -= PAGE_SIZE; | |
f538d4da CH |
1223 | } |
1224 | ||
3e85c868 DC |
1225 | /* |
1226 | * Limit the IO size to the length of the current vector, and update the | |
1227 | * remaining IO count for the next time around. | |
1228 | */ | |
1229 | size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count); | |
1230 | *count -= size; | |
1231 | *buf_offset += size; | |
34951f5c | 1232 | |
1da177e4 | 1233 | next_chunk: |
ce8e922c | 1234 | atomic_inc(&bp->b_io_remaining); |
c908e380 | 1235 | nr_pages = min(total_nr_pages, BIO_MAX_PAGES); |
1da177e4 LT |
1236 | |
1237 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1238 | bio->bi_bdev = bp->b_target->bt_bdev; |
4f024f37 | 1239 | bio->bi_iter.bi_sector = sector; |
ce8e922c NS |
1240 | bio->bi_end_io = xfs_buf_bio_end_io; |
1241 | bio->bi_private = bp; | |
50bfcd0c | 1242 | bio_set_op_attrs(bio, op, op_flags); |
0e6e847f | 1243 | |
3e85c868 | 1244 | for (; size && nr_pages; nr_pages--, page_index++) { |
0e6e847f | 1245 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1246 | |
1247 | if (nbytes > size) | |
1248 | nbytes = size; | |
1249 | ||
3e85c868 DC |
1250 | rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes, |
1251 | offset); | |
ce8e922c | 1252 | if (rbytes < nbytes) |
1da177e4 LT |
1253 | break; |
1254 | ||
1255 | offset = 0; | |
aa0e8833 | 1256 | sector += BTOBB(nbytes); |
1da177e4 LT |
1257 | size -= nbytes; |
1258 | total_nr_pages--; | |
1259 | } | |
1260 | ||
4f024f37 | 1261 | if (likely(bio->bi_iter.bi_size)) { |
73c77e2c JB |
1262 | if (xfs_buf_is_vmapped(bp)) { |
1263 | flush_kernel_vmap_range(bp->b_addr, | |
1264 | xfs_buf_vmap_len(bp)); | |
1265 | } | |
4e49ea4a | 1266 | submit_bio(bio); |
1da177e4 LT |
1267 | if (size) |
1268 | goto next_chunk; | |
1269 | } else { | |
37eb17e6 DC |
1270 | /* |
1271 | * This is guaranteed not to be the last io reference count | |
595bff75 | 1272 | * because the caller (xfs_buf_submit) holds a count itself. |
37eb17e6 DC |
1273 | */ |
1274 | atomic_dec(&bp->b_io_remaining); | |
2451337d | 1275 | xfs_buf_ioerror(bp, -EIO); |
ec53d1db | 1276 | bio_put(bio); |
1da177e4 | 1277 | } |
3e85c868 DC |
1278 | |
1279 | } | |
1280 | ||
1281 | STATIC void | |
1282 | _xfs_buf_ioapply( | |
1283 | struct xfs_buf *bp) | |
1284 | { | |
1285 | struct blk_plug plug; | |
50bfcd0c MC |
1286 | int op; |
1287 | int op_flags = 0; | |
3e85c868 DC |
1288 | int offset; |
1289 | int size; | |
1290 | int i; | |
1291 | ||
c163f9a1 DC |
1292 | /* |
1293 | * Make sure we capture only current IO errors rather than stale errors | |
1294 | * left over from previous use of the buffer (e.g. failed readahead). | |
1295 | */ | |
1296 | bp->b_error = 0; | |
1297 | ||
b29c70f5 BF |
1298 | /* |
1299 | * Initialize the I/O completion workqueue if we haven't yet or the | |
1300 | * submitter has not opted to specify a custom one. | |
1301 | */ | |
1302 | if (!bp->b_ioend_wq) | |
1303 | bp->b_ioend_wq = bp->b_target->bt_mount->m_buf_workqueue; | |
1304 | ||
3e85c868 | 1305 | if (bp->b_flags & XBF_WRITE) { |
50bfcd0c | 1306 | op = REQ_OP_WRITE; |
3e85c868 | 1307 | if (bp->b_flags & XBF_SYNCIO) |
50bfcd0c | 1308 | op_flags = WRITE_SYNC; |
3e85c868 | 1309 | if (bp->b_flags & XBF_FUA) |
50bfcd0c | 1310 | op_flags |= REQ_FUA; |
3e85c868 | 1311 | if (bp->b_flags & XBF_FLUSH) |
28a8f0d3 | 1312 | op_flags |= REQ_PREFLUSH; |
1813dd64 DC |
1313 | |
1314 | /* | |
1315 | * Run the write verifier callback function if it exists. If | |
1316 | * this function fails it will mark the buffer with an error and | |
1317 | * the IO should not be dispatched. | |
1318 | */ | |
1319 | if (bp->b_ops) { | |
1320 | bp->b_ops->verify_write(bp); | |
1321 | if (bp->b_error) { | |
1322 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1323 | SHUTDOWN_CORRUPT_INCORE); | |
1324 | return; | |
1325 | } | |
400b9d88 DC |
