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 | |
b7963133 | 37 | #include "xfs_sb.h" |
ed3b4d6c | 38 | #include "xfs_log.h" |
b7963133 | 39 | #include "xfs_ag.h" |
b7963133 | 40 | #include "xfs_mount.h" |
0b1b213f | 41 | #include "xfs_trace.h" |
b7963133 | 42 | |
7989cb8e | 43 | static kmem_zone_t *xfs_buf_zone; |
23ea4032 | 44 | |
7989cb8e | 45 | static struct workqueue_struct *xfslogd_workqueue; |
1da177e4 | 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 | ||
1da177e4 | 82 | /* |
430cbeb8 DC |
83 | * xfs_buf_lru_add - add a buffer to the LRU. |
84 | * | |
85 | * The LRU takes a new reference to the buffer so that it will only be freed | |
86 | * once the shrinker takes the buffer off the LRU. | |
87 | */ | |
88 | STATIC void | |
89 | xfs_buf_lru_add( | |
90 | struct xfs_buf *bp) | |
91 | { | |
92 | struct xfs_buftarg *btp = bp->b_target; | |
93 | ||
94 | spin_lock(&btp->bt_lru_lock); | |
95 | if (list_empty(&bp->b_lru)) { | |
96 | atomic_inc(&bp->b_hold); | |
97 | list_add_tail(&bp->b_lru, &btp->bt_lru); | |
98 | btp->bt_lru_nr++; | |
6fb8a90a | 99 | bp->b_lru_flags &= ~_XBF_LRU_DISPOSE; |
430cbeb8 DC |
100 | } |
101 | spin_unlock(&btp->bt_lru_lock); | |
102 | } | |
103 | ||
104 | /* | |
105 | * xfs_buf_lru_del - remove a buffer from the LRU | |
106 | * | |
107 | * The unlocked check is safe here because it only occurs when there are not | |
108 | * b_lru_ref counts left on the inode under the pag->pag_buf_lock. it is there | |
109 | * to optimise the shrinker removing the buffer from the LRU and calling | |
25985edc | 110 | * xfs_buf_free(). i.e. it removes an unnecessary round trip on the |
430cbeb8 | 111 | * bt_lru_lock. |
1da177e4 | 112 | */ |
430cbeb8 DC |
113 | STATIC void |
114 | xfs_buf_lru_del( | |
115 | struct xfs_buf *bp) | |
116 | { | |
117 | struct xfs_buftarg *btp = bp->b_target; | |
118 | ||
119 | if (list_empty(&bp->b_lru)) | |
120 | return; | |
121 | ||
122 | spin_lock(&btp->bt_lru_lock); | |
123 | if (!list_empty(&bp->b_lru)) { | |
124 | list_del_init(&bp->b_lru); | |
125 | btp->bt_lru_nr--; | |
126 | } | |
127 | spin_unlock(&btp->bt_lru_lock); | |
128 | } | |
129 | ||
130 | /* | |
131 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
132 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
133 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
134 | * to remove the reference that LRU holds on the buffer. | |
135 | * | |
136 | * This prevents build-up of stale buffers on the LRU. | |
137 | */ | |
138 | void | |
139 | xfs_buf_stale( | |
140 | struct xfs_buf *bp) | |
141 | { | |
43ff2122 CH |
142 | ASSERT(xfs_buf_islocked(bp)); |
143 | ||
430cbeb8 | 144 | bp->b_flags |= XBF_STALE; |
43ff2122 CH |
145 | |
146 | /* | |
147 | * Clear the delwri status so that a delwri queue walker will not | |
148 | * flush this buffer to disk now that it is stale. The delwri queue has | |
149 | * a reference to the buffer, so this is safe to do. | |
150 | */ | |
151 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
152 | ||
430cbeb8 DC |
153 | atomic_set(&(bp)->b_lru_ref, 0); |
154 | if (!list_empty(&bp->b_lru)) { | |
155 | struct xfs_buftarg *btp = bp->b_target; | |
156 | ||
157 | spin_lock(&btp->bt_lru_lock); | |
6fb8a90a CM |
158 | if (!list_empty(&bp->b_lru) && |
159 | !(bp->b_lru_flags & _XBF_LRU_DISPOSE)) { | |
430cbeb8 DC |
160 | list_del_init(&bp->b_lru); |
161 | btp->bt_lru_nr--; | |
162 | atomic_dec(&bp->b_hold); | |
163 | } | |
164 | spin_unlock(&btp->bt_lru_lock); | |
165 | } | |
166 | ASSERT(atomic_read(&bp->b_hold) >= 1); | |
167 | } | |
1da177e4 | 168 | |
3e85c868 DC |
169 | static int |
170 | xfs_buf_get_maps( | |
171 | struct xfs_buf *bp, | |
172 | int map_count) | |
173 | { | |
174 | ASSERT(bp->b_maps == NULL); | |
175 | bp->b_map_count = map_count; | |
176 | ||
177 | if (map_count == 1) { | |
178 | bp->b_maps = &bp->b_map; | |
179 | return 0; | |
180 | } | |
181 | ||
182 | bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map), | |
183 | KM_NOFS); | |
184 | if (!bp->b_maps) | |
185 | return ENOMEM; | |
186 | return 0; | |
187 | } | |
188 | ||
189 | /* | |
190 | * Frees b_pages if it was allocated. | |
191 | */ | |
192 | static void | |
193 | xfs_buf_free_maps( | |
194 | struct xfs_buf *bp) | |
195 | { | |
196 | if (bp->b_maps != &bp->b_map) { | |
197 | kmem_free(bp->b_maps); | |
198 | bp->b_maps = NULL; | |
199 | } | |
200 | } | |
201 | ||
4347b9d7 | 202 | struct xfs_buf * |
3e85c868 | 203 | _xfs_buf_alloc( |
4347b9d7 | 204 | struct xfs_buftarg *target, |
3e85c868 DC |
205 | struct xfs_buf_map *map, |
206 | int nmaps, | |
ce8e922c | 207 | xfs_buf_flags_t flags) |
1da177e4 | 208 | { |
4347b9d7 | 209 | struct xfs_buf *bp; |
3e85c868 DC |
210 | int error; |
211 | int i; | |
4347b9d7 | 212 | |
aa5c158e | 213 | bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS); |
4347b9d7 CH |
214 | if (unlikely(!bp)) |
215 | return NULL; | |
216 | ||
1da177e4 | 217 | /* |
12bcb3f7 DC |
218 | * We don't want certain flags to appear in b_flags unless they are |
219 | * specifically set by later operations on the buffer. | |
1da177e4 | 220 | */ |
611c9946 | 221 | flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD); |
ce8e922c | 222 | |
ce8e922c | 223 | atomic_set(&bp->b_hold, 1); |
430cbeb8 | 224 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 225 | init_completion(&bp->b_iowait); |
430cbeb8 | 226 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 227 | INIT_LIST_HEAD(&bp->b_list); |
74f75a0c | 228 | RB_CLEAR_NODE(&bp->b_rbnode); |
a731cd11 | 229 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
ce8e922c NS |
230 | XB_SET_OWNER(bp); |
231 | bp->b_target = target; | |
3e85c868 | 232 | bp->b_flags = flags; |
de1cbee4 | 233 | |
1da177e4 | 234 | /* |
aa0e8833 DC |
235 | * Set length and io_length to the same value initially. |
236 | * I/O routines should use io_length, which will be the same in | |
1da177e4 LT |
237 | * most cases but may be reset (e.g. XFS recovery). |
238 | */ | |
3e85c868 DC |
239 | error = xfs_buf_get_maps(bp, nmaps); |
240 | if (error) { | |
241 | kmem_zone_free(xfs_buf_zone, bp); | |
242 | return NULL; | |
243 | } | |
244 | ||
245 | bp->b_bn = map[0].bm_bn; | |
246 | bp->b_length = 0; | |
247 | for (i = 0; i < nmaps; i++) { | |
248 | bp->b_maps[i].bm_bn = map[i].bm_bn; | |
249 | bp->b_maps[i].bm_len = map[i].bm_len; | |
250 | bp->b_length += map[i].bm_len; | |
251 | } | |
252 | bp->b_io_length = bp->b_length; | |
253 | ||
ce8e922c NS |
254 | atomic_set(&bp->b_pin_count, 0); |
255 | init_waitqueue_head(&bp->b_waiters); | |
256 | ||
257 | XFS_STATS_INC(xb_create); | |
0b1b213f | 258 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 CH |
259 | |
260 | return bp; | |
1da177e4 LT |
261 | } |
262 | ||
263 | /* | |
ce8e922c NS |
264 | * Allocate a page array capable of holding a specified number |
265 | * of pages, and point the page buf at it. | |
1da177e4 LT |
266 | */ |
267 | STATIC int | |
ce8e922c NS |
268 | _xfs_buf_get_pages( |
269 | xfs_buf_t *bp, | |
1da177e4 | 270 | int page_count, |
ce8e922c | 271 | xfs_buf_flags_t flags) |
1da177e4 LT |
272 | { |
273 | /* Make sure that we have a page list */ | |
ce8e922c | 274 | if (bp->b_pages == NULL) { |
ce8e922c NS |
275 | bp->b_page_count = page_count; |
276 | if (page_count <= XB_PAGES) { | |
277 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 278 | } else { |
ce8e922c | 279 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
aa5c158e | 280 | page_count, KM_NOFS); |
ce8e922c | 281 | if (bp->b_pages == NULL) |
1da177e4 LT |
282 | return -ENOMEM; |
283 | } | |
ce8e922c | 284 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
285 | } |
286 | return 0; | |
287 | } | |
288 | ||
289 | /* | |
ce8e922c | 290 | * Frees b_pages if it was allocated. |
