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