1326 | } else if (bp->b_bn != XFS_BUF_DADDR_NULL) { |
1327 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
1328 | ||
1329 | /* | |
1330 | * non-crc filesystems don't attach verifiers during | |
1331 | * log recovery, so don't warn for such filesystems. | |
1332 | */ | |
1333 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
1334 | xfs_warn(mp, | |
1335 | "%s: no ops on block 0x%llx/0x%x", | |
1336 | __func__, bp->b_bn, bp->b_length); | |
1337 | xfs_hex_dump(bp->b_addr, 64); | |
1338 | dump_stack(); | |
1339 | } | |
1813dd64 | 1340 | } |
3e85c868 | 1341 | } else if (bp->b_flags & XBF_READ_AHEAD) { |
50bfcd0c MC |
1342 | op = REQ_OP_READ; |
1343 | op_flags = REQ_RAHEAD; | |
3e85c868 | 1344 | } else { |
50bfcd0c | 1345 | op = REQ_OP_READ; |
3e85c868 DC |
1346 | } |
1347 | ||
1348 | /* we only use the buffer cache for meta-data */ | |
50bfcd0c | 1349 | op_flags |= REQ_META; |
3e85c868 DC |
1350 | |
1351 | /* | |
1352 | * Walk all the vectors issuing IO on them. Set up the initial offset | |
1353 | * into the buffer and the desired IO size before we start - | |
1354 | * _xfs_buf_ioapply_vec() will modify them appropriately for each | |
1355 | * subsequent call. | |
1356 | */ | |
1357 | offset = bp->b_offset; | |
1358 | size = BBTOB(bp->b_io_length); | |
1359 | blk_start_plug(&plug); | |
1360 | for (i = 0; i < bp->b_map_count; i++) { | |
50bfcd0c | 1361 | xfs_buf_ioapply_map(bp, i, &offset, &size, op, op_flags); |
3e85c868 DC |
1362 | if (bp->b_error) |
1363 | break; | |
1364 | if (size <= 0) | |
1365 | break; /* all done */ | |
1366 | } | |
1367 | blk_finish_plug(&plug); | |
1da177e4 LT |
1368 | } |
1369 | ||
595bff75 DC |
1370 | /* |
1371 | * Asynchronous IO submission path. This transfers the buffer lock ownership and | |
1372 | * the current reference to the IO. It is not safe to reference the buffer after | |
1373 | * a call to this function unless the caller holds an additional reference | |
1374 | * itself. | |
1375 | */ | |
0e95f19a | 1376 | void |
595bff75 DC |
1377 | xfs_buf_submit( |
1378 | struct xfs_buf *bp) | |
1da177e4 | 1379 | { |
595bff75 | 1380 | trace_xfs_buf_submit(bp, _RET_IP_); |
1da177e4 | 1381 | |
43ff2122 | 1382 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
595bff75 DC |
1383 | ASSERT(bp->b_flags & XBF_ASYNC); |
1384 | ||
1385 | /* on shutdown we stale and complete the buffer immediately */ | |
1386 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { | |
1387 | xfs_buf_ioerror(bp, -EIO); | |
1388 | bp->b_flags &= ~XBF_DONE; | |
1389 | xfs_buf_stale(bp); | |
1390 | xfs_buf_ioend(bp); | |
1391 | return; | |
1392 | } | |
1da177e4 | 1393 | |
375ec69d | 1394 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1395 | xfs_buf_wait_unpin(bp); |
e11bb805 | 1396 | |
61be9c52 DC |
1397 | /* clear the internal error state to avoid spurious errors */ |
1398 | bp->b_io_error = 0; | |
1399 | ||
e11bb805 | 1400 | /* |
595bff75 DC |
1401 | * The caller's reference is released during I/O completion. |
1402 | * This occurs some time after the last b_io_remaining reference is | |
1403 | * released, so after we drop our Io reference we have to have some | |
1404 | * other reference to ensure the buffer doesn't go away from underneath | |
1405 | * us. Take a direct reference to ensure we have safe access to the | |
1406 | * buffer until we are finished with it. | |
e11bb805 | 1407 | */ |
ce8e922c | 1408 | xfs_buf_hold(bp); |
1da177e4 | 1409 | |
8d6c1210 | 1410 | /* |
e11bb805 DC |
1411 | * Set the count to 1 initially, this will stop an I/O completion |
1412 | * callout which happens before we have started all the I/O from calling | |
1413 | * xfs_buf_ioend too early. | |
1da177e4 | 1414 | */ |
ce8e922c | 1415 | atomic_set(&bp->b_io_remaining, 1); |
9c7504aa | 1416 | xfs_buf_ioacct_inc(bp); |
ce8e922c | 1417 | _xfs_buf_ioapply(bp); |
e11bb805 | 1418 | |
8d6c1210 | 1419 | /* |