1da177e4 LT |
291 | */ |
292 | STATIC void | |
ce8e922c | 293 | _xfs_buf_free_pages( |
1da177e4 LT |
294 | xfs_buf_t *bp) |
295 | { | |
ce8e922c | 296 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 297 | kmem_free(bp->b_pages); |
3fc98b1a | 298 | bp->b_pages = NULL; |
1da177e4 LT |
299 | } |
300 | } | |
301 | ||
302 | /* | |
303 | * Releases the specified buffer. | |
304 | * | |
305 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 306 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
307 | * hashed and refcounted buffers |
308 | */ | |
309 | void | |
ce8e922c | 310 | xfs_buf_free( |
1da177e4 LT |
311 | xfs_buf_t *bp) |
312 | { | |
0b1b213f | 313 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 314 | |
430cbeb8 DC |
315 | ASSERT(list_empty(&bp->b_lru)); |
316 | ||
0e6e847f | 317 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
318 | uint i; |
319 | ||
73c77e2c | 320 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
321 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
322 | bp->b_page_count); | |
1da177e4 | 323 | |
948ecdb4 NS |
324 | for (i = 0; i < bp->b_page_count; i++) { |
325 | struct page *page = bp->b_pages[i]; | |
326 | ||
0e6e847f | 327 | __free_page(page); |
948ecdb4 | 328 | } |
0e6e847f DC |
329 | } else if (bp->b_flags & _XBF_KMEM) |
330 | kmem_free(bp->b_addr); | |
3fc98b1a | 331 | _xfs_buf_free_pages(bp); |
3e85c868 | 332 | xfs_buf_free_maps(bp); |
4347b9d7 | 333 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
334 | } |
335 | ||
336 | /* | |
0e6e847f | 337 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
338 | */ |
339 | STATIC int | |
0e6e847f | 340 | xfs_buf_allocate_memory( |
1da177e4 LT |
341 | xfs_buf_t *bp, |
342 | uint flags) | |
343 | { | |
aa0e8833 | 344 | size_t size; |
1da177e4 | 345 | size_t nbytes, offset; |
ce8e922c | 346 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 347 | unsigned short page_count, i; |
795cac72 | 348 | xfs_off_t start, end; |
1da177e4 LT |
349 | int error; |
350 | ||
0e6e847f DC |
351 | /* |
352 | * for buffers that are contained within a single page, just allocate | |
353 | * the memory from the heap - there's no need for the complexity of | |
354 | * page arrays to keep allocation down to order 0. | |
355 | */ | |
795cac72 DC |
356 | size = BBTOB(bp->b_length); |
357 | if (size < PAGE_SIZE) { | |
aa5c158e | 358 | bp->b_addr = kmem_alloc(size, KM_NOFS); |
0e6e847f DC |
359 | if (!bp->b_addr) { |
360 | /* low memory - use alloc_page loop instead */ | |
361 | goto use_alloc_page; | |
362 | } | |
363 | ||
795cac72 | 364 | if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) != |
0e6e847f DC |
365 | ((unsigned long)bp->b_addr & PAGE_MASK)) { |
366 | /* b_addr spans two pages - use alloc_page instead */ | |
367 | kmem_free(bp->b_addr); | |
368 | bp->b_addr = NULL; | |
369 | goto use_alloc_page; | |
370 | } | |
371 | bp->b_offset = offset_in_page(bp->b_addr); | |
372 | bp->b_pages = bp->b_page_array; | |
373 | bp->b_pages[0] = virt_to_page(bp->b_addr); | |
374 | bp->b_page_count = 1; | |
611c9946 | 375 | bp->b_flags |= _XBF_KMEM; |
0e6e847f DC |
376 | return 0; |
377 | } | |
378 | ||
379 | use_alloc_page: | |
cbb7baab DC |
380 | start = BBTOB(bp->b_map.bm_bn) >> PAGE_SHIFT; |
381 | end = (BBTOB(bp->b_map.bm_bn + bp->b_length) + PAGE_SIZE - 1) | |
382 | >> PAGE_SHIFT; | |
795cac72 | 383 | page_count = end - start; |
ce8e922c | 384 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
385 | if (unlikely(error)) |
386 | return error; | |
1da177e4 | 387 | |
ce8e922c | 388 | offset = bp->b_offset; |
0e6e847f | 389 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 390 | |
ce8e922c | 391 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
392 | struct page *page; |
393 | uint retries = 0; | |
0e6e847f DC |
394 | retry: |
395 | page = alloc_page(gfp_mask); | |
1da177e4 | 396 | if (unlikely(page == NULL)) { |
ce8e922c NS |
397 | if (flags & XBF_READ_AHEAD) { |
398 | bp->b_page_count = i; | |
0e6e847f DC |
399 | error = ENOMEM; |
400 | goto out_free_pages; | |
1da177e4 LT |
401 | } |
402 | ||
403 | /* | |
404 | * This could deadlock. | |
405 | * | |
406 | * But until all the XFS lowlevel code is revamped to | |
407 | * handle buffer allocation failures we can't do much. | |
408 | */ | |
409 | if (!(++retries % 100)) | |
4f10700a DC |
410 | xfs_err(NULL, |
411 | "possible memory allocation deadlock in %s (mode:0x%x)", | |
34a622b2 | 412 | __func__, gfp_mask); |
1da177e4 | 413 | |
ce8e922c | 414 | XFS_STATS_INC(xb_page_retries); |
8aa7e847 | 415 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
416 | goto retry; |
417 | } | |
418 | ||
ce8e922c | 419 | XFS_STATS_INC(xb_page_found); |
1da177e4 | 420 | |
0e6e847f | 421 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 422 | size -= nbytes; |
ce8e922c | 423 | bp->b_pages[i] = page; |
1da177e4 LT |
424 | offset = 0; |
425 | } | |
0e6e847f | 426 | return 0; |
1da177e4 | 427 | |
0e6e847f DC |
428 | out_free_pages: |
429 | for (i = 0; i < bp->b_page_count; i++) | |
430 | __free_page(bp->b_pages[i]); | |
1da177e4 LT |
431 | return error; |
432 | } | |
433 | ||
434 | /* | |
25985edc | 435 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
436 | */ |
437 | STATIC int | |
ce8e922c | 438 | _xfs_buf_map_pages( |
1da177e4 LT |
439 | xfs_buf_t *bp, |
440 | uint flags) | |
441 | { | |
0e6e847f | 442 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 443 | if (bp->b_page_count == 1) { |
0e6e847f | 444 | /* A single page buffer is always mappable */ |
ce8e922c | 445 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
611c9946 DC |
446 | } else if (flags & XBF_UNMAPPED) { |
447 | bp->b_addr = NULL; | |
448 | } else { | |
a19fb380 DC |
449 | int retried = 0; |
450 | ||
451 | do { | |
452 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
453 | -1, PAGE_KERNEL); | |
454 | if (bp->b_addr) | |
455 | break; | |
456 | vm_unmap_aliases(); | |
457 | } while (retried++ <= 1); | |
458 | ||
459 | if (!bp->b_addr) | |
1da177e4 | 460 | return -ENOMEM; |
ce8e922c | 461 | bp->b_addr += bp->b_offset; |
1da177e4 LT |
462 | } |
463 | ||
464 | return 0; | |
465 | } | |
466 | ||
467 | /* | |
468 | * Finding and Reading Buffers | |
469 | */ | |
470 | ||
471 | /* | |
ce8e922c | 472 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 | 473 | * a given range of an inode. The buffer is returned |
eabbaf11 | 474 | * locked. No I/O is implied by this call. |
1da177e4 LT |
475 | */ |
476 | xfs_buf_t * | |
ce8e922c | 477 | _xfs_buf_find( |
e70b73f8 | 478 | struct xfs_buftarg *btp, |
3e85c868 DC |
479 | struct xfs_buf_map *map, |
480 | int nmaps, | |
ce8e922c NS |
481 | xfs_buf_flags_t flags, |
482 | xfs_buf_t *new_bp) | |
1da177e4 | 483 | { |
e70b73f8 | 484 | size_t numbytes; |
74f75a0c DC |
485 | struct xfs_perag *pag; |
486 | struct rb_node **rbp; | |
487 | struct rb_node *parent; | |
488 | xfs_buf_t *bp; | |
3e85c868 DC |
489 | xfs_daddr_t blkno = map[0].bm_bn; |
490 | int numblks = 0; | |
491 | int i; | |
1da177e4 | 492 | |
3e85c868 DC |
493 | for (i = 0; i < nmaps; i++) |
494 | numblks += map[i].bm_len; | |
e70b73f8 | 495 | numbytes = BBTOB(numblks); |
1da177e4 LT |
496 | |
497 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
e70b73f8 | 498 | ASSERT(!(numbytes < (1 << btp->bt_sshift))); |
de1cbee4 | 499 | ASSERT(!(BBTOB(blkno) & (xfs_off_t)btp->bt_smask)); |
1da177e4 | 500 | |
74f75a0c DC |
501 | /* get tree root */ |
502 | pag = xfs_perag_get(btp->bt_mount, | |
e70b73f8 | 503 | xfs_daddr_to_agno(btp->bt_mount, blkno)); |
74f75a0c DC |
504 | |
505 | /* walk tree */ | |
506 | spin_lock(&pag->pag_buf_lock); | |
507 | rbp = &pag->pag_buf_tree.rb_node; | |
508 | parent = NULL; | |
509 | bp = NULL; | |
510 | while (*rbp) { | |
