]>
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 | |
239880ef | 37 | #include "xfs_log_format.h" |
7fd36c44 | 38 | #include "xfs_trans_resv.h" |
239880ef | 39 | #include "xfs_sb.h" |
b7963133 | 40 | #include "xfs_ag.h" |
b7963133 | 41 | #include "xfs_mount.h" |
0b1b213f | 42 | #include "xfs_trace.h" |
239880ef | 43 | #include "xfs_log.h" |
b7963133 | 44 | |
7989cb8e | 45 | static kmem_zone_t *xfs_buf_zone; |
23ea4032 | 46 | |
7989cb8e | 47 | static struct workqueue_struct *xfslogd_workqueue; |
1da177e4 | 48 | |
ce8e922c NS |
49 | #ifdef XFS_BUF_LOCK_TRACKING |
50 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
51 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
52 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 53 | #else |
ce8e922c NS |
54 | # define XB_SET_OWNER(bp) do { } while (0) |
55 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
56 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
57 | #endif |
58 | ||
ce8e922c | 59 | #define xb_to_gfp(flags) \ |
aa5c158e | 60 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN) |
1da177e4 | 61 | |
1da177e4 | 62 | |
73c77e2c JB |
63 | static inline int |
64 | xfs_buf_is_vmapped( | |
65 | struct xfs_buf *bp) | |
66 | { | |
67 | /* | |
68 | * Return true if the buffer is vmapped. | |
69 | * | |
611c9946 DC |
70 | * b_addr is null if the buffer is not mapped, but the code is clever |
71 | * enough to know it doesn't have to map a single page, so the check has | |
72 | * to be both for b_addr and bp->b_page_count > 1. | |
73c77e2c | 73 | */ |
611c9946 | 74 | return bp->b_addr && bp->b_page_count > 1; |
73c77e2c JB |
75 | } |
76 | ||
77 | static inline int | |
78 | xfs_buf_vmap_len( | |
79 | struct xfs_buf *bp) | |
80 | { | |
81 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
82 | } | |
83 | ||
430cbeb8 DC |
84 | /* |
85 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
86 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
87 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
88 | * to remove the reference that LRU holds on the buffer. | |
89 | * | |
90 | * This prevents build-up of stale buffers on the LRU. | |
91 | */ | |
92 | void | |
93 | xfs_buf_stale( | |
94 | struct xfs_buf *bp) | |
95 | { | |
43ff2122 CH |
96 | ASSERT(xfs_buf_islocked(bp)); |
97 | ||
430cbeb8 | 98 | bp->b_flags |= XBF_STALE; |
43ff2122 CH |
99 | |
100 | /* | |
101 | * Clear the delwri status so that a delwri queue walker will not | |
102 | * flush this buffer to disk now that it is stale. The delwri queue has | |
103 | * a reference to the buffer, so this is safe to do. | |
104 | */ | |
105 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
106 | ||
a4082357 DC |
107 | spin_lock(&bp->b_lock); |
108 | atomic_set(&bp->b_lru_ref, 0); | |
109 | if (!(bp->b_state & XFS_BSTATE_DISPOSE) && | |
e80dfa19 DC |
110 | (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru))) |
111 | atomic_dec(&bp->b_hold); | |
112 | ||
430cbeb8 | 113 | ASSERT(atomic_read(&bp->b_hold) >= 1); |
a4082357 | 114 | spin_unlock(&bp->b_lock); |
430cbeb8 | 115 | } |
1da177e4 | 116 | |
3e85c868 DC |
117 | static int |
118 | xfs_buf_get_maps( | |
119 | struct xfs_buf *bp, | |
120 | int map_count) | |
121 | { | |
122 | ASSERT(bp->b_maps == NULL); | |
123 | bp->b_map_count = map_count; | |
124 | ||
125 | if (map_count == 1) { | |
f4b42421 | 126 | bp->b_maps = &bp->__b_map; |
3e85c868 DC |
127 | return 0; |
128 | } | |
129 | ||
130 | bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map), | |
131 | KM_NOFS); | |
132 | if (!bp->b_maps) | |
2451337d | 133 | return -ENOMEM; |
3e85c868 DC |
134 | return 0; |
135 | } | |
136 | ||
137 | /* | |
138 | * Frees b_pages if it was allocated. | |
139 | */ | |
140 | static void | |
141 | xfs_buf_free_maps( | |
142 | struct xfs_buf *bp) | |
143 | { | |
f4b42421 | 144 | if (bp->b_maps != &bp->__b_map) { |
3e85c868 DC |
145 | kmem_free(bp->b_maps); |
146 | bp->b_maps = NULL; | |
147 | } | |
148 | } | |
149 | ||
4347b9d7 | 150 | struct xfs_buf * |
3e85c868 | 151 | _xfs_buf_alloc( |
4347b9d7 | 152 | struct xfs_buftarg *target, |
3e85c868 DC |
153 | struct xfs_buf_map *map, |
154 | int nmaps, | |
ce8e922c | 155 | xfs_buf_flags_t flags) |
1da177e4 | 156 | { |
4347b9d7 | 157 | struct xfs_buf *bp; |
3e85c868 DC |
158 | int error; |
159 | int i; | |
4347b9d7 | 160 | |
aa5c158e | 161 | bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS); |
4347b9d7 CH |
162 | if (unlikely(!bp)) |
163 | return NULL; | |
164 | ||
1da177e4 | 165 | /* |
12bcb3f7 DC |
166 | * We don't want certain flags to appear in b_flags unless they are |
167 | * specifically set by later operations on the buffer. | |
1da177e4 | 168 | */ |
611c9946 | 169 | flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD); |
ce8e922c | 170 | |
ce8e922c | 171 | atomic_set(&bp->b_hold, 1); |
430cbeb8 | 172 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 173 | init_completion(&bp->b_iowait); |
430cbeb8 | 174 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 175 | INIT_LIST_HEAD(&bp->b_list); |
74f75a0c | 176 | RB_CLEAR_NODE(&bp->b_rbnode); |
a731cd11 | 177 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
a4082357 | 178 | spin_lock_init(&bp->b_lock); |
ce8e922c NS |
179 | XB_SET_OWNER(bp); |
180 | bp->b_target = target; | |
3e85c868 | 181 | bp->b_flags = flags; |
de1cbee4 | 182 | |
1da177e4 | 183 | /* |
aa0e8833 DC |
184 | * Set length and io_length to the same value initially. |
185 | * I/O routines should use io_length, which will be the same in | |
1da177e4 LT |
186 | * most cases but may be reset (e.g. XFS recovery). |
187 | */ | |
3e85c868 DC |
188 | error = xfs_buf_get_maps(bp, nmaps); |
189 | if (error) { | |
190 | kmem_zone_free(xfs_buf_zone, bp); | |
191 | return NULL; | |
192 | } | |
193 | ||
194 | bp->b_bn = map[0].bm_bn; | |
195 | bp->b_length = 0; | |
196 | for (i = 0; i < nmaps; i++) { | |
197 | bp->b_maps[i].bm_bn = map[i].bm_bn; | |
198 | bp->b_maps[i].bm_len = map[i].bm_len; | |
199 | bp->b_length += map[i].bm_len; | |
200 | } | |
201 | bp->b_io_length = bp->b_length; | |
202 | ||
ce8e922c NS |
203 | atomic_set(&bp->b_pin_count, 0); |
204 | init_waitqueue_head(&bp->b_waiters); | |
205 | ||
206 | XFS_STATS_INC(xb_create); | |
0b1b213f | 207 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 CH |
208 | |
209 | return bp; | |
1da177e4 LT |
210 | } |
211 | ||
212 | /* | |
ce8e922c NS |
213 | * Allocate a page array capable of holding a specified number |
214 | * of pages, and point the page buf at it. | |
1da177e4 LT |
215 | */ |
216 | STATIC int | |
ce8e922c NS |
217 | _xfs_buf_get_pages( |
218 | xfs_buf_t *bp, | |
87937bf8 | 219 | int page_count) |
1da177e4 LT |
220 | { |
221 | /* Make sure that we have a page list */ | |
ce8e922c | 222 | if (bp->b_pages == NULL) { |
ce8e922c NS |
223 | bp->b_page_count = page_count; |
224 | if (page_count <= XB_PAGES) { | |
225 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 226 | } else { |
ce8e922c | 227 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
aa5c158e | 228 | page_count, KM_NOFS); |
ce8e922c | 229 | if (bp->b_pages == NULL) |
1da177e4 LT |
230 | return -ENOMEM; |
231 | } | |
ce8e922c | 232 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
233 | } |
234 | return 0; | |
235 | } | |
236 | ||
237 | /* | |
ce8e922c | 238 | * Frees b_pages if it was allocated. |
1da177e4 LT |
239 | */ |
240 | STATIC void | |
ce8e922c | 241 | _xfs_buf_free_pages( |
1da177e4 LT |
242 | xfs_buf_t *bp) |
243 | { | |
ce8e922c | 244 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 245 | kmem_free(bp->b_pages); |
3fc98b1a | 246 | bp->b_pages = NULL; |
1da177e4 LT |
247 | } |
248 | } | |
249 | ||
250 | /* | |
251 | * Releases the specified buffer. | |
252 | * | |
253 | * The modification state of any associated pages is left unchanged. | |
b46fe825 | 254 | * The buffer must not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
255 | * hashed and refcounted buffers |
256 | */ | |
257 | void | |
ce8e922c | 258 | xfs_buf_free( |
1da177e4 LT |
259 | xfs_buf_t *bp) |
260 | { | |
0b1b213f | 261 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 262 | |
430cbeb8 DC |
263 | ASSERT(list_empty(&bp->b_lru)); |
264 | ||
0e6e847f | 265 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
266 | uint i; |
267 | ||
73c77e2c | 268 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
269 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
270 | bp->b_page_count); | |
1da177e4 | 271 | |
948ecdb4 NS |
272 | for (i = 0; i < bp->b_page_count; i++) { |
273 | struct page *page = bp->b_pages[i]; | |
