]>
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> |
5b3cc15a | 36 | #include <linux/sched/mm.h> |
1da177e4 | 37 | |
4fb6e8ad | 38 | #include "xfs_format.h" |
239880ef | 39 | #include "xfs_log_format.h" |
7fd36c44 | 40 | #include "xfs_trans_resv.h" |
239880ef | 41 | #include "xfs_sb.h" |
b7963133 | 42 | #include "xfs_mount.h" |
0b1b213f | 43 | #include "xfs_trace.h" |
239880ef | 44 | #include "xfs_log.h" |
b7963133 | 45 | |
7989cb8e | 46 | static kmem_zone_t *xfs_buf_zone; |
23ea4032 | 47 | |
ce8e922c NS |
48 | #ifdef XFS_BUF_LOCK_TRACKING |
49 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
50 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
51 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 52 | #else |
ce8e922c NS |
53 | # define XB_SET_OWNER(bp) do { } while (0) |
54 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
55 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
56 | #endif |
57 | ||
ce8e922c | 58 | #define xb_to_gfp(flags) \ |
aa5c158e | 59 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN) |
1da177e4 | 60 | |
1da177e4 | 61 | |
73c77e2c JB |
62 | static inline int |
63 | xfs_buf_is_vmapped( | |
64 | struct xfs_buf *bp) | |
65 | { | |
66 | /* | |
67 | * Return true if the buffer is vmapped. | |
68 | * | |
611c9946 DC |
69 | * b_addr is null if the buffer is not mapped, but the code is clever |
70 | * enough to know it doesn't have to map a single page, so the check has | |
71 | * to be both for b_addr and bp->b_page_count > 1. | |
73c77e2c | 72 | */ |
611c9946 | 73 | return bp->b_addr && bp->b_page_count > 1; |
73c77e2c JB |
74 | } |
75 | ||
76 | static inline int | |
77 | xfs_buf_vmap_len( | |
78 | struct xfs_buf *bp) | |
79 | { | |
80 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
81 | } | |
82 | ||
9c7504aa BF |
83 | /* |
84 | * Bump the I/O in flight count on the buftarg if we haven't yet done so for | |
85 | * this buffer. The count is incremented once per buffer (per hold cycle) | |
86 | * because the corresponding decrement is deferred to buffer release. Buffers | |
87 | * can undergo I/O multiple times in a hold-release cycle and per buffer I/O | |
88 | * tracking adds unnecessary overhead. This is used for sychronization purposes | |
89 | * with unmount (see xfs_wait_buftarg()), so all we really need is a count of | |
90 | * in-flight buffers. | |
91 | * | |
92 | * Buffers that are never released (e.g., superblock, iclog buffers) must set | |
93 | * the XBF_NO_IOACCT flag before I/O submission. Otherwise, the buftarg count | |
94 | * never reaches zero and unmount hangs indefinitely. | |
95 | */ | |
96 | static inline void | |
97 | xfs_buf_ioacct_inc( | |
98 | struct xfs_buf *bp) | |
99 | { | |
63db7c81 | 100 | if (bp->b_flags & XBF_NO_IOACCT) |
9c7504aa BF |
101 | return; |
102 | ||
103 | ASSERT(bp->b_flags & XBF_ASYNC); | |
63db7c81 BF |
104 | spin_lock(&bp->b_lock); |
105 | if (!(bp->b_state & XFS_BSTATE_IN_FLIGHT)) { | |
106 | bp->b_state |= XFS_BSTATE_IN_FLIGHT; | |
107 | percpu_counter_inc(&bp->b_target->bt_io_count); | |
108 | } | |
109 | spin_unlock(&bp->b_lock); | |
9c7504aa BF |
110 | } |
111 | ||
112 | /* | |
113 | * Clear the in-flight state on a buffer about to be released to the LRU or | |
114 | * freed and unaccount from the buftarg. | |
115 | */ | |
116 | static inline void | |
63db7c81 | 117 | __xfs_buf_ioacct_dec( |
9c7504aa BF |
118 | struct xfs_buf *bp) |
119 | { | |
95989c46 | 120 | lockdep_assert_held(&bp->b_lock); |
9c7504aa | 121 | |
63db7c81 BF |
122 | if (bp->b_state & XFS_BSTATE_IN_FLIGHT) { |
123 | bp->b_state &= ~XFS_BSTATE_IN_FLIGHT; | |
124 | percpu_counter_dec(&bp->b_target->bt_io_count); | |
125 | } | |
126 | } | |
127 | ||
128 | static inline void | |
129 | xfs_buf_ioacct_dec( | |
130 | struct xfs_buf *bp) | |
131 | { | |
132 | spin_lock(&bp->b_lock); | |
133 | __xfs_buf_ioacct_dec(bp); | |
134 | spin_unlock(&bp->b_lock); | |
9c7504aa BF |
135 | } |
136 | ||
430cbeb8 DC |
137 | /* |
138 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
139 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
140 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
141 | * to remove the reference that LRU holds on the buffer. | |
142 | * | |
143 | * This prevents build-up of stale buffers on the LRU. | |
144 | */ | |
145 | void | |
146 | xfs_buf_stale( | |
147 | struct xfs_buf *bp) | |
148 | { | |
43ff2122 CH |
149 | ASSERT(xfs_buf_islocked(bp)); |
150 | ||
430cbeb8 | 151 | bp->b_flags |= XBF_STALE; |
43ff2122 CH |
152 | |
153 | /* | |
154 | * Clear the delwri status so that a delwri queue walker will not | |
155 | * flush this buffer to disk now that it is stale. The delwri queue has | |
156 | * a reference to the buffer, so this is safe to do. | |
157 | */ | |
158 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
159 | ||
9c7504aa BF |
160 | /* |
161 | * Once the buffer is marked stale and unlocked, a subsequent lookup | |
162 | * could reset b_flags. There is no guarantee that the buffer is | |
163 | * unaccounted (released to LRU) before that occurs. Drop in-flight | |
164 | * status now to preserve accounting consistency. | |
165 | */ | |
a4082357 | 166 | spin_lock(&bp->b_lock); |
63db7c81 BF |
167 | __xfs_buf_ioacct_dec(bp); |
168 | ||
a4082357 DC |
169 | atomic_set(&bp->b_lru_ref, 0); |
170 | if (!(bp->b_state & XFS_BSTATE_DISPOSE) && | |
e80dfa19 DC |
171 | (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru))) |
172 | atomic_dec(&bp->b_hold); | |
173 | ||
430cbeb8 | 174 | ASSERT(atomic_read(&bp->b_hold) >= 1); |
a4082357 | 175 | spin_unlock(&bp->b_lock); |
430cbeb8 | 176 | } |
1da177e4 | 177 | |
3e85c868 DC |
178 | static int |
179 | xfs_buf_get_maps( | |
180 | struct xfs_buf *bp, | |
181 | int map_count) | |
182 | { | |
183 | ASSERT(bp->b_maps == NULL); | |
184 | bp->b_map_count = map_count; | |
185 | ||
186 | if (map_count == 1) { | |
f4b42421 | 187 | bp->b_maps = &bp->__b_map; |
3e85c868 DC |
188 | return 0; |
189 | } | |
190 | ||
191 | bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map), | |
192 | KM_NOFS); | |
193 | if (!bp->b_maps) | |
2451337d | 194 | return -ENOMEM; |
3e85c868 DC |
195 | return 0; |
196 | } | |
197 | ||
198 | /* | |
199 | * Frees b_pages if it was allocated. | |
200 | */ | |
201 | static void | |
202 | xfs_buf_free_maps( | |
203 | struct xfs_buf *bp) | |
204 | { | |
f4b42421 | 205 | if (bp->b_maps != &bp->__b_map) { |
3e85c868 DC |
206 | kmem_free(bp->b_maps); |
207 | bp->b_maps = NULL; | |
208 | } | |
209 | } | |
210 | ||
4347b9d7 | 211 | struct xfs_buf * |
3e85c868 | 212 | _xfs_buf_alloc( |
4347b9d7 | 213 | struct xfs_buftarg *target, |
3e85c868 DC |
214 | struct xfs_buf_map *map, |
215 | int nmaps, | |
ce8e922c | 216 | xfs_buf_flags_t flags) |
1da177e4 | 217 | { |
4347b9d7 | 218 | struct xfs_buf *bp; |
3e85c868 DC |
219 | int error; |
220 | int i; | |
4347b9d7 | 221 | |
aa5c158e | 222 | bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS); |
4347b9d7 CH |
223 | if (unlikely(!bp)) |
224 | return NULL; | |
225 | ||
1da177e4 | 226 | /* |
12bcb3f7 DC |
227 | * We don't want certain flags to appear in b_flags unless they are |
228 | * specifically set by later operations on the buffer. | |
1da177e4 | 229 | */ |
611c9946 | 230 | flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD); |
ce8e922c | 231 | |
ce8e922c | 232 | atomic_set(&bp->b_hold, 1); |
430cbeb8 | 233 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 234 | init_completion(&bp->b_iowait); |
430cbeb8 | 235 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 236 | INIT_LIST_HEAD(&bp->b_list); |
a731cd11 | 237 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
a4082357 | 238 | spin_lock_init(&bp->b_lock); |
ce8e922c NS |
239 | XB_SET_OWNER(bp); |
240 | bp->b_target = target; | |
3e85c868 | 241 | bp->b_flags = flags; |
de1cbee4 | 242 | |
1da177e4 | 243 | /* |
aa0e8833 DC |
244 | * Set length and io_length to the same value initially. |
245 | * I/O routines should use io_length, which will be the same in | |
1da177e4 LT |
246 | * most cases but may be reset (e.g. XFS recovery). |
247 | */ | |
3e85c868 DC |
248 | error = xfs_buf_get_maps(bp, nmaps); |
249 | if (error) { | |
250 | kmem_zone_free(xfs_buf_zone, bp); | |
251 | return NULL; | |
252 | } | |
253 | ||
254 | bp->b_bn = map[0].bm_bn; | |
255 | bp->b_length = 0; | |
256 | for (i = 0; i < nmaps; i++) { | |
257 | bp->b_maps[i].bm_bn = map[i].bm_bn; | |
258 | bp->b_maps[i].bm_len = map[i].bm_len; | |
259 | bp->b_length += map[i].bm_len; | |
260 | } | |
261 | bp->b_io_length = bp->b_length; | |
262 | ||
ce8e922c NS |
263 | atomic_set(&bp->b_pin_count, 0); |
264 | init_waitqueue_head(&bp->b_waiters); | |
265 | ||
ff6d6af2 | 266 | XFS_STATS_INC(target->bt_mount, xb_create); |
0b1b213f | 267 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 CH |
268 | |
269 | return bp; | |
1da177e4 LT |
270 | } |
271 | ||
272 | /* | |
ce8e922c NS |
273 | * Allocate a page array capable of holding a specified number |
274 | * of pages, and point the page buf at it. | |
1da177e4 LT |
275 | */ |
276 | STATIC int | |
ce8e922c NS |
277 | _xfs_buf_get_pages( |
278 | xfs_buf_t *bp, | |
87937bf8 | 279 | int page_count) |
1da177e4 LT |
280 | { |
281 | /* Make sure that we have a page list */ | |
ce8e922c | 282 | if (bp->b_pages == NULL) { |
ce8e922c NS |
283 | bp->b_page_count = page_count; |
284 | if (page_count <= XB_PAGES) { | |
285 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 286 | } else { |
ce8e922c | 287 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
aa5c158e | 288 | page_count, KM_NOFS); |
ce8e922c | 289 | if (bp->b_pages == NULL) |
