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) | |
133 | return ENOMEM; | |
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, | |
1da177e4 | 219 | int page_count, |
ce8e922c | 220 | xfs_buf_flags_t flags) |
1da177e4 LT |
221 | { |
222 | /* Make sure that we have a page list */ | |
ce8e922c | 223 | if (bp->b_pages == NULL) { |
ce8e922c NS |
224 | bp->b_page_count = page_count; |
225 | if (page_count <= XB_PAGES) { | |
226 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 227 | } else { |
ce8e922c | 228 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
aa5c158e | 229 | page_count, KM_NOFS); |
ce8e922c | 230 | if (bp->b_pages == NULL) |
1da177e4 LT |
231 | return -ENOMEM; |
232 | } | |
ce8e922c | 233 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
234 | } |
235 | return 0; | |
236 | } | |
237 | ||
238 | /* | |
ce8e922c | 239 | * Frees b_pages if it was allocated. |
1da177e4 LT |
240 | */ |
241 | STATIC void | |
ce8e922c | 242 | _xfs_buf_free_pages( |
1da177e4 LT |
243 | xfs_buf_t *bp) |
244 | { | |
ce8e922c | 245 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 246 | kmem_free(bp->b_pages); |
3fc98b1a | 247 | bp->b_pages = NULL; |
1da177e4 LT |
248 | } |
249 | } | |
250 | ||
251 | /* | |
252 | * Releases the specified buffer. | |
253 | * | |
254 | * The modification state of any associated pages is left unchanged. | |
b46fe825 | 255 | * The buffer must not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
256 | * hashed and refcounted buffers |
257 | */ | |
258 | void | |
ce8e922c | 259 | xfs_buf_free( |
1da177e4 LT |
260 | xfs_buf_t *bp) |
261 | { | |
0b1b213f | 262 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 263 | |
430cbeb8 DC |
264 | ASSERT(list_empty(&bp->b_lru)); |
265 | ||
0e6e847f | 266 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
267 | uint i; |
268 | ||
73c77e2c | 269 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
270 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
271 | bp->b_page_count); | |
1da177e4 | 272 | |
948ecdb4 NS |
273 | for (i = 0; i < bp->b_page_count; i++) { |
274 | struct page *page = bp->b_pages[i]; | |
275 | ||
0e6e847f | 276 | __free_page(page); |
948ecdb4 | 277 | } |
0e6e847f DC |
278 | } else if (bp->b_flags & _XBF_KMEM) |
279 | kmem_free(bp->b_addr); | |
3fc98b1a | 280 | _xfs_buf_free_pages(bp); |
3e85c868 | 281 | xfs_buf_free_maps(bp); |
4347b9d7 | 282 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
283 | } |
284 | ||
285 | /* | |
0e6e847f | 286 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
287 | */ |
288 | STATIC int | |
0e6e847f | 289 | xfs_buf_allocate_memory( |
1da177e4 LT |
290 | xfs_buf_t *bp, |
291 | uint flags) | |
292 | { | |
aa0e8833 | 293 | size_t size; |
1da177e4 | 294 | size_t nbytes, offset; |
ce8e922c | 295 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 296 | unsigned short page_count, i; |
795cac72 | 297 | xfs_off_t start, end; |
1da177e4 LT |
298 | int error; |
299 | ||
0e6e847f DC |
300 | /* |
301 | * for buffers that are contained within a single page, just allocate | |
302 | * the memory from the heap - there's no need for the complexity of | |
303 | * page arrays to keep allocation down to order 0. | |
304 | */ | |
795cac72 DC |
305 | size = BBTOB(bp->b_length); |
306 | if (size < PAGE_SIZE) { | |
aa5c158e | 307 | bp->b_addr = kmem_alloc(size, KM_NOFS); |
0e6e847f DC |