274 | ||
0e6e847f | 275 | __free_page(page); |
948ecdb4 | 276 | } |
0e6e847f DC |
277 | } else if (bp->b_flags & _XBF_KMEM) |
278 | kmem_free(bp->b_addr); | |
3fc98b1a | 279 | _xfs_buf_free_pages(bp); |
3e85c868 | 280 | xfs_buf_free_maps(bp); |
4347b9d7 | 281 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
282 | } |
283 | ||
284 | /* | |
0e6e847f | 285 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
286 | */ |
287 | STATIC int | |
0e6e847f | 288 | xfs_buf_allocate_memory( |
1da177e4 LT |
289 | xfs_buf_t *bp, |
290 | uint flags) | |
291 | { | |
aa0e8833 | 292 | size_t size; |
1da177e4 | 293 | size_t nbytes, offset; |
ce8e922c | 294 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 295 | unsigned short page_count, i; |
795cac72 | 296 | xfs_off_t start, end; |
1da177e4 LT |
297 | int error; |
298 | ||
0e6e847f DC |
299 | /* |
300 | * for buffers that are contained within a single page, just allocate | |
301 | * the memory from the heap - there's no need for the complexity of | |
302 | * page arrays to keep allocation down to order 0. | |
303 | */ | |
795cac72 DC |
304 | size = BBTOB(bp->b_length); |
305 | if (size < PAGE_SIZE) { | |
aa5c158e | 306 | bp->b_addr = kmem_alloc(size, KM_NOFS); |
0e6e847f DC |
307 | if (!bp->b_addr) { |
308 | /* low memory - use alloc_page loop instead */ | |
309 | goto use_alloc_page; | |
310 | } | |
311 | ||
795cac72 | 312 | if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) != |
0e6e847f DC |
313 | ((unsigned long)bp->b_addr & PAGE_MASK)) { |
314 | /* b_addr spans two pages - use alloc_page instead */ | |
315 | kmem_free(bp->b_addr); | |
316 | bp->b_addr = NULL; | |
317 | goto use_alloc_page; | |
318 | } | |
319 | bp->b_offset = offset_in_page(bp->b_addr); | |
320 | bp->b_pages = bp->b_page_array; | |
321 | bp->b_pages[0] = virt_to_page(bp->b_addr); | |
322 | bp->b_page_count = 1; | |
611c9946 | 323 | bp->b_flags |= _XBF_KMEM; |
0e6e847f DC |
324 | return 0; |
325 | } | |
326 | ||
327 | use_alloc_page: | |
f4b42421 MT |
328 | start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT; |
329 | end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1) | |
cbb7baab | 330 | >> PAGE_SHIFT; |
795cac72 | 331 | page_count = end - start; |
87937bf8 | 332 | error = _xfs_buf_get_pages(bp, page_count); |
1da177e4 LT |
333 | if (unlikely(error)) |
334 | return error; | |
1da177e4 | 335 | |
ce8e922c | 336 | offset = bp->b_offset; |
0e6e847f | 337 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 338 | |
ce8e922c | 339 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
340 | struct page *page; |
341 | uint retries = 0; | |
0e6e847f DC |
342 | retry: |
343 | page = alloc_page(gfp_mask); | |
1da177e4 | 344 | if (unlikely(page == NULL)) { |
ce8e922c NS |
345 | if (flags & XBF_READ_AHEAD) { |
346 | bp->b_page_count = i; | |
2451337d | 347 | error = -ENOMEM; |
0e6e847f | 348 | goto out_free_pages; |
1da177e4 LT |
349 | } |
350 | ||
351 | /* | |
352 | * This could deadlock. | |
353 | * | |
354 | * But until all the XFS lowlevel code is revamped to | |
355 | * handle buffer allocation failures we can't do much. | |
356 | */ | |
357 | if (!(++retries % 100)) | |
4f10700a DC |
358 | xfs_err(NULL, |
359 | "possible memory allocation deadlock in %s (mode:0x%x)", | |
34a622b2 | 360 | __func__, gfp_mask); |
1da177e4 | 361 | |
ce8e922c | 362 | XFS_STATS_INC(xb_page_retries); |
8aa7e847 | 363 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
364 | goto retry; |
365 | } | |
366 | ||
ce8e922c | 367 | XFS_STATS_INC(xb_page_found); |
1da177e4 | 368 | |
0e6e847f | 369 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 370 | size -= nbytes; |
ce8e922c | 371 | bp->b_pages[i] = page; |
1da177e4 LT |
372 | offset = 0; |
373 | } | |
0e6e847f | 374 | return 0; |
1da177e4 | 375 | |
0e6e847f DC |
376 | out_free_pages: |
377 | for (i = 0; i < bp->b_page_count; i++) | |
378 | __free_page(bp->b_pages[i]); | |
1da177e4 LT |
379 | return error; |
380 | } | |
381 | ||
382 | /* | |
25985edc | 383 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
384 | */ |
385 | STATIC int | |
ce8e922c | 386 | _xfs_buf_map_pages( |
1da177e4 LT |
387 | xfs_buf_t *bp, |
388 | uint flags) | |
389 | { | |
0e6e847f | 390 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 391 | if (bp->b_page_count == 1) { |
0e6e847f | 392 | /* A single page buffer is always mappable */ |
ce8e922c | 393 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
611c9946 DC |
394 | } else if (flags & XBF_UNMAPPED) { |
395 | bp->b_addr = NULL; | |
396 | } else { | |
a19fb380 | 397 | int retried = 0; |
ae687e58 DC |
398 | unsigned noio_flag; |
399 | ||
400 | /* | |
401 | * vm_map_ram() will allocate auxillary structures (e.g. | |
402 | * pagetables) with GFP_KERNEL, yet we are likely to be under | |
403 | * GFP_NOFS context here. Hence we need to tell memory reclaim | |
404 | * that we are in such a context via PF_MEMALLOC_NOIO to prevent | |
405 | * memory reclaim re-entering the filesystem here and | |
406 | * potentially deadlocking. | |
407 | */ | |
408 | noio_flag = memalloc_noio_save(); | |
a19fb380 DC |
409 | do { |
410 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
411 | -1, PAGE_KERNEL); | |
412 | if (bp->b_addr) | |
413 | break; | |
414 | vm_unmap_aliases(); | |
415 | } while (retried++ <= 1); | |
ae687e58 | 416 | memalloc_noio_restore(noio_flag); |
a19fb380 DC |
417 | |
418 | if (!bp->b_addr) | |
1da177e4 | 419 | return -ENOMEM; |
ce8e922c | 420 | bp->b_addr += bp->b_offset; |
1da177e4 LT |
421 | } |
422 | ||
423 | return 0; | |
424 | } | |
425 | ||
426 | /* | |
427 | * Finding and Reading Buffers | |
428 | */ | |
429 | ||
430 | /* | |
ce8e922c | 431 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 | 432 | * a given range of an inode. The buffer is returned |
eabbaf11 | 433 | * locked. No I/O is implied by this call. |
1da177e4 LT |
434 | */ |
435 | xfs_buf_t * | |
ce8e922c | 436 | _xfs_buf_find( |
e70b73f8 | 437 | struct xfs_buftarg *btp, |
3e85c868 DC |
438 | struct xfs_buf_map *map, |
439 | int nmaps, | |
ce8e922c NS |
440 | xfs_buf_flags_t flags, |
441 | xfs_buf_t *new_bp) | |
1da177e4 | 442 | { |
e70b73f8 | 443 | size_t numbytes; |
74f75a0c DC |
444 | struct xfs_perag *pag; |
445 | struct rb_node **rbp; | |
446 | struct rb_node *parent; | |
447 | xfs_buf_t *bp; | |
3e85c868 | 448 | xfs_daddr_t blkno = map[0].bm_bn; |
10616b80 | 449 | xfs_daddr_t eofs; |
3e85c868 DC |
450 | int numblks = 0; |
451 | int i; | |
1da177e4 | 452 | |
3e85c868 DC |
453 | for (i = 0; i < nmaps; i++) |
454 | numblks += map[i].bm_len; | |
e70b73f8 | 455 | numbytes = BBTOB(numblks); |
1da177e4 LT |
456 | |
457 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
6da54179 ES |
458 | ASSERT(!(numbytes < btp->bt_meta_sectorsize)); |
459 | ASSERT(!(BBTOB(blkno) & (xfs_off_t)btp->bt_meta_sectormask)); | |
1da177e4 | 460 | |
10616b80 DC |
461 | /* |
462 | * Corrupted block numbers can get through to here, unfortunately, so we | |
463 | * have to check that the buffer falls within the filesystem bounds. | |
464 | */ | |
465 | eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks); | |
466 | if (blkno >= eofs) { | |
467 | /* | |
2451337d | 468 | * XXX (dgc): we should really be returning -EFSCORRUPTED here, |
10616b80 DC |
469 | * but none of the higher level infrastructure supports |
470 | * returning a specific error on buffer lookup failures. | |
471 | */ | |
472 | xfs_alert(btp->bt_mount, | |
473 | "%s: Block out of range: block 0x%llx, EOFS 0x%llx ", | |
474 | __func__, blkno, eofs); | |
7bc0dc27 | 475 | WARN_ON(1); |
10616b80 DC |
476 | return NULL; |
477 | } | |
478 | ||
74f75a0c DC |
479 | /* get tree root */ |
480 | pag = xfs_perag_get(btp->bt_mount, | |
e70b73f8 | 481 | xfs_daddr_to_agno(btp->bt_mount, blkno)); |
74f75a0c DC |
482 | |
483 | /* walk tree */ | |
484 | spin_lock(&pag->pag_buf_lock); | |
485 | rbp = &pag->pag_buf_tree.rb_node; | |
486 | parent = NULL; | |
487 | bp = NULL; | |
488 | while (*rbp) { | |
489 | parent = *rbp; | |
490 | bp = rb_entry(parent, struct xfs_buf, b_rbnode); | |
491 | ||
de1cbee4 | 492 | if (blkno < bp->b_bn) |
74f75a0c | 493 | rbp = &(*rbp)->rb_left; |
de1cbee4 | 494 | else if (blkno > bp->b_bn) |
74f75a0c DC |
495 | rbp = &(*rbp)->rb_right; |
496 | else { | |
497 | /* | |
de1cbee4 | 498 | * found a block number match. If the range doesn't |
74f75a0c DC |
499 | * match, the only way this is allowed is if the buffer |
500 | * in the cache is stale and the transaction that made | |
501 | * it stale has not yet committed. i.e. we are | |