1da177e4 LT |
290 | return -ENOMEM; |
291 | } | |
ce8e922c | 292 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
293 | } |
294 | return 0; | |
295 | } | |
296 | ||
297 | /* | |
ce8e922c | 298 | * Frees b_pages if it was allocated. |
1da177e4 LT |
299 | */ |
300 | STATIC void | |
ce8e922c | 301 | _xfs_buf_free_pages( |
1da177e4 LT |
302 | xfs_buf_t *bp) |
303 | { | |
ce8e922c | 304 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 305 | kmem_free(bp->b_pages); |
3fc98b1a | 306 | bp->b_pages = NULL; |
1da177e4 LT |
307 | } |
308 | } | |
309 | ||
310 | /* | |
311 | * Releases the specified buffer. | |
312 | * | |
313 | * The modification state of any associated pages is left unchanged. | |
b46fe825 | 314 | * The buffer must not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
315 | * hashed and refcounted buffers |
316 | */ | |
317 | void | |
ce8e922c | 318 | xfs_buf_free( |
1da177e4 LT |
319 | xfs_buf_t *bp) |
320 | { | |
0b1b213f | 321 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 322 | |
430cbeb8 DC |
323 | ASSERT(list_empty(&bp->b_lru)); |
324 | ||
0e6e847f | 325 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
326 | uint i; |
327 | ||
73c77e2c | 328 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
329 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
330 | bp->b_page_count); | |
1da177e4 | 331 | |
948ecdb4 NS |
332 | for (i = 0; i < bp->b_page_count; i++) { |
333 | struct page *page = bp->b_pages[i]; | |
334 | ||
0e6e847f | 335 | __free_page(page); |
948ecdb4 | 336 | } |
0e6e847f DC |
337 | } else if (bp->b_flags & _XBF_KMEM) |
338 | kmem_free(bp->b_addr); | |
3fc98b1a | 339 | _xfs_buf_free_pages(bp); |
3e85c868 | 340 | xfs_buf_free_maps(bp); |
4347b9d7 | 341 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
342 | } |
343 | ||
344 | /* | |
0e6e847f | 345 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
346 | */ |
347 | STATIC int | |
0e6e847f | 348 | xfs_buf_allocate_memory( |
1da177e4 LT |
349 | xfs_buf_t *bp, |
350 | uint flags) | |
351 | { | |
aa0e8833 | 352 | size_t size; |
1da177e4 | 353 | size_t nbytes, offset; |
ce8e922c | 354 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 355 | unsigned short page_count, i; |
795cac72 | 356 | xfs_off_t start, end; |
1da177e4 LT |
357 | int error; |
358 | ||
0e6e847f DC |
359 | /* |
360 | * for buffers that are contained within a single page, just allocate | |
361 | * the memory from the heap - there's no need for the complexity of | |
362 | * page arrays to keep allocation down to order 0. | |
363 | */ | |
795cac72 DC |
364 | size = BBTOB(bp->b_length); |
365 | if (size < PAGE_SIZE) { | |
aa5c158e | 366 | bp->b_addr = kmem_alloc(size, KM_NOFS); |
0e6e847f DC |
367 | if (!bp->b_addr) { |
368 | /* low memory - use alloc_page loop instead */ | |
369 | goto use_alloc_page; | |
370 | } | |
371 | ||
795cac72 | 372 | if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) != |
0e6e847f DC |
373 | ((unsigned long)bp->b_addr & PAGE_MASK)) { |
374 | /* b_addr spans two pages - use alloc_page instead */ | |
375 | kmem_free(bp->b_addr); | |
376 | bp->b_addr = NULL; | |
377 | goto use_alloc_page; | |
378 | } | |
379 | bp->b_offset = offset_in_page(bp->b_addr); | |
380 | bp->b_pages = bp->b_page_array; | |
381 | bp->b_pages[0] = virt_to_page(bp->b_addr); | |
382 | bp->b_page_count = 1; | |
611c9946 | 383 | bp->b_flags |= _XBF_KMEM; |
0e6e847f DC |
384 | return 0; |
385 | } | |
386 | ||
387 | use_alloc_page: | |
f4b42421 MT |
388 | start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT; |
389 | end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1) | |
cbb7baab | 390 | >> PAGE_SHIFT; |
795cac72 | 391 | page_count = end - start; |
87937bf8 | 392 | error = _xfs_buf_get_pages(bp, page_count); |
1da177e4 LT |
393 | if (unlikely(error)) |
394 | return error; | |
1da177e4 | 395 | |
ce8e922c | 396 | offset = bp->b_offset; |
0e6e847f | 397 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 398 | |
ce8e922c | 399 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
400 | struct page *page; |
401 | uint retries = 0; | |
0e6e847f DC |
402 | retry: |
403 | page = alloc_page(gfp_mask); | |
1da177e4 | 404 | if (unlikely(page == NULL)) { |
ce8e922c NS |
405 | if (flags & XBF_READ_AHEAD) { |
406 | bp->b_page_count = i; | |
2451337d | 407 | error = -ENOMEM; |
0e6e847f | 408 | goto out_free_pages; |
1da177e4 LT |
409 | } |
410 | ||
411 | /* | |
412 | * This could deadlock. | |
413 | * | |
414 | * But until all the XFS lowlevel code is revamped to | |
415 | * handle buffer allocation failures we can't do much. | |
416 | */ | |
417 | if (!(++retries % 100)) | |
4f10700a | 418 | xfs_err(NULL, |
5bf97b1c TH |
419 | "%s(%u) possible memory allocation deadlock in %s (mode:0x%x)", |
420 | current->comm, current->pid, | |
34a622b2 | 421 | __func__, gfp_mask); |
1da177e4 | 422 | |
ff6d6af2 | 423 | XFS_STATS_INC(bp->b_target->bt_mount, xb_page_retries); |
8aa7e847 | 424 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
425 | goto retry; |
426 | } | |
427 | ||
ff6d6af2 | 428 | XFS_STATS_INC(bp->b_target->bt_mount, xb_page_found); |
1da177e4 | 429 | |
0e6e847f | 430 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 431 | size -= nbytes; |
ce8e922c | 432 | bp->b_pages[i] = page; |
1da177e4 LT |
433 | offset = 0; |
434 | } | |
0e6e847f | 435 | return 0; |
1da177e4 | 436 | |
0e6e847f DC |
437 | out_free_pages: |
438 | for (i = 0; i < bp->b_page_count; i++) | |
439 | __free_page(bp->b_pages[i]); | |
2aa6ba7b | 440 | bp->b_flags &= ~_XBF_PAGES; |
1da177e4 LT |
441 | return error; |
442 | } | |
443 | ||
444 | /* | |
25985edc | 445 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
446 | */ |
447 | STATIC int | |
ce8e922c | 448 | _xfs_buf_map_pages( |
1da177e4 LT |
449 | xfs_buf_t *bp, |
450 | uint flags) | |
451 | { | |
0e6e847f | 452 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 453 | if (bp->b_page_count == 1) { |
0e6e847f | 454 | /* A single page buffer is always mappable */ |
ce8e922c | 455 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
611c9946 DC |
456 | } else if (flags & XBF_UNMAPPED) { |
457 | bp->b_addr = NULL; | |
458 | } else { | |
a19fb380 | 459 | int retried = 0; |
9ba1fb2c | 460 | unsigned nofs_flag; |
ae687e58 DC |
461 | |
462 | /* | |
463 | * vm_map_ram() will allocate auxillary structures (e.g. | |
464 | * pagetables) with GFP_KERNEL, yet we are likely to be under | |
465 | * GFP_NOFS context here. Hence we need to tell memory reclaim | |
9ba1fb2c | 466 | * that we are in such a context via PF_MEMALLOC_NOFS to prevent |
ae687e58 DC |
467 | * memory reclaim re-entering the filesystem here and |
468 | * potentially deadlocking. | |
469 | */ | |
9ba1fb2c | 470 | nofs_flag = memalloc_nofs_save(); |
a19fb380 DC |
471 | do { |
472 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
473 | -1, PAGE_KERNEL); | |
474 | if (bp->b_addr) | |
475 | break; | |
476 | vm_unmap_aliases(); | |
477 | } while (retried++ <= 1); | |
9ba1fb2c | 478 | memalloc_nofs_restore(nofs_flag); |
a19fb380 DC |
479 | |
480 | if (!bp->b_addr) | |
1da177e4 | 481 | return -ENOMEM; |
ce8e922c | 482 | bp->b_addr += bp->b_offset; |
1da177e4 LT |
483 | } |
484 | ||
485 | return 0; | |
486 | } | |
487 | ||
488 | /* | |
489 | * Finding and Reading Buffers | |
490 | */ | |
6031e73a LS |
491 | static int |
492 | _xfs_buf_obj_cmp( | |
493 | struct rhashtable_compare_arg *arg, | |
494 | const void *obj) | |
495 | { | |
496 | const struct xfs_buf_map *map = arg->key; | |
497 | const struct xfs_buf *bp = obj; | |
498 | ||
499 | /* | |
500 | * The key hashing in the lookup path depends on the key being the | |
501 | * first element of the compare_arg, make sure to assert this. | |
502 | */ | |
503 | BUILD_BUG_ON(offsetof(struct xfs_buf_map, bm_bn) != 0); | |
504 | ||
505 | if (bp->b_bn != map->bm_bn) | |
506 | return 1; | |
507 | ||
508 | if (unlikely(bp->b_length != map->bm_len)) { | |
509 | /* | |
510 | * found a block number match. If the range doesn't | |
511 | * match, the only way this is allowed is if the buffer | |
512 | * in the cache is stale and the transaction that made | |
513 | * it stale has not yet committed. i.e. we are | |
514 | * reallocating a busy extent. Skip this buffer and | |
515 | * continue searching for an exact match. | |
516 | */ | |
517 | ASSERT(bp->b_flags & XBF_STALE); | |
518 | return 1; | |
519 | } | |
520 | return 0; | |
521 | } | |
522 | ||
523 | static const struct rhashtable_params xfs_buf_hash_params = { | |
524 | .min_size = 32, /* empty AGs have minimal footprint */ | |
525 | .nelem_hint = 16, | |
526 | .key_len = sizeof(xfs_daddr_t), | |
527 | .key_offset = offsetof(struct xfs_buf, b_bn), | |
528 | .head_offset = offsetof(struct xfs_buf, b_rhash_head), | |
529 | .automatic_shrinking = true, | |
530 | .obj_cmpfn = _xfs_buf_obj_cmp, | |
531 | }; | |
532 | ||
533 | int | |
534 | xfs_buf_hash_init( | |
535 | struct xfs_perag *pag) | |
536 | { | |
537 | spin_lock_init(&pag->pag_buf_lock); | |
538 | return rhashtable_init(&pag->pag_buf_hash, &xfs_buf_hash_params); | |
539 | } | |
540 | ||
541 | void | |
542 | xfs_buf_hash_destroy( | |
543 | struct xfs_perag *pag) | |
544 | { | |
545 | rhashtable_destroy(&pag->pag_buf_hash); | |
546 | } | |
1da177e4 LT |
547 | |
548 | /* | |
ce8e922c | 549 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 | 550 | * a given range of an inode. The buffer is returned |
eabbaf11 | 551 | * locked. No I/O is implied by this call. |
1da177e4 LT |
552 | */ |
553 | xfs_buf_t * | |
ce8e922c | 554 | _xfs_buf_find( |
e70b73f8 | 555 | struct xfs_buftarg *btp, |