595bff75 DC |
1420 | * If _xfs_buf_ioapply failed, we can get back here with only the IO |
1421 | * reference we took above. If we drop it to zero, run completion so | |
1422 | * that we don't return to the caller with completion still pending. | |
8d6c1210 | 1423 | */ |
e8aaba9a | 1424 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
595bff75 | 1425 | if (bp->b_error) |
e8aaba9a DC |
1426 | xfs_buf_ioend(bp); |
1427 | else | |
1428 | xfs_buf_ioend_async(bp); | |
1429 | } | |
1da177e4 | 1430 | |
ce8e922c | 1431 | xfs_buf_rele(bp); |
595bff75 | 1432 | /* Note: it is not safe to reference bp now we've dropped our ref */ |
1da177e4 LT |
1433 | } |
1434 | ||
1435 | /* | |
595bff75 | 1436 | * Synchronous buffer IO submission path, read or write. |
1da177e4 LT |
1437 | */ |
1438 | int | |
595bff75 DC |
1439 | xfs_buf_submit_wait( |
1440 | struct xfs_buf *bp) | |
1da177e4 | 1441 | { |
595bff75 | 1442 | int error; |
0b1b213f | 1443 | |
595bff75 DC |
1444 | trace_xfs_buf_submit_wait(bp, _RET_IP_); |
1445 | ||
1446 | ASSERT(!(bp->b_flags & (_XBF_DELWRI_Q | XBF_ASYNC))); | |
0b1b213f | 1447 | |
595bff75 DC |
1448 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
1449 | xfs_buf_ioerror(bp, -EIO); | |
1450 | xfs_buf_stale(bp); | |
1451 | bp->b_flags &= ~XBF_DONE; | |
1452 | return -EIO; | |
1453 | } | |
1454 | ||
1455 | if (bp->b_flags & XBF_WRITE) | |
1456 | xfs_buf_wait_unpin(bp); | |
1457 | ||
1458 | /* clear the internal error state to avoid spurious errors */ | |
1459 | bp->b_io_error = 0; | |
1460 | ||
1461 | /* | |
1462 | * For synchronous IO, the IO does not inherit the submitters reference | |
1463 | * count, nor the buffer lock. Hence we cannot release the reference we | |
1464 | * are about to take until we've waited for all IO completion to occur, | |
1465 | * including any xfs_buf_ioend_async() work that may be pending. | |
1466 | */ | |
1467 | xfs_buf_hold(bp); | |
1468 | ||
1469 | /* | |
1470 | * Set the count to 1 initially, this will stop an I/O completion | |
1471 | * callout which happens before we have started all the I/O from calling | |
1472 | * xfs_buf_ioend too early. | |
1473 | */ | |
1474 | atomic_set(&bp->b_io_remaining, 1); | |
1475 | _xfs_buf_ioapply(bp); | |
1476 | ||
1477 | /* | |
1478 | * make sure we run completion synchronously if it raced with us and is | |
1479 | * already complete. | |
1480 | */ | |
1481 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) | |
1482 | xfs_buf_ioend(bp); | |
0b1b213f | 1483 | |
595bff75 DC |
1484 | /* wait for completion before gathering the error from the buffer */ |
1485 | trace_xfs_buf_iowait(bp, _RET_IP_); | |
1486 | wait_for_completion(&bp->b_iowait); | |
0b1b213f | 1487 | trace_xfs_buf_iowait_done(bp, _RET_IP_); |
595bff75 DC |
1488 | error = bp->b_error; |
1489 | ||
1490 | /* | |
1491 | * all done now, we can release the hold that keeps the buffer | |
1492 | * referenced for the entire IO. | |
1493 | */ | |
1494 | xfs_buf_rele(bp); | |
1495 | return error; | |
1da177e4 LT |
1496 | } |
1497 | ||
88ee2df7 | 1498 | void * |
ce8e922c | 1499 | xfs_buf_offset( |
88ee2df7 | 1500 | struct xfs_buf *bp, |
1da177e4 LT |
1501 | size_t offset) |
1502 | { | |
1503 | struct page *page; | |
1504 | ||
611c9946 | 1505 | if (bp->b_addr) |
62926044 | 1506 | return bp->b_addr + offset; |
1da177e4 | 1507 | |
ce8e922c | 1508 | offset += bp->b_offset; |
0e6e847f | 1509 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
88ee2df7 | 1510 | return page_address(page) + (offset & (PAGE_SIZE-1)); |
1da177e4 LT |
1511 | } |
1512 | ||
1513 | /* | |
1da177e4 LT |
1514 | * Move data into or out of a buffer. |
1515 | */ | |
1516 | void | |
ce8e922c NS |
1517 | xfs_buf_iomove( |
1518 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1519 | size_t boff, /* starting buffer offset */ |
1520 | size_t bsize, /* length to copy */ | |