511 | parent = *rbp; | |
512 | bp = rb_entry(parent, struct xfs_buf, b_rbnode); | |
513 | ||
de1cbee4 | 514 | if (blkno < bp->b_bn) |
74f75a0c | 515 | rbp = &(*rbp)->rb_left; |
de1cbee4 | 516 | else if (blkno > bp->b_bn) |
74f75a0c DC |
517 | rbp = &(*rbp)->rb_right; |
518 | else { | |
519 | /* | |
de1cbee4 | 520 | * found a block number match. If the range doesn't |
74f75a0c DC |
521 | * match, the only way this is allowed is if the buffer |
522 | * in the cache is stale and the transaction that made | |
523 | * it stale has not yet committed. i.e. we are | |
524 | * reallocating a busy extent. Skip this buffer and | |
525 | * continue searching to the right for an exact match. | |
526 | */ | |
4e94b71b | 527 | if (bp->b_length != numblks) { |
74f75a0c DC |
528 | ASSERT(bp->b_flags & XBF_STALE); |
529 | rbp = &(*rbp)->rb_right; | |
530 | continue; | |
531 | } | |
ce8e922c | 532 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
533 | goto found; |
534 | } | |
535 | } | |
536 | ||
537 | /* No match found */ | |
ce8e922c | 538 | if (new_bp) { |
74f75a0c DC |
539 | rb_link_node(&new_bp->b_rbnode, parent, rbp); |
540 | rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree); | |
541 | /* the buffer keeps the perag reference until it is freed */ | |
542 | new_bp->b_pag = pag; | |
543 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 544 | } else { |
ce8e922c | 545 | XFS_STATS_INC(xb_miss_locked); |
74f75a0c DC |
546 | spin_unlock(&pag->pag_buf_lock); |
547 | xfs_perag_put(pag); | |
1da177e4 | 548 | } |
ce8e922c | 549 | return new_bp; |
1da177e4 LT |
550 | |
551 | found: | |
74f75a0c DC |
552 | spin_unlock(&pag->pag_buf_lock); |
553 | xfs_perag_put(pag); | |
1da177e4 | 554 | |
0c842ad4 CH |
555 | if (!xfs_buf_trylock(bp)) { |
556 | if (flags & XBF_TRYLOCK) { | |
ce8e922c NS |
557 | xfs_buf_rele(bp); |
558 | XFS_STATS_INC(xb_busy_locked); | |
559 | return NULL; | |
1da177e4 | 560 | } |
0c842ad4 CH |
561 | xfs_buf_lock(bp); |
562 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
563 | } |
564 | ||
0e6e847f DC |
565 | /* |
566 | * if the buffer is stale, clear all the external state associated with | |
567 | * it. We need to keep flags such as how we allocated the buffer memory | |
568 | * intact here. | |
569 | */ | |
ce8e922c NS |
570 | if (bp->b_flags & XBF_STALE) { |
571 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
611c9946 | 572 | bp->b_flags &= _XBF_KMEM | _XBF_PAGES; |
2f926587 | 573 | } |
0b1b213f CH |
574 | |
575 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ce8e922c NS |
576 | XFS_STATS_INC(xb_get_locked); |
577 | return bp; | |
1da177e4 LT |
578 | } |
579 | ||
580 | /* | |
3815832a DC |
581 | * Assembles a buffer covering the specified range. The code is optimised for |
582 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
583 | * more hits than misses. | |
1da177e4 | 584 | */ |
3815832a | 585 | struct xfs_buf * |
6dde2707 DC |
586 | xfs_buf_get_map( |
587 | struct xfs_buftarg *target, | |
588 | struct xfs_buf_map *map, | |
589 | int nmaps, | |
ce8e922c | 590 | xfs_buf_flags_t flags) |
1da177e4 | 591 | { |
3815832a DC |
592 | struct xfs_buf *bp; |
593 | struct xfs_buf *new_bp; | |
0e6e847f | 594 | int error = 0; |
1da177e4 | 595 | |
6dde2707 | 596 | bp = _xfs_buf_find(target, map, nmaps, flags, NULL); |
3815832a DC |
597 | if (likely(bp)) |
598 | goto found; | |
599 | ||
6dde2707 | 600 | new_bp = _xfs_buf_alloc(target, map, nmaps, flags); |
ce8e922c | 601 | if (unlikely(!new_bp)) |
1da177e4 LT |
602 | return NULL; |
603 | ||
fe2429b0 DC |
604 | error = xfs_buf_allocate_memory(new_bp, flags); |
605 | if (error) { | |
3e85c868 | 606 | xfs_buf_free(new_bp); |
fe2429b0 DC |
607 | return NULL; |
608 | } | |
609 | ||
6dde2707 | 610 | bp = _xfs_buf_find(target, map, nmaps, flags, new_bp); |
3815832a | 611 | if (!bp) { |
fe2429b0 | 612 | xfs_buf_free(new_bp); |
3815832a DC |
613 | return NULL; |
614 | } | |
615 | ||
fe2429b0 DC |
616 | if (bp != new_bp) |
617 | xfs_buf_free(new_bp); | |
1da177e4 | 618 | |
3815832a | 619 | found: |
611c9946 | 620 | if (!bp->b_addr) { |
ce8e922c | 621 | error = _xfs_buf_map_pages(bp, flags); |
1da177e4 | 622 | if (unlikely(error)) { |
4f10700a DC |
623 | xfs_warn(target->bt_mount, |
624 | "%s: failed to map pages\n", __func__); | |
a8acad70 DC |
625 | xfs_buf_relse(bp); |
626 | return NULL; | |
1da177e4 LT |
627 | } |
628 | } | |
629 | ||
ce8e922c | 630 | XFS_STATS_INC(xb_get); |
0b1b213f | 631 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 632 | return bp; |
1da177e4 LT |
633 | } |
634 | ||
5d765b97 CH |
635 | STATIC int |
636 | _xfs_buf_read( | |
637 | xfs_buf_t *bp, | |
638 | xfs_buf_flags_t flags) | |
639 | { | |
43ff2122 | 640 | ASSERT(!(flags & XBF_WRITE)); |
cbb7baab | 641 | ASSERT(bp->b_map.bm_bn != XFS_BUF_DADDR_NULL); |
5d765b97 | 642 | |
43ff2122 | 643 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD); |
1d5ae5df | 644 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 645 | |
0e95f19a DC |
646 | xfs_buf_iorequest(bp); |
647 | if (flags & XBF_ASYNC) | |
648 | return 0; | |
ec53d1db | 649 | return xfs_buf_iowait(bp); |
5d765b97 CH |
650 | } |
651 | ||
1da177e4 | 652 | xfs_buf_t * |
6dde2707 DC |
653 | xfs_buf_read_map( |
654 | struct xfs_buftarg *target, | |
655 | struct xfs_buf_map *map, | |
656 | int nmaps, | |
ce8e922c | 657 | xfs_buf_flags_t flags) |
1da177e4 | 658 | { |
6dde2707 | 659 | struct xfs_buf *bp; |
ce8e922c NS |
660 | |
661 | flags |= XBF_READ; | |
662 | ||
6dde2707 | 663 | bp = xfs_buf_get_map(target, map, nmaps, flags); |
ce8e922c | 664 | if (bp) { |
0b1b213f CH |
665 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
666 | ||
ce8e922c | 667 | if (!XFS_BUF_ISDONE(bp)) { |
ce8e922c | 668 | XFS_STATS_INC(xb_get_read); |
5d765b97 | 669 | _xfs_buf_read(bp, flags); |
ce8e922c | 670 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
671 | /* |
672 | * Read ahead call which is already satisfied, | |
673 | * drop the buffer | |
674 | */ | |
a8acad70 DC |
675 | xfs_buf_relse(bp); |
676 | return NULL; | |
1da177e4 | 677 | } else { |
1da177e4 | 678 | /* We do not want read in the flags */ |
ce8e922c | 679 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
680 | } |
681 | } | |
682 | ||
ce8e922c | 683 | return bp; |
1da177e4 LT |
684 | } |
685 | ||
1da177e4 | 686 | /* |
ce8e922c NS |
687 | * If we are not low on memory then do the readahead in a deadlock |
688 | * safe manner. | |
1da177e4 LT |
689 | */ |
690 | void | |
6dde2707 DC |
691 | xfs_buf_readahead_map( |
692 | struct xfs_buftarg *target, | |
693 | struct xfs_buf_map *map, | |
694 | int nmaps) | |
1da177e4 | 695 | { |
0e6e847f | 696 | if (bdi_read_congested(target->bt_bdi)) |
1da177e4 LT |
697 | return; |
698 | ||
6dde2707 | 699 | xfs_buf_read_map(target, map, nmaps, |
aa5c158e | 700 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD); |
1da177e4 LT |
701 | } |
702 | ||
5adc94c2 DC |
703 | /* |
704 | * Read an uncached buffer from disk. Allocates and returns a locked | |
705 | * buffer containing the disk contents or nothing. | |
706 | */ | |
707 | struct xfs_buf * | |
708 | xfs_buf_read_uncached( | |
5adc94c2 DC |
709 | struct xfs_buftarg *target, |
710 | xfs_daddr_t daddr, | |
e70b73f8 | 711 | size_t numblks, |
5adc94c2 DC |
712 | int flags) |
713 | { | |
714 | xfs_buf_t *bp; | |
715 | int error; | |
716 | ||
e70b73f8 | 717 | bp = xfs_buf_get_uncached(target, numblks, flags); |
5adc94c2 DC |
718 | if (!bp) |
719 | return NULL; | |
720 | ||
721 | /* set up the buffer for a read IO */ | |
3e85c868 DC |
722 | ASSERT(bp->b_map_count == 1); |
723 | bp->b_bn = daddr; | |
724 | bp->b_maps[0].bm_bn = daddr; | |
cbb7baab | 725 | bp->b_flags |= XBF_READ; |
5adc94c2 | 726 | |