3e85c868 DC |
556 | struct xfs_buf_map *map, |
557 | int nmaps, | |
ce8e922c NS |
558 | xfs_buf_flags_t flags, |
559 | xfs_buf_t *new_bp) | |
1da177e4 | 560 | { |
74f75a0c | 561 | struct xfs_perag *pag; |
74f75a0c | 562 | xfs_buf_t *bp; |
6031e73a | 563 | struct xfs_buf_map cmap = { .bm_bn = map[0].bm_bn }; |
10616b80 | 564 | xfs_daddr_t eofs; |
3e85c868 | 565 | int i; |
1da177e4 | 566 | |
3e85c868 | 567 | for (i = 0; i < nmaps; i++) |
6031e73a | 568 | cmap.bm_len += map[i].bm_len; |
1da177e4 LT |
569 | |
570 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
6031e73a LS |
571 | ASSERT(!(BBTOB(cmap.bm_len) < btp->bt_meta_sectorsize)); |
572 | ASSERT(!(BBTOB(cmap.bm_bn) & (xfs_off_t)btp->bt_meta_sectormask)); | |
1da177e4 | 573 | |
10616b80 DC |
574 | /* |
575 | * Corrupted block numbers can get through to here, unfortunately, so we | |
576 | * have to check that the buffer falls within the filesystem bounds. | |
577 | */ | |
578 | eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks); | |
6031e73a | 579 | if (cmap.bm_bn < 0 || cmap.bm_bn >= eofs) { |
10616b80 | 580 | /* |
2451337d | 581 | * XXX (dgc): we should really be returning -EFSCORRUPTED here, |
10616b80 DC |
582 | * but none of the higher level infrastructure supports |
583 | * returning a specific error on buffer lookup failures. | |
584 | */ | |
585 | xfs_alert(btp->bt_mount, | |
586 | "%s: Block out of range: block 0x%llx, EOFS 0x%llx ", | |
6031e73a | 587 | __func__, cmap.bm_bn, eofs); |
7bc0dc27 | 588 | WARN_ON(1); |
10616b80 DC |
589 | return NULL; |
590 | } | |
591 | ||
74f75a0c | 592 | pag = xfs_perag_get(btp->bt_mount, |
6031e73a | 593 | xfs_daddr_to_agno(btp->bt_mount, cmap.bm_bn)); |
74f75a0c | 594 | |
74f75a0c | 595 | spin_lock(&pag->pag_buf_lock); |
6031e73a LS |
596 | bp = rhashtable_lookup_fast(&pag->pag_buf_hash, &cmap, |
597 | xfs_buf_hash_params); | |
598 | if (bp) { | |
599 | atomic_inc(&bp->b_hold); | |
600 | goto found; | |
1da177e4 LT |
601 | } |
602 | ||
603 | /* No match found */ | |
ce8e922c | 604 | if (new_bp) { |
74f75a0c DC |
605 | /* the buffer keeps the perag reference until it is freed */ |
606 | new_bp->b_pag = pag; | |
6031e73a LS |
607 | rhashtable_insert_fast(&pag->pag_buf_hash, |
608 | &new_bp->b_rhash_head, | |
609 | xfs_buf_hash_params); | |
74f75a0c | 610 | spin_unlock(&pag->pag_buf_lock); |
1da177e4 | 611 | } else { |
ff6d6af2 | 612 | XFS_STATS_INC(btp->bt_mount, xb_miss_locked); |
74f75a0c DC |
613 | spin_unlock(&pag->pag_buf_lock); |
614 | xfs_perag_put(pag); | |
1da177e4 | 615 | } |
ce8e922c | 616 | return new_bp; |
1da177e4 LT |
617 | |
618 | found: | |
74f75a0c DC |
619 | spin_unlock(&pag->pag_buf_lock); |
620 | xfs_perag_put(pag); | |
1da177e4 | 621 | |
0c842ad4 CH |
622 | if (!xfs_buf_trylock(bp)) { |
623 | if (flags & XBF_TRYLOCK) { | |
ce8e922c | 624 | xfs_buf_rele(bp); |
ff6d6af2 | 625 | XFS_STATS_INC(btp->bt_mount, xb_busy_locked); |
ce8e922c | 626 | return NULL; |
1da177e4 | 627 | } |
0c842ad4 | 628 | xfs_buf_lock(bp); |
ff6d6af2 | 629 | XFS_STATS_INC(btp->bt_mount, xb_get_locked_waited); |
1da177e4 LT |
630 | } |
631 | ||
0e6e847f DC |
632 | /* |
633 | * if the buffer is stale, clear all the external state associated with | |
634 | * it. We need to keep flags such as how we allocated the buffer memory | |
635 | * intact here. | |
636 | */ | |
ce8e922c NS |
637 | if (bp->b_flags & XBF_STALE) { |
638 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
cfb02852 | 639 | ASSERT(bp->b_iodone == NULL); |
611c9946 | 640 | bp->b_flags &= _XBF_KMEM | _XBF_PAGES; |
1813dd64 | 641 | bp->b_ops = NULL; |
2f926587 | 642 | } |
0b1b213f CH |
643 | |
644 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ff6d6af2 | 645 | XFS_STATS_INC(btp->bt_mount, xb_get_locked); |
ce8e922c | 646 | return bp; |
1da177e4 LT |
647 | } |
648 | ||
649 | /* | |
3815832a DC |
650 | * Assembles a buffer covering the specified range. The code is optimised for |
651 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
652 | * more hits than misses. | |
1da177e4 | 653 | */ |
3815832a | 654 | struct xfs_buf * |
6dde2707 DC |
655 | xfs_buf_get_map( |
656 | struct xfs_buftarg *target, | |
657 | struct xfs_buf_map *map, | |
658 | int nmaps, | |
ce8e922c | 659 | xfs_buf_flags_t flags) |
1da177e4 | 660 | { |
3815832a DC |
661 | struct xfs_buf *bp; |
662 | struct xfs_buf *new_bp; | |
0e6e847f | 663 | int error = 0; |
1da177e4 | 664 | |
6dde2707 | 665 | bp = _xfs_buf_find(target, map, nmaps, flags, NULL); |
3815832a DC |
666 | if (likely(bp)) |
667 | goto found; | |
668 | ||
6dde2707 | 669 | new_bp = _xfs_buf_alloc(target, map, nmaps, flags); |
ce8e922c | 670 | if (unlikely(!new_bp)) |
1da177e4 LT |
671 | return NULL; |
672 | ||
fe2429b0 DC |
673 | error = xfs_buf_allocate_memory(new_bp, flags); |
674 | if (error) { | |
3e85c868 | 675 | xfs_buf_free(new_bp); |
fe2429b0 DC |
676 | return NULL; |
677 | } | |
678 | ||
6dde2707 | 679 | bp = _xfs_buf_find(target, map, nmaps, flags, new_bp); |
3815832a | 680 | if (!bp) { |
fe2429b0 | 681 | xfs_buf_free(new_bp); |
3815832a DC |
682 | return NULL; |
683 | } | |
684 | ||
fe2429b0 DC |
685 | if (bp != new_bp) |
686 | xfs_buf_free(new_bp); | |
1da177e4 | 687 | |
3815832a | 688 | found: |
611c9946 | 689 | if (!bp->b_addr) { |
ce8e922c | 690 | error = _xfs_buf_map_pages(bp, flags); |
1da177e4 | 691 | if (unlikely(error)) { |
4f10700a | 692 | xfs_warn(target->bt_mount, |
08e96e1a | 693 | "%s: failed to map pagesn", __func__); |
a8acad70 DC |
694 | xfs_buf_relse(bp); |
695 | return NULL; | |
1da177e4 LT |
696 | } |
697 | } | |
698 | ||
b79f4a1c DC |
699 | /* |
700 | * Clear b_error if this is a lookup from a caller that doesn't expect | |
701 | * valid data to be found in the buffer. | |
702 | */ | |
703 | if (!(flags & XBF_READ)) | |
704 | xfs_buf_ioerror(bp, 0); | |
705 | ||
ff6d6af2 | 706 | XFS_STATS_INC(target->bt_mount, xb_get); |
0b1b213f | 707 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 708 | return bp; |
1da177e4 LT |
709 | } |
710 | ||
5d765b97 CH |
711 | STATIC int |
712 | _xfs_buf_read( | |
713 | xfs_buf_t *bp, | |
714 | xfs_buf_flags_t flags) | |
715 | { | |
43ff2122 | 716 | ASSERT(!(flags & XBF_WRITE)); |
f4b42421 | 717 | ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL); |
5d765b97 | 718 | |
43ff2122 | 719 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD); |
1d5ae5df | 720 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 721 | |
595bff75 DC |
722 | if (flags & XBF_ASYNC) { |
723 | xfs_buf_submit(bp); | |
0e95f19a | 724 | return 0; |
595bff75 DC |
725 | } |
726 | return xfs_buf_submit_wait(bp); | |
5d765b97 CH |
727 | } |
728 | ||
1da177e4 | 729 | xfs_buf_t * |
6dde2707 DC |
730 | xfs_buf_read_map( |
731 | struct xfs_buftarg *target, | |
732 | struct xfs_buf_map *map, | |
733 | int nmaps, | |
c3f8fc73 | 734 | xfs_buf_flags_t flags, |
1813dd64 | 735 | const struct xfs_buf_ops *ops) |
1da177e4 | 736 | { |
6dde2707 | 737 | struct xfs_buf *bp; |
ce8e922c NS |
738 | |
739 | flags |= XBF_READ; | |
740 | ||
6dde2707 | 741 | bp = xfs_buf_get_map(target, map, nmaps, flags); |
ce8e922c | 742 | if (bp) { |
0b1b213f CH |
743 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
744 | ||
b0388bf1 | 745 | if (!(bp->b_flags & XBF_DONE)) { |
ff6d6af2 | 746 | XFS_STATS_INC(target->bt_mount, xb_get_read); |
1813dd64 | 747 | bp->b_ops = ops; |
5d765b97 | 748 | _xfs_buf_read(bp, flags); |
ce8e922c | 749 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
750 | /* |
751 | * Read ahead call which is already satisfied, | |
752 | * drop the buffer | |
753 | */ | |
a8acad70 DC |
754 | xfs_buf_relse(bp); |
755 | return NULL; | |
1da177e4 | 756 | } else { |
1da177e4 | 757 | /* We do not want read in the flags */ |
ce8e922c | 758 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
759 | } |
760 | } | |
761 | ||
ce8e922c | 762 | return bp; |
1da177e4 LT |
763 | } |
764 | ||
1da177e4 | 765 | /* |
ce8e922c NS |
766 | * If we are not low on memory then do the readahead in a deadlock |
767 | * safe manner. | |
1da177e4 LT |
768 | */ |
769 | void | |
6dde2707 DC |
770 | xfs_buf_readahead_map( |
771 | struct xfs_buftarg *target, | |
772 | struct xfs_buf_map *map, | |
c3f8fc73 | 773 | int nmaps, |
1813dd64 | 774 | const struct xfs_buf_ops *ops) |
1da177e4 | 775 | { |
efa7c9f9 | 776 | if (bdi_read_congested(target->bt_bdev->bd_bdi)) |
1da177e4 LT |
777 | return; |
778 | ||
6dde2707 | 779 | xfs_buf_read_map(target, map, nmaps, |
1813dd64 | 780 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD, ops); |
1da177e4 LT |
781 | } |
782 | ||
5adc94c2 DC |
783 | /* |
784 | * Read an uncached buffer from disk. Allocates and returns a locked | |
785 | * buffer containing the disk contents or nothing. | |
786 | */ | |
ba372674 | 787 | int |
5adc94c2 | 788 | xfs_buf_read_uncached( |
5adc94c2 DC |
789 | struct xfs_buftarg *target, |
790 | xfs_daddr_t daddr, | |
e70b73f8 | 791 | size_t numblks, |
c3f8fc73 | 792 | int flags, |
ba372674 | 793 | struct xfs_buf **bpp, |
1813dd64 | 794 | const struct xfs_buf_ops *ops) |
5adc94c2 | 795 | { |
eab4e633 | 796 | struct xfs_buf *bp; |
5adc94c2 | 797 | |
ba372674 DC |
798 | *bpp = NULL; |
799 | ||
e70b73f8 | 800 | bp = xfs_buf_get_uncached(target, numblks, flags); |
5adc94c2 | 801 | if (!bp) |
ba372674 | 802 | return -ENOMEM; |
5adc94c2 DC |
803 | |
804 | /* set up the buffer for a read IO */ | |
3e85c868 | 805 | ASSERT(bp->b_map_count == 1); |
ba372674 | 806 | bp->b_bn = XFS_BUF_DADDR_NULL; /* always null for uncached buffers */ |
3e85c868 | 807 | bp->b_maps[0].bm_bn = daddr; |
cbb7baab | 808 | bp->b_flags |= XBF_READ; |