b9c48649 | 1521 | void *data, /* data address */ |
ce8e922c | 1522 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 | 1523 | { |
795cac72 | 1524 | size_t bend; |
1da177e4 LT |
1525 | |
1526 | bend = boff + bsize; | |
1527 | while (boff < bend) { | |
795cac72 DC |
1528 | struct page *page; |
1529 | int page_index, page_offset, csize; | |
1530 | ||
1531 | page_index = (boff + bp->b_offset) >> PAGE_SHIFT; | |
1532 | page_offset = (boff + bp->b_offset) & ~PAGE_MASK; | |
1533 | page = bp->b_pages[page_index]; | |
1534 | csize = min_t(size_t, PAGE_SIZE - page_offset, | |
1535 | BBTOB(bp->b_io_length) - boff); | |
1da177e4 | 1536 | |
795cac72 | 1537 | ASSERT((csize + page_offset) <= PAGE_SIZE); |
1da177e4 LT |
1538 | |
1539 | switch (mode) { | |
ce8e922c | 1540 | case XBRW_ZERO: |
795cac72 | 1541 | memset(page_address(page) + page_offset, 0, csize); |
1da177e4 | 1542 | break; |
ce8e922c | 1543 | case XBRW_READ: |
795cac72 | 1544 | memcpy(data, page_address(page) + page_offset, csize); |
1da177e4 | 1545 | break; |
ce8e922c | 1546 | case XBRW_WRITE: |
795cac72 | 1547 | memcpy(page_address(page) + page_offset, data, csize); |
1da177e4 LT |
1548 | } |
1549 | ||
1550 | boff += csize; | |
1551 | data += csize; | |
1552 | } | |
1553 | } | |
1554 | ||
1555 | /* | |
ce8e922c | 1556 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1557 | */ |
1558 | ||
1559 | /* | |
430cbeb8 DC |
1560 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1561 | * returned. These buffers will have an elevated hold count, so wait on those | |
1562 | * while freeing all the buffers only held by the LRU. | |
1da177e4 | 1563 | */ |
e80dfa19 DC |
1564 | static enum lru_status |
1565 | xfs_buftarg_wait_rele( | |
1566 | struct list_head *item, | |
3f97b163 | 1567 | struct list_lru_one *lru, |
e80dfa19 DC |
1568 | spinlock_t *lru_lock, |
1569 | void *arg) | |
1570 | ||
1da177e4 | 1571 | { |
e80dfa19 | 1572 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); |
a4082357 | 1573 | struct list_head *dispose = arg; |
430cbeb8 | 1574 | |
e80dfa19 | 1575 | if (atomic_read(&bp->b_hold) > 1) { |
a4082357 | 1576 | /* need to wait, so skip it this pass */ |
e80dfa19 | 1577 | trace_xfs_buf_wait_buftarg(bp, _RET_IP_); |
a4082357 | 1578 | return LRU_SKIP; |
1da177e4 | 1579 | } |
a4082357 DC |
1580 | if (!spin_trylock(&bp->b_lock)) |
1581 | return LRU_SKIP; | |
e80dfa19 | 1582 | |
a4082357 DC |
1583 | /* |
1584 | * clear the LRU reference count so the buffer doesn't get | |
1585 | * ignored in xfs_buf_rele(). | |
1586 | */ | |
1587 | atomic_set(&bp->b_lru_ref, 0); | |
1588 | bp->b_state |= XFS_BSTATE_DISPOSE; | |
3f97b163 | 1589 | list_lru_isolate_move(lru, item, dispose); |
a4082357 DC |
1590 | spin_unlock(&bp->b_lock); |
1591 | return LRU_REMOVED; | |
1da177e4 LT |
1592 | } |
1593 | ||
e80dfa19 DC |
1594 | void |
1595 | xfs_wait_buftarg( | |
1596 | struct xfs_buftarg *btp) | |
1597 | { | |
a4082357 DC |
1598 | LIST_HEAD(dispose); |
1599 | int loop = 0; | |
1600 | ||
85bec546 | 1601 | /* |
9c7504aa BF |
1602 | * First wait on the buftarg I/O count for all in-flight buffers to be |
1603 | * released. This is critical as new buffers do not make the LRU until | |
1604 | * they are released. | |
1605 | * | |
1606 | * Next, flush the buffer workqueue to ensure all completion processing | |
1607 | * has finished. Just waiting on buffer locks is not sufficient for | |
1608 | * async IO as the reference count held over IO is not released until | |
1609 | * after the buffer lock is dropped. Hence we need to ensure here that | |
1610 | * all reference counts have been dropped before we start walking the | |
1611 | * LRU list. | |
85bec546 | 1612 | */ |
9c7504aa BF |
1613 | while (percpu_counter_sum(&btp->bt_io_count)) |
1614 | delay(100); | |
85bec546 DC |
1615 | drain_workqueue(btp->bt_mount->m_buf_workqueue); |
1616 | ||
a4082357 DC |