502 | * reallocating a busy extent. Skip this buffer and | |
503 | * continue searching to the right for an exact match. | |
504 | */ | |
4e94b71b | 505 | if (bp->b_length != numblks) { |
74f75a0c DC |
506 | ASSERT(bp->b_flags & XBF_STALE); |
507 | rbp = &(*rbp)->rb_right; | |
508 | continue; | |
509 | } | |
ce8e922c | 510 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
511 | goto found; |
512 | } | |
513 | } | |
514 | ||
515 | /* No match found */ | |
ce8e922c | 516 | if (new_bp) { |
74f75a0c DC |
517 | rb_link_node(&new_bp->b_rbnode, parent, rbp); |
518 | rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree); | |
519 | /* the buffer keeps the perag reference until it is freed */ | |
520 | new_bp->b_pag = pag; | |
521 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 522 | } else { |
ce8e922c | 523 | XFS_STATS_INC(xb_miss_locked); |
74f75a0c DC |
524 | spin_unlock(&pag->pag_buf_lock); |
525 | xfs_perag_put(pag); | |
1da177e4 | 526 | } |
ce8e922c | 527 | return new_bp; |
1da177e4 LT |
528 | |
529 | found: | |
74f75a0c DC |
530 | spin_unlock(&pag->pag_buf_lock); |
531 | xfs_perag_put(pag); | |
1da177e4 | 532 | |
0c842ad4 CH |
533 | if (!xfs_buf_trylock(bp)) { |
534 | if (flags & XBF_TRYLOCK) { | |
ce8e922c NS |
535 | xfs_buf_rele(bp); |
536 | XFS_STATS_INC(xb_busy_locked); | |
537 | return NULL; | |
1da177e4 | 538 | } |
0c842ad4 CH |
539 | xfs_buf_lock(bp); |
540 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
541 | } |
542 | ||
0e6e847f DC |
543 | /* |
544 | * if the buffer is stale, clear all the external state associated with | |
545 | * it. We need to keep flags such as how we allocated the buffer memory | |
546 | * intact here. | |
547 | */ | |
ce8e922c NS |
548 | if (bp->b_flags & XBF_STALE) { |
549 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
cfb02852 | 550 | ASSERT(bp->b_iodone == NULL); |
611c9946 | 551 | bp->b_flags &= _XBF_KMEM | _XBF_PAGES; |
1813dd64 | 552 | bp->b_ops = NULL; |
2f926587 | 553 | } |
0b1b213f CH |
554 | |
555 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ce8e922c NS |
556 | XFS_STATS_INC(xb_get_locked); |
557 | return bp; | |
1da177e4 LT |
558 | } |
559 | ||
560 | /* | |
3815832a DC |
561 | * Assembles a buffer covering the specified range. The code is optimised for |
562 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
563 | * more hits than misses. | |
1da177e4 | 564 | */ |
3815832a | 565 | struct xfs_buf * |
6dde2707 DC |
566 | xfs_buf_get_map( |
567 | struct xfs_buftarg *target, | |
568 | struct xfs_buf_map *map, | |
569 | int nmaps, | |
ce8e922c | 570 | xfs_buf_flags_t flags) |
1da177e4 | 571 | { |
3815832a DC |
572 | struct xfs_buf *bp; |
573 | struct xfs_buf *new_bp; | |
0e6e847f | 574 | int error = 0; |
1da177e4 | 575 | |
6dde2707 | 576 | bp = _xfs_buf_find(target, map, nmaps, flags, NULL); |
3815832a DC |
577 | if (likely(bp)) |
578 | goto found; | |
579 | ||
6dde2707 | 580 | new_bp = _xfs_buf_alloc(target, map, nmaps, flags); |
ce8e922c | 581 | if (unlikely(!new_bp)) |
1da177e4 LT |
582 | return NULL; |
583 | ||
fe2429b0 DC |
584 | error = xfs_buf_allocate_memory(new_bp, flags); |
585 | if (error) { | |
3e85c868 | 586 | xfs_buf_free(new_bp); |
fe2429b0 DC |
587 | return NULL; |
588 | } | |
589 | ||
6dde2707 | 590 | bp = _xfs_buf_find(target, map, nmaps, flags, new_bp); |
3815832a | 591 | if (!bp) { |
fe2429b0 | 592 | xfs_buf_free(new_bp); |
3815832a DC |
593 | return NULL; |
594 | } | |
595 | ||
fe2429b0 DC |
596 | if (bp != new_bp) |
597 | xfs_buf_free(new_bp); | |
1da177e4 | 598 | |
3815832a | 599 | found: |
611c9946 | 600 | if (!bp->b_addr) { |
ce8e922c | 601 | error = _xfs_buf_map_pages(bp, flags); |
1da177e4 | 602 | if (unlikely(error)) { |
4f10700a | 603 | xfs_warn(target->bt_mount, |
08e96e1a | 604 | "%s: failed to map pagesn", __func__); |
a8acad70 DC |
605 | xfs_buf_relse(bp); |
606 | return NULL; | |
1da177e4 LT |
607 | } |
608 | } | |
609 | ||
ce8e922c | 610 | XFS_STATS_INC(xb_get); |
0b1b213f | 611 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 612 | return bp; |
1da177e4 LT |
613 | } |
614 | ||
5d765b97 CH |
615 | STATIC int |
616 | _xfs_buf_read( | |
617 | xfs_buf_t *bp, | |
618 | xfs_buf_flags_t flags) | |
619 | { | |
43ff2122 | 620 | ASSERT(!(flags & XBF_WRITE)); |
f4b42421 | 621 | ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL); |
5d765b97 | 622 | |
43ff2122 | 623 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD); |
1d5ae5df | 624 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 625 | |
595bff75 DC |
626 | if (flags & XBF_ASYNC) { |
627 | xfs_buf_submit(bp); | |
0e95f19a | 628 | return 0; |
595bff75 DC |
629 | } |
630 | return xfs_buf_submit_wait(bp); | |
5d765b97 CH |
631 | } |
632 | ||
1da177e4 | 633 | xfs_buf_t * |
6dde2707 DC |
634 | xfs_buf_read_map( |
635 | struct xfs_buftarg *target, | |
636 | struct xfs_buf_map *map, | |
637 | int nmaps, | |
c3f8fc73 | 638 | xfs_buf_flags_t flags, |
1813dd64 | 639 | const struct xfs_buf_ops *ops) |
1da177e4 | 640 | { |
6dde2707 | 641 | struct xfs_buf *bp; |
ce8e922c NS |
642 | |
643 | flags |= XBF_READ; | |
644 | ||
6dde2707 | 645 | bp = xfs_buf_get_map(target, map, nmaps, flags); |
ce8e922c | 646 | if (bp) { |
0b1b213f CH |
647 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
648 | ||
ce8e922c | 649 | if (!XFS_BUF_ISDONE(bp)) { |
ce8e922c | 650 | XFS_STATS_INC(xb_get_read); |
1813dd64 | 651 | bp->b_ops = ops; |
5d765b97 | 652 | _xfs_buf_read(bp, flags); |
ce8e922c | 653 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
654 | /* |
655 | * Read ahead call which is already satisfied, | |
656 | * drop the buffer | |
657 | */ | |
a8acad70 DC |
658 | xfs_buf_relse(bp); |
659 | return NULL; | |
1da177e4 | 660 | } else { |
1da177e4 | 661 | /* We do not want read in the flags */ |
ce8e922c | 662 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
663 | } |
664 | } | |
665 | ||
ce8e922c | 666 | return bp; |
1da177e4 LT |
667 | } |
668 | ||
1da177e4 | 669 | /* |
ce8e922c NS |
670 | * If we are not low on memory then do the readahead in a deadlock |
671 | * safe manner. | |
1da177e4 LT |
672 | */ |
673 | void | |
6dde2707 DC |
674 | xfs_buf_readahead_map( |
675 | struct xfs_buftarg *target, | |
676 | struct xfs_buf_map *map, | |
c3f8fc73 | 677 | int nmaps, |
1813dd64 | 678 | const struct xfs_buf_ops *ops) |
1da177e4 | 679 | { |
0e6e847f | 680 | if (bdi_read_congested(target->bt_bdi)) |
1da177e4 LT |
681 | return; |
682 | ||
6dde2707 | 683 | xfs_buf_read_map(target, map, nmaps, |
1813dd64 | 684 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD, ops); |
1da177e4 LT |
685 | } |
686 | ||
5adc94c2 DC |
687 | /* |
688 | * Read an uncached buffer from disk. Allocates and returns a locked | |
689 | * buffer containing the disk contents or nothing. | |
690 | */ | |
691 | struct xfs_buf * | |
692 | xfs_buf_read_uncached( | |
5adc94c2 DC |
693 | struct xfs_buftarg *target, |
694 | xfs_daddr_t daddr, | |
e70b73f8 | 695 | size_t numblks, |
c3f8fc73 | 696 | int flags, |
1813dd64 | 697 | const struct xfs_buf_ops *ops) |
5adc94c2 | 698 | { |
eab4e633 | 699 | struct xfs_buf *bp; |
5adc94c2 | 700 | |
e70b73f8 | 701 | bp = xfs_buf_get_uncached(target, numblks, flags); |
5adc94c2 DC |
702 | if (!bp) |
703 | return NULL; | |
704 | ||
705 | /* set up the buffer for a read IO */ | |
3e85c868 DC |
706 | ASSERT(bp->b_map_count == 1); |
707 | bp->b_bn = daddr; | |
708 | bp->b_maps[0].bm_bn = daddr; | |
cbb7baab | 709 | bp->b_flags |= XBF_READ; |
1813dd64 | 710 | bp->b_ops = ops; |
5adc94c2 | 711 | |
595bff75 | 712 | xfs_buf_submit_wait(bp); |
5adc94c2 | 713 | return bp; |
1da177e4 LT |
714 | } |
715 | ||
44396476 DC |
716 | /* |
717 | * Return a buffer allocated as an empty buffer and associated to external | |
718 | * memory via xfs_buf_associate_memory() back to it's empty state. | |
719 | */ | |
720 | void | |
721 | xfs_buf_set_empty( | |
722 | struct xfs_buf *bp, | |
e70b73f8 | 723 | size_t numblks) |
44396476 DC |
724 | { |
725 | if (bp->b_pages) | |
726 | _xfs_buf_free_pages(bp); | |
727 | ||
728 | bp->b_pages = NULL; | |
729 | bp->b_page_count = 0; | |
730 | bp->b_addr = NULL; | |
4e94b71b | 731 | bp->b_length = numblks; |
aa0e8833 | 732 | bp->b_io_length = numblks; |
3e85c868 DC |
733 | |
734 | ASSERT(bp->b_map_count == 1); | |
44396476 | 735 | bp->b_bn = XFS_BUF_DADDR_NULL; |
3e85c868 DC |
736 | bp->b_maps[0].bm_bn = XFS_BUF_DADDR_NULL; |
737 | bp->b_maps[0].bm_len = bp->b_length; | |
44396476 DC |
738 | } |
739 | ||
1da177e4 LT |
740 | static inline struct page * |
741 | mem_to_page( | |