308 | if (!bp->b_addr) { |
309 | /* low memory - use alloc_page loop instead */ | |
310 | goto use_alloc_page; | |
311 | } | |
312 | ||
795cac72 | 313 | if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) != |
0e6e847f DC |
314 | ((unsigned long)bp->b_addr & PAGE_MASK)) { |
315 | /* b_addr spans two pages - use alloc_page instead */ | |
316 | kmem_free(bp->b_addr); | |
317 | bp->b_addr = NULL; | |
318 | goto use_alloc_page; | |
319 | } | |
320 | bp->b_offset = offset_in_page(bp->b_addr); | |
321 | bp->b_pages = bp->b_page_array; | |
322 | bp->b_pages[0] = virt_to_page(bp->b_addr); | |
323 | bp->b_page_count = 1; | |
611c9946 | 324 | bp->b_flags |= _XBF_KMEM; |
0e6e847f DC |
325 | return 0; |
326 | } | |
327 | ||
328 | use_alloc_page: | |
f4b42421 MT |
329 | start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT; |
330 | end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1) | |
cbb7baab | 331 | >> PAGE_SHIFT; |
795cac72 | 332 | page_count = end - start; |
ce8e922c | 333 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
334 | if (unlikely(error)) |
335 | return error; | |
1da177e4 | 336 | |
ce8e922c | 337 | offset = bp->b_offset; |
0e6e847f | 338 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 339 | |
ce8e922c | 340 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
341 | struct page *page; |
342 | uint retries = 0; | |
0e6e847f DC |
343 | retry: |
344 | page = alloc_page(gfp_mask); | |
1da177e4 | 345 | if (unlikely(page == NULL)) { |
ce8e922c NS |
346 | if (flags & XBF_READ_AHEAD) { |
347 | bp->b_page_count = i; | |
0e6e847f DC |
348 | error = ENOMEM; |
349 | goto out_free_pages; | |
1da177e4 LT |
350 | } |
351 | ||
352 | /* | |
353 | * This could deadlock. | |
354 | * | |
355 | * But until all the XFS lowlevel code is revamped to | |
356 | * handle buffer allocation failures we can't do much. | |
357 | */ | |
358 | if (!(++retries % 100)) | |
4f10700a DC |
359 | xfs_err(NULL, |
360 | "possible memory allocation deadlock in %s (mode:0x%x)", | |
34a622b2 | 361 | __func__, gfp_mask); |
1da177e4 | 362 | |
ce8e922c | 363 | XFS_STATS_INC(xb_page_retries); |
8aa7e847 | 364 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
365 | goto retry; |
366 | } | |
367 | ||
ce8e922c | 368 | XFS_STATS_INC(xb_page_found); |
1da177e4 | 369 | |
0e6e847f | 370 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 371 | size -= nbytes; |
ce8e922c | 372 | bp->b_pages[i] = page; |
1da177e4 LT |
373 | offset = 0; |
374 | } | |
0e6e847f | 375 | return 0; |
1da177e4 | 376 | |
0e6e847f DC |
377 | out_free_pages: |
378 | for (i = 0; i < bp->b_page_count; i++) | |
379 | __free_page(bp->b_pages[i]); | |
1da177e4 LT |
380 | return error; |
381 | } | |
382 | ||
383 | /* | |
25985edc | 384 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
385 | */ |
386 | STATIC int | |
ce8e922c | 387 | _xfs_buf_map_pages( |
1da177e4 LT |
388 | xfs_buf_t *bp, |
389 | uint flags) | |
390 | { | |
0e6e847f | 391 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 392 | if (bp->b_page_count == 1) { |
0e6e847f | 393 | /* A single page buffer is always mappable */ |
ce8e922c | 394 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
611c9946 DC |
395 | } else if (flags & XBF_UNMAPPED) { |
396 | bp->b_addr = NULL; | |
397 | } else { | |
a19fb380 | 398 | int retried = 0; |
ae687e58 DC |
399 | unsigned noio_flag; |
400 | ||
401 | /* | |