1617 | /* loop until there is nothing left on the lru list. */ |
1618 | while (list_lru_count(&btp->bt_lru)) { | |
e80dfa19 | 1619 | list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele, |
a4082357 DC |
1620 | &dispose, LONG_MAX); |
1621 | ||
1622 | while (!list_empty(&dispose)) { | |
1623 | struct xfs_buf *bp; | |
1624 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1625 | list_del_init(&bp->b_lru); | |
ac8809f9 DC |
1626 | if (bp->b_flags & XBF_WRITE_FAIL) { |
1627 | xfs_alert(btp->bt_mount, | |
f41febd2 | 1628 | "Corruption Alert: Buffer at block 0x%llx had permanent write failures!", |
ac8809f9 | 1629 | (long long)bp->b_bn); |
f41febd2 JP |
1630 | xfs_alert(btp->bt_mount, |
1631 | "Please run xfs_repair to determine the extent of the problem."); | |
ac8809f9 | 1632 | } |
a4082357 DC |
1633 | xfs_buf_rele(bp); |
1634 | } | |
1635 | if (loop++ != 0) | |
1636 | delay(100); | |
1637 | } | |
e80dfa19 DC |
1638 | } |
1639 | ||
1640 | static enum lru_status | |
1641 | xfs_buftarg_isolate( | |
1642 | struct list_head *item, | |
3f97b163 | 1643 | struct list_lru_one *lru, |
e80dfa19 DC |
1644 | spinlock_t *lru_lock, |
1645 | void *arg) | |
1646 | { | |
1647 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); | |
1648 | struct list_head *dispose = arg; | |
1649 | ||
a4082357 DC |
1650 | /* |
1651 | * we are inverting the lru lock/bp->b_lock here, so use a trylock. | |
1652 | * If we fail to get the lock, just skip it. | |
1653 | */ | |
1654 | if (!spin_trylock(&bp->b_lock)) | |
1655 | return LRU_SKIP; | |
e80dfa19 DC |
1656 | /* |
1657 | * Decrement the b_lru_ref count unless the value is already | |
1658 | * zero. If the value is already zero, we need to reclaim the | |
1659 | * buffer, otherwise it gets another trip through the LRU. | |
1660 | */ | |
a4082357 DC |
1661 | if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) { |
1662 | spin_unlock(&bp->b_lock); | |
e80dfa19 | 1663 | return LRU_ROTATE; |
a4082357 | 1664 | } |
e80dfa19 | 1665 | |
a4082357 | 1666 | bp->b_state |= XFS_BSTATE_DISPOSE; |
3f97b163 | 1667 | list_lru_isolate_move(lru, item, dispose); |
a4082357 | 1668 | spin_unlock(&bp->b_lock); |
e80dfa19 DC |
1669 | return LRU_REMOVED; |
1670 | } | |
1671 | ||
addbda40 | 1672 | static unsigned long |
e80dfa19 | 1673 | xfs_buftarg_shrink_scan( |
ff57ab21 | 1674 | struct shrinker *shrink, |
1495f230 | 1675 | struct shrink_control *sc) |
a6867a68 | 1676 | { |
ff57ab21 DC |
1677 | struct xfs_buftarg *btp = container_of(shrink, |
1678 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1679 | LIST_HEAD(dispose); |
addbda40 | 1680 | unsigned long freed; |
430cbeb8 | 1681 | |
503c358c VD |
1682 | freed = list_lru_shrink_walk(&btp->bt_lru, sc, |
1683 | xfs_buftarg_isolate, &dispose); | |
430cbeb8 DC |
1684 | |
1685 | while (!list_empty(&dispose)) { | |
e80dfa19 | 1686 | struct xfs_buf *bp; |
430cbeb8 DC |
1687 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); |
1688 | list_del_init(&bp->b_lru); | |
1689 | xfs_buf_rele(bp); | |
1690 | } | |
1691 | ||
e80dfa19 DC |
1692 | return freed; |
1693 | } | |
1694 | ||
addbda40 | 1695 | static unsigned long |
e80dfa19 DC |
1696 | xfs_buftarg_shrink_count( |
1697 | struct shrinker *shrink, | |
1698 | struct shrink_control *sc) | |
1699 | { | |
1700 | struct xfs_buftarg *btp = container_of(shrink, | |
1701 | struct xfs_buftarg, bt_shrinker); | |
503c358c | 1702 | return list_lru_shrink_count(&btp->bt_lru, sc); |
a6867a68 DC |
1703 | } |
1704 | ||
1da177e4 LT |
1705 | void |
1706 | xfs_free_buftarg( | |
b7963133 CH |
1707 | struct xfs_mount *mp, |
1708 | struct xfs_buftarg *btp) | |
1da177e4 | 1709 | { |
ff57ab21 | 1710 | unregister_shrinker(&btp->bt_shrinker); |
9c7504aa BF |
1711 | ASSERT(percpu_counter_sum(&btp->bt_io_count) == 0); |
1712 | percpu_counter_destroy(&btp->bt_io_count); | |