742 | void *addr) | |
743 | { | |
9e2779fa | 744 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
745 | return virt_to_page(addr); |
746 | } else { | |
747 | return vmalloc_to_page(addr); | |
748 | } | |
749 | } | |
750 | ||
751 | int | |
ce8e922c NS |
752 | xfs_buf_associate_memory( |
753 | xfs_buf_t *bp, | |
1da177e4 LT |
754 | void *mem, |
755 | size_t len) | |
756 | { | |
757 | int rval; | |
758 | int i = 0; | |
d1afb678 LM |
759 | unsigned long pageaddr; |
760 | unsigned long offset; | |
761 | size_t buflen; | |
1da177e4 LT |
762 | int page_count; |
763 | ||
0e6e847f | 764 | pageaddr = (unsigned long)mem & PAGE_MASK; |
d1afb678 | 765 | offset = (unsigned long)mem - pageaddr; |
0e6e847f DC |
766 | buflen = PAGE_ALIGN(len + offset); |
767 | page_count = buflen >> PAGE_SHIFT; | |
1da177e4 LT |
768 | |
769 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
770 | if (bp->b_pages) |
771 | _xfs_buf_free_pages(bp); | |
1da177e4 | 772 | |
ce8e922c NS |
773 | bp->b_pages = NULL; |
774 | bp->b_addr = mem; | |
1da177e4 | 775 | |
87937bf8 | 776 | rval = _xfs_buf_get_pages(bp, page_count); |
1da177e4 LT |
777 | if (rval) |
778 | return rval; | |
779 | ||
ce8e922c | 780 | bp->b_offset = offset; |
d1afb678 LM |
781 | |
782 | for (i = 0; i < bp->b_page_count; i++) { | |
783 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
0e6e847f | 784 | pageaddr += PAGE_SIZE; |
1da177e4 | 785 | } |
1da177e4 | 786 | |
aa0e8833 | 787 | bp->b_io_length = BTOBB(len); |
4e94b71b | 788 | bp->b_length = BTOBB(buflen); |
1da177e4 LT |
789 | |
790 | return 0; | |
791 | } | |
792 | ||
793 | xfs_buf_t * | |
686865f7 DC |
794 | xfs_buf_get_uncached( |
795 | struct xfs_buftarg *target, | |
e70b73f8 | 796 | size_t numblks, |
686865f7 | 797 | int flags) |
1da177e4 | 798 | { |
e70b73f8 | 799 | unsigned long page_count; |
1fa40b01 | 800 | int error, i; |
3e85c868 DC |
801 | struct xfs_buf *bp; |
802 | DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks); | |
1da177e4 | 803 | |
3e85c868 | 804 | bp = _xfs_buf_alloc(target, &map, 1, 0); |
1da177e4 LT |
805 | if (unlikely(bp == NULL)) |
806 | goto fail; | |
1da177e4 | 807 | |
e70b73f8 | 808 | page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT; |
87937bf8 | 809 | error = _xfs_buf_get_pages(bp, page_count); |
1fa40b01 | 810 | if (error) |
1da177e4 LT |
811 | goto fail_free_buf; |
812 | ||
1fa40b01 | 813 | for (i = 0; i < page_count; i++) { |
686865f7 | 814 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
815 | if (!bp->b_pages[i]) |
816 | goto fail_free_mem; | |
1da177e4 | 817 | } |
1fa40b01 | 818 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 819 | |
611c9946 | 820 | error = _xfs_buf_map_pages(bp, 0); |
1fa40b01 | 821 | if (unlikely(error)) { |
4f10700a | 822 | xfs_warn(target->bt_mount, |
08e96e1a | 823 | "%s: failed to map pages", __func__); |
1da177e4 | 824 | goto fail_free_mem; |
1fa40b01 | 825 | } |
1da177e4 | 826 | |
686865f7 | 827 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 828 | return bp; |
1fa40b01 | 829 | |
1da177e4 | 830 | fail_free_mem: |
1fa40b01 CH |
831 | while (--i >= 0) |
832 | __free_page(bp->b_pages[i]); | |
ca165b88 | 833 | _xfs_buf_free_pages(bp); |
1da177e4 | 834 | fail_free_buf: |
3e85c868 | 835 | xfs_buf_free_maps(bp); |
4347b9d7 | 836 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
837 | fail: |
838 | return NULL; | |
839 | } | |
840 | ||
841 | /* | |
1da177e4 LT |
842 | * Increment reference count on buffer, to hold the buffer concurrently |
843 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
844 | * Must hold the buffer already to call this function. |
845 | */ | |
846 | void | |
ce8e922c NS |
847 | xfs_buf_hold( |
848 | xfs_buf_t *bp) | |
1da177e4 | 849 | { |
0b1b213f | 850 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 851 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
852 | } |
853 | ||
854 | /* | |
ce8e922c NS |
855 | * Releases a hold on the specified buffer. If the |
856 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
857 | */ |
858 | void | |
ce8e922c NS |
859 | xfs_buf_rele( |
860 | xfs_buf_t *bp) | |
1da177e4 | 861 | { |
74f75a0c | 862 | struct xfs_perag *pag = bp->b_pag; |
1da177e4 | 863 | |
0b1b213f | 864 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 865 | |
74f75a0c | 866 | if (!pag) { |
430cbeb8 | 867 | ASSERT(list_empty(&bp->b_lru)); |
74f75a0c | 868 | ASSERT(RB_EMPTY_NODE(&bp->b_rbnode)); |
fad3aa1e NS |
869 | if (atomic_dec_and_test(&bp->b_hold)) |
870 | xfs_buf_free(bp); | |
871 | return; | |
872 | } | |
873 | ||
74f75a0c | 874 | ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode)); |
430cbeb8 | 875 | |
3790689f | 876 | ASSERT(atomic_read(&bp->b_hold) > 0); |
74f75a0c | 877 | if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) { |
a4082357 DC |
878 | spin_lock(&bp->b_lock); |
879 | if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) { | |
880 | /* | |
881 | * If the buffer is added to the LRU take a new | |
882 | * reference to the buffer for the LRU and clear the | |
883 | * (now stale) dispose list state flag | |
884 | */ | |
885 | if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) { | |
886 | bp->b_state &= ~XFS_BSTATE_DISPOSE; | |
887 | atomic_inc(&bp->b_hold); | |
888 | } | |
889 | spin_unlock(&bp->b_lock); | |
430cbeb8 | 890 | spin_unlock(&pag->pag_buf_lock); |
1da177e4 | 891 | } else { |
a4082357 DC |
892 | /* |
893 | * most of the time buffers will already be removed from | |
894 | * the LRU, so optimise that case by checking for the | |
895 | * XFS_BSTATE_DISPOSE flag indicating the last list the | |
896 | * buffer was on was the disposal list | |
897 | */ | |
898 | if (!(bp->b_state & XFS_BSTATE_DISPOSE)) { | |
899 | list_lru_del(&bp->b_target->bt_lru, &bp->b_lru); | |
900 | } else { | |
901 | ASSERT(list_empty(&bp->b_lru)); | |
902 | } | |
903 | spin_unlock(&bp->b_lock); | |
904 | ||
43ff2122 | 905 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
74f75a0c DC |
906 | rb_erase(&bp->b_rbnode, &pag->pag_buf_tree); |
907 | spin_unlock(&pag->pag_buf_lock); | |
908 | xfs_perag_put(pag); | |
ce8e922c | 909 | xfs_buf_free(bp); |
1da177e4 LT |
910 | } |
911 | } | |
912 | } | |
913 | ||
914 | ||
915 | /* | |
0e6e847f | 916 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
917 | * |
918 | * If we come across a stale, pinned, locked buffer, we know that we are | |
919 | * being asked to lock a buffer that has been reallocated. Because it is | |
920 | * pinned, we know that the log has not been pushed to disk and hence it | |
921 | * will still be locked. Rather than continuing to have trylock attempts | |
922 | * fail until someone else pushes the log, push it ourselves before | |
923 | * returning. This means that the xfsaild will not get stuck trying | |
924 | * to push on stale inode buffers. | |
1da177e4 LT |
925 | */ |
926 | int | |
0c842ad4 CH |
927 | xfs_buf_trylock( |
928 | struct xfs_buf *bp) | |
1da177e4 LT |
929 | { |
930 | int locked; | |
931 | ||
ce8e922c | 932 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 933 | if (locked) |
ce8e922c | 934 | XB_SET_OWNER(bp); |
0b1b213f | 935 | |
0c842ad4 CH |
936 | trace_xfs_buf_trylock(bp, _RET_IP_); |
937 | return locked; | |
1da177e4 | 938 | } |
1da177e4 LT |
939 | |
940 | /* | |
0e6e847f | 941 | * Lock a buffer object. |
ed3b4d6c DC |
942 | * |
943 | * If we come across a stale, pinned, locked buffer, we know that we | |
944 | * are being asked to lock a buffer that has been reallocated. Because | |
945 | * it is pinned, we know that the log has not been pushed to disk and | |
946 | * hence it will still be locked. Rather than sleeping until someone | |
947 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 948 | */ |
ce8e922c NS |
949 | void |
950 | xfs_buf_lock( | |
0c842ad4 | 951 | struct xfs_buf *bp) |
1da177e4 | 952 | { |
0b1b213f CH |
953 | trace_xfs_buf_lock(bp, _RET_IP_); |
954 | ||
ed3b4d6c | 955 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 956 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
957 | down(&bp->b_sema); |
958 | XB_SET_OWNER(bp); | |
0b1b213f CH |
959 | |
960 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
961 | } |
962 | ||
1da177e4 | 963 | void |
ce8e922c | 964 | xfs_buf_unlock( |
0c842ad4 | 965 | struct xfs_buf *bp) |
1da177e4 | 966 | { |
ce8e922c NS |
967 | XB_CLEAR_OWNER(bp); |
968 | up(&bp->b_sema); | |
0b1b213f CH |
969 | |
970 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
971 | } |
972 | ||
ce8e922c NS |
973 | STATIC void |
974 | xfs_buf_wait_unpin( | |
975 | xfs_buf_t *bp) | |