402 | * vm_map_ram() will allocate auxillary structures (e.g. | |
403 | * pagetables) with GFP_KERNEL, yet we are likely to be under | |
404 | * GFP_NOFS context here. Hence we need to tell memory reclaim | |
405 | * that we are in such a context via PF_MEMALLOC_NOIO to prevent | |
406 | * memory reclaim re-entering the filesystem here and | |
407 | * potentially deadlocking. | |
408 | */ | |
409 | noio_flag = memalloc_noio_save(); | |
a19fb380 DC |
410 | do { |
411 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
412 | -1, PAGE_KERNEL); | |
413 | if (bp->b_addr) | |
414 | break; | |
415 | vm_unmap_aliases(); | |
416 | } while (retried++ <= 1); | |
ae687e58 | 417 | memalloc_noio_restore(noio_flag); |
a19fb380 DC |
418 | |
419 | if (!bp->b_addr) | |
1da177e4 | 420 | return -ENOMEM; |
ce8e922c | 421 | bp->b_addr += bp->b_offset; |
1da177e4 LT |
422 | } |
423 | ||
424 | return 0; | |
425 | } | |
426 | ||
427 | /* | |
428 | * Finding and Reading Buffers | |
429 | */ | |
430 | ||
431 | /* | |
ce8e922c | 432 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 | 433 | * a given range of an inode. The buffer is returned |
eabbaf11 | 434 | * locked. No I/O is implied by this call. |
1da177e4 LT |
435 | */ |
436 | xfs_buf_t * | |
ce8e922c | 437 | _xfs_buf_find( |
e70b73f8 | 438 | struct xfs_buftarg *btp, |
3e85c868 DC |
439 | struct xfs_buf_map *map, |
440 | int nmaps, | |
ce8e922c NS |
441 | xfs_buf_flags_t flags, |
442 | xfs_buf_t *new_bp) | |
1da177e4 | 443 | { |
e70b73f8 | 444 | size_t numbytes; |
74f75a0c DC |
445 | struct xfs_perag *pag; |
446 | struct rb_node **rbp; | |
447 | struct rb_node *parent; | |
448 | xfs_buf_t *bp; | |
3e85c868 | 449 | xfs_daddr_t blkno = map[0].bm_bn; |
10616b80 | 450 | xfs_daddr_t eofs; |
3e85c868 DC |
451 | int numblks = 0; |
452 | int i; | |
1da177e4 | 453 | |
3e85c868 DC |
454 | for (i = 0; i < nmaps; i++) |
455 | numblks += map[i].bm_len; | |
e70b73f8 | 456 | numbytes = BBTOB(numblks); |
1da177e4 LT |
457 | |
458 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
6da54179 ES |
459 | ASSERT(!(numbytes < btp->bt_meta_sectorsize)); |
460 | ASSERT(!(BBTOB(blkno) & (xfs_off_t)btp->bt_meta_sectormask)); | |
1da177e4 | 461 | |
10616b80 DC |
462 | /* |
463 | * Corrupted block numbers can get through to here, unfortunately, so we | |
464 | * have to check that the buffer falls within the filesystem bounds. | |
465 | */ | |
466 | eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks); | |
467 | if (blkno >= eofs) { | |
468 | /* | |
469 | * XXX (dgc): we should really be returning EFSCORRUPTED here, | |
470 | * but none of the higher level infrastructure supports | |
471 | * returning a specific error on buffer lookup failures. | |
472 | */ | |
473 | xfs_alert(btp->bt_mount, | |
474 | "%s: Block out of range: block 0x%llx, EOFS 0x%llx ", | |
475 | __func__, blkno, eofs); | |
7bc0dc27 | 476 | WARN_ON(1); |
10616b80 DC |
477 | return NULL; |
478 | } | |
479 | ||
74f75a0c DC |
480 | /* get tree root */ |
481 | pag = xfs_perag_get(btp->bt_mount, | |
e70b73f8 | 482 | xfs_daddr_to_agno(btp->bt_mount, blkno)); |
74f75a0c DC |
483 | |
484 | /* walk tree */ | |
485 | spin_lock(&pag->pag_buf_lock); | |
486 | rbp = &pag->pag_buf_tree.rb_node; | |
487 | parent = NULL; | |
488 | bp = NULL; | |
489 | while (*rbp) { | |
490 | parent = *rbp; | |
491 | bp = rb_entry(parent, struct xfs_buf, b_rbnode); | |
492 | ||