1813dd64 | 809 | bp->b_ops = ops; |
5adc94c2 | 810 | |
595bff75 | 811 | xfs_buf_submit_wait(bp); |
ba372674 DC |
812 | if (bp->b_error) { |
813 | int error = bp->b_error; | |
83a0adc3 | 814 | xfs_buf_relse(bp); |
ba372674 | 815 | return error; |
83a0adc3 | 816 | } |
ba372674 DC |
817 | |
818 | *bpp = bp; | |
819 | return 0; | |
1da177e4 LT |
820 | } |
821 | ||
44396476 DC |
822 | /* |
823 | * Return a buffer allocated as an empty buffer and associated to external | |
824 | * memory via xfs_buf_associate_memory() back to it's empty state. | |
825 | */ | |
826 | void | |
827 | xfs_buf_set_empty( | |
828 | struct xfs_buf *bp, | |
e70b73f8 | 829 | size_t numblks) |
44396476 DC |
830 | { |
831 | if (bp->b_pages) | |
832 | _xfs_buf_free_pages(bp); | |
833 | ||
834 | bp->b_pages = NULL; | |
835 | bp->b_page_count = 0; | |
836 | bp->b_addr = NULL; | |
4e94b71b | 837 | bp->b_length = numblks; |
aa0e8833 | 838 | bp->b_io_length = numblks; |
3e85c868 DC |
839 | |
840 | ASSERT(bp->b_map_count == 1); | |
44396476 | 841 | bp->b_bn = XFS_BUF_DADDR_NULL; |
3e85c868 DC |
842 | bp->b_maps[0].bm_bn = XFS_BUF_DADDR_NULL; |
843 | bp->b_maps[0].bm_len = bp->b_length; | |
44396476 DC |
844 | } |
845 | ||
1da177e4 LT |
846 | static inline struct page * |
847 | mem_to_page( | |
848 | void *addr) | |
849 | { | |
9e2779fa | 850 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
851 | return virt_to_page(addr); |
852 | } else { | |
853 | return vmalloc_to_page(addr); | |
854 | } | |
855 | } | |
856 | ||
857 | int | |
ce8e922c NS |
858 | xfs_buf_associate_memory( |
859 | xfs_buf_t *bp, | |
1da177e4 LT |
860 | void *mem, |
861 | size_t len) | |
862 | { | |
863 | int rval; | |
864 | int i = 0; | |
d1afb678 LM |
865 | unsigned long pageaddr; |
866 | unsigned long offset; | |
867 | size_t buflen; | |
1da177e4 LT |
868 | int page_count; |
869 | ||
0e6e847f | 870 | pageaddr = (unsigned long)mem & PAGE_MASK; |
d1afb678 | 871 | offset = (unsigned long)mem - pageaddr; |
0e6e847f DC |
872 | buflen = PAGE_ALIGN(len + offset); |
873 | page_count = buflen >> PAGE_SHIFT; | |
1da177e4 LT |
874 | |
875 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
876 | if (bp->b_pages) |
877 | _xfs_buf_free_pages(bp); | |
1da177e4 | 878 | |
ce8e922c NS |
879 | bp->b_pages = NULL; |
880 | bp->b_addr = mem; | |
1da177e4 | 881 | |
87937bf8 | 882 | rval = _xfs_buf_get_pages(bp, page_count); |
1da177e4 LT |
883 | if (rval) |
884 | return rval; | |
885 | ||
ce8e922c | 886 | bp->b_offset = offset; |
d1afb678 LM |
887 | |
888 | for (i = 0; i < bp->b_page_count; i++) { | |
889 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
0e6e847f | 890 | pageaddr += PAGE_SIZE; |
1da177e4 | 891 | } |
1da177e4 | 892 | |
aa0e8833 | 893 | bp->b_io_length = BTOBB(len); |
4e94b71b | 894 | bp->b_length = BTOBB(buflen); |
1da177e4 LT |
895 | |
896 | return 0; | |
897 | } | |
898 | ||
899 | xfs_buf_t * | |
686865f7 DC |
900 | xfs_buf_get_uncached( |
901 | struct xfs_buftarg *target, | |
e70b73f8 | 902 | size_t numblks, |
686865f7 | 903 | int flags) |
1da177e4 | 904 | { |
e70b73f8 | 905 | unsigned long page_count; |
1fa40b01 | 906 | int error, i; |
3e85c868 DC |
907 | struct xfs_buf *bp; |
908 | DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks); | |
1da177e4 | 909 | |
c891c30a BF |
910 | /* flags might contain irrelevant bits, pass only what we care about */ |
911 | bp = _xfs_buf_alloc(target, &map, 1, flags & XBF_NO_IOACCT); | |
1da177e4 LT |
912 | if (unlikely(bp == NULL)) |
913 | goto fail; | |
1da177e4 | 914 | |
e70b73f8 | 915 | page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT; |
87937bf8 | 916 | error = _xfs_buf_get_pages(bp, page_count); |
1fa40b01 | 917 | if (error) |
1da177e4 LT |
918 | goto fail_free_buf; |
919 | ||
1fa40b01 | 920 | for (i = 0; i < page_count; i++) { |
686865f7 | 921 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
922 | if (!bp->b_pages[i]) |
923 | goto fail_free_mem; | |
1da177e4 | 924 | } |
1fa40b01 | 925 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 926 | |
611c9946 | 927 | error = _xfs_buf_map_pages(bp, 0); |
1fa40b01 | 928 | if (unlikely(error)) { |
4f10700a | 929 | xfs_warn(target->bt_mount, |
08e96e1a | 930 | "%s: failed to map pages", __func__); |
1da177e4 | 931 | goto fail_free_mem; |
1fa40b01 | 932 | } |
1da177e4 | 933 | |
686865f7 | 934 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 935 | return bp; |
1fa40b01 | 936 | |
1da177e4 | 937 | fail_free_mem: |
1fa40b01 CH |
938 | while (--i >= 0) |
939 | __free_page(bp->b_pages[i]); | |
ca165b88 | 940 | _xfs_buf_free_pages(bp); |
1da177e4 | 941 | fail_free_buf: |
3e85c868 | 942 | xfs_buf_free_maps(bp); |
4347b9d7 | 943 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
944 | fail: |
945 | return NULL; | |
946 | } | |
947 | ||
948 | /* | |
1da177e4 LT |
949 | * Increment reference count on buffer, to hold the buffer concurrently |
950 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
951 | * Must hold the buffer already to call this function. |
952 | */ | |
953 | void | |
ce8e922c NS |
954 | xfs_buf_hold( |
955 | xfs_buf_t *bp) | |
1da177e4 | 956 | { |
0b1b213f | 957 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 958 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
959 | } |
960 | ||
961 | /* | |
9c7504aa BF |
962 | * Release a hold on the specified buffer. If the hold count is 1, the buffer is |
963 | * placed on LRU or freed (depending on b_lru_ref). | |
1da177e4 LT |
964 | */ |
965 | void | |
ce8e922c NS |
966 | xfs_buf_rele( |
967 | xfs_buf_t *bp) | |
1da177e4 | 968 | { |
74f75a0c | 969 | struct xfs_perag *pag = bp->b_pag; |
9c7504aa BF |
970 | bool release; |
971 | bool freebuf = false; | |
1da177e4 | 972 | |
0b1b213f | 973 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 974 | |
74f75a0c | 975 | if (!pag) { |
430cbeb8 | 976 | ASSERT(list_empty(&bp->b_lru)); |
9c7504aa BF |
977 | if (atomic_dec_and_test(&bp->b_hold)) { |
978 | xfs_buf_ioacct_dec(bp); | |
fad3aa1e | 979 | xfs_buf_free(bp); |
9c7504aa | 980 | } |
fad3aa1e NS |
981 | return; |
982 | } | |
983 | ||
3790689f | 984 | ASSERT(atomic_read(&bp->b_hold) > 0); |
a4082357 | 985 | |
9c7504aa BF |
986 | release = atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock); |
987 | spin_lock(&bp->b_lock); | |
988 | if (!release) { | |
989 | /* | |
990 | * Drop the in-flight state if the buffer is already on the LRU | |
991 | * and it holds the only reference. This is racy because we | |
992 | * haven't acquired the pag lock, but the use of _XBF_IN_FLIGHT | |
993 | * ensures the decrement occurs only once per-buf. | |
994 | */ | |
995 | if ((atomic_read(&bp->b_hold) == 1) && !list_empty(&bp->b_lru)) | |
63db7c81 | 996 | __xfs_buf_ioacct_dec(bp); |
9c7504aa BF |
997 | goto out_unlock; |
998 | } | |
999 | ||
1000 | /* the last reference has been dropped ... */ | |
63db7c81 | 1001 | __xfs_buf_ioacct_dec(bp); |
9c7504aa BF |
1002 | if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) { |
1003 | /* | |
1004 | * If the buffer is added to the LRU take a new reference to the | |
1005 | * buffer for the LRU and clear the (now stale) dispose list | |
1006 | * state flag | |
1007 | */ | |
1008 | if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) { | |
1009 | bp->b_state &= ~XFS_BSTATE_DISPOSE; | |
1010 | atomic_inc(&bp->b_hold); | |
1da177e4 | 1011 | } |
9c7504aa BF |
1012 | spin_unlock(&pag->pag_buf_lock); |
1013 | } else { | |
1014 | /* | |
1015 | * most of the time buffers will already be removed from the | |
1016 | * LRU, so optimise that case by checking for the | |
1017 | * XFS_BSTATE_DISPOSE flag indicating the last list the buffer | |
1018 | * was on was the disposal list | |
1019 | */ | |
1020 | if (!(bp->b_state & XFS_BSTATE_DISPOSE)) { | |
1021 | list_lru_del(&bp->b_target->bt_lru, &bp->b_lru); | |
1022 | } else { | |
1023 | ASSERT(list_empty(&bp->b_lru)); | |
1da177e4 | 1024 | } |
9c7504aa BF |
1025 | |
1026 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); | |
6031e73a LS |
1027 | rhashtable_remove_fast(&pag->pag_buf_hash, &bp->b_rhash_head, |
1028 | xfs_buf_hash_params); | |
9c7504aa BF |
1029 | spin_unlock(&pag->pag_buf_lock); |
1030 | xfs_perag_put(pag); | |
1031 | freebuf = true; | |
1da177e4 | 1032 | } |
9c7504aa BF |
1033 | |
1034 | out_unlock: | |
1035 | spin_unlock(&bp->b_lock); | |
1036 | ||
1037 | if (freebuf) | |
1038 | xfs_buf_free(bp); | |
1da177e4 LT |
1039 | } |
1040 | ||
1041 | ||
1042 | /* | |
0e6e847f | 1043 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
1044 | * |
1045 | * If we come across a stale, pinned, locked buffer, we know that we are | |
1046 | * being asked to lock a buffer that has been reallocated. Because it is | |
1047 | * pinned, we know that the log has not been pushed to disk and hence it | |
1048 | * will still be locked. Rather than continuing to have trylock attempts | |
1049 | * fail until someone else pushes the log, push it ourselves before | |
1050 | * returning. This means that the xfsaild will not get stuck trying | |
1051 | * to push on stale inode buffers. | |
1da177e4 LT |
1052 | */ |
1053 | int | |
0c842ad4 CH |
1054 | xfs_buf_trylock( |
1055 | struct xfs_buf *bp) | |
1da177e4 LT |
1056 | { |
1057 | int locked; | |
1058 | ||
ce8e922c | 1059 | locked = down_trylock(&bp->b_sema) == 0; |
479c6412 | 1060 | if (locked) { |
ce8e922c | 1061 | XB_SET_OWNER(bp); |
479c6412 DW |
1062 | trace_xfs_buf_trylock(bp, _RET_IP_); |
1063 | } else { | |
1064 | trace_xfs_buf_trylock_fail(bp, _RET_IP_); | |
1065 | } | |
0c842ad4 | 1066 | return locked; |
1da177e4 | 1067 | } |
1da177e4 LT |