1da177e4 LT |
976 | { |
977 | DECLARE_WAITQUEUE (wait, current); | |
978 | ||
ce8e922c | 979 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
980 | return; |
981 | ||
ce8e922c | 982 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
983 | for (;;) { |
984 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 985 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 986 | break; |
7eaceacc | 987 | io_schedule(); |
1da177e4 | 988 | } |
ce8e922c | 989 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
990 | set_current_state(TASK_RUNNING); |
991 | } | |
992 | ||
993 | /* | |
994 | * Buffer Utility Routines | |
995 | */ | |
996 | ||
e8aaba9a DC |
997 | void |
998 | xfs_buf_ioend( | |
999 | struct xfs_buf *bp) | |
1da177e4 | 1000 | { |
e8aaba9a DC |
1001 | bool read = bp->b_flags & XBF_READ; |
1002 | ||
1003 | trace_xfs_buf_iodone(bp, _RET_IP_); | |
1813dd64 DC |
1004 | |
1005 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); | |
d5929de8 | 1006 | |
61be9c52 DC |
1007 | /* |
1008 | * Pull in IO completion errors now. We are guaranteed to be running | |
1009 | * single threaded, so we don't need the lock to read b_io_error. | |
1010 | */ | |
1011 | if (!bp->b_error && bp->b_io_error) | |
1012 | xfs_buf_ioerror(bp, bp->b_io_error); | |
1013 | ||
e8aaba9a DC |
1014 | /* Only validate buffers that were read without errors */ |
1015 | if (read && !bp->b_error && bp->b_ops) { | |
1016 | ASSERT(!bp->b_iodone); | |
1813dd64 | 1017 | bp->b_ops->verify_read(bp); |
e8aaba9a DC |
1018 | } |
1019 | ||
1020 | if (!bp->b_error) | |
1021 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1022 | |
80f6c29d | 1023 | if (bp->b_iodone) |
ce8e922c NS |
1024 | (*(bp->b_iodone))(bp); |
1025 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 | 1026 | xfs_buf_relse(bp); |
595bff75 | 1027 | else |
1813dd64 | 1028 | complete(&bp->b_iowait); |
1da177e4 LT |
1029 | } |
1030 | ||
e8aaba9a DC |
1031 | static void |
1032 | xfs_buf_ioend_work( | |
1033 | struct work_struct *work) | |
1da177e4 | 1034 | { |
e8aaba9a DC |
1035 | struct xfs_buf *bp = |
1036 | container_of(work, xfs_buf_t, b_iodone_work); | |
0b1b213f | 1037 | |
e8aaba9a DC |
1038 | xfs_buf_ioend(bp); |
1039 | } | |
1da177e4 | 1040 | |
e8aaba9a DC |
1041 | void |
1042 | xfs_buf_ioend_async( | |
1043 | struct xfs_buf *bp) | |
1044 | { | |
1045 | INIT_WORK(&bp->b_iodone_work, xfs_buf_ioend_work); | |
1046 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); | |
1da177e4 LT |
1047 | } |
1048 | ||
1da177e4 | 1049 | void |
ce8e922c NS |
1050 | xfs_buf_ioerror( |
1051 | xfs_buf_t *bp, | |
1052 | int error) | |
1da177e4 | 1053 | { |
2451337d DC |
1054 | ASSERT(error <= 0 && error >= -1000); |
1055 | bp->b_error = error; | |
0b1b213f | 1056 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1057 | } |
1058 | ||
901796af CH |
1059 | void |
1060 | xfs_buf_ioerror_alert( | |
1061 | struct xfs_buf *bp, | |
1062 | const char *func) | |
1063 | { | |
1064 | xfs_alert(bp->b_target->bt_mount, | |
aa0e8833 | 1065 | "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d", |
2451337d | 1066 | (__uint64_t)XFS_BUF_ADDR(bp), func, -bp->b_error, bp->b_length); |
901796af CH |
1067 | } |
1068 | ||
a2dcf5df CH |
1069 | int |
1070 | xfs_bwrite( | |
1071 | struct xfs_buf *bp) | |
1072 | { | |
1073 | int error; | |
1074 | ||
1075 | ASSERT(xfs_buf_islocked(bp)); | |
1076 | ||
1077 | bp->b_flags |= XBF_WRITE; | |
27187754 DC |
1078 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q | |
1079 | XBF_WRITE_FAIL | XBF_DONE); | |
a2dcf5df | 1080 | |
595bff75 | 1081 | error = xfs_buf_submit_wait(bp); |
a2dcf5df CH |
1082 | if (error) { |
1083 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1084 | SHUTDOWN_META_IO_ERROR); | |
1085 | } | |
1086 | return error; | |
1087 | } | |
1088 | ||
782e3b3b | 1089 | STATIC void |
ce8e922c | 1090 | xfs_buf_bio_end_io( |
1da177e4 | 1091 | struct bio *bio, |
1da177e4 LT |
1092 | int error) |
1093 | { | |
ce8e922c | 1094 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1da177e4 | 1095 | |
37eb17e6 DC |
1096 | /* |
1097 | * don't overwrite existing errors - otherwise we can lose errors on | |
1098 | * buffers that require multiple bios to complete. | |
1099 | */ | |
61be9c52 DC |
1100 | if (error) { |
1101 | spin_lock(&bp->b_lock); | |
1102 | if (!bp->b_io_error) | |
1103 | bp->b_io_error = error; | |
1104 | spin_unlock(&bp->b_lock); | |
1105 | } | |
1da177e4 | 1106 | |
37eb17e6 | 1107 | if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
73c77e2c JB |
1108 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); |
1109 | ||
e8aaba9a DC |
1110 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
1111 | xfs_buf_ioend_async(bp); | |
1da177e4 | 1112 | bio_put(bio); |
1da177e4 LT |
1113 | } |
1114 | ||
3e85c868 DC |
1115 | static void |
1116 | xfs_buf_ioapply_map( | |
1117 | struct xfs_buf *bp, | |
1118 | int map, | |
1119 | int *buf_offset, | |
1120 | int *count, | |
1121 | int rw) | |
1da177e4 | 1122 | { |
3e85c868 DC |
1123 | int page_index; |
1124 | int total_nr_pages = bp->b_page_count; | |
1125 | int nr_pages; | |
1126 | struct bio *bio; | |
1127 | sector_t sector = bp->b_maps[map].bm_bn; | |
1128 | int size; | |
1129 | int offset; | |
1da177e4 | 1130 | |
ce8e922c | 1131 | total_nr_pages = bp->b_page_count; |
1da177e4 | 1132 | |
3e85c868 DC |
1133 | /* skip the pages in the buffer before the start offset */ |
1134 | page_index = 0; | |
1135 | offset = *buf_offset; | |
1136 | while (offset >= PAGE_SIZE) { | |
1137 | page_index++; | |
1138 | offset -= PAGE_SIZE; | |
f538d4da CH |
1139 | } |
1140 | ||
3e85c868 DC |
1141 | /* |
1142 | * Limit the IO size to the length of the current vector, and update the | |
1143 | * remaining IO count for the next time around. | |
1144 | */ | |
1145 | size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count); | |
1146 | *count -= size; | |
1147 | *buf_offset += size; | |
34951f5c | 1148 | |
1da177e4 | 1149 | next_chunk: |
ce8e922c | 1150 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1151 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1152 | if (nr_pages > total_nr_pages) | |
1153 | nr_pages = total_nr_pages; | |
1154 | ||
1155 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1156 | bio->bi_bdev = bp->b_target->bt_bdev; |
4f024f37 | 1157 | bio->bi_iter.bi_sector = sector; |
ce8e922c NS |
1158 | bio->bi_end_io = xfs_buf_bio_end_io; |
1159 | bio->bi_private = bp; | |
1da177e4 | 1160 | |
0e6e847f | 1161 | |
3e85c868 | 1162 | for (; size && nr_pages; nr_pages--, page_index++) { |
0e6e847f | 1163 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1164 | |
1165 | if (nbytes > size) | |
1166 | nbytes = size; | |
1167 | ||
3e85c868 DC |
1168 | rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes, |
1169 | offset); | |
ce8e922c | 1170 | if (rbytes < nbytes) |
1da177e4 LT |
1171 | break; |
1172 | ||
1173 | offset = 0; | |
aa0e8833 | 1174 | sector += BTOBB(nbytes); |
1da177e4 LT |
1175 | size -= nbytes; |
1176 | total_nr_pages--; | |
1177 | } | |
1178 | ||
4f024f37 | 1179 | if (likely(bio->bi_iter.bi_size)) { |
73c77e2c JB |
1180 | if (xfs_buf_is_vmapped(bp)) { |
1181 | flush_kernel_vmap_range(bp->b_addr, | |
1182 | xfs_buf_vmap_len(bp)); | |
1183 | } | |
1da177e4 LT |
1184 | submit_bio(rw, bio); |
1185 | if (size) | |
1186 | goto next_chunk; | |
1187 | } else { | |
37eb17e6 DC |
1188 | /* |
1189 | * This is guaranteed not to be the last io reference count | |
595bff75 | 1190 | * because the caller (xfs_buf_submit) holds a count itself. |
37eb17e6 DC |
1191 | */ |
1192 | atomic_dec(&bp->b_io_remaining); | |
2451337d | 1193 | xfs_buf_ioerror(bp, -EIO); |
ec53d1db | 1194 | bio_put(bio); |
1da177e4 | 1195 | } |
3e85c868 DC |
1196 | |
1197 | } | |
1198 | ||
1199 | STATIC void | |
1200 | _xfs_buf_ioapply( | |
1201 | struct xfs_buf *bp) | |
1202 | { | |
1203 | struct blk_plug plug; | |
1204 | int rw; | |
1205 | int offset; | |
1206 | int size; | |
1207 | int i; | |
1208 | ||
c163f9a1 DC |
1209 | /* |
1210 | * Make sure we capture only current IO errors rather than stale errors | |
1211 | * left over from previous use of the buffer (e.g. failed readahead). | |
1212 | */ | |
1213 | bp->b_error = 0; | |
1214 | ||
3e85c868 DC |
1215 | if (bp->b_flags & XBF_WRITE) { |
1216 | if (bp->b_flags & XBF_SYNCIO) | |
1217 | rw = WRITE_SYNC; | |
1218 | else | |
1219 | rw = WRITE; | |
1220 | if (bp->b_flags & XBF_FUA) | |
1221 | rw |= REQ_FUA; | |
1222 | if (bp->b_flags & XBF_FLUSH) | |
1223 | rw |= REQ_FLUSH; | |
1813dd64 DC |
1224 | |
1225 | /* | |