e70b73f8 | 727 | xfsbdstrat(target->bt_mount, bp); |
1a1a3e97 | 728 | error = xfs_buf_iowait(bp); |
0e95f19a | 729 | if (error) { |
5adc94c2 DC |
730 | xfs_buf_relse(bp); |
731 | return NULL; | |
732 | } | |
733 | return bp; | |
1da177e4 LT |
734 | } |
735 | ||
44396476 DC |
736 | /* |
737 | * Return a buffer allocated as an empty buffer and associated to external | |
738 | * memory via xfs_buf_associate_memory() back to it's empty state. | |
739 | */ | |
740 | void | |
741 | xfs_buf_set_empty( | |
742 | struct xfs_buf *bp, | |
e70b73f8 | 743 | size_t numblks) |
44396476 DC |
744 | { |
745 | if (bp->b_pages) | |
746 | _xfs_buf_free_pages(bp); | |
747 | ||
748 | bp->b_pages = NULL; | |
749 | bp->b_page_count = 0; | |
750 | bp->b_addr = NULL; | |
4e94b71b | 751 | bp->b_length = numblks; |
aa0e8833 | 752 | bp->b_io_length = numblks; |
3e85c868 DC |
753 | |
754 | ASSERT(bp->b_map_count == 1); | |
44396476 | 755 | bp->b_bn = XFS_BUF_DADDR_NULL; |
3e85c868 DC |
756 | bp->b_maps[0].bm_bn = XFS_BUF_DADDR_NULL; |
757 | bp->b_maps[0].bm_len = bp->b_length; | |
44396476 DC |
758 | } |
759 | ||
1da177e4 LT |
760 | static inline struct page * |
761 | mem_to_page( | |
762 | void *addr) | |
763 | { | |
9e2779fa | 764 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
765 | return virt_to_page(addr); |
766 | } else { | |
767 | return vmalloc_to_page(addr); | |
768 | } | |
769 | } | |
770 | ||
771 | int | |
ce8e922c NS |
772 | xfs_buf_associate_memory( |
773 | xfs_buf_t *bp, | |
1da177e4 LT |
774 | void *mem, |
775 | size_t len) | |
776 | { | |
777 | int rval; | |
778 | int i = 0; | |
d1afb678 LM |
779 | unsigned long pageaddr; |
780 | unsigned long offset; | |
781 | size_t buflen; | |
1da177e4 LT |
782 | int page_count; |
783 | ||
0e6e847f | 784 | pageaddr = (unsigned long)mem & PAGE_MASK; |
d1afb678 | 785 | offset = (unsigned long)mem - pageaddr; |
0e6e847f DC |
786 | buflen = PAGE_ALIGN(len + offset); |
787 | page_count = buflen >> PAGE_SHIFT; | |
1da177e4 LT |
788 | |
789 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
790 | if (bp->b_pages) |
791 | _xfs_buf_free_pages(bp); | |
1da177e4 | 792 | |
ce8e922c NS |
793 | bp->b_pages = NULL; |
794 | bp->b_addr = mem; | |
1da177e4 | 795 | |
aa5c158e | 796 | rval = _xfs_buf_get_pages(bp, page_count, 0); |
1da177e4 LT |
797 | if (rval) |
798 | return rval; | |
799 | ||
ce8e922c | 800 | bp->b_offset = offset; |
d1afb678 LM |
801 | |
802 | for (i = 0; i < bp->b_page_count; i++) { | |
803 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
0e6e847f | 804 | pageaddr += PAGE_SIZE; |
1da177e4 | 805 | } |
1da177e4 | 806 | |
aa0e8833 | 807 | bp->b_io_length = BTOBB(len); |
4e94b71b | 808 | bp->b_length = BTOBB(buflen); |
1da177e4 LT |
809 | |
810 | return 0; | |
811 | } | |
812 | ||
813 | xfs_buf_t * | |
686865f7 DC |
814 | xfs_buf_get_uncached( |
815 | struct xfs_buftarg *target, | |
e70b73f8 | 816 | size_t numblks, |
686865f7 | 817 | int flags) |
1da177e4 | 818 | { |
e70b73f8 | 819 | unsigned long page_count; |
1fa40b01 | 820 | int error, i; |
3e85c868 DC |
821 | struct xfs_buf *bp; |
822 | DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks); | |
1da177e4 | 823 | |
3e85c868 | 824 | bp = _xfs_buf_alloc(target, &map, 1, 0); |
1da177e4 LT |
825 | if (unlikely(bp == NULL)) |
826 | goto fail; | |
1da177e4 | 827 | |
e70b73f8 | 828 | page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT; |
1fa40b01 CH |
829 | error = _xfs_buf_get_pages(bp, page_count, 0); |
830 | if (error) | |
1da177e4 LT |
831 | goto fail_free_buf; |
832 | ||
1fa40b01 | 833 | for (i = 0; i < page_count; i++) { |
686865f7 | 834 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
835 | if (!bp->b_pages[i]) |
836 | goto fail_free_mem; | |
1da177e4 | 837 | } |
1fa40b01 | 838 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 839 | |
611c9946 | 840 | error = _xfs_buf_map_pages(bp, 0); |
1fa40b01 | 841 | if (unlikely(error)) { |
4f10700a DC |
842 | xfs_warn(target->bt_mount, |
843 | "%s: failed to map pages\n", __func__); | |
1da177e4 | 844 | goto fail_free_mem; |
1fa40b01 | 845 | } |
1da177e4 | 846 | |
686865f7 | 847 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 848 | return bp; |
1fa40b01 | 849 | |
1da177e4 | 850 | fail_free_mem: |
1fa40b01 CH |
851 | while (--i >= 0) |
852 | __free_page(bp->b_pages[i]); | |
ca165b88 | 853 | _xfs_buf_free_pages(bp); |
1da177e4 | 854 | fail_free_buf: |
3e85c868 | 855 | xfs_buf_free_maps(bp); |
4347b9d7 | 856 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
857 | fail: |
858 | return NULL; | |
859 | } | |
860 | ||
861 | /* | |
1da177e4 LT |
862 | * Increment reference count on buffer, to hold the buffer concurrently |
863 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
864 | * Must hold the buffer already to call this function. |
865 | */ | |
866 | void | |
ce8e922c NS |
867 | xfs_buf_hold( |
868 | xfs_buf_t *bp) | |
1da177e4 | 869 | { |
0b1b213f | 870 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 871 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
872 | } |
873 | ||
874 | /* | |
ce8e922c NS |
875 | * Releases a hold on the specified buffer. If the |
876 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
877 | */ |
878 | void | |
ce8e922c NS |
879 | xfs_buf_rele( |
880 | xfs_buf_t *bp) | |
1da177e4 | 881 | { |
74f75a0c | 882 | struct xfs_perag *pag = bp->b_pag; |
1da177e4 | 883 | |
0b1b213f | 884 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 885 | |
74f75a0c | 886 | if (!pag) { |
430cbeb8 | 887 | ASSERT(list_empty(&bp->b_lru)); |
74f75a0c | 888 | ASSERT(RB_EMPTY_NODE(&bp->b_rbnode)); |
fad3aa1e NS |
889 | if (atomic_dec_and_test(&bp->b_hold)) |
890 | xfs_buf_free(bp); | |
891 | return; | |
892 | } | |
893 | ||
74f75a0c | 894 | ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode)); |
430cbeb8 | 895 | |
3790689f | 896 | ASSERT(atomic_read(&bp->b_hold) > 0); |
74f75a0c | 897 | if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) { |
bfc60177 | 898 | if (!(bp->b_flags & XBF_STALE) && |
430cbeb8 DC |
899 | atomic_read(&bp->b_lru_ref)) { |
900 | xfs_buf_lru_add(bp); | |
901 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 902 | } else { |
430cbeb8 | 903 | xfs_buf_lru_del(bp); |
43ff2122 | 904 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
74f75a0c DC |
905 | rb_erase(&bp->b_rbnode, &pag->pag_buf_tree); |
906 | spin_unlock(&pag->pag_buf_lock); | |
907 | xfs_perag_put(pag); | |
ce8e922c | 908 | xfs_buf_free(bp); |
1da177e4 LT |
909 | } |
910 | } | |
911 | } | |
912 | ||
913 | ||
914 | /* | |
0e6e847f | 915 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
916 | * |
917 | * If we come across a stale, pinned, locked buffer, we know that we are | |
918 | * being asked to lock a buffer that has been reallocated. Because it is | |
919 | * pinned, we know that the log has not been pushed to disk and hence it | |
920 | * will still be locked. Rather than continuing to have trylock attempts | |
921 | * fail until someone else pushes the log, push it ourselves before | |
922 | * returning. This means that the xfsaild will not get stuck trying | |
923 | * to push on stale inode buffers. | |
1da177e4 LT |
924 | */ |
925 | int | |
0c842ad4 CH |
926 | xfs_buf_trylock( |
927 | struct xfs_buf *bp) | |
1da177e4 LT |
928 | { |
929 | int locked; | |
930 | ||
ce8e922c | 931 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 932 | if (locked) |
ce8e922c | 933 | XB_SET_OWNER(bp); |
90810b9e DC |
934 | else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
935 | xfs_log_force(bp->b_target->bt_mount, 0); | |
0b1b213f | 936 | |
0c842ad4 CH |
937 | trace_xfs_buf_trylock(bp, _RET_IP_); |
938 | return locked; | |
1da177e4 | 939 | } |
1da177e4 LT |
940 | |
941 | /* | |
0e6e847f | 942 | * Lock a buffer object. |