f5e1dd34 | 1713 | list_lru_destroy(&btp->bt_lru); |
ff57ab21 | 1714 | |
b7963133 CH |
1715 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1716 | xfs_blkdev_issue_flush(btp); | |
a6867a68 | 1717 | |
f0e2d93c | 1718 | kmem_free(btp); |
1da177e4 LT |
1719 | } |
1720 | ||
3fefdeee ES |
1721 | int |
1722 | xfs_setsize_buftarg( | |
1da177e4 | 1723 | xfs_buftarg_t *btp, |
3fefdeee | 1724 | unsigned int sectorsize) |
1da177e4 | 1725 | { |
7c71ee78 | 1726 | /* Set up metadata sector size info */ |
6da54179 ES |
1727 | btp->bt_meta_sectorsize = sectorsize; |
1728 | btp->bt_meta_sectormask = sectorsize - 1; | |
1da177e4 | 1729 | |
ce8e922c | 1730 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
4f10700a | 1731 | xfs_warn(btp->bt_mount, |
a1c6f057 DM |
1732 | "Cannot set_blocksize to %u on device %pg", |
1733 | sectorsize, btp->bt_bdev); | |
2451337d | 1734 | return -EINVAL; |
1da177e4 LT |
1735 | } |
1736 | ||
7c71ee78 ES |
1737 | /* Set up device logical sector size mask */ |
1738 | btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev); | |
1739 | btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1; | |
1740 | ||
1da177e4 LT |
1741 | return 0; |
1742 | } | |
1743 | ||
1744 | /* | |
3fefdeee ES |
1745 | * When allocating the initial buffer target we have not yet |
1746 | * read in the superblock, so don't know what sized sectors | |
1747 | * are being used at this early stage. Play safe. | |
ce8e922c | 1748 | */ |
1da177e4 LT |
1749 | STATIC int |
1750 | xfs_setsize_buftarg_early( | |
1751 | xfs_buftarg_t *btp, | |
1752 | struct block_device *bdev) | |
1753 | { | |
a96c4151 | 1754 | return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev)); |
1da177e4 LT |
1755 | } |
1756 | ||
1da177e4 LT |
1757 | xfs_buftarg_t * |
1758 | xfs_alloc_buftarg( | |
ebad861b | 1759 | struct xfs_mount *mp, |
34dcefd7 | 1760 | struct block_device *bdev) |
1da177e4 LT |
1761 | { |
1762 | xfs_buftarg_t *btp; | |
1763 | ||
b17cb364 | 1764 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP | KM_NOFS); |
1da177e4 | 1765 | |
ebad861b | 1766 | btp->bt_mount = mp; |
ce8e922c NS |
1767 | btp->bt_dev = bdev->bd_dev; |
1768 | btp->bt_bdev = bdev; | |
0e6e847f | 1769 | btp->bt_bdi = blk_get_backing_dev_info(bdev); |
0e6e847f | 1770 | |
1da177e4 LT |
1771 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1772 | goto error; | |
5ca302c8 GC |
1773 | |
1774 | if (list_lru_init(&btp->bt_lru)) | |
1775 | goto error; | |
1776 | ||
9c7504aa BF |
1777 | if (percpu_counter_init(&btp->bt_io_count, 0, GFP_KERNEL)) |
1778 | goto error; | |
1779 | ||
e80dfa19 DC |
1780 | btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count; |
1781 | btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan; | |
ff57ab21 | 1782 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; |
e80dfa19 | 1783 | btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE; |
ff57ab21 | 1784 | register_shrinker(&btp->bt_shrinker); |
1da177e4 LT |
1785 | return btp; |
1786 | ||
1787 | error: | |
f0e2d93c | 1788 | kmem_free(btp); |
1da177e4 LT |
1789 | return NULL; |
1790 | } | |
1791 | ||
1da177e4 | 1792 | /* |
43ff2122 CH |
1793 | * Add a buffer to the delayed write list. |
1794 | * | |
1795 | * This queues a buffer for writeout if it hasn't already been. Note that | |
1796 | * neither this routine nor the buffer list submission functions perform | |
1797 | * any internal synchronization. It is expected that the lists are thread-local | |
1798 | * to the callers. | |
1799 | * | |
1800 | * Returns true if we queued up the buffer, or false if it already had | |
1801 | * been on the buffer list. | |
1da177e4 | 1802 | */ |
43ff2122 | 1803 | bool |
ce8e922c | 1804 | xfs_buf_delwri_queue( |
43ff2122 CH |
1805 | struct xfs_buf *bp, |
1806 | struct list_head *list) | |
1da177e4 | 1807 | { |