1226 | * Run the write verifier callback function if it exists. If | |
1227 | * this function fails it will mark the buffer with an error and | |
1228 | * the IO should not be dispatched. | |
1229 | */ | |
1230 | if (bp->b_ops) { | |
1231 | bp->b_ops->verify_write(bp); | |
1232 | if (bp->b_error) { | |
1233 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1234 | SHUTDOWN_CORRUPT_INCORE); | |
1235 | return; | |
1236 | } | |
400b9d88 DC |
1237 | } else if (bp->b_bn != XFS_BUF_DADDR_NULL) { |
1238 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
1239 | ||
1240 | /* | |
1241 | * non-crc filesystems don't attach verifiers during | |
1242 | * log recovery, so don't warn for such filesystems. | |
1243 | */ | |
1244 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
1245 | xfs_warn(mp, | |
1246 | "%s: no ops on block 0x%llx/0x%x", | |
1247 | __func__, bp->b_bn, bp->b_length); | |
1248 | xfs_hex_dump(bp->b_addr, 64); | |
1249 | dump_stack(); | |
1250 | } | |
1813dd64 | 1251 | } |
3e85c868 DC |
1252 | } else if (bp->b_flags & XBF_READ_AHEAD) { |
1253 | rw = READA; | |
1254 | } else { | |
1255 | rw = READ; | |
1256 | } | |
1257 | ||
1258 | /* we only use the buffer cache for meta-data */ | |
1259 | rw |= REQ_META; | |
1260 | ||
1261 | /* | |
1262 | * Walk all the vectors issuing IO on them. Set up the initial offset | |
1263 | * into the buffer and the desired IO size before we start - | |
1264 | * _xfs_buf_ioapply_vec() will modify them appropriately for each | |
1265 | * subsequent call. | |
1266 | */ | |
1267 | offset = bp->b_offset; | |
1268 | size = BBTOB(bp->b_io_length); | |
1269 | blk_start_plug(&plug); | |
1270 | for (i = 0; i < bp->b_map_count; i++) { | |
1271 | xfs_buf_ioapply_map(bp, i, &offset, &size, rw); | |
1272 | if (bp->b_error) | |
1273 | break; | |
1274 | if (size <= 0) | |
1275 | break; /* all done */ | |
1276 | } | |
1277 | blk_finish_plug(&plug); | |
1da177e4 LT |
1278 | } |
1279 | ||
595bff75 DC |
1280 | /* |
1281 | * Asynchronous IO submission path. This transfers the buffer lock ownership and | |
1282 | * the current reference to the IO. It is not safe to reference the buffer after | |
1283 | * a call to this function unless the caller holds an additional reference | |
1284 | * itself. | |
1285 | */ | |
0e95f19a | 1286 | void |
595bff75 DC |
1287 | xfs_buf_submit( |
1288 | struct xfs_buf *bp) | |
1da177e4 | 1289 | { |
595bff75 | 1290 | trace_xfs_buf_submit(bp, _RET_IP_); |
1da177e4 | 1291 | |
43ff2122 | 1292 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
595bff75 DC |
1293 | ASSERT(bp->b_flags & XBF_ASYNC); |
1294 | ||
1295 | /* on shutdown we stale and complete the buffer immediately */ | |
1296 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { | |
1297 | xfs_buf_ioerror(bp, -EIO); | |
1298 | bp->b_flags &= ~XBF_DONE; | |
1299 | xfs_buf_stale(bp); | |
1300 | xfs_buf_ioend(bp); | |
1301 | return; | |
1302 | } | |
1da177e4 | 1303 | |
375ec69d | 1304 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1305 | xfs_buf_wait_unpin(bp); |
e11bb805 | 1306 | |
61be9c52 DC |
1307 | /* clear the internal error state to avoid spurious errors */ |
1308 | bp->b_io_error = 0; | |
1309 | ||
e11bb805 | 1310 | /* |
595bff75 DC |
1311 | * The caller's reference is released during I/O completion. |
1312 | * This occurs some time after the last b_io_remaining reference is | |
1313 | * released, so after we drop our Io reference we have to have some | |
1314 | * other reference to ensure the buffer doesn't go away from underneath | |
1315 | * us. Take a direct reference to ensure we have safe access to the | |
1316 | * buffer until we are finished with it. | |
e11bb805 | 1317 | */ |
ce8e922c | 1318 | xfs_buf_hold(bp); |
1da177e4 | 1319 | |
8d6c1210 | 1320 | /* |
e11bb805 DC |
1321 | * Set the count to 1 initially, this will stop an I/O completion |
1322 | * callout which happens before we have started all the I/O from calling | |
1323 | * xfs_buf_ioend too early. | |
1da177e4 | 1324 | */ |
ce8e922c NS |
1325 | atomic_set(&bp->b_io_remaining, 1); |
1326 | _xfs_buf_ioapply(bp); | |
e11bb805 | 1327 | |
8d6c1210 | 1328 | /* |
595bff75 DC |
1329 | * If _xfs_buf_ioapply failed, we can get back here with only the IO |
1330 | * reference we took above. If we drop it to zero, run completion so | |
1331 | * that we don't return to the caller with completion still pending. | |
8d6c1210 | 1332 | */ |
e8aaba9a | 1333 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
595bff75 | 1334 | if (bp->b_error) |
e8aaba9a DC |
1335 | xfs_buf_ioend(bp); |
1336 | else | |
1337 | xfs_buf_ioend_async(bp); | |
1338 | } | |
1da177e4 | 1339 | |
ce8e922c | 1340 | xfs_buf_rele(bp); |
595bff75 | 1341 | /* Note: it is not safe to reference bp now we've dropped our ref */ |
1da177e4 LT |
1342 | } |
1343 | ||
1344 | /* | |
595bff75 | 1345 | * Synchronous buffer IO submission path, read or write. |
1da177e4 LT |
1346 | */ |
1347 | int | |
595bff75 DC |
1348 | xfs_buf_submit_wait( |
1349 | struct xfs_buf *bp) | |
1da177e4 | 1350 | { |
595bff75 | 1351 | int error; |
0b1b213f | 1352 | |
595bff75 DC |
1353 | trace_xfs_buf_submit_wait(bp, _RET_IP_); |
1354 | ||
1355 | ASSERT(!(bp->b_flags & (_XBF_DELWRI_Q | XBF_ASYNC))); | |
0b1b213f | 1356 | |
595bff75 DC |
1357 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
1358 | xfs_buf_ioerror(bp, -EIO); | |
1359 | xfs_buf_stale(bp); | |
1360 | bp->b_flags &= ~XBF_DONE; | |
1361 | return -EIO; | |
1362 | } | |
1363 | ||
1364 | if (bp->b_flags & XBF_WRITE) | |
1365 | xfs_buf_wait_unpin(bp); | |
1366 | ||
1367 | /* clear the internal error state to avoid spurious errors */ | |
1368 | bp->b_io_error = 0; | |
1369 | ||
1370 | /* | |
1371 | * For synchronous IO, the IO does not inherit the submitters reference | |
1372 | * count, nor the buffer lock. Hence we cannot release the reference we | |
1373 | * are about to take until we've waited for all IO completion to occur, | |
1374 | * including any xfs_buf_ioend_async() work that may be pending. | |
1375 | */ | |
1376 | xfs_buf_hold(bp); | |
1377 | ||
1378 | /* | |
1379 | * Set the count to 1 initially, this will stop an I/O completion | |
1380 | * callout which happens before we have started all the I/O from calling | |
1381 | * xfs_buf_ioend too early. | |
1382 | */ | |
1383 | atomic_set(&bp->b_io_remaining, 1); | |
1384 | _xfs_buf_ioapply(bp); | |
1385 | ||
1386 | /* | |
1387 | * make sure we run completion synchronously if it raced with us and is | |
1388 | * already complete. | |
1389 | */ | |
1390 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) | |
1391 | xfs_buf_ioend(bp); | |
1392 | ||
1393 | /* wait for completion before gathering the error from the buffer */ | |
1394 | trace_xfs_buf_iowait(bp, _RET_IP_); | |
1395 | wait_for_completion(&bp->b_iowait); | |
0b1b213f | 1396 | trace_xfs_buf_iowait_done(bp, _RET_IP_); |
595bff75 DC |
1397 | error = bp->b_error; |
1398 | ||
1399 | /* | |
1400 | * all done now, we can release the hold that keeps the buffer | |
1401 | * referenced for the entire IO. | |
1402 | */ | |
1403 | xfs_buf_rele(bp); | |
1404 | return error; | |
1da177e4 LT |
1405 | } |
1406 | ||
ce8e922c NS |
1407 | xfs_caddr_t |
1408 | xfs_buf_offset( | |
1409 | xfs_buf_t *bp, | |
1da177e4 LT |
1410 | size_t offset) |
1411 | { | |
1412 | struct page *page; | |
1413 | ||
611c9946 | 1414 | if (bp->b_addr) |
62926044 | 1415 | return bp->b_addr + offset; |
1da177e4 | 1416 | |
ce8e922c | 1417 | offset += bp->b_offset; |
0e6e847f DC |
1418 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
1419 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_SIZE-1)); | |
1da177e4 LT |
1420 | } |
1421 | ||
1422 | /* | |
1da177e4 LT |
1423 | * Move data into or out of a buffer. |
1424 | */ | |
1425 | void | |
ce8e922c NS |
1426 | xfs_buf_iomove( |
1427 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1428 | size_t boff, /* starting buffer offset */ |
1429 | size_t bsize, /* length to copy */ | |
b9c48649 | 1430 | void *data, /* data address */ |
ce8e922c | 1431 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 | 1432 | { |
795cac72 | 1433 | size_t bend; |
1da177e4 LT |
1434 | |
1435 | bend = boff + bsize; | |
1436 | while (boff < bend) { | |
795cac72 DC |
1437 | struct page *page; |
1438 | int page_index, page_offset, csize; | |
1439 | ||
1440 | page_index = (boff + bp->b_offset) >> PAGE_SHIFT; | |
1441 | page_offset = (boff + bp->b_offset) & ~PAGE_MASK; | |
1442 | page = bp->b_pages[page_index]; | |
1443 | csize = min_t(size_t, PAGE_SIZE - page_offset, | |
1444 | BBTOB(bp->b_io_length) - boff); | |