1068 | |
1069 | /* | |
0e6e847f | 1070 | * Lock a buffer object. |
ed3b4d6c DC |
1071 | * |
1072 | * If we come across a stale, pinned, locked buffer, we know that we | |
1073 | * are being asked to lock a buffer that has been reallocated. Because | |
1074 | * it is pinned, we know that the log has not been pushed to disk and | |
1075 | * hence it will still be locked. Rather than sleeping until someone | |
1076 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 1077 | */ |
ce8e922c NS |
1078 | void |
1079 | xfs_buf_lock( | |
0c842ad4 | 1080 | struct xfs_buf *bp) |
1da177e4 | 1081 | { |
0b1b213f CH |
1082 | trace_xfs_buf_lock(bp, _RET_IP_); |
1083 | ||
ed3b4d6c | 1084 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 1085 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
1086 | down(&bp->b_sema); |
1087 | XB_SET_OWNER(bp); | |
0b1b213f CH |
1088 | |
1089 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
1090 | } |
1091 | ||
1da177e4 | 1092 | void |
ce8e922c | 1093 | xfs_buf_unlock( |
0c842ad4 | 1094 | struct xfs_buf *bp) |
1da177e4 | 1095 | { |
20e8a063 BF |
1096 | ASSERT(xfs_buf_islocked(bp)); |
1097 | ||
ce8e922c NS |
1098 | XB_CLEAR_OWNER(bp); |
1099 | up(&bp->b_sema); | |
0b1b213f CH |
1100 | |
1101 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
1102 | } |
1103 | ||
ce8e922c NS |
1104 | STATIC void |
1105 | xfs_buf_wait_unpin( | |
1106 | xfs_buf_t *bp) | |
1da177e4 LT |
1107 | { |
1108 | DECLARE_WAITQUEUE (wait, current); | |
1109 | ||
ce8e922c | 1110 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
1111 | return; |
1112 | ||
ce8e922c | 1113 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1114 | for (;;) { |
1115 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 1116 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 1117 | break; |
7eaceacc | 1118 | io_schedule(); |
1da177e4 | 1119 | } |
ce8e922c | 1120 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1121 | set_current_state(TASK_RUNNING); |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * Buffer Utility Routines | |
1126 | */ | |
1127 | ||
e8aaba9a DC |
1128 | void |
1129 | xfs_buf_ioend( | |
1130 | struct xfs_buf *bp) | |
1da177e4 | 1131 | { |
e8aaba9a DC |
1132 | bool read = bp->b_flags & XBF_READ; |
1133 | ||
1134 | trace_xfs_buf_iodone(bp, _RET_IP_); | |
1813dd64 DC |
1135 | |
1136 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); | |
d5929de8 | 1137 | |
61be9c52 DC |
1138 | /* |
1139 | * Pull in IO completion errors now. We are guaranteed to be running | |
1140 | * single threaded, so we don't need the lock to read b_io_error. | |
1141 | */ | |
1142 | if (!bp->b_error && bp->b_io_error) | |
1143 | xfs_buf_ioerror(bp, bp->b_io_error); | |
1144 | ||
e8aaba9a DC |
1145 | /* Only validate buffers that were read without errors */ |
1146 | if (read && !bp->b_error && bp->b_ops) { | |
1147 | ASSERT(!bp->b_iodone); | |
1813dd64 | 1148 | bp->b_ops->verify_read(bp); |
e8aaba9a DC |
1149 | } |
1150 | ||
1151 | if (!bp->b_error) | |
1152 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1153 | |
80f6c29d | 1154 | if (bp->b_iodone) |
ce8e922c NS |
1155 | (*(bp->b_iodone))(bp); |
1156 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 | 1157 | xfs_buf_relse(bp); |
595bff75 | 1158 | else |
1813dd64 | 1159 | complete(&bp->b_iowait); |
1da177e4 LT |
1160 | } |
1161 | ||
e8aaba9a DC |
1162 | static void |
1163 | xfs_buf_ioend_work( | |
1164 | struct work_struct *work) | |
1da177e4 | 1165 | { |
e8aaba9a | 1166 | struct xfs_buf *bp = |
b29c70f5 | 1167 | container_of(work, xfs_buf_t, b_ioend_work); |
0b1b213f | 1168 | |
e8aaba9a DC |
1169 | xfs_buf_ioend(bp); |
1170 | } | |
1da177e4 | 1171 | |
211fe1a4 | 1172 | static void |
e8aaba9a DC |
1173 | xfs_buf_ioend_async( |
1174 | struct xfs_buf *bp) | |
1175 | { | |
b29c70f5 BF |
1176 | INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work); |
1177 | queue_work(bp->b_ioend_wq, &bp->b_ioend_work); | |
1da177e4 LT |
1178 | } |
1179 | ||
1da177e4 | 1180 | void |
ce8e922c NS |
1181 | xfs_buf_ioerror( |
1182 | xfs_buf_t *bp, | |
1183 | int error) | |
1da177e4 | 1184 | { |
2451337d DC |
1185 | ASSERT(error <= 0 && error >= -1000); |
1186 | bp->b_error = error; | |
0b1b213f | 1187 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1188 | } |
1189 | ||
901796af CH |
1190 | void |
1191 | xfs_buf_ioerror_alert( | |
1192 | struct xfs_buf *bp, | |
1193 | const char *func) | |
1194 | { | |
1195 | xfs_alert(bp->b_target->bt_mount, | |
aa0e8833 | 1196 | "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d", |
2451337d | 1197 | (__uint64_t)XFS_BUF_ADDR(bp), func, -bp->b_error, bp->b_length); |
901796af CH |
1198 | } |
1199 | ||
a2dcf5df CH |
1200 | int |
1201 | xfs_bwrite( | |
1202 | struct xfs_buf *bp) | |
1203 | { | |
1204 | int error; | |
1205 | ||
1206 | ASSERT(xfs_buf_islocked(bp)); | |
1207 | ||
1208 | bp->b_flags |= XBF_WRITE; | |
27187754 DC |
1209 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q | |
1210 | XBF_WRITE_FAIL | XBF_DONE); | |
a2dcf5df | 1211 | |
595bff75 | 1212 | error = xfs_buf_submit_wait(bp); |
a2dcf5df CH |
1213 | if (error) { |
1214 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1215 | SHUTDOWN_META_IO_ERROR); | |
1216 | } | |
1217 | return error; | |
1218 | } | |
1219 | ||
9bdd9bd6 | 1220 | static void |
ce8e922c | 1221 | xfs_buf_bio_end_io( |
4246a0b6 | 1222 | struct bio *bio) |
1da177e4 | 1223 | { |
9bdd9bd6 | 1224 | struct xfs_buf *bp = (struct xfs_buf *)bio->bi_private; |
1da177e4 | 1225 | |
37eb17e6 DC |
1226 | /* |
1227 | * don't overwrite existing errors - otherwise we can lose errors on | |
1228 | * buffers that require multiple bios to complete. | |
1229 | */ | |
9bdd9bd6 BF |
1230 | if (bio->bi_error) |
1231 | cmpxchg(&bp->b_io_error, 0, bio->bi_error); | |
1da177e4 | 1232 | |
37eb17e6 | 1233 | if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
73c77e2c JB |
1234 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); |
1235 | ||
e8aaba9a DC |
1236 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
1237 | xfs_buf_ioend_async(bp); | |
1da177e4 | 1238 | bio_put(bio); |
1da177e4 LT |
1239 | } |
1240 | ||
3e85c868 DC |
1241 | static void |
1242 | xfs_buf_ioapply_map( | |
1243 | struct xfs_buf *bp, | |
1244 | int map, | |
1245 | int *buf_offset, | |
1246 | int *count, | |
50bfcd0c MC |
1247 | int op, |
1248 | int op_flags) | |
1da177e4 | 1249 | { |
3e85c868 DC |
1250 | int page_index; |
1251 | int total_nr_pages = bp->b_page_count; | |
1252 | int nr_pages; | |
1253 | struct bio *bio; | |
1254 | sector_t sector = bp->b_maps[map].bm_bn; | |
1255 | int size; | |
1256 | int offset; | |
1da177e4 | 1257 | |
ce8e922c | 1258 | total_nr_pages = bp->b_page_count; |
1da177e4 | 1259 | |
3e85c868 DC |
1260 | /* skip the pages in the buffer before the start offset */ |
1261 | page_index = 0; | |
1262 | offset = *buf_offset; | |
1263 | while (offset >= PAGE_SIZE) { | |
1264 | page_index++; | |
1265 | offset -= PAGE_SIZE; | |
f538d4da CH |
1266 | } |
1267 | ||
3e85c868 DC |
1268 | /* |
1269 | * Limit the IO size to the length of the current vector, and update the | |
1270 | * remaining IO count for the next time around. | |
1271 | */ | |
1272 | size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count); | |
1273 | *count -= size; | |
1274 | *buf_offset += size; | |
34951f5c | 1275 | |
1da177e4 | 1276 | next_chunk: |
ce8e922c | 1277 | atomic_inc(&bp->b_io_remaining); |
c908e380 | 1278 | nr_pages = min(total_nr_pages, BIO_MAX_PAGES); |
1da177e4 LT |
1279 | |
1280 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1281 | bio->bi_bdev = bp->b_target->bt_bdev; |
4f024f37 | 1282 | bio->bi_iter.bi_sector = sector; |
ce8e922c NS |
1283 | bio->bi_end_io = xfs_buf_bio_end_io; |
1284 | bio->bi_private = bp; | |
50bfcd0c | 1285 | bio_set_op_attrs(bio, op, op_flags); |
0e6e847f | 1286 | |
3e85c868 | 1287 | for (; size && nr_pages; nr_pages--, page_index++) { |
0e6e847f | 1288 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1289 | |
1290 | if (nbytes > size) | |
1291 | nbytes = size; | |
1292 | ||
3e85c868 DC |
1293 | rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes, |
1294 | offset); | |
ce8e922c | 1295 | if (rbytes < nbytes) |
1da177e4 LT |
1296 | break; |
1297 | ||
1298 | offset = 0; | |
aa0e8833 | 1299 | sector += BTOBB(nbytes); |
1da177e4 LT |
1300 | size -= nbytes; |
1301 | total_nr_pages--; | |
1302 | } | |
1303 | ||
4f024f37 | 1304 | if (likely(bio->bi_iter.bi_size)) { |
73c77e2c JB |
1305 | if (xfs_buf_is_vmapped(bp)) { |
1306 | flush_kernel_vmap_range(bp->b_addr, | |
1307 | xfs_buf_vmap_len(bp)); | |
1308 | } | |
4e49ea4a | 1309 | submit_bio(bio); |
1da177e4 LT |
1310 | if (size) |
1311 | goto next_chunk; | |
1312 | } else { | |
37eb17e6 DC |
1313 | /* |
1314 | * This is guaranteed not to be the last io reference count | |
595bff75 | 1315 | * because the caller (xfs_buf_submit) holds a count itself. |
37eb17e6 DC |
1316 | */ |
1317 | atomic_dec(&bp->b_io_remaining); | |
2451337d | 1318 | xfs_buf_ioerror(bp, -EIO); |
ec53d1db | 1319 | bio_put(bio); |
1da177e4 | 1320 | } |
3e85c868 DC |
1321 | |
1322 | } | |
1323 | ||
1324 | STATIC void | |
1325 | _xfs_buf_ioapply( | |
1326 | struct xfs_buf *bp) | |
1327 | { | |
1328 | struct blk_plug plug; | |
50bfcd0c MC |
1329 | int op; |
1330 | int op_flags = 0; | |
3e85c868 DC |
1331 | int offset; |
1332 | int size; | |
1333 | int i; | |
1334 | ||
c163f9a1 DC |
1335 | /* |
1336 | * Make sure we capture only current IO errors rather than stale errors | |
1337 | * left over from previous use of the buffer (e.g. failed readahead). | |
1338 | */ | |