43ff2122 | 1808 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 1809 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1810 | |
43ff2122 CH |
1811 | /* |
1812 | * If the buffer is already marked delwri it already is queued up | |
1813 | * by someone else for imediate writeout. Just ignore it in that | |
1814 | * case. | |
1815 | */ | |
1816 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
1817 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
1818 | return false; | |
1da177e4 | 1819 | } |
1da177e4 | 1820 | |
43ff2122 | 1821 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
1822 | |
1823 | /* | |
43ff2122 CH |
1824 | * If a buffer gets written out synchronously or marked stale while it |
1825 | * is on a delwri list we lazily remove it. To do this, the other party | |
1826 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
1827 | * It remains referenced and on the list. In a rare corner case it | |
1828 | * might get readded to a delwri list after the synchronous writeout, in | |
1829 | * which case we need just need to re-add the flag here. | |
d808f617 | 1830 | */ |
43ff2122 CH |
1831 | bp->b_flags |= _XBF_DELWRI_Q; |
1832 | if (list_empty(&bp->b_list)) { | |
1833 | atomic_inc(&bp->b_hold); | |
1834 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 1835 | } |
585e6d88 | 1836 | |
43ff2122 | 1837 | return true; |
585e6d88 DC |
1838 | } |
1839 | ||
089716aa DC |
1840 | /* |
1841 | * Compare function is more complex than it needs to be because | |
1842 | * the return value is only 32 bits and we are doing comparisons | |
1843 | * on 64 bit values | |
1844 | */ | |
1845 | static int | |
1846 | xfs_buf_cmp( | |
1847 | void *priv, | |
1848 | struct list_head *a, | |
1849 | struct list_head *b) | |
1850 | { | |
1851 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1852 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1853 | xfs_daddr_t diff; | |
1854 | ||
f4b42421 | 1855 | diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn; |
089716aa DC |
1856 | if (diff < 0) |
1857 | return -1; | |
1858 | if (diff > 0) | |
1859 | return 1; | |
1860 | return 0; | |
1861 | } | |
1862 | ||
26f1fe85 DC |
1863 | /* |
1864 | * submit buffers for write. | |
1865 | * | |
1866 | * When we have a large buffer list, we do not want to hold all the buffers | |
1867 | * locked while we block on the request queue waiting for IO dispatch. To avoid | |
1868 | * this problem, we lock and submit buffers in groups of 50, thereby minimising | |
1869 | * the lock hold times for lists which may contain thousands of objects. | |
1870 | * | |
1871 | * To do this, we sort the buffer list before we walk the list to lock and | |
1872 | * submit buffers, and we plug and unplug around each group of buffers we | |
1873 | * submit. | |
1874 | */ | |
43ff2122 | 1875 | static int |
26f1fe85 | 1876 | xfs_buf_delwri_submit_buffers( |
43ff2122 | 1877 | struct list_head *buffer_list, |
26f1fe85 | 1878 | struct list_head *wait_list) |
1da177e4 | 1879 | { |
43ff2122 | 1880 | struct xfs_buf *bp, *n; |
26f1fe85 | 1881 | LIST_HEAD (submit_list); |
43ff2122 | 1882 | int pinned = 0; |
26f1fe85 | 1883 | struct blk_plug plug; |
43ff2122 | 1884 | |
26f1fe85 | 1885 | list_sort(NULL, buffer_list, xfs_buf_cmp); |
43ff2122 | 1886 | |
26f1fe85 | 1887 | blk_start_plug(&plug); |
43ff2122 | 1888 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { |
26f1fe85 | 1889 | if (!wait_list) { |
43ff2122 CH |
1890 | if (xfs_buf_ispinned(bp)) { |
1891 | pinned++; | |
1892 | continue; | |
1893 | } | |
1894 | if (!xfs_buf_trylock(bp)) | |
1895 | continue; | |
1896 | } else { | |
1897 | xfs_buf_lock(bp); | |
1898 | } | |
978c7b2f | 1899 | |
43ff2122 CH |
1900 | /* |
1901 | * Someone else might have written the buffer synchronously or | |
1902 | * marked it stale in the meantime. In that case only the | |
1903 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