1da177e4 | 1445 | |
795cac72 | 1446 | ASSERT((csize + page_offset) <= PAGE_SIZE); |
1da177e4 LT |
1447 | |
1448 | switch (mode) { | |
ce8e922c | 1449 | case XBRW_ZERO: |
795cac72 | 1450 | memset(page_address(page) + page_offset, 0, csize); |
1da177e4 | 1451 | break; |
ce8e922c | 1452 | case XBRW_READ: |
795cac72 | 1453 | memcpy(data, page_address(page) + page_offset, csize); |
1da177e4 | 1454 | break; |
ce8e922c | 1455 | case XBRW_WRITE: |
795cac72 | 1456 | memcpy(page_address(page) + page_offset, data, csize); |
1da177e4 LT |
1457 | } |
1458 | ||
1459 | boff += csize; | |
1460 | data += csize; | |
1461 | } | |
1462 | } | |
1463 | ||
1464 | /* | |
ce8e922c | 1465 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1466 | */ |
1467 | ||
1468 | /* | |
430cbeb8 DC |
1469 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1470 | * returned. These buffers will have an elevated hold count, so wait on those | |
1471 | * while freeing all the buffers only held by the LRU. | |
1da177e4 | 1472 | */ |
e80dfa19 DC |
1473 | static enum lru_status |
1474 | xfs_buftarg_wait_rele( | |
1475 | struct list_head *item, | |
1476 | spinlock_t *lru_lock, | |
1477 | void *arg) | |
1478 | ||
1da177e4 | 1479 | { |
e80dfa19 | 1480 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); |
a4082357 | 1481 | struct list_head *dispose = arg; |
430cbeb8 | 1482 | |
e80dfa19 | 1483 | if (atomic_read(&bp->b_hold) > 1) { |
a4082357 | 1484 | /* need to wait, so skip it this pass */ |
e80dfa19 | 1485 | trace_xfs_buf_wait_buftarg(bp, _RET_IP_); |
a4082357 | 1486 | return LRU_SKIP; |
1da177e4 | 1487 | } |
a4082357 DC |
1488 | if (!spin_trylock(&bp->b_lock)) |
1489 | return LRU_SKIP; | |
e80dfa19 | 1490 | |
a4082357 DC |
1491 | /* |
1492 | * clear the LRU reference count so the buffer doesn't get | |
1493 | * ignored in xfs_buf_rele(). | |
1494 | */ | |
1495 | atomic_set(&bp->b_lru_ref, 0); | |
1496 | bp->b_state |= XFS_BSTATE_DISPOSE; | |
1497 | list_move(item, dispose); | |
1498 | spin_unlock(&bp->b_lock); | |
1499 | return LRU_REMOVED; | |
1da177e4 LT |
1500 | } |
1501 | ||
e80dfa19 DC |
1502 | void |
1503 | xfs_wait_buftarg( | |
1504 | struct xfs_buftarg *btp) | |
1505 | { | |
a4082357 DC |
1506 | LIST_HEAD(dispose); |
1507 | int loop = 0; | |
1508 | ||
1509 | /* loop until there is nothing left on the lru list. */ | |
1510 | while (list_lru_count(&btp->bt_lru)) { | |
e80dfa19 | 1511 | list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele, |
a4082357 DC |
1512 | &dispose, LONG_MAX); |
1513 | ||
1514 | while (!list_empty(&dispose)) { | |
1515 | struct xfs_buf *bp; | |
1516 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1517 | list_del_init(&bp->b_lru); | |
ac8809f9 DC |
1518 | if (bp->b_flags & XBF_WRITE_FAIL) { |
1519 | xfs_alert(btp->bt_mount, | |
1520 | "Corruption Alert: Buffer at block 0x%llx had permanent write failures!\n" | |
1521 | "Please run xfs_repair to determine the extent of the problem.", | |
1522 | (long long)bp->b_bn); | |
1523 | } | |
a4082357 DC |
1524 | xfs_buf_rele(bp); |
1525 | } | |
1526 | if (loop++ != 0) | |
1527 | delay(100); | |
1528 | } | |
e80dfa19 DC |
1529 | } |
1530 | ||
1531 | static enum lru_status | |
1532 | xfs_buftarg_isolate( | |
1533 | struct list_head *item, | |
1534 | spinlock_t *lru_lock, | |
1535 | void *arg) | |
1536 | { | |
1537 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); | |
1538 | struct list_head *dispose = arg; | |
1539 | ||
a4082357 DC |
1540 | /* |
1541 | * we are inverting the lru lock/bp->b_lock here, so use a trylock. | |
1542 | * If we fail to get the lock, just skip it. | |
1543 | */ | |
1544 | if (!spin_trylock(&bp->b_lock)) | |
1545 | return LRU_SKIP; | |
e80dfa19 DC |
1546 | /* |
1547 | * Decrement the b_lru_ref count unless the value is already | |
1548 | * zero. If the value is already zero, we need to reclaim the | |
1549 | * buffer, otherwise it gets another trip through the LRU. | |
1550 | */ | |
a4082357 DC |
1551 | if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) { |
1552 | spin_unlock(&bp->b_lock); | |
e80dfa19 | 1553 | return LRU_ROTATE; |
a4082357 | 1554 | } |
e80dfa19 | 1555 | |
a4082357 | 1556 | bp->b_state |= XFS_BSTATE_DISPOSE; |
e80dfa19 | 1557 | list_move(item, dispose); |
a4082357 | 1558 | spin_unlock(&bp->b_lock); |
e80dfa19 DC |
1559 | return LRU_REMOVED; |
1560 | } | |
1561 | ||
addbda40 | 1562 | static unsigned long |
e80dfa19 | 1563 | xfs_buftarg_shrink_scan( |
ff57ab21 | 1564 | struct shrinker *shrink, |
1495f230 | 1565 | struct shrink_control *sc) |
a6867a68 | 1566 | { |
ff57ab21 DC |
1567 | struct xfs_buftarg *btp = container_of(shrink, |
1568 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1569 | LIST_HEAD(dispose); |
addbda40 | 1570 | unsigned long freed; |
e80dfa19 | 1571 | unsigned long nr_to_scan = sc->nr_to_scan; |
430cbeb8 | 1572 | |
e80dfa19 DC |
1573 | freed = list_lru_walk_node(&btp->bt_lru, sc->nid, xfs_buftarg_isolate, |
1574 | &dispose, &nr_to_scan); | |
430cbeb8 DC |
1575 | |
1576 | while (!list_empty(&dispose)) { | |
e80dfa19 | 1577 | struct xfs_buf *bp; |
430cbeb8 DC |
1578 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); |
1579 | list_del_init(&bp->b_lru); | |
1580 | xfs_buf_rele(bp); | |
1581 | } | |
1582 | ||
e80dfa19 DC |
1583 | return freed; |
1584 | } | |
1585 | ||
addbda40 | 1586 | static unsigned long |
e80dfa19 DC |
1587 | xfs_buftarg_shrink_count( |
1588 | struct shrinker *shrink, | |
1589 | struct shrink_control *sc) | |
1590 | { | |
1591 | struct xfs_buftarg *btp = container_of(shrink, | |
1592 | struct xfs_buftarg, bt_shrinker); | |
1593 | return list_lru_count_node(&btp->bt_lru, sc->nid); | |
a6867a68 DC |
1594 | } |
1595 | ||
1da177e4 LT |
1596 | void |
1597 | xfs_free_buftarg( | |
b7963133 CH |
1598 | struct xfs_mount *mp, |
1599 | struct xfs_buftarg *btp) | |
1da177e4 | 1600 | { |
ff57ab21 | 1601 | unregister_shrinker(&btp->bt_shrinker); |
f5e1dd34 | 1602 | list_lru_destroy(&btp->bt_lru); |
ff57ab21 | 1603 | |
b7963133 CH |
1604 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1605 | xfs_blkdev_issue_flush(btp); | |
a6867a68 | 1606 | |
f0e2d93c | 1607 | kmem_free(btp); |
1da177e4 LT |
1608 | } |
1609 | ||
3fefdeee ES |
1610 | int |
1611 | xfs_setsize_buftarg( | |
1da177e4 | 1612 | xfs_buftarg_t *btp, |
3fefdeee | 1613 | unsigned int sectorsize) |
1da177e4 | 1614 | { |
7c71ee78 | 1615 | /* Set up metadata sector size info */ |
6da54179 ES |
1616 | btp->bt_meta_sectorsize = sectorsize; |
1617 | btp->bt_meta_sectormask = sectorsize - 1; | |
1da177e4 | 1618 | |
ce8e922c | 1619 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
02b102df CH |
1620 | char name[BDEVNAME_SIZE]; |
1621 | ||
1622 | bdevname(btp->bt_bdev, name); | |
1623 | ||
4f10700a | 1624 | xfs_warn(btp->bt_mount, |
08e96e1a | 1625 | "Cannot set_blocksize to %u on device %s", |
02b102df | 1626 | sectorsize, name); |
2451337d | 1627 | return -EINVAL; |
1da177e4 LT |
1628 | } |
1629 | ||
7c71ee78 ES |
1630 | /* Set up device logical sector size mask */ |
1631 | btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev); | |
1632 | btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1; | |
1633 | ||
1da177e4 LT |
1634 | return 0; |
1635 | } | |
1636 | ||
1637 | /* | |
3fefdeee ES |
1638 | * When allocating the initial buffer target we have not yet |
1639 | * read in the superblock, so don't know what sized sectors | |
1640 | * are being used at this early stage. Play safe. | |
ce8e922c | 1641 | */ |
1da177e4 LT |
1642 | STATIC int |
1643 | xfs_setsize_buftarg_early( | |
1644 | xfs_buftarg_t *btp, | |
1645 | struct block_device *bdev) | |
1646 | { | |
a96c4151 | 1647 | return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev)); |
1da177e4 LT |
1648 | } |
1649 | ||
1da177e4 LT |
1650 | xfs_buftarg_t * |
1651 | xfs_alloc_buftarg( | |
ebad861b | 1652 | struct xfs_mount *mp, |
34dcefd7 | 1653 | struct block_device *bdev) |
1da177e4 LT |
1654 | { |
1655 | xfs_buftarg_t *btp; | |
1656 | ||
b17cb364 | 1657 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP | KM_NOFS); |
1da177e4 | 1658 | |
ebad861b | 1659 | btp->bt_mount = mp; |
ce8e922c NS |
1660 | btp->bt_dev = bdev->bd_dev; |
1661 | btp->bt_bdev = bdev; | |
0e6e847f DC |
1662 | btp->bt_bdi = blk_get_backing_dev_info(bdev); |
1663 | if (!btp->bt_bdi) | |
1664 | goto error; | |
1665 | ||
1da177e4 LT |
1666 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1667 | goto error; | |
5ca302c8 GC |
1668 | |
1669 | if (list_lru_init(&btp->bt_lru)) | |
1670 | goto error; | |
1671 | ||
e80dfa19 DC |