de1cbee4 | 493 | if (blkno < bp->b_bn) |
74f75a0c | 494 | rbp = &(*rbp)->rb_left; |
de1cbee4 | 495 | else if (blkno > bp->b_bn) |
74f75a0c DC |
496 | rbp = &(*rbp)->rb_right; |
497 | else { | |
498 | /* | |
de1cbee4 | 499 | * found a block number match. If the range doesn't |
74f75a0c DC |
500 | * match, the only way this is allowed is if the buffer |
501 | * in the cache is stale and the transaction that made | |
502 | * it stale has not yet committed. i.e. we are | |
503 | * reallocating a busy extent. Skip this buffer and | |
504 | * continue searching to the right for an exact match. | |
505 | */ | |
4e94b71b | 506 | if (bp->b_length != numblks) { |
74f75a0c DC |
507 | ASSERT(bp->b_flags & XBF_STALE); |
508 | rbp = &(*rbp)->rb_right; | |
509 | continue; | |
510 | } | |
ce8e922c | 511 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
512 | goto found; |
513 | } | |
514 | } | |
515 | ||
516 | /* No match found */ | |
ce8e922c | 517 | if (new_bp) { |
74f75a0c DC |
518 | rb_link_node(&new_bp->b_rbnode, parent, rbp); |
519 | rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree); | |
520 | /* the buffer keeps the perag reference until it is freed */ | |
521 | new_bp->b_pag = pag; | |
522 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 523 | } else { |
ce8e922c | 524 | XFS_STATS_INC(xb_miss_locked); |
74f75a0c DC |
525 | spin_unlock(&pag->pag_buf_lock); |
526 | xfs_perag_put(pag); | |
1da177e4 | 527 | } |
ce8e922c | 528 | return new_bp; |
1da177e4 LT |
529 | |
530 | found: | |
74f75a0c DC |
531 | spin_unlock(&pag->pag_buf_lock); |
532 | xfs_perag_put(pag); | |
1da177e4 | 533 | |
0c842ad4 CH |
534 | if (!xfs_buf_trylock(bp)) { |
535 | if (flags & XBF_TRYLOCK) { | |
ce8e922c NS |
536 | xfs_buf_rele(bp); |
537 | XFS_STATS_INC(xb_busy_locked); | |
538 | return NULL; | |
1da177e4 | 539 | } |
0c842ad4 CH |
540 | xfs_buf_lock(bp); |
541 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
542 | } |
543 | ||
0e6e847f DC |
544 | /* |
545 | * if the buffer is stale, clear all the external state associated with | |
546 | * it. We need to keep flags such as how we allocated the buffer memory | |
547 | * intact here. | |
548 | */ | |
ce8e922c NS |
549 | if (bp->b_flags & XBF_STALE) { |
550 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
cfb02852 | 551 | ASSERT(bp->b_iodone == NULL); |
611c9946 | 552 | bp->b_flags &= _XBF_KMEM | _XBF_PAGES; |
1813dd64 | 553 | bp->b_ops = NULL; |
2f926587 | 554 | } |
0b1b213f CH |
555 | |
556 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ce8e922c NS |
557 | XFS_STATS_INC(xb_get_locked); |
558 | return bp; | |
1da177e4 LT |
559 | } |
560 | ||
561 | /* | |
3815832a DC |
562 | * Assembles a buffer covering the specified range. The code is optimised for |
563 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
564 | * more hits than misses. | |
1da177e4 | 565 | */ |
3815832a | 566 | struct xfs_buf * |
6dde2707 DC |
567 | xfs_buf_get_map( |
568 | struct xfs_buftarg *target, | |
569 | struct xfs_buf_map *map, | |
570 | int nmaps, | |
ce8e922c | 571 | xfs_buf_flags_t flags) |
1da177e4 | 572 | { |
3815832a DC |
573 | struct xfs_buf *bp; |
574 | struct xfs_buf *new_bp; | |
0e6e847f | 575 | int error = 0; |
1da177e4 | 576 | |
6dde2707 | 577 | bp = _xfs_buf_find(target, map, nmaps, flags, NULL); |
3815832a DC |
578 | if (likely(bp)) |
579 | goto found; | |
580 | ||