1339 | bp->b_error = 0; | |
1340 | ||
b29c70f5 BF |
1341 | /* |
1342 | * Initialize the I/O completion workqueue if we haven't yet or the | |
1343 | * submitter has not opted to specify a custom one. | |
1344 | */ | |
1345 | if (!bp->b_ioend_wq) | |
1346 | bp->b_ioend_wq = bp->b_target->bt_mount->m_buf_workqueue; | |
1347 | ||
3e85c868 | 1348 | if (bp->b_flags & XBF_WRITE) { |
50bfcd0c | 1349 | op = REQ_OP_WRITE; |
3e85c868 | 1350 | if (bp->b_flags & XBF_SYNCIO) |
70fd7614 | 1351 | op_flags = REQ_SYNC; |
3e85c868 | 1352 | if (bp->b_flags & XBF_FUA) |
50bfcd0c | 1353 | op_flags |= REQ_FUA; |
3e85c868 | 1354 | if (bp->b_flags & XBF_FLUSH) |
28a8f0d3 | 1355 | op_flags |= REQ_PREFLUSH; |
1813dd64 DC |
1356 | |
1357 | /* | |
1358 | * Run the write verifier callback function if it exists. If | |
1359 | * this function fails it will mark the buffer with an error and | |
1360 | * the IO should not be dispatched. | |
1361 | */ | |
1362 | if (bp->b_ops) { | |
1363 | bp->b_ops->verify_write(bp); | |
1364 | if (bp->b_error) { | |
1365 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1366 | SHUTDOWN_CORRUPT_INCORE); | |
1367 | return; | |
1368 | } | |
400b9d88 DC |
1369 | } else if (bp->b_bn != XFS_BUF_DADDR_NULL) { |
1370 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
1371 | ||
1372 | /* | |
1373 | * non-crc filesystems don't attach verifiers during | |
1374 | * log recovery, so don't warn for such filesystems. | |
1375 | */ | |
1376 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
1377 | xfs_warn(mp, | |
1378 | "%s: no ops on block 0x%llx/0x%x", | |
1379 | __func__, bp->b_bn, bp->b_length); | |
1380 | xfs_hex_dump(bp->b_addr, 64); | |
1381 | dump_stack(); | |
1382 | } | |
1813dd64 | 1383 | } |
3e85c868 | 1384 | } else if (bp->b_flags & XBF_READ_AHEAD) { |
50bfcd0c MC |
1385 | op = REQ_OP_READ; |
1386 | op_flags = REQ_RAHEAD; | |
3e85c868 | 1387 | } else { |
50bfcd0c | 1388 | op = REQ_OP_READ; |
3e85c868 DC |
1389 | } |
1390 | ||
1391 | /* we only use the buffer cache for meta-data */ | |
50bfcd0c | 1392 | op_flags |= REQ_META; |
3e85c868 DC |
1393 | |
1394 | /* | |
1395 | * Walk all the vectors issuing IO on them. Set up the initial offset | |
1396 | * into the buffer and the desired IO size before we start - | |
1397 | * _xfs_buf_ioapply_vec() will modify them appropriately for each | |
1398 | * subsequent call. | |
1399 | */ | |
1400 | offset = bp->b_offset; | |
1401 | size = BBTOB(bp->b_io_length); | |
1402 | blk_start_plug(&plug); | |
1403 | for (i = 0; i < bp->b_map_count; i++) { | |
50bfcd0c | 1404 | xfs_buf_ioapply_map(bp, i, &offset, &size, op, op_flags); |
3e85c868 DC |
1405 | if (bp->b_error) |
1406 | break; | |
1407 | if (size <= 0) | |
1408 | break; /* all done */ | |
1409 | } | |
1410 | blk_finish_plug(&plug); | |
1da177e4 LT |
1411 | } |
1412 | ||
595bff75 DC |
1413 | /* |
1414 | * Asynchronous IO submission path. This transfers the buffer lock ownership and | |
1415 | * the current reference to the IO. It is not safe to reference the buffer after | |
1416 | * a call to this function unless the caller holds an additional reference | |
1417 | * itself. | |
1418 | */ | |
0e95f19a | 1419 | void |
595bff75 DC |
1420 | xfs_buf_submit( |
1421 | struct xfs_buf *bp) | |
1da177e4 | 1422 | { |
595bff75 | 1423 | trace_xfs_buf_submit(bp, _RET_IP_); |
1da177e4 | 1424 | |
43ff2122 | 1425 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
595bff75 DC |
1426 | ASSERT(bp->b_flags & XBF_ASYNC); |
1427 | ||
1428 | /* on shutdown we stale and complete the buffer immediately */ | |
1429 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { | |
1430 | xfs_buf_ioerror(bp, -EIO); | |
1431 | bp->b_flags &= ~XBF_DONE; | |
1432 | xfs_buf_stale(bp); | |
1433 | xfs_buf_ioend(bp); | |
1434 | return; | |
1435 | } | |
1da177e4 | 1436 | |
375ec69d | 1437 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1438 | xfs_buf_wait_unpin(bp); |
e11bb805 | 1439 | |
61be9c52 DC |
1440 | /* clear the internal error state to avoid spurious errors */ |
1441 | bp->b_io_error = 0; | |
1442 | ||
e11bb805 | 1443 | /* |
595bff75 DC |
1444 | * The caller's reference is released during I/O completion. |
1445 | * This occurs some time after the last b_io_remaining reference is | |
1446 | * released, so after we drop our Io reference we have to have some | |
1447 | * other reference to ensure the buffer doesn't go away from underneath | |
1448 | * us. Take a direct reference to ensure we have safe access to the | |
1449 | * buffer until we are finished with it. | |
e11bb805 | 1450 | */ |
ce8e922c | 1451 | xfs_buf_hold(bp); |
1da177e4 | 1452 | |
8d6c1210 | 1453 | /* |
e11bb805 DC |
1454 | * Set the count to 1 initially, this will stop an I/O completion |
1455 | * callout which happens before we have started all the I/O from calling | |
1456 | * xfs_buf_ioend too early. | |
1da177e4 | 1457 | */ |
ce8e922c | 1458 | atomic_set(&bp->b_io_remaining, 1); |
9c7504aa | 1459 | xfs_buf_ioacct_inc(bp); |
ce8e922c | 1460 | _xfs_buf_ioapply(bp); |
e11bb805 | 1461 | |
8d6c1210 | 1462 | /* |
595bff75 DC |
1463 | * If _xfs_buf_ioapply failed, we can get back here with only the IO |
1464 | * reference we took above. If we drop it to zero, run completion so | |
1465 | * that we don't return to the caller with completion still pending. | |
8d6c1210 | 1466 | */ |
e8aaba9a | 1467 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
595bff75 | 1468 | if (bp->b_error) |
e8aaba9a DC |
1469 | xfs_buf_ioend(bp); |
1470 | else | |
1471 | xfs_buf_ioend_async(bp); | |
1472 | } | |
1da177e4 | 1473 | |
ce8e922c | 1474 | xfs_buf_rele(bp); |
595bff75 | 1475 | /* Note: it is not safe to reference bp now we've dropped our ref */ |
1da177e4 LT |
1476 | } |
1477 | ||
1478 | /* | |
595bff75 | 1479 | * Synchronous buffer IO submission path, read or write. |
1da177e4 LT |
1480 | */ |
1481 | int | |
595bff75 DC |
1482 | xfs_buf_submit_wait( |
1483 | struct xfs_buf *bp) | |
1da177e4 | 1484 | { |
595bff75 | 1485 | int error; |
0b1b213f | 1486 | |
595bff75 DC |
1487 | trace_xfs_buf_submit_wait(bp, _RET_IP_); |
1488 | ||
1489 | ASSERT(!(bp->b_flags & (_XBF_DELWRI_Q | XBF_ASYNC))); | |
0b1b213f | 1490 | |
595bff75 DC |
1491 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
1492 | xfs_buf_ioerror(bp, -EIO); | |
1493 | xfs_buf_stale(bp); | |
1494 | bp->b_flags &= ~XBF_DONE; | |
1495 | return -EIO; | |
1496 | } | |
1497 | ||
1498 | if (bp->b_flags & XBF_WRITE) | |
1499 | xfs_buf_wait_unpin(bp); | |
1500 | ||
1501 | /* clear the internal error state to avoid spurious errors */ | |
1502 | bp->b_io_error = 0; | |
1503 | ||
1504 | /* | |
1505 | * For synchronous IO, the IO does not inherit the submitters reference | |
1506 | * count, nor the buffer lock. Hence we cannot release the reference we | |
1507 | * are about to take until we've waited for all IO completion to occur, | |
1508 | * including any xfs_buf_ioend_async() work that may be pending. | |
1509 | */ | |
1510 | xfs_buf_hold(bp); | |
1511 | ||
1512 | /* | |
1513 | * Set the count to 1 initially, this will stop an I/O completion | |
1514 | * callout which happens before we have started all the I/O from calling | |
1515 | * xfs_buf_ioend too early. | |
1516 | */ | |
1517 | atomic_set(&bp->b_io_remaining, 1); | |
1518 | _xfs_buf_ioapply(bp); | |
1519 | ||
1520 | /* | |
1521 | * make sure we run completion synchronously if it raced with us and is | |
1522 | * already complete. | |
1523 | */ | |
1524 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) | |
1525 | xfs_buf_ioend(bp); | |
0b1b213f | 1526 | |
595bff75 DC |
1527 | /* wait for completion before gathering the error from the buffer */ |
1528 | trace_xfs_buf_iowait(bp, _RET_IP_); | |
1529 | wait_for_completion(&bp->b_iowait); | |
0b1b213f | 1530 | trace_xfs_buf_iowait_done(bp, _RET_IP_); |
595bff75 DC |
1531 | error = bp->b_error; |
1532 | ||
1533 | /* | |
1534 | * all done now, we can release the hold that keeps the buffer | |
1535 | * referenced for the entire IO. | |
1536 | */ | |
1537 | xfs_buf_rele(bp); | |
1538 | return error; | |
1da177e4 LT |
1539 | } |
1540 | ||
88ee2df7 | 1541 | void * |
ce8e922c | 1542 | xfs_buf_offset( |
88ee2df7 | 1543 | struct xfs_buf *bp, |
1da177e4 LT |
1544 | size_t offset) |
1545 | { | |
1546 | struct page *page; | |
1547 | ||
611c9946 | 1548 | if (bp->b_addr) |
62926044 | 1549 | return bp->b_addr + offset; |
1da177e4 | 1550 | |
ce8e922c | 1551 | offset += bp->b_offset; |
0e6e847f | 1552 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
88ee2df7 | 1553 | return page_address(page) + (offset & (PAGE_SIZE-1)); |
1da177e4 LT |
1554 | } |
1555 | ||
1556 | /* | |
1da177e4 LT |
1557 | * Move data into or out of a buffer. |
1558 | */ | |
1559 | void | |
ce8e922c NS |
1560 | xfs_buf_iomove( |
1561 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1562 | size_t boff, /* starting buffer offset */ |
1563 | size_t bsize, /* length to copy */ | |
b9c48649 | 1564 | void *data, /* data address */ |
ce8e922c | 1565 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 | 1566 | { |
795cac72 | 1567 | size_t bend; |
1da177e4 LT |
1568 | |
1569 | bend = boff + bsize; | |
1570 | while (boff < bend) { | |
795cac72 DC |
1571 | struct page *page; |
1572 | int page_index, page_offset, csize; | |
1573 | ||
1574 | page_index = (boff + bp->b_offset) >> PAGE_SHIFT; | |
1575 | page_offset = (boff + bp->b_offset) & ~PAGE_MASK; | |
1576 | page = bp->b_pages[page_index]; | |
1577 | csize = min_t(size_t, PAGE_SIZE - page_offset, | |
1578 | BBTOB(bp->b_io_length) - boff); | |
1da177e4 | 1579 | |