1672 | btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count; |
1673 | btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan; | |
ff57ab21 | 1674 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; |
e80dfa19 | 1675 | btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE; |
ff57ab21 | 1676 | register_shrinker(&btp->bt_shrinker); |
1da177e4 LT |
1677 | return btp; |
1678 | ||
1679 | error: | |
f0e2d93c | 1680 | kmem_free(btp); |
1da177e4 LT |
1681 | return NULL; |
1682 | } | |
1683 | ||
1da177e4 | 1684 | /* |
43ff2122 CH |
1685 | * Add a buffer to the delayed write list. |
1686 | * | |
1687 | * This queues a buffer for writeout if it hasn't already been. Note that | |
1688 | * neither this routine nor the buffer list submission functions perform | |
1689 | * any internal synchronization. It is expected that the lists are thread-local | |
1690 | * to the callers. | |
1691 | * | |
1692 | * Returns true if we queued up the buffer, or false if it already had | |
1693 | * been on the buffer list. | |
1da177e4 | 1694 | */ |
43ff2122 | 1695 | bool |
ce8e922c | 1696 | xfs_buf_delwri_queue( |
43ff2122 CH |
1697 | struct xfs_buf *bp, |
1698 | struct list_head *list) | |
1da177e4 | 1699 | { |
43ff2122 | 1700 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 1701 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1702 | |
43ff2122 CH |
1703 | /* |
1704 | * If the buffer is already marked delwri it already is queued up | |
1705 | * by someone else for imediate writeout. Just ignore it in that | |
1706 | * case. | |
1707 | */ | |
1708 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
1709 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
1710 | return false; | |
1da177e4 | 1711 | } |
1da177e4 | 1712 | |
43ff2122 | 1713 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
1714 | |
1715 | /* | |
43ff2122 CH |
1716 | * If a buffer gets written out synchronously or marked stale while it |
1717 | * is on a delwri list we lazily remove it. To do this, the other party | |
1718 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
1719 | * It remains referenced and on the list. In a rare corner case it | |
1720 | * might get readded to a delwri list after the synchronous writeout, in | |
1721 | * which case we need just need to re-add the flag here. | |
d808f617 | 1722 | */ |
43ff2122 CH |
1723 | bp->b_flags |= _XBF_DELWRI_Q; |
1724 | if (list_empty(&bp->b_list)) { | |
1725 | atomic_inc(&bp->b_hold); | |
1726 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 1727 | } |
585e6d88 | 1728 | |
43ff2122 | 1729 | return true; |
585e6d88 DC |
1730 | } |
1731 | ||
089716aa DC |
1732 | /* |
1733 | * Compare function is more complex than it needs to be because | |
1734 | * the return value is only 32 bits and we are doing comparisons | |
1735 | * on 64 bit values | |
1736 | */ | |
1737 | static int | |
1738 | xfs_buf_cmp( | |
1739 | void *priv, | |
1740 | struct list_head *a, | |
1741 | struct list_head *b) | |
1742 | { | |
1743 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1744 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1745 | xfs_daddr_t diff; | |
1746 | ||
f4b42421 | 1747 | diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn; |
089716aa DC |
1748 | if (diff < 0) |
1749 | return -1; | |
1750 | if (diff > 0) | |
1751 | return 1; | |
1752 | return 0; | |
1753 | } | |
1754 | ||
43ff2122 CH |
1755 | static int |
1756 | __xfs_buf_delwri_submit( | |
1757 | struct list_head *buffer_list, | |
1758 | struct list_head *io_list, | |
1759 | bool wait) | |
1da177e4 | 1760 | { |
43ff2122 CH |
1761 | struct blk_plug plug; |
1762 | struct xfs_buf *bp, *n; | |
1763 | int pinned = 0; | |
1764 | ||
1765 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { | |
1766 | if (!wait) { | |
1767 | if (xfs_buf_ispinned(bp)) { | |
1768 | pinned++; | |
1769 | continue; | |
1770 | } | |
1771 | if (!xfs_buf_trylock(bp)) | |
1772 | continue; | |
1773 | } else { | |
1774 | xfs_buf_lock(bp); | |
1775 | } | |
978c7b2f | 1776 | |
43ff2122 CH |
1777 | /* |
1778 | * Someone else might have written the buffer synchronously or | |
1779 | * marked it stale in the meantime. In that case only the | |
1780 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
1781 | * reference and remove it from the list here. | |
1782 | */ | |
1783 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
1784 | list_del_init(&bp->b_list); | |
1785 | xfs_buf_relse(bp); | |
1786 | continue; | |
1787 | } | |
c9c12971 | 1788 | |
43ff2122 CH |
1789 | list_move_tail(&bp->b_list, io_list); |
1790 | trace_xfs_buf_delwri_split(bp, _RET_IP_); | |
1791 | } | |
1da177e4 | 1792 | |
43ff2122 | 1793 | list_sort(NULL, io_list, xfs_buf_cmp); |
1da177e4 | 1794 | |
43ff2122 CH |
1795 | blk_start_plug(&plug); |
1796 | list_for_each_entry_safe(bp, n, io_list, b_list) { | |
ac8809f9 | 1797 | bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_ASYNC | XBF_WRITE_FAIL); |
cf53e99d | 1798 | bp->b_flags |= XBF_WRITE | XBF_ASYNC; |
a1b7ea5d | 1799 | |
cf53e99d DC |
1800 | /* |
1801 | * we do all Io submission async. This means if we need to wait | |
1802 | * for IO completion we need to take an extra reference so the | |
1803 | * buffer is still valid on the other side. | |
1804 | */ | |
1805 | if (wait) | |
1806 | xfs_buf_hold(bp); | |
1807 | else | |
ce8e922c | 1808 | list_del_init(&bp->b_list); |
8dac3921 | 1809 | |
595bff75 | 1810 | xfs_buf_submit(bp); |
43ff2122 CH |
1811 | } |
1812 | blk_finish_plug(&plug); | |
1da177e4 | 1813 | |
43ff2122 | 1814 | return pinned; |
1da177e4 LT |
1815 | } |
1816 | ||
1817 | /* | |
43ff2122 CH |
1818 | * Write out a buffer list asynchronously. |
1819 | * | |
1820 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
1821 | * out and not wait for I/O completion on any of the buffers. This interface | |
1822 | * is only safely useable for callers that can track I/O completion by higher | |
1823 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
1824 | * function. | |
1da177e4 LT |
1825 | */ |
1826 | int | |
43ff2122 CH |
1827 | xfs_buf_delwri_submit_nowait( |
1828 | struct list_head *buffer_list) | |
1da177e4 | 1829 | { |
43ff2122 CH |
1830 | LIST_HEAD (io_list); |
1831 | return __xfs_buf_delwri_submit(buffer_list, &io_list, false); | |
1832 | } | |
1da177e4 | 1833 | |
43ff2122 CH |
1834 | /* |
1835 | * Write out a buffer list synchronously. | |
1836 | * | |
1837 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
1838 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
1839 | * so callers must have some other way of tracking buffers if they require such | |
1840 | * functionality. | |
1841 | */ | |
1842 | int | |
1843 | xfs_buf_delwri_submit( | |
1844 | struct list_head *buffer_list) | |
1845 | { | |
1846 | LIST_HEAD (io_list); | |
1847 | int error = 0, error2; | |
1848 | struct xfs_buf *bp; | |
1da177e4 | 1849 | |
43ff2122 | 1850 | __xfs_buf_delwri_submit(buffer_list, &io_list, true); |
1da177e4 | 1851 | |
43ff2122 CH |
1852 | /* Wait for IO to complete. */ |
1853 | while (!list_empty(&io_list)) { | |
1854 | bp = list_first_entry(&io_list, struct xfs_buf, b_list); | |
a1b7ea5d | 1855 | |
089716aa | 1856 | list_del_init(&bp->b_list); |
cf53e99d DC |
1857 | |
1858 | /* locking the buffer will wait for async IO completion. */ | |
1859 | xfs_buf_lock(bp); | |
1860 | error2 = bp->b_error; | |
43ff2122 CH |
1861 | xfs_buf_relse(bp); |
1862 | if (!error) | |
1863 | error = error2; | |
1da177e4 LT |
1864 | } |
1865 | ||
43ff2122 | 1866 | return error; |
1da177e4 LT |
1867 | } |
1868 | ||
04d8b284 | 1869 | int __init |
ce8e922c | 1870 | xfs_buf_init(void) |
1da177e4 | 1871 | { |
8758280f NS |
1872 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1873 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1874 | if (!xfs_buf_zone) |
0b1b213f | 1875 | goto out; |
04d8b284 | 1876 | |
51749e47 | 1877 | xfslogd_workqueue = alloc_workqueue("xfslogd", |
6370a6ad | 1878 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 1); |
23ea4032 | 1879 | if (!xfslogd_workqueue) |
04d8b284 | 1880 | goto out_free_buf_zone; |
1da177e4 | 1881 | |
23ea4032 | 1882 | return 0; |
1da177e4 | 1883 | |
23ea4032 | 1884 | out_free_buf_zone: |
ce8e922c | 1885 | kmem_zone_destroy(xfs_buf_zone); |
0b1b213f | 1886 | out: |
8758280f | 1887 | return -ENOMEM; |
1da177e4 LT |
1888 | } |
1889 | ||
1da177e4 | 1890 | void |
ce8e922c | 1891 | xfs_buf_terminate(void) |
1da177e4 | 1892 | { |
04d8b284 | 1893 | destroy_workqueue(xfslogd_workqueue); |
ce8e922c | 1894 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 1895 | } |