ed3b4d6c DC |
943 | * |
944 | * If we come across a stale, pinned, locked buffer, we know that we | |
945 | * are being asked to lock a buffer that has been reallocated. Because | |
946 | * it is pinned, we know that the log has not been pushed to disk and | |
947 | * hence it will still be locked. Rather than sleeping until someone | |
948 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 949 | */ |
ce8e922c NS |
950 | void |
951 | xfs_buf_lock( | |
0c842ad4 | 952 | struct xfs_buf *bp) |
1da177e4 | 953 | { |
0b1b213f CH |
954 | trace_xfs_buf_lock(bp, _RET_IP_); |
955 | ||
ed3b4d6c | 956 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 957 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
958 | down(&bp->b_sema); |
959 | XB_SET_OWNER(bp); | |
0b1b213f CH |
960 | |
961 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
962 | } |
963 | ||
1da177e4 | 964 | void |
ce8e922c | 965 | xfs_buf_unlock( |
0c842ad4 | 966 | struct xfs_buf *bp) |
1da177e4 | 967 | { |
ce8e922c NS |
968 | XB_CLEAR_OWNER(bp); |
969 | up(&bp->b_sema); | |
0b1b213f CH |
970 | |
971 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
972 | } |
973 | ||
ce8e922c NS |
974 | STATIC void |
975 | xfs_buf_wait_unpin( | |
976 | xfs_buf_t *bp) | |
1da177e4 LT |
977 | { |
978 | DECLARE_WAITQUEUE (wait, current); | |
979 | ||
ce8e922c | 980 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
981 | return; |
982 | ||
ce8e922c | 983 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
984 | for (;;) { |
985 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 986 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 987 | break; |
7eaceacc | 988 | io_schedule(); |
1da177e4 | 989 | } |
ce8e922c | 990 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
991 | set_current_state(TASK_RUNNING); |
992 | } | |
993 | ||
994 | /* | |
995 | * Buffer Utility Routines | |
996 | */ | |
997 | ||
1da177e4 | 998 | STATIC void |
ce8e922c | 999 | xfs_buf_iodone_work( |
c4028958 | 1000 | struct work_struct *work) |
1da177e4 | 1001 | { |
c4028958 DH |
1002 | xfs_buf_t *bp = |
1003 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 1004 | |
80f6c29d | 1005 | if (bp->b_iodone) |
ce8e922c NS |
1006 | (*(bp->b_iodone))(bp); |
1007 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
1008 | xfs_buf_relse(bp); |
1009 | } | |
1010 | ||
1011 | void | |
ce8e922c NS |
1012 | xfs_buf_ioend( |
1013 | xfs_buf_t *bp, | |
1da177e4 LT |
1014 | int schedule) |
1015 | { | |
0b1b213f CH |
1016 | trace_xfs_buf_iodone(bp, _RET_IP_); |
1017 | ||
77be55a5 | 1018 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
ce8e922c NS |
1019 | if (bp->b_error == 0) |
1020 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1021 | |
ce8e922c | 1022 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 1023 | if (schedule) { |
c4028958 | 1024 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 1025 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 1026 | } else { |
c4028958 | 1027 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
1028 | } |
1029 | } else { | |
b4dd330b | 1030 | complete(&bp->b_iowait); |
1da177e4 LT |
1031 | } |
1032 | } | |
1033 | ||
1da177e4 | 1034 | void |
ce8e922c NS |
1035 | xfs_buf_ioerror( |
1036 | xfs_buf_t *bp, | |
1037 | int error) | |
1da177e4 LT |
1038 | { |
1039 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c | 1040 | bp->b_error = (unsigned short)error; |
0b1b213f | 1041 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1042 | } |
1043 | ||
901796af CH |
1044 | void |
1045 | xfs_buf_ioerror_alert( | |
1046 | struct xfs_buf *bp, | |
1047 | const char *func) | |
1048 | { | |
1049 | xfs_alert(bp->b_target->bt_mount, | |
aa0e8833 DC |
1050 | "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d", |
1051 | (__uint64_t)XFS_BUF_ADDR(bp), func, bp->b_error, bp->b_length); | |
901796af CH |
1052 | } |
1053 | ||
4e23471a CH |
1054 | /* |
1055 | * Called when we want to stop a buffer from getting written or read. | |
1a1a3e97 | 1056 | * We attach the EIO error, muck with its flags, and call xfs_buf_ioend |
4e23471a CH |
1057 | * so that the proper iodone callbacks get called. |
1058 | */ | |
1059 | STATIC int | |
1060 | xfs_bioerror( | |
1061 | xfs_buf_t *bp) | |
1062 | { | |
1063 | #ifdef XFSERRORDEBUG | |
1064 | ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone); | |
1065 | #endif | |
1066 | ||
1067 | /* | |
1068 | * No need to wait until the buffer is unpinned, we aren't flushing it. | |
1069 | */ | |
5a52c2a5 | 1070 | xfs_buf_ioerror(bp, EIO); |
4e23471a CH |
1071 | |
1072 | /* | |
1a1a3e97 | 1073 | * We're calling xfs_buf_ioend, so delete XBF_DONE flag. |
4e23471a CH |
1074 | */ |
1075 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1076 | XFS_BUF_UNDONE(bp); |
c867cb61 | 1077 | xfs_buf_stale(bp); |
4e23471a | 1078 | |
1a1a3e97 | 1079 | xfs_buf_ioend(bp, 0); |
4e23471a CH |
1080 | |
1081 | return EIO; | |
1082 | } | |
1083 | ||
1084 | /* | |
1085 | * Same as xfs_bioerror, except that we are releasing the buffer | |
1a1a3e97 | 1086 | * here ourselves, and avoiding the xfs_buf_ioend call. |
4e23471a CH |
1087 | * This is meant for userdata errors; metadata bufs come with |
1088 | * iodone functions attached, so that we can track down errors. | |
1089 | */ | |
1090 | STATIC int | |
1091 | xfs_bioerror_relse( | |
1092 | struct xfs_buf *bp) | |
1093 | { | |
ed43233b | 1094 | int64_t fl = bp->b_flags; |
4e23471a CH |
1095 | /* |
1096 | * No need to wait until the buffer is unpinned. | |
1097 | * We aren't flushing it. | |
1098 | * | |
1099 | * chunkhold expects B_DONE to be set, whether | |
1100 | * we actually finish the I/O or not. We don't want to | |
1101 | * change that interface. | |
1102 | */ | |
1103 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1104 | XFS_BUF_DONE(bp); |
c867cb61 | 1105 | xfs_buf_stale(bp); |
cb669ca5 | 1106 | bp->b_iodone = NULL; |
0cadda1c | 1107 | if (!(fl & XBF_ASYNC)) { |
4e23471a CH |
1108 | /* |
1109 | * Mark b_error and B_ERROR _both_. | |
1110 | * Lot's of chunkcache code assumes that. | |
1111 | * There's no reason to mark error for | |
1112 | * ASYNC buffers. | |
1113 | */ | |
5a52c2a5 | 1114 | xfs_buf_ioerror(bp, EIO); |
5fde0326 | 1115 | complete(&bp->b_iowait); |
4e23471a CH |
1116 | } else { |
1117 | xfs_buf_relse(bp); | |
1118 | } | |
1119 | ||
1120 | return EIO; | |
1121 | } | |
1122 | ||
a2dcf5df | 1123 | STATIC int |
4e23471a CH |
1124 | xfs_bdstrat_cb( |
1125 | struct xfs_buf *bp) | |
1126 | { | |
ebad861b | 1127 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
4e23471a CH |
1128 | trace_xfs_bdstrat_shut(bp, _RET_IP_); |
1129 | /* | |
1130 | * Metadata write that didn't get logged but | |
1131 | * written delayed anyway. These aren't associated | |
1132 | * with a transaction, and can be ignored. | |
1133 | */ | |
1134 | if (!bp->b_iodone && !XFS_BUF_ISREAD(bp)) | |
1135 | return xfs_bioerror_relse(bp); | |
1136 | else | |
1137 | return xfs_bioerror(bp); | |
1138 | } | |
1139 | ||
1140 | xfs_buf_iorequest(bp); | |
1141 | return 0; | |
1142 | } | |
1143 | ||
a2dcf5df CH |
1144 | int |
1145 | xfs_bwrite( | |
1146 | struct xfs_buf *bp) | |
1147 | { | |
1148 | int error; | |
1149 | ||
1150 | ASSERT(xfs_buf_islocked(bp)); | |
1151 | ||
1152 | bp->b_flags |= XBF_WRITE; | |
1153 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q); | |
1154 | ||
1155 | xfs_bdstrat_cb(bp); | |
1156 | ||
1157 | error = xfs_buf_iowait(bp); | |
1158 | if (error) { | |
1159 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1160 | SHUTDOWN_META_IO_ERROR); | |
1161 | } | |
1162 | return error; | |
1163 | } | |
1164 | ||
4e23471a CH |
1165 | /* |
1166 | * Wrapper around bdstrat so that we can stop data from going to disk in case | |