795cac72 | 1580 | ASSERT((csize + page_offset) <= PAGE_SIZE); |
1da177e4 LT |
1581 | |
1582 | switch (mode) { | |
ce8e922c | 1583 | case XBRW_ZERO: |
795cac72 | 1584 | memset(page_address(page) + page_offset, 0, csize); |
1da177e4 | 1585 | break; |
ce8e922c | 1586 | case XBRW_READ: |
795cac72 | 1587 | memcpy(data, page_address(page) + page_offset, csize); |
1da177e4 | 1588 | break; |
ce8e922c | 1589 | case XBRW_WRITE: |
795cac72 | 1590 | memcpy(page_address(page) + page_offset, data, csize); |
1da177e4 LT |
1591 | } |
1592 | ||
1593 | boff += csize; | |
1594 | data += csize; | |
1595 | } | |
1596 | } | |
1597 | ||
1598 | /* | |
ce8e922c | 1599 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1600 | */ |
1601 | ||
1602 | /* | |
430cbeb8 DC |
1603 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1604 | * returned. These buffers will have an elevated hold count, so wait on those | |
1605 | * while freeing all the buffers only held by the LRU. | |
1da177e4 | 1606 | */ |
e80dfa19 DC |
1607 | static enum lru_status |
1608 | xfs_buftarg_wait_rele( | |
1609 | struct list_head *item, | |
3f97b163 | 1610 | struct list_lru_one *lru, |
e80dfa19 DC |
1611 | spinlock_t *lru_lock, |
1612 | void *arg) | |
1613 | ||
1da177e4 | 1614 | { |
e80dfa19 | 1615 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); |
a4082357 | 1616 | struct list_head *dispose = arg; |
430cbeb8 | 1617 | |
e80dfa19 | 1618 | if (atomic_read(&bp->b_hold) > 1) { |
a4082357 | 1619 | /* need to wait, so skip it this pass */ |
e80dfa19 | 1620 | trace_xfs_buf_wait_buftarg(bp, _RET_IP_); |
a4082357 | 1621 | return LRU_SKIP; |
1da177e4 | 1622 | } |
a4082357 DC |
1623 | if (!spin_trylock(&bp->b_lock)) |
1624 | return LRU_SKIP; | |
e80dfa19 | 1625 | |
a4082357 DC |
1626 | /* |
1627 | * clear the LRU reference count so the buffer doesn't get | |
1628 | * ignored in xfs_buf_rele(). | |
1629 | */ | |
1630 | atomic_set(&bp->b_lru_ref, 0); | |
1631 | bp->b_state |= XFS_BSTATE_DISPOSE; | |
3f97b163 | 1632 | list_lru_isolate_move(lru, item, dispose); |
a4082357 DC |
1633 | spin_unlock(&bp->b_lock); |
1634 | return LRU_REMOVED; | |
1da177e4 LT |
1635 | } |
1636 | ||
e80dfa19 DC |
1637 | void |
1638 | xfs_wait_buftarg( | |
1639 | struct xfs_buftarg *btp) | |
1640 | { | |
a4082357 DC |
1641 | LIST_HEAD(dispose); |
1642 | int loop = 0; | |
1643 | ||
85bec546 | 1644 | /* |
9c7504aa BF |
1645 | * First wait on the buftarg I/O count for all in-flight buffers to be |
1646 | * released. This is critical as new buffers do not make the LRU until | |
1647 | * they are released. | |
1648 | * | |
1649 | * Next, flush the buffer workqueue to ensure all completion processing | |
1650 | * has finished. Just waiting on buffer locks is not sufficient for | |
1651 | * async IO as the reference count held over IO is not released until | |
1652 | * after the buffer lock is dropped. Hence we need to ensure here that | |
1653 | * all reference counts have been dropped before we start walking the | |
1654 | * LRU list. | |
85bec546 | 1655 | */ |
9c7504aa BF |
1656 | while (percpu_counter_sum(&btp->bt_io_count)) |
1657 | delay(100); | |
800b2694 | 1658 | flush_workqueue(btp->bt_mount->m_buf_workqueue); |
85bec546 | 1659 | |
a4082357 DC |
1660 | /* loop until there is nothing left on the lru list. */ |
1661 | while (list_lru_count(&btp->bt_lru)) { | |
e80dfa19 | 1662 | list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele, |
a4082357 DC |
1663 | &dispose, LONG_MAX); |
1664 | ||
1665 | while (!list_empty(&dispose)) { | |
1666 | struct xfs_buf *bp; | |
1667 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1668 | list_del_init(&bp->b_lru); | |
ac8809f9 DC |
1669 | if (bp->b_flags & XBF_WRITE_FAIL) { |
1670 | xfs_alert(btp->bt_mount, | |
f41febd2 | 1671 | "Corruption Alert: Buffer at block 0x%llx had permanent write failures!", |
ac8809f9 | 1672 | (long long)bp->b_bn); |
f41febd2 JP |
1673 | xfs_alert(btp->bt_mount, |
1674 | "Please run xfs_repair to determine the extent of the problem."); | |
ac8809f9 | 1675 | } |
a4082357 DC |
1676 | xfs_buf_rele(bp); |
1677 | } | |
1678 | if (loop++ != 0) | |
1679 | delay(100); | |
1680 | } | |
e80dfa19 DC |
1681 | } |
1682 | ||
1683 | static enum lru_status | |
1684 | xfs_buftarg_isolate( | |
1685 | struct list_head *item, | |
3f97b163 | 1686 | struct list_lru_one *lru, |
e80dfa19 DC |
1687 | spinlock_t *lru_lock, |
1688 | void *arg) | |
1689 | { | |
1690 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); | |
1691 | struct list_head *dispose = arg; | |
1692 | ||
a4082357 DC |
1693 | /* |
1694 | * we are inverting the lru lock/bp->b_lock here, so use a trylock. | |
1695 | * If we fail to get the lock, just skip it. | |
1696 | */ | |
1697 | if (!spin_trylock(&bp->b_lock)) | |
1698 | return LRU_SKIP; | |
e80dfa19 DC |
1699 | /* |
1700 | * Decrement the b_lru_ref count unless the value is already | |
1701 | * zero. If the value is already zero, we need to reclaim the | |
1702 | * buffer, otherwise it gets another trip through the LRU. | |
1703 | */ | |
a4082357 DC |
1704 | if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) { |
1705 | spin_unlock(&bp->b_lock); | |
e80dfa19 | 1706 | return LRU_ROTATE; |
a4082357 | 1707 | } |
e80dfa19 | 1708 | |
a4082357 | 1709 | bp->b_state |= XFS_BSTATE_DISPOSE; |
3f97b163 | 1710 | list_lru_isolate_move(lru, item, dispose); |
a4082357 | 1711 | spin_unlock(&bp->b_lock); |
e80dfa19 DC |
1712 | return LRU_REMOVED; |
1713 | } | |
1714 | ||
addbda40 | 1715 | static unsigned long |
e80dfa19 | 1716 | xfs_buftarg_shrink_scan( |
ff57ab21 | 1717 | struct shrinker *shrink, |
1495f230 | 1718 | struct shrink_control *sc) |
a6867a68 | 1719 | { |
ff57ab21 DC |
1720 | struct xfs_buftarg *btp = container_of(shrink, |
1721 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1722 | LIST_HEAD(dispose); |
addbda40 | 1723 | unsigned long freed; |
430cbeb8 | 1724 | |
503c358c VD |
1725 | freed = list_lru_shrink_walk(&btp->bt_lru, sc, |
1726 | xfs_buftarg_isolate, &dispose); | |
430cbeb8 DC |
1727 | |
1728 | while (!list_empty(&dispose)) { | |
e80dfa19 | 1729 | struct xfs_buf *bp; |
430cbeb8 DC |
1730 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); |
1731 | list_del_init(&bp->b_lru); | |
1732 | xfs_buf_rele(bp); | |
1733 | } | |
1734 | ||
e80dfa19 DC |
1735 | return freed; |
1736 | } | |
1737 | ||
addbda40 | 1738 | static unsigned long |
e80dfa19 DC |
1739 | xfs_buftarg_shrink_count( |
1740 | struct shrinker *shrink, | |
1741 | struct shrink_control *sc) | |
1742 | { | |
1743 | struct xfs_buftarg *btp = container_of(shrink, | |
1744 | struct xfs_buftarg, bt_shrinker); | |
503c358c | 1745 | return list_lru_shrink_count(&btp->bt_lru, sc); |
a6867a68 DC |
1746 | } |
1747 | ||
1da177e4 LT |
1748 | void |
1749 | xfs_free_buftarg( | |
b7963133 CH |
1750 | struct xfs_mount *mp, |
1751 | struct xfs_buftarg *btp) | |
1da177e4 | 1752 | { |
ff57ab21 | 1753 | unregister_shrinker(&btp->bt_shrinker); |
9c7504aa BF |
1754 | ASSERT(percpu_counter_sum(&btp->bt_io_count) == 0); |
1755 | percpu_counter_destroy(&btp->bt_io_count); | |
f5e1dd34 | 1756 | list_lru_destroy(&btp->bt_lru); |
ff57ab21 | 1757 | |
2291dab2 | 1758 | xfs_blkdev_issue_flush(btp); |
a6867a68 | 1759 | |
f0e2d93c | 1760 | kmem_free(btp); |
1da177e4 LT |
1761 | } |
1762 | ||
3fefdeee ES |
1763 | int |
1764 | xfs_setsize_buftarg( | |
1da177e4 | 1765 | xfs_buftarg_t *btp, |
3fefdeee | 1766 | unsigned int sectorsize) |
1da177e4 | 1767 | { |
7c71ee78 | 1768 | /* Set up metadata sector size info */ |
6da54179 ES |
1769 | btp->bt_meta_sectorsize = sectorsize; |
1770 | btp->bt_meta_sectormask = sectorsize - 1; | |
1da177e4 | 1771 | |
ce8e922c | 1772 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
4f10700a | 1773 | xfs_warn(btp->bt_mount, |
a1c6f057 DM |
1774 | "Cannot set_blocksize to %u on device %pg", |
1775 | sectorsize, btp->bt_bdev); | |
2451337d | 1776 | return -EINVAL; |
1da177e4 LT |
1777 | } |
1778 | ||
7c71ee78 ES |
1779 | /* Set up device logical sector size mask */ |
1780 | btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev); | |
1781 | btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1; | |
1782 | ||
1da177e4 LT |
1783 | return 0; |
1784 | } | |
1785 | ||
1786 | /* | |
3fefdeee ES |
1787 | * When allocating the initial buffer target we have not yet |
1788 | * read in the superblock, so don't know what sized sectors | |
1789 | * are being used at this early stage. Play safe. | |
ce8e922c | 1790 | */ |
1da177e4 LT |
1791 | STATIC int |
1792 | xfs_setsize_buftarg_early( | |
1793 | xfs_buftarg_t *btp, | |
1794 | struct block_device *bdev) | |
1795 | { | |
a96c4151 | 1796 | return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev)); |
1da177e4 LT |
1797 | } |
1798 | ||
1da177e4 LT |
1799 | xfs_buftarg_t * |
1800 | xfs_alloc_buftarg( | |
ebad861b | 1801 | struct xfs_mount *mp, |
34dcefd7 | 1802 | struct block_device *bdev) |
1da177e4 LT |
1803 | { |
1804 | xfs_buftarg_t *btp; | |
1805 | ||
b17cb364 | 1806 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP | KM_NOFS); |
1da177e4 | 1807 | |
ebad861b | 1808 | btp->bt_mount = mp; |
ce8e922c NS |
1809 | btp->bt_dev = bdev->bd_dev; |
1810 | btp->bt_bdev = bdev; | |
0e6e847f | 1811 | |
1da177e4 LT |
1812 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1813 | goto error; | |
5ca302c8 GC |
1814 | |
1815 | if (list_lru_init(&btp->bt_lru)) | |
1816 | goto error; | |
1817 | ||
9c7504aa BF |
1818 | if (percpu_counter_init(&btp->bt_io_count, 0, GFP_KERNEL)) |
1819 | goto error; | |
1820 | ||
e80dfa19 DC |
1821 | btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count; |