6dde2707 | 581 | new_bp = _xfs_buf_alloc(target, map, nmaps, flags); |
ce8e922c | 582 | if (unlikely(!new_bp)) |
1da177e4 LT |
583 | return NULL; |
584 | ||
fe2429b0 DC |
585 | error = xfs_buf_allocate_memory(new_bp, flags); |
586 | if (error) { | |
3e85c868 | 587 | xfs_buf_free(new_bp); |
fe2429b0 DC |
588 | return NULL; |
589 | } | |
590 | ||
6dde2707 | 591 | bp = _xfs_buf_find(target, map, nmaps, flags, new_bp); |
3815832a | 592 | if (!bp) { |
fe2429b0 | 593 | xfs_buf_free(new_bp); |
3815832a DC |
594 | return NULL; |
595 | } | |
596 | ||
fe2429b0 DC |
597 | if (bp != new_bp) |
598 | xfs_buf_free(new_bp); | |
1da177e4 | 599 | |
3815832a | 600 | found: |
611c9946 | 601 | if (!bp->b_addr) { |
ce8e922c | 602 | error = _xfs_buf_map_pages(bp, flags); |
1da177e4 | 603 | if (unlikely(error)) { |
4f10700a | 604 | xfs_warn(target->bt_mount, |
08e96e1a | 605 | "%s: failed to map pagesn", __func__); |
a8acad70 DC |
606 | xfs_buf_relse(bp); |
607 | return NULL; | |
1da177e4 LT |
608 | } |
609 | } | |
610 | ||
ce8e922c | 611 | XFS_STATS_INC(xb_get); |
0b1b213f | 612 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 613 | return bp; |
1da177e4 LT |
614 | } |
615 | ||
5d765b97 CH |
616 | STATIC int |
617 | _xfs_buf_read( | |
618 | xfs_buf_t *bp, | |
619 | xfs_buf_flags_t flags) | |
620 | { | |
43ff2122 | 621 | ASSERT(!(flags & XBF_WRITE)); |
f4b42421 | 622 | ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL); |
5d765b97 | 623 | |
43ff2122 | 624 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD); |
1d5ae5df | 625 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 626 | |
0e95f19a DC |
627 | xfs_buf_iorequest(bp); |
628 | if (flags & XBF_ASYNC) | |
629 | return 0; | |
ec53d1db | 630 | return xfs_buf_iowait(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 | |
83a0adc3 CH |
712 | if (XFS_FORCED_SHUTDOWN(target->bt_mount)) { |
713 | xfs_buf_relse(bp); | |
714 | return NULL; | |
715 | } | |
716 | xfs_buf_iorequest(bp); | |
eab4e633 | 717 | xfs_buf_iowait(bp); |
5adc94c2 | 718 | return bp; |
1da177e4 LT |
719 | } |
720 | ||
44396476 DC |
721 | /* |
722 | * Return a buffer allocated as an empty buffer and associated to external | |
723 | * memory via xfs_buf_associate_memory() back to it's empty state. | |
724 | */ | |
725 | void | |
726 | xfs_buf_set_empty( | |
727 | struct xfs_buf *bp, | |
e70b73f8 | 728 | size_t numblks) |
44396476 DC |
729 | { |
730 | if (bp->b_pages) | |
731 | _xfs_buf_free_pages(bp); | |
732 | ||
733 | bp->b_pages = NULL; | |
734 | bp->b_page_count = 0; | |
735 | bp->b_addr = NULL; | |
4e94b71b | 736 | bp->b_length = numblks; |
aa0e8833 | 737 | bp->b_io_length = numblks; |
3e85c868 DC |
738 | |
739 | ASSERT(bp->b_map_count == 1); | |
44396476 | 740 | bp->b_bn = XFS_BUF_DADDR_NULL; |
3e85c868 DC |
741 | bp->b_maps[0].bm_bn = XFS_BUF_DADDR_NULL; |
742 | bp->b_maps[0].bm_len = bp->b_length; | |
44396476 DC |
743 | } |
744 | ||
1da177e4 LT |
745 | static inline struct page * |
746 | mem_to_page( | |
747 | void *addr) | |
748 | { | |
9e2779fa | 749 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
750 | return virt_to_page(addr); |
751 | } else { | |
752 | return vmalloc_to_page(addr); | |
753 | } | |
754 | } | |
755 | ||
756 | int | |
ce8e922c NS |
757 | xfs_buf_associate_memory( |
758 | xfs_buf_t *bp, | |
1da177e4 LT |
759 | void *mem, |
760 | size_t len) | |
761 | { | |
762 | int rval; | |
763 | int i = 0; | |