1167 | * we are shutting down the filesystem. Typically user data goes thru this | |
1168 | * path; one of the exceptions is the superblock. | |
1169 | */ | |
1170 | void | |
1171 | xfsbdstrat( | |
1172 | struct xfs_mount *mp, | |
1173 | struct xfs_buf *bp) | |
1174 | { | |
1175 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1176 | trace_xfs_bdstrat_shut(bp, _RET_IP_); | |
1177 | xfs_bioerror_relse(bp); | |
1178 | return; | |
1179 | } | |
1180 | ||
1181 | xfs_buf_iorequest(bp); | |
1182 | } | |
1183 | ||
b8f82a4a | 1184 | STATIC void |
ce8e922c NS |
1185 | _xfs_buf_ioend( |
1186 | xfs_buf_t *bp, | |
1da177e4 LT |
1187 | int schedule) |
1188 | { | |
0e6e847f | 1189 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
ce8e922c | 1190 | xfs_buf_ioend(bp, schedule); |
1da177e4 LT |
1191 | } |
1192 | ||
782e3b3b | 1193 | STATIC void |
ce8e922c | 1194 | xfs_buf_bio_end_io( |
1da177e4 | 1195 | struct bio *bio, |
1da177e4 LT |
1196 | int error) |
1197 | { | |
ce8e922c | 1198 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1da177e4 | 1199 | |
cfbe5267 | 1200 | xfs_buf_ioerror(bp, -error); |
1da177e4 | 1201 | |
73c77e2c JB |
1202 | if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
1203 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); | |
1204 | ||
ce8e922c | 1205 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1206 | bio_put(bio); |
1da177e4 LT |
1207 | } |
1208 | ||
3e85c868 DC |
1209 | static void |
1210 | xfs_buf_ioapply_map( | |
1211 | struct xfs_buf *bp, | |
1212 | int map, | |
1213 | int *buf_offset, | |
1214 | int *count, | |
1215 | int rw) | |
1da177e4 | 1216 | { |
3e85c868 DC |
1217 | int page_index; |
1218 | int total_nr_pages = bp->b_page_count; | |
1219 | int nr_pages; | |
1220 | struct bio *bio; | |
1221 | sector_t sector = bp->b_maps[map].bm_bn; | |
1222 | int size; | |
1223 | int offset; | |
1da177e4 | 1224 | |
ce8e922c | 1225 | total_nr_pages = bp->b_page_count; |
1da177e4 | 1226 | |
3e85c868 DC |
1227 | /* skip the pages in the buffer before the start offset */ |
1228 | page_index = 0; | |
1229 | offset = *buf_offset; | |
1230 | while (offset >= PAGE_SIZE) { | |
1231 | page_index++; | |
1232 | offset -= PAGE_SIZE; | |
f538d4da CH |
1233 | } |
1234 | ||
3e85c868 DC |
1235 | /* |
1236 | * Limit the IO size to the length of the current vector, and update the | |
1237 | * remaining IO count for the next time around. | |
1238 | */ | |
1239 | size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count); | |
1240 | *count -= size; | |
1241 | *buf_offset += size; | |
34951f5c | 1242 | |
1da177e4 | 1243 | next_chunk: |
ce8e922c | 1244 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1245 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1246 | if (nr_pages > total_nr_pages) | |
1247 | nr_pages = total_nr_pages; | |
1248 | ||
1249 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1250 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1251 | bio->bi_sector = sector; |
ce8e922c NS |
1252 | bio->bi_end_io = xfs_buf_bio_end_io; |
1253 | bio->bi_private = bp; | |
1da177e4 | 1254 | |
0e6e847f | 1255 | |
3e85c868 | 1256 | for (; size && nr_pages; nr_pages--, page_index++) { |
0e6e847f | 1257 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1258 | |
1259 | if (nbytes > size) | |
1260 | nbytes = size; | |
1261 | ||
3e85c868 DC |
1262 | rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes, |
1263 | offset); | |
ce8e922c | 1264 | if (rbytes < nbytes) |
1da177e4 LT |
1265 | break; |
1266 | ||
1267 | offset = 0; | |
aa0e8833 | 1268 | sector += BTOBB(nbytes); |
1da177e4 LT |
1269 | size -= nbytes; |
1270 | total_nr_pages--; | |
1271 | } | |
1272 | ||
1da177e4 | 1273 | if (likely(bio->bi_size)) { |
73c77e2c JB |
1274 | if (xfs_buf_is_vmapped(bp)) { |
1275 | flush_kernel_vmap_range(bp->b_addr, | |
1276 | xfs_buf_vmap_len(bp)); | |
1277 | } | |
1da177e4 LT |
1278 | submit_bio(rw, bio); |
1279 | if (size) | |
1280 | goto next_chunk; | |
1281 | } else { | |
ce8e922c | 1282 | xfs_buf_ioerror(bp, EIO); |
ec53d1db | 1283 | bio_put(bio); |
1da177e4 | 1284 | } |
3e85c868 DC |
1285 | |
1286 | } | |
1287 | ||
1288 | STATIC void | |
1289 | _xfs_buf_ioapply( | |
1290 | struct xfs_buf *bp) | |
1291 | { | |
1292 | struct blk_plug plug; | |
1293 | int rw; | |
1294 | int offset; | |
1295 | int size; | |
1296 | int i; | |
1297 | ||
1298 | if (bp->b_flags & XBF_WRITE) { | |
1299 | if (bp->b_flags & XBF_SYNCIO) | |
1300 | rw = WRITE_SYNC; | |
1301 | else | |
1302 | rw = WRITE; | |
1303 | if (bp->b_flags & XBF_FUA) | |
1304 | rw |= REQ_FUA; | |
1305 | if (bp->b_flags & XBF_FLUSH) | |
1306 | rw |= REQ_FLUSH; | |
1307 | } else if (bp->b_flags & XBF_READ_AHEAD) { | |
1308 | rw = READA; | |
1309 | } else { | |
1310 | rw = READ; | |
1311 | } | |
1312 | ||
1313 | /* we only use the buffer cache for meta-data */ | |
1314 | rw |= REQ_META; | |
1315 | ||
1316 | /* | |
1317 | * Walk all the vectors issuing IO on them. Set up the initial offset | |
1318 | * into the buffer and the desired IO size before we start - | |
1319 | * _xfs_buf_ioapply_vec() will modify them appropriately for each | |
1320 | * subsequent call. | |
1321 | */ | |
1322 | offset = bp->b_offset; | |
1323 | size = BBTOB(bp->b_io_length); | |
1324 | blk_start_plug(&plug); | |
1325 | for (i = 0; i < bp->b_map_count; i++) { | |
1326 | xfs_buf_ioapply_map(bp, i, &offset, &size, rw); | |
1327 | if (bp->b_error) | |
1328 | break; | |
1329 | if (size <= 0) | |
1330 | break; /* all done */ | |
1331 | } | |
1332 | blk_finish_plug(&plug); | |
1da177e4 LT |
1333 | } |
1334 | ||
0e95f19a | 1335 | void |
ce8e922c NS |
1336 | xfs_buf_iorequest( |
1337 | xfs_buf_t *bp) | |
1da177e4 | 1338 | { |
0b1b213f | 1339 | trace_xfs_buf_iorequest(bp, _RET_IP_); |
1da177e4 | 1340 | |
43ff2122 | 1341 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
1da177e4 | 1342 | |
375ec69d | 1343 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1344 | xfs_buf_wait_unpin(bp); |
ce8e922c | 1345 | xfs_buf_hold(bp); |
1da177e4 LT |
1346 | |
1347 | /* Set the count to 1 initially, this will stop an I/O | |
1348 | * completion callout which happens before we have started | |
ce8e922c | 1349 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1350 | */ |
ce8e922c NS |
1351 | atomic_set(&bp->b_io_remaining, 1); |
1352 | _xfs_buf_ioapply(bp); | |
08023d6d | 1353 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1354 | |
ce8e922c | 1355 | xfs_buf_rele(bp); |
1da177e4 LT |
1356 | } |
1357 | ||
1358 | /* | |
0e95f19a DC |
1359 | * Waits for I/O to complete on the buffer supplied. It returns immediately if |
1360 | * no I/O is pending or there is already a pending error on the buffer. It | |
1361 | * returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1362 | */ |
1363 | int | |
ce8e922c NS |
1364 | xfs_buf_iowait( |
1365 | xfs_buf_t *bp) | |
1da177e4 | 1366 | { |
0b1b213f CH |
1367 | trace_xfs_buf_iowait(bp, _RET_IP_); |
1368 | ||
0e95f19a DC |
1369 | if (!bp->b_error) |
1370 | wait_for_completion(&bp->b_iowait); | |
0b1b213f CH |
1371 | |
1372 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
ce8e922c | 1373 | return bp->b_error; |
1da177e4 LT |
1374 | } |
1375 | ||
ce8e922c NS |
1376 | xfs_caddr_t |
1377 | xfs_buf_offset( | |
1378 | xfs_buf_t *bp, | |
1da177e4 LT |
1379 | size_t offset) |
1380 | { | |
1381 | struct page *page; | |
1382 | ||
611c9946 | 1383 | if (bp->b_addr) |
62926044 | 1384 | return bp->b_addr + offset; |
1da177e4 | 1385 | |
ce8e922c | 1386 | offset += bp->b_offset; |
0e6e847f DC |
1387 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
1388 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_SIZE-1)); | |
1da177e4 LT |
1389 | } |
1390 | ||
1391 | /* | |
1da177e4 LT |