1822 | btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan; | |
ff57ab21 | 1823 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; |
e80dfa19 | 1824 | btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE; |
ff57ab21 | 1825 | register_shrinker(&btp->bt_shrinker); |
1da177e4 LT |
1826 | return btp; |
1827 | ||
1828 | error: | |
f0e2d93c | 1829 | kmem_free(btp); |
1da177e4 LT |
1830 | return NULL; |
1831 | } | |
1832 | ||
20e8a063 BF |
1833 | /* |
1834 | * Cancel a delayed write list. | |
1835 | * | |
1836 | * Remove each buffer from the list, clear the delwri queue flag and drop the | |
1837 | * associated buffer reference. | |
1838 | */ | |
1839 | void | |
1840 | xfs_buf_delwri_cancel( | |
1841 | struct list_head *list) | |
1842 | { | |
1843 | struct xfs_buf *bp; | |
1844 | ||
1845 | while (!list_empty(list)) { | |
1846 | bp = list_first_entry(list, struct xfs_buf, b_list); | |
1847 | ||
1848 | xfs_buf_lock(bp); | |
1849 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
1850 | list_del_init(&bp->b_list); | |
1851 | xfs_buf_relse(bp); | |
1852 | } | |
1853 | } | |
1854 | ||
1da177e4 | 1855 | /* |
43ff2122 CH |
1856 | * Add a buffer to the delayed write list. |
1857 | * | |
1858 | * This queues a buffer for writeout if it hasn't already been. Note that | |
1859 | * neither this routine nor the buffer list submission functions perform | |
1860 | * any internal synchronization. It is expected that the lists are thread-local | |
1861 | * to the callers. | |
1862 | * | |
1863 | * Returns true if we queued up the buffer, or false if it already had | |
1864 | * been on the buffer list. | |
1da177e4 | 1865 | */ |
43ff2122 | 1866 | bool |
ce8e922c | 1867 | xfs_buf_delwri_queue( |
43ff2122 CH |
1868 | struct xfs_buf *bp, |
1869 | struct list_head *list) | |
1da177e4 | 1870 | { |
43ff2122 | 1871 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 1872 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1873 | |
43ff2122 CH |
1874 | /* |
1875 | * If the buffer is already marked delwri it already is queued up | |
1876 | * by someone else for imediate writeout. Just ignore it in that | |
1877 | * case. | |
1878 | */ | |
1879 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
1880 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
1881 | return false; | |
1da177e4 | 1882 | } |
1da177e4 | 1883 | |
43ff2122 | 1884 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
1885 | |
1886 | /* | |
43ff2122 CH |
1887 | * If a buffer gets written out synchronously or marked stale while it |
1888 | * is on a delwri list we lazily remove it. To do this, the other party | |
1889 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
1890 | * It remains referenced and on the list. In a rare corner case it | |
1891 | * might get readded to a delwri list after the synchronous writeout, in | |
1892 | * which case we need just need to re-add the flag here. | |
d808f617 | 1893 | */ |
43ff2122 CH |
1894 | bp->b_flags |= _XBF_DELWRI_Q; |
1895 | if (list_empty(&bp->b_list)) { | |
1896 | atomic_inc(&bp->b_hold); | |
1897 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 1898 | } |
585e6d88 | 1899 | |
43ff2122 | 1900 | return true; |
585e6d88 DC |
1901 | } |
1902 | ||
089716aa DC |
1903 | /* |
1904 | * Compare function is more complex than it needs to be because | |
1905 | * the return value is only 32 bits and we are doing comparisons | |
1906 | * on 64 bit values | |
1907 | */ | |
1908 | static int | |
1909 | xfs_buf_cmp( | |
1910 | void *priv, | |
1911 | struct list_head *a, | |
1912 | struct list_head *b) | |
1913 | { | |
1914 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1915 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1916 | xfs_daddr_t diff; | |
1917 | ||
f4b42421 | 1918 | diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn; |
089716aa DC |
1919 | if (diff < 0) |
1920 | return -1; | |
1921 | if (diff > 0) | |
1922 | return 1; | |
1923 | return 0; | |
1924 | } | |
1925 | ||
26f1fe85 DC |
1926 | /* |
1927 | * submit buffers for write. | |
1928 | * | |
1929 | * When we have a large buffer list, we do not want to hold all the buffers | |
1930 | * locked while we block on the request queue waiting for IO dispatch. To avoid | |
1931 | * this problem, we lock and submit buffers in groups of 50, thereby minimising | |
1932 | * the lock hold times for lists which may contain thousands of objects. | |
1933 | * | |
1934 | * To do this, we sort the buffer list before we walk the list to lock and | |
1935 | * submit buffers, and we plug and unplug around each group of buffers we | |
1936 | * submit. | |
1937 | */ | |
43ff2122 | 1938 | static int |
26f1fe85 | 1939 | xfs_buf_delwri_submit_buffers( |
43ff2122 | 1940 | struct list_head *buffer_list, |
26f1fe85 | 1941 | struct list_head *wait_list) |
1da177e4 | 1942 | { |
43ff2122 | 1943 | struct xfs_buf *bp, *n; |
26f1fe85 | 1944 | LIST_HEAD (submit_list); |
43ff2122 | 1945 | int pinned = 0; |
26f1fe85 | 1946 | struct blk_plug plug; |
43ff2122 | 1947 | |
26f1fe85 | 1948 | list_sort(NULL, buffer_list, xfs_buf_cmp); |
43ff2122 | 1949 | |
26f1fe85 | 1950 | blk_start_plug(&plug); |
43ff2122 | 1951 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { |
26f1fe85 | 1952 | if (!wait_list) { |
43ff2122 CH |
1953 | if (xfs_buf_ispinned(bp)) { |
1954 | pinned++; | |
1955 | continue; | |
1956 | } | |
1957 | if (!xfs_buf_trylock(bp)) | |
1958 | continue; | |
1959 | } else { | |
1960 | xfs_buf_lock(bp); | |
1961 | } | |
978c7b2f | 1962 | |
43ff2122 CH |
1963 | /* |
1964 | * Someone else might have written the buffer synchronously or | |
1965 | * marked it stale in the meantime. In that case only the | |
1966 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
1967 | * reference and remove it from the list here. | |
1968 | */ | |
1969 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
1970 | list_del_init(&bp->b_list); | |
1971 | xfs_buf_relse(bp); | |
1972 | continue; | |
1973 | } | |
c9c12971 | 1974 | |
43ff2122 | 1975 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
a1b7ea5d | 1976 | |
cf53e99d | 1977 | /* |
26f1fe85 DC |
1978 | * We do all IO submission async. This means if we need |
1979 | * to wait for IO completion we need to take an extra | |
1980 | * reference so the buffer is still valid on the other | |
1981 | * side. We need to move the buffer onto the io_list | |
1982 | * at this point so the caller can still access it. | |
cf53e99d | 1983 | */ |
bbfeb614 | 1984 | bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_WRITE_FAIL); |
26f1fe85 DC |
1985 | bp->b_flags |= XBF_WRITE | XBF_ASYNC; |
1986 | if (wait_list) { | |
cf53e99d | 1987 | xfs_buf_hold(bp); |
26f1fe85 DC |
1988 | list_move_tail(&bp->b_list, wait_list); |
1989 | } else | |
ce8e922c | 1990 | list_del_init(&bp->b_list); |
8dac3921 | 1991 | |
595bff75 | 1992 | xfs_buf_submit(bp); |
43ff2122 CH |
1993 | } |
1994 | blk_finish_plug(&plug); | |
1da177e4 | 1995 | |
43ff2122 | 1996 | return pinned; |
1da177e4 LT |
1997 | } |
1998 | ||
1999 | /* | |
43ff2122 CH |
2000 | * Write out a buffer list asynchronously. |
2001 | * | |
2002 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
2003 | * out and not wait for I/O completion on any of the buffers. This interface | |
2004 | * is only safely useable for callers that can track I/O completion by higher | |
2005 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
2006 | * function. | |
1da177e4 LT |
2007 | */ |
2008 | int | |
43ff2122 CH |
2009 | xfs_buf_delwri_submit_nowait( |
2010 | struct list_head *buffer_list) | |
1da177e4 | 2011 | { |
26f1fe85 | 2012 | return xfs_buf_delwri_submit_buffers(buffer_list, NULL); |
43ff2122 | 2013 | } |
1da177e4 | 2014 | |
43ff2122 CH |
2015 | /* |
2016 | * Write out a buffer list synchronously. | |
2017 | * | |
2018 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
2019 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
2020 | * so callers must have some other way of tracking buffers if they require such | |
2021 | * functionality. | |
2022 | */ | |
2023 | int | |
2024 | xfs_buf_delwri_submit( | |
2025 | struct list_head *buffer_list) | |
2026 | { | |
26f1fe85 | 2027 | LIST_HEAD (wait_list); |
43ff2122 CH |
2028 | int error = 0, error2; |
2029 | struct xfs_buf *bp; | |
1da177e4 | 2030 | |
26f1fe85 | 2031 | xfs_buf_delwri_submit_buffers(buffer_list, &wait_list); |
1da177e4 | 2032 | |
43ff2122 | 2033 | /* Wait for IO to complete. */ |
26f1fe85 DC |
2034 | while (!list_empty(&wait_list)) { |
2035 | bp = list_first_entry(&wait_list, struct xfs_buf, b_list); | |
a1b7ea5d | 2036 | |
089716aa | 2037 | list_del_init(&bp->b_list); |
cf53e99d DC |
2038 | |
2039 | /* locking the buffer will wait for async IO completion. */ | |
2040 | xfs_buf_lock(bp); | |
2041 | error2 = bp->b_error; | |
43ff2122 CH |
2042 | xfs_buf_relse(bp); |
2043 | if (!error) | |
2044 | error = error2; | |
1da177e4 LT |
2045 | } |
2046 | ||
43ff2122 | 2047 | return error; |
1da177e4 LT |
2048 | } |
2049 | ||
04d8b284 | 2050 | int __init |
ce8e922c | 2051 | xfs_buf_init(void) |
1da177e4 | 2052 | { |
8758280f NS |
2053 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
2054 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 2055 | if (!xfs_buf_zone) |
0b1b213f | 2056 | goto out; |
04d8b284 | 2057 | |
23ea4032 | 2058 | return 0; |
1da177e4 | 2059 | |
0b1b213f | 2060 | out: |
8758280f | 2061 | return -ENOMEM; |
1da177e4 LT |
2062 | } |
2063 | ||
1da177e4 | 2064 | void |
ce8e922c | 2065 | xfs_buf_terminate(void) |
1da177e4 | 2066 | { |
ce8e922c | 2067 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 2068 | } |