d1afb678 LM |
764 | unsigned long pageaddr; |
765 | unsigned long offset; | |
766 | size_t buflen; | |
1da177e4 LT |
767 | int page_count; |
768 | ||
0e6e847f | 769 | pageaddr = (unsigned long)mem & PAGE_MASK; |
d1afb678 | 770 | offset = (unsigned long)mem - pageaddr; |
0e6e847f DC |
771 | buflen = PAGE_ALIGN(len + offset); |
772 | page_count = buflen >> PAGE_SHIFT; | |
1da177e4 LT |
773 | |
774 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
775 | if (bp->b_pages) |
776 | _xfs_buf_free_pages(bp); | |
1da177e4 | 777 | |
ce8e922c NS |
778 | bp->b_pages = NULL; |
779 | bp->b_addr = mem; | |
1da177e4 | 780 | |
aa5c158e | 781 | rval = _xfs_buf_get_pages(bp, page_count, 0); |
1da177e4 LT |
782 | if (rval) |
783 | return rval; | |
784 | ||
ce8e922c | 785 | bp->b_offset = offset; |
d1afb678 LM |
786 | |
787 | for (i = 0; i < bp->b_page_count; i++) { | |
788 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
0e6e847f | 789 | pageaddr += PAGE_SIZE; |
1da177e4 | 790 | } |
1da177e4 | 791 | |
aa0e8833 | 792 | bp->b_io_length = BTOBB(len); |
4e94b71b | 793 | bp->b_length = BTOBB(buflen); |
1da177e4 LT |
794 | |
795 | return 0; | |
796 | } | |
797 | ||
798 | xfs_buf_t * | |
686865f7 DC |
799 | xfs_buf_get_uncached( |
800 | struct xfs_buftarg *target, | |
e70b73f8 | 801 | size_t numblks, |
686865f7 | 802 | int flags) |
1da177e4 | 803 | { |
e70b73f8 | 804 | unsigned long page_count; |
1fa40b01 | 805 | int error, i; |
3e85c868 DC |
806 | struct xfs_buf *bp; |
807 | DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks); | |
1da177e4 | 808 | |
3e85c868 | 809 | bp = _xfs_buf_alloc(target, &map, 1, 0); |
1da177e4 LT |
810 | if (unlikely(bp == NULL)) |
811 | goto fail; | |
1da177e4 | 812 | |
e70b73f8 | 813 | page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT; |
1fa40b01 CH |
814 | error = _xfs_buf_get_pages(bp, page_count, 0); |
815 | if (error) | |
1da177e4 LT |
816 | goto fail_free_buf; |
817 | ||
1fa40b01 | 818 | for (i = 0; i < page_count; i++) { |
686865f7 | 819 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
820 | if (!bp->b_pages[i]) |
821 | goto fail_free_mem; | |
1da177e4 | 822 | } |
1fa40b01 | 823 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 824 | |
611c9946 | 825 | error = _xfs_buf_map_pages(bp, 0); |
1fa40b01 | 826 | if (unlikely(error)) { |
4f10700a | 827 | xfs_warn(target->bt_mount, |
08e96e1a | 828 | "%s: failed to map pages", __func__); |
1da177e4 | 829 | goto fail_free_mem; |
1fa40b01 | 830 | } |
1da177e4 | 831 | |
686865f7 | 832 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 833 | return bp; |
1fa40b01 | 834 | |
1da177e4 | 835 | fail_free_mem: |
1fa40b01 CH |
836 | while (--i >= 0) |
837 | __free_page(bp->b_pages[i]); | |
ca165b88 | 838 | _xfs_buf_free_pages(bp); |
1da177e4 | 839 | fail_free_buf: |
3e85c868 | 840 | xfs_buf_free_maps(bp); |
4347b9d7 | 841 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
842 | fail: |
843 | return NULL; | |
844 | } | |
845 | ||
846 | /* | |
1da177e4 LT |
847 | * Increment reference count on buffer, to hold the buffer concurrently |
848 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
849 | * Must hold the buffer already to call this function. |
850 | */ | |
851 | void | |
ce8e922c NS |
852 | xfs_buf_hold( |
853 | xfs_buf_t *bp) | |
1da177e4 | 854 | { |
0b1b213f | 855 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 856 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