1392 | * Move data into or out of a buffer. |
1393 | */ | |
1394 | void | |
ce8e922c NS |
1395 | xfs_buf_iomove( |
1396 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1397 | size_t boff, /* starting buffer offset */ |
1398 | size_t bsize, /* length to copy */ | |
b9c48649 | 1399 | void *data, /* data address */ |
ce8e922c | 1400 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 | 1401 | { |
795cac72 | 1402 | size_t bend; |
1da177e4 LT |
1403 | |
1404 | bend = boff + bsize; | |
1405 | while (boff < bend) { | |
795cac72 DC |
1406 | struct page *page; |
1407 | int page_index, page_offset, csize; | |
1408 | ||
1409 | page_index = (boff + bp->b_offset) >> PAGE_SHIFT; | |
1410 | page_offset = (boff + bp->b_offset) & ~PAGE_MASK; | |
1411 | page = bp->b_pages[page_index]; | |
1412 | csize = min_t(size_t, PAGE_SIZE - page_offset, | |
1413 | BBTOB(bp->b_io_length) - boff); | |
1da177e4 | 1414 | |
795cac72 | 1415 | ASSERT((csize + page_offset) <= PAGE_SIZE); |
1da177e4 LT |
1416 | |
1417 | switch (mode) { | |
ce8e922c | 1418 | case XBRW_ZERO: |
795cac72 | 1419 | memset(page_address(page) + page_offset, 0, csize); |
1da177e4 | 1420 | break; |
ce8e922c | 1421 | case XBRW_READ: |
795cac72 | 1422 | memcpy(data, page_address(page) + page_offset, csize); |
1da177e4 | 1423 | break; |
ce8e922c | 1424 | case XBRW_WRITE: |
795cac72 | 1425 | memcpy(page_address(page) + page_offset, data, csize); |
1da177e4 LT |
1426 | } |
1427 | ||
1428 | boff += csize; | |
1429 | data += csize; | |
1430 | } | |
1431 | } | |
1432 | ||
1433 | /* | |
ce8e922c | 1434 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1435 | */ |
1436 | ||
1437 | /* | |
430cbeb8 DC |
1438 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1439 | * returned. These buffers will have an elevated hold count, so wait on those | |
1440 | * while freeing all the buffers only held by the LRU. | |
1da177e4 LT |
1441 | */ |
1442 | void | |
1443 | xfs_wait_buftarg( | |
74f75a0c | 1444 | struct xfs_buftarg *btp) |
1da177e4 | 1445 | { |
430cbeb8 DC |
1446 | struct xfs_buf *bp; |
1447 | ||
1448 | restart: | |
1449 | spin_lock(&btp->bt_lru_lock); | |
1450 | while (!list_empty(&btp->bt_lru)) { | |
1451 | bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru); | |
1452 | if (atomic_read(&bp->b_hold) > 1) { | |
1453 | spin_unlock(&btp->bt_lru_lock); | |
26af6552 | 1454 | delay(100); |
430cbeb8 | 1455 | goto restart; |
1da177e4 | 1456 | } |
430cbeb8 | 1457 | /* |
90802ed9 | 1458 | * clear the LRU reference count so the buffer doesn't get |
430cbeb8 DC |
1459 | * ignored in xfs_buf_rele(). |
1460 | */ | |
1461 | atomic_set(&bp->b_lru_ref, 0); | |
1462 | spin_unlock(&btp->bt_lru_lock); | |
1463 | xfs_buf_rele(bp); | |
1464 | spin_lock(&btp->bt_lru_lock); | |
1da177e4 | 1465 | } |
430cbeb8 | 1466 | spin_unlock(&btp->bt_lru_lock); |
1da177e4 LT |
1467 | } |
1468 | ||
ff57ab21 DC |
1469 | int |
1470 | xfs_buftarg_shrink( | |
1471 | struct shrinker *shrink, | |
1495f230 | 1472 | struct shrink_control *sc) |
a6867a68 | 1473 | { |
ff57ab21 DC |
1474 | struct xfs_buftarg *btp = container_of(shrink, |
1475 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1476 | struct xfs_buf *bp; |
1495f230 | 1477 | int nr_to_scan = sc->nr_to_scan; |
430cbeb8 DC |
1478 | LIST_HEAD(dispose); |
1479 | ||
1480 | if (!nr_to_scan) | |
1481 | return btp->bt_lru_nr; | |
1482 | ||
1483 | spin_lock(&btp->bt_lru_lock); | |
1484 | while (!list_empty(&btp->bt_lru)) { | |
1485 | if (nr_to_scan-- <= 0) | |
1486 | break; | |
1487 | ||
1488 | bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru); | |
1489 | ||
1490 | /* | |
1491 | * Decrement the b_lru_ref count unless the value is already | |
1492 | * zero. If the value is already zero, we need to reclaim the | |
1493 | * buffer, otherwise it gets another trip through the LRU. | |
1494 | */ | |
1495 | if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) { | |
1496 | list_move_tail(&bp->b_lru, &btp->bt_lru); | |
1497 | continue; | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * remove the buffer from the LRU now to avoid needing another | |
1502 | * lock round trip inside xfs_buf_rele(). | |
1503 | */ | |
1504 | list_move(&bp->b_lru, &dispose); | |
1505 | btp->bt_lru_nr--; | |
6fb8a90a | 1506 | bp->b_lru_flags |= _XBF_LRU_DISPOSE; |
ff57ab21 | 1507 | } |
430cbeb8 DC |
1508 | spin_unlock(&btp->bt_lru_lock); |
1509 | ||
1510 | while (!list_empty(&dispose)) { | |
1511 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1512 | list_del_init(&bp->b_lru); | |
1513 | xfs_buf_rele(bp); | |
1514 | } | |
1515 | ||
1516 | return btp->bt_lru_nr; | |
a6867a68 DC |
1517 | } |
1518 | ||
1da177e4 LT |
1519 | void |
1520 | xfs_free_buftarg( | |
b7963133 CH |
1521 | struct xfs_mount *mp, |
1522 | struct xfs_buftarg *btp) | |
1da177e4 | 1523 | { |
ff57ab21 DC |
1524 | unregister_shrinker(&btp->bt_shrinker); |
1525 | ||
b7963133 CH |
1526 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1527 | xfs_blkdev_issue_flush(btp); | |
a6867a68 | 1528 | |
f0e2d93c | 1529 | kmem_free(btp); |
1da177e4 LT |
1530 | } |
1531 | ||
1da177e4 LT |
1532 | STATIC int |
1533 | xfs_setsize_buftarg_flags( | |
1534 | xfs_buftarg_t *btp, | |
1535 | unsigned int blocksize, | |
1536 | unsigned int sectorsize, | |
1537 | int verbose) | |
1538 | { | |
ce8e922c NS |
1539 | btp->bt_bsize = blocksize; |
1540 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1541 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1542 | |
ce8e922c | 1543 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
02b102df CH |
1544 | char name[BDEVNAME_SIZE]; |
1545 | ||
1546 | bdevname(btp->bt_bdev, name); | |
1547 | ||
4f10700a DC |
1548 | xfs_warn(btp->bt_mount, |
1549 | "Cannot set_blocksize to %u on device %s\n", | |
02b102df | 1550 | sectorsize, name); |
1da177e4 LT |
1551 | return EINVAL; |
1552 | } | |
1553 | ||
1da177e4 LT |
1554 | return 0; |
1555 | } | |
1556 | ||
1557 | /* | |
ce8e922c NS |
1558 | * When allocating the initial buffer target we have not yet |
1559 | * read in the superblock, so don't know what sized sectors | |
1560 | * are being used is at this early stage. Play safe. | |
1561 | */ | |
1da177e4 LT |
1562 | STATIC int |
1563 | xfs_setsize_buftarg_early( | |
1564 | xfs_buftarg_t *btp, | |
1565 | struct block_device *bdev) | |
1566 | { | |
1567 | return xfs_setsize_buftarg_flags(btp, | |
0e6e847f | 1568 | PAGE_SIZE, bdev_logical_block_size(bdev), 0); |
1da177e4 LT |
1569 | } |
1570 | ||
1571 | int | |
1572 | xfs_setsize_buftarg( | |
1573 | xfs_buftarg_t *btp, | |
1574 | unsigned int blocksize, | |
1575 | unsigned int sectorsize) | |
1576 | { | |
1577 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1578 | } | |
1579 | ||
1da177e4 LT |
1580 | xfs_buftarg_t * |
1581 | xfs_alloc_buftarg( | |
ebad861b | 1582 | struct xfs_mount *mp, |
1da177e4 | 1583 | struct block_device *bdev, |
e2a07812 JE |
1584 | int external, |
1585 | const char *fsname) | |
1da177e4 LT |
1586 | { |
1587 | xfs_buftarg_t *btp; | |
1588 | ||
1589 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1590 | ||
ebad861b | 1591 | btp->bt_mount = mp; |
ce8e922c NS |
1592 | btp->bt_dev = bdev->bd_dev; |
1593 | btp->bt_bdev = bdev; | |
0e6e847f DC |
1594 | btp->bt_bdi = blk_get_backing_dev_info(bdev); |
1595 | if (!btp->bt_bdi) | |
1596 | goto error; | |
1597 | ||
430cbeb8 DC |
1598 | INIT_LIST_HEAD(&btp->bt_lru); |
1599 | spin_lock_init(&btp->bt_lru_lock); | |
1da177e4 LT |
1600 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1601 | goto error; | |
ff57ab21 DC |
1602 | btp->bt_shrinker.shrink = xfs_buftarg_shrink; |
1603 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; | |
1604 | register_shrinker(&btp->bt_shrinker); | |
1da177e4 LT |
1605 | return btp; |
1606 | ||
1607 | error: | |