1904 | * reference and remove it from the list here. | |
1905 | */ | |
1906 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
1907 | list_del_init(&bp->b_list); | |
1908 | xfs_buf_relse(bp); | |
1909 | continue; | |
1910 | } | |
c9c12971 | 1911 | |
43ff2122 | 1912 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
a1b7ea5d | 1913 | |
cf53e99d | 1914 | /* |
26f1fe85 DC |
1915 | * We do all IO submission async. This means if we need |
1916 | * to wait for IO completion we need to take an extra | |
1917 | * reference so the buffer is still valid on the other | |
1918 | * side. We need to move the buffer onto the io_list | |
1919 | * at this point so the caller can still access it. | |
cf53e99d | 1920 | */ |
bbfeb614 | 1921 | bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_WRITE_FAIL); |
26f1fe85 DC |
1922 | bp->b_flags |= XBF_WRITE | XBF_ASYNC; |
1923 | if (wait_list) { | |
cf53e99d | 1924 | xfs_buf_hold(bp); |
26f1fe85 DC |
1925 | list_move_tail(&bp->b_list, wait_list); |
1926 | } else | |
ce8e922c | 1927 | list_del_init(&bp->b_list); |
8dac3921 | 1928 | |
595bff75 | 1929 | xfs_buf_submit(bp); |
43ff2122 CH |
1930 | } |
1931 | blk_finish_plug(&plug); | |
1da177e4 | 1932 | |
43ff2122 | 1933 | return pinned; |
1da177e4 LT |
1934 | } |
1935 | ||
1936 | /* | |
43ff2122 CH |
1937 | * Write out a buffer list asynchronously. |
1938 | * | |
1939 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
1940 | * out and not wait for I/O completion on any of the buffers. This interface | |
1941 | * is only safely useable for callers that can track I/O completion by higher | |
1942 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
1943 | * function. | |
1da177e4 LT |
1944 | */ |
1945 | int | |
43ff2122 CH |
1946 | xfs_buf_delwri_submit_nowait( |
1947 | struct list_head *buffer_list) | |
1da177e4 | 1948 | { |
26f1fe85 | 1949 | return xfs_buf_delwri_submit_buffers(buffer_list, NULL); |
43ff2122 | 1950 | } |
1da177e4 | 1951 | |
43ff2122 CH |
1952 | /* |
1953 | * Write out a buffer list synchronously. | |
1954 | * | |
1955 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
1956 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
1957 | * so callers must have some other way of tracking buffers if they require such | |
1958 | * functionality. | |
1959 | */ | |
1960 | int | |
1961 | xfs_buf_delwri_submit( | |
1962 | struct list_head *buffer_list) | |
1963 | { | |
26f1fe85 | 1964 | LIST_HEAD (wait_list); |
43ff2122 CH |
1965 | int error = 0, error2; |
1966 | struct xfs_buf *bp; | |
1da177e4 | 1967 | |
26f1fe85 | 1968 | xfs_buf_delwri_submit_buffers(buffer_list, &wait_list); |
1da177e4 | 1969 | |
43ff2122 | 1970 | /* Wait for IO to complete. */ |
26f1fe85 DC |
1971 | while (!list_empty(&wait_list)) { |
1972 | bp = list_first_entry(&wait_list, struct xfs_buf, b_list); | |
a1b7ea5d | 1973 | |
089716aa | 1974 | list_del_init(&bp->b_list); |
cf53e99d DC |
1975 | |
1976 | /* locking the buffer will wait for async IO completion. */ | |
1977 | xfs_buf_lock(bp); | |
1978 | error2 = bp->b_error; | |
43ff2122 CH |
1979 | xfs_buf_relse(bp); |
1980 | if (!error) | |
1981 | error = error2; | |
1da177e4 LT |
1982 | } |
1983 | ||
43ff2122 | 1984 | return error; |
1da177e4 LT |
1985 | } |
1986 | ||
04d8b284 | 1987 | int __init |
ce8e922c | 1988 | xfs_buf_init(void) |
1da177e4 | 1989 | { |
8758280f NS |
1990 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1991 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1992 | if (!xfs_buf_zone) |
0b1b213f | 1993 | goto out; |
04d8b284 | 1994 | |
23ea4032 | 1995 | return 0; |
1da177e4 | 1996 | |
0b1b213f | 1997 | out: |
8758280f | 1998 | return -ENOMEM; |
1da177e4 LT |
1999 | } |
2000 | ||
1da177e4 | 2001 | void |
ce8e922c | 2002 | xfs_buf_terminate(void) |
1da177e4 | 2003 | { |
ce8e922c | 2004 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 2005 | } |