857 | } |
858 | ||
859 | /* | |
ce8e922c NS |
860 | * Releases a hold on the specified buffer. If the |
861 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
862 | */ |
863 | void | |
ce8e922c NS |
864 | xfs_buf_rele( |
865 | xfs_buf_t *bp) | |
1da177e4 | 866 | { |
74f75a0c | 867 | struct xfs_perag *pag = bp->b_pag; |
1da177e4 | 868 | |
0b1b213f | 869 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 870 | |
74f75a0c | 871 | if (!pag) { |
430cbeb8 | 872 | ASSERT(list_empty(&bp->b_lru)); |
74f75a0c | 873 | ASSERT(RB_EMPTY_NODE(&bp->b_rbnode)); |
fad3aa1e NS |
874 | if (atomic_dec_and_test(&bp->b_hold)) |
875 | xfs_buf_free(bp); | |
876 | return; | |
877 | } | |
878 | ||
74f75a0c | 879 | ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode)); |
430cbeb8 | 880 | |
3790689f | 881 | ASSERT(atomic_read(&bp->b_hold) > 0); |
74f75a0c | 882 | if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) { |
a4082357 DC |
883 | spin_lock(&bp->b_lock); |
884 | if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) { | |
885 | /* | |
886 | * If the buffer is added to the LRU take a new | |
887 | * reference to the buffer for the LRU and clear the | |
888 | * (now stale) dispose list state flag | |
889 | */ | |
890 | if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) { | |
891 | bp->b_state &= ~XFS_BSTATE_DISPOSE; | |
892 | atomic_inc(&bp->b_hold); | |
893 | } | |
894 | spin_unlock(&bp->b_lock); | |
430cbeb8 | 895 | spin_unlock(&pag->pag_buf_lock); |
1da177e4 | 896 | } else { |
a4082357 DC |
897 | /* |
898 | * most of the time buffers will already be removed from | |
899 | * the LRU, so optimise that case by checking for the | |
900 | * XFS_BSTATE_DISPOSE flag indicating the last list the | |
901 | * buffer was on was the disposal list | |
902 | */ | |
903 | if (!(bp->b_state & XFS_BSTATE_DISPOSE)) { | |
904 | list_lru_del(&bp->b_target->bt_lru, &bp->b_lru); | |
905 | } else { | |
906 | ASSERT(list_empty(&bp->b_lru)); | |
907 | } | |
908 | spin_unlock(&bp->b_lock); | |
909 | ||
43ff2122 | 910 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
74f75a0c DC |
911 | rb_erase(&bp->b_rbnode, &pag->pag_buf_tree); |
912 | spin_unlock(&pag->pag_buf_lock); | |
913 | xfs_perag_put(pag); | |
ce8e922c | 914 | xfs_buf_free(bp); |
1da177e4 LT |
915 | } |
916 | } | |
917 | } | |
918 | ||
919 | ||
920 | /* | |
0e6e847f | 921 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
922 | * |
923 | * If we come across a stale, pinned, locked buffer, we know that we are | |
924 | * being asked to lock a buffer that has been reallocated. Because it is | |
925 | * pinned, we know that the log has not been pushed to disk and hence it | |
926 | * will still be locked. Rather than continuing to have trylock attempts | |
927 | * fail until someone else pushes the log, push it ourselves before | |
928 | * returning. This means that the xfsaild will not get stuck trying | |
929 | * to push on stale inode buffers. | |
1da177e4 LT |
930 | */ |
931 | int | |
0c842ad4 CH |
932 | xfs_buf_trylock( |
933 | struct xfs_buf *bp) | |
1da177e4 LT |
934 | { |
935 | int locked; | |
936 | ||
ce8e922c | 937 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 938 | if (locked) |
ce8e922c | 939 | XB_SET_OWNER(bp); |
0b1b213f | 940 | |
0c842ad4 CH |
941 | trace_xfs_buf_trylock(bp, _RET_IP_); |
942 | return locked; | |
1da177e4 | 943 | } |
1da177e4 LT |
944 | |
945 | /* | |
0e6e847f | 946 | * Lock a buffer object. |
ed3b4d6c DC |
947 | * |
948 | * If we come across a stale, pinned, locked buffer, we know that we | |
949 | * are being asked to lock a buffer that has been reallocated. Because | |
950 | * it is pinned, we know that the log has not been pushed to disk and | |
951 | * hence it will still be locked. Rather than sleeping until someone | |
952 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 953 | */ |
ce8e922c NS |
954 | void |
955 | xfs_buf_lock( | |
0c842ad4 | 956 | struct xfs_buf *bp) |
1da177e4 | 957 | { |
0b1b213f CH |
958 | trace_xfs_buf_lock(bp, _RET_IP_); |
959 | ||
ed3b4d6c | 960 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 961 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
962 | down(&bp->b_sema); |
963 | XB_SET_OWNER(bp); | |
0b1b213f CH |
964 | |
965 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
966 | } |
967 | ||
1da177e4 | 968 | void |
ce8e922c | 969 | xfs_buf_unlock( |
0c842ad4 | 970 | struct xfs_buf *bp) |
1da177e4 | 971 | { |
ce8e922c NS |
972 | XB_CLEAR_OWNER(bp); |
973 | up(&bp->b_sema); | |
0b1b213f CH |
974 | |
975 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
976 | } |
977 | ||
ce8e922c NS |
978 | STATIC void |
979 | xfs_buf_wait_unpin( | |
980 | xfs_buf_t *bp) | |
1da177e4 LT |
981 | { |
982 | DECLARE_WAITQUEUE (wait, current); | |
983 | ||
ce8e922c | 984 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
985 | return; |
986 | ||
ce8e922c | 987 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
988 | for (;;) { |
989 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 990 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 991 | break; |
7eaceacc | 992 | io_schedule(); |
1da177e4 | 993 | } |
ce8e922c | 994 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
995 | set_current_state(TASK_RUNNING); |
996 | } | |
997 | ||
998 | /* | |
999 | * Buffer Utility Routines | |
1000 | */ | |
1001 | ||
1da177e4 | 1002 | STATIC void |
ce8e922c | 1003 | xfs_buf_iodone_work( |
c4028958 | 1004 | struct work_struct *work) |
1da177e4 | 1005 | { |
1813dd64 | 1006 | struct xfs_buf *bp = |
c4028958 | 1007 | container_of(work, xfs_buf_t, b_iodone_work); |
1813dd64 DC |
1008 | bool read = !!(bp->b_flags & XBF_READ); |
1009 | ||
1010 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); | |
d5929de8 DC |
1011 | |
1012 | /* only validate buffers that were read without errors */ | |
1013 | if (read && bp->b_ops && !bp->b_error && (bp->b_flags & XBF_DONE)) | |
1813dd64 | 1014 | bp->b_ops->verify_read(bp); |
1da177e4 | 1015 | |
80f6c29d | 1016 | if (bp->b_iodone) |
ce8e922c NS |
1017 | (*(bp->b_iodone))(bp); |
1018 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 | 1019 | xfs_buf_relse(bp); |
1813dd64 DC |
1020 | else { |
1021 | ASSERT(read && bp->b_ops); | |
1022 | complete(&bp->b_iowait); | |
1023 | } | |
1da177e4 LT |
1024 | } |
1025 | ||
1026 | void | |
ce8e922c | 1027 | xfs_buf_ioend( |
1813dd64 DC |
1028 | struct xfs_buf *bp, |
1029 | int schedule) | |
1da177e4 | 1030 | { |
1813dd64 DC |
1031 | bool read = !!(bp->b_flags & XBF_READ); |
1032 | ||
0b1b213f CH |
1033 | trace_xfs_buf_iodone(bp, _RET_IP_); |
1034 | ||
ce8e922c NS |
1035 | if (bp->b_error == 0) |
1036 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1037 | |
1813dd64 | 1038 | if (bp->b_iodone || (read && bp->b_