f0e2d93c | 1608 | kmem_free(btp); |
1da177e4 LT |
1609 | return NULL; |
1610 | } | |
1611 | ||
1da177e4 | 1612 | /* |
43ff2122 CH |
1613 | * Add a buffer to the delayed write list. |
1614 | * | |
1615 | * This queues a buffer for writeout if it hasn't already been. Note that | |
1616 | * neither this routine nor the buffer list submission functions perform | |
1617 | * any internal synchronization. It is expected that the lists are thread-local | |
1618 | * to the callers. | |
1619 | * | |
1620 | * Returns true if we queued up the buffer, or false if it already had | |
1621 | * been on the buffer list. | |
1da177e4 | 1622 | */ |
43ff2122 | 1623 | bool |
ce8e922c | 1624 | xfs_buf_delwri_queue( |
43ff2122 CH |
1625 | struct xfs_buf *bp, |
1626 | struct list_head *list) | |
1da177e4 | 1627 | { |
43ff2122 | 1628 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 1629 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1630 | |
43ff2122 CH |
1631 | /* |
1632 | * If the buffer is already marked delwri it already is queued up | |
1633 | * by someone else for imediate writeout. Just ignore it in that | |
1634 | * case. | |
1635 | */ | |
1636 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
1637 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
1638 | return false; | |
1da177e4 | 1639 | } |
1da177e4 | 1640 | |
43ff2122 | 1641 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
1642 | |
1643 | /* | |
43ff2122 CH |
1644 | * If a buffer gets written out synchronously or marked stale while it |
1645 | * is on a delwri list we lazily remove it. To do this, the other party | |
1646 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
1647 | * It remains referenced and on the list. In a rare corner case it | |
1648 | * might get readded to a delwri list after the synchronous writeout, in | |
1649 | * which case we need just need to re-add the flag here. | |
d808f617 | 1650 | */ |
43ff2122 CH |
1651 | bp->b_flags |= _XBF_DELWRI_Q; |
1652 | if (list_empty(&bp->b_list)) { | |
1653 | atomic_inc(&bp->b_hold); | |
1654 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 1655 | } |
585e6d88 | 1656 | |
43ff2122 | 1657 | return true; |
585e6d88 DC |
1658 | } |
1659 | ||
089716aa DC |
1660 | /* |
1661 | * Compare function is more complex than it needs to be because | |
1662 | * the return value is only 32 bits and we are doing comparisons | |
1663 | * on 64 bit values | |
1664 | */ | |
1665 | static int | |
1666 | xfs_buf_cmp( | |
1667 | void *priv, | |
1668 | struct list_head *a, | |
1669 | struct list_head *b) | |
1670 | { | |
1671 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1672 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1673 | xfs_daddr_t diff; | |
1674 | ||
cbb7baab | 1675 | diff = ap->b_map.bm_bn - bp->b_map.bm_bn; |
089716aa DC |
1676 | if (diff < 0) |
1677 | return -1; | |
1678 | if (diff > 0) | |
1679 | return 1; | |
1680 | return 0; | |
1681 | } | |
1682 | ||
43ff2122 CH |
1683 | static int |
1684 | __xfs_buf_delwri_submit( | |
1685 | struct list_head *buffer_list, | |
1686 | struct list_head *io_list, | |
1687 | bool wait) | |
1da177e4 | 1688 | { |
43ff2122 CH |
1689 | struct blk_plug plug; |
1690 | struct xfs_buf *bp, *n; | |
1691 | int pinned = 0; | |
1692 | ||
1693 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { | |
1694 | if (!wait) { | |
1695 | if (xfs_buf_ispinned(bp)) { | |
1696 | pinned++; | |
1697 | continue; | |
1698 | } | |
1699 | if (!xfs_buf_trylock(bp)) | |
1700 | continue; | |
1701 | } else { | |
1702 | xfs_buf_lock(bp); | |
1703 | } | |
978c7b2f | 1704 | |
43ff2122 CH |
1705 | /* |
1706 | * Someone else might have written the buffer synchronously or | |
1707 | * marked it stale in the meantime. In that case only the | |
1708 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
1709 | * reference and remove it from the list here. | |
1710 | */ | |
1711 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
1712 | list_del_init(&bp->b_list); | |
1713 | xfs_buf_relse(bp); | |
1714 | continue; | |
1715 | } | |
c9c12971 | 1716 | |
43ff2122 CH |
1717 | list_move_tail(&bp->b_list, io_list); |
1718 | trace_xfs_buf_delwri_split(bp, _RET_IP_); | |
1719 | } | |
1da177e4 | 1720 | |
43ff2122 | 1721 | list_sort(NULL, io_list, xfs_buf_cmp); |
1da177e4 | 1722 | |
43ff2122 CH |
1723 | blk_start_plug(&plug); |
1724 | list_for_each_entry_safe(bp, n, io_list, b_list) { | |
1725 | bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_ASYNC); | |
1726 | bp->b_flags |= XBF_WRITE; | |
a1b7ea5d | 1727 | |
43ff2122 CH |
1728 | if (!wait) { |
1729 | bp->b_flags |= XBF_ASYNC; | |
ce8e922c | 1730 | list_del_init(&bp->b_list); |
1da177e4 | 1731 | } |
43ff2122 CH |
1732 | xfs_bdstrat_cb(bp); |
1733 | } | |
1734 | blk_finish_plug(&plug); | |
1da177e4 | 1735 | |
43ff2122 | 1736 | return pinned; |
1da177e4 LT |
1737 | } |
1738 | ||
1739 | /* | |
43ff2122 CH |
1740 | * Write out a buffer list asynchronously. |
1741 | * | |
1742 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
1743 | * out and not wait for I/O completion on any of the buffers. This interface | |
1744 | * is only safely useable for callers that can track I/O completion by higher | |
1745 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
1746 | * function. | |
1da177e4 LT |
1747 | */ |
1748 | int | |
43ff2122 CH |
1749 | xfs_buf_delwri_submit_nowait( |
1750 | struct list_head *buffer_list) | |
1da177e4 | 1751 | { |
43ff2122 CH |
1752 | LIST_HEAD (io_list); |
1753 | return __xfs_buf_delwri_submit(buffer_list, &io_list, false); | |
1754 | } | |
1da177e4 | 1755 | |
43ff2122 CH |
1756 | /* |
1757 | * Write out a buffer list synchronously. | |
1758 | * | |
1759 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
1760 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
1761 | * so callers must have some other way of tracking buffers if they require such | |
1762 | * functionality. | |
1763 | */ | |
1764 | int | |
1765 | xfs_buf_delwri_submit( | |
1766 | struct list_head *buffer_list) | |
1767 | { | |
1768 | LIST_HEAD (io_list); | |
1769 | int error = 0, error2; | |
1770 | struct xfs_buf *bp; | |
1da177e4 | 1771 | |
43ff2122 | 1772 | __xfs_buf_delwri_submit(buffer_list, &io_list, true); |
1da177e4 | 1773 | |
43ff2122 CH |
1774 | /* Wait for IO to complete. */ |
1775 | while (!list_empty(&io_list)) { | |
1776 | bp = list_first_entry(&io_list, struct xfs_buf, b_list); | |
a1b7ea5d | 1777 | |
089716aa | 1778 | list_del_init(&bp->b_list); |
43ff2122 CH |
1779 | error2 = xfs_buf_iowait(bp); |
1780 | xfs_buf_relse(bp); | |
1781 | if (!error) | |
1782 | error = error2; | |
1da177e4 LT |
1783 | } |
1784 | ||
43ff2122 | 1785 | return error; |
1da177e4 LT |
1786 | } |
1787 | ||
04d8b284 | 1788 | int __init |
ce8e922c | 1789 | xfs_buf_init(void) |
1da177e4 | 1790 | { |
8758280f NS |
1791 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1792 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1793 | if (!xfs_buf_zone) |
0b1b213f | 1794 | goto out; |
04d8b284 | 1795 | |
51749e47 | 1796 | xfslogd_workqueue = alloc_workqueue("xfslogd", |
6370a6ad | 1797 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 1); |
23ea4032 | 1798 | if (!xfslogd_workqueue) |
04d8b284 | 1799 | goto out_free_buf_zone; |
1da177e4 | 1800 | |
23ea4032 | 1801 | return 0; |
1da177e4 | 1802 | |
23ea4032 | 1803 | out_free_buf_zone: |
ce8e922c | 1804 | kmem_zone_destroy(xfs_buf_zone); |
0b1b213f | 1805 | out: |
8758280f | 1806 | return -ENOMEM; |
1da177e4 LT |
1807 | } |
1808 | ||
1da177e4 | 1809 | void |
ce8e922c | 1810 | xfs_buf_terminate(void) |
1da177e4 | 1811 | { |
04d8b284 | 1812 | destroy_workqueue(xfslogd_workqueue); |
ce8e922c | 1813 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 1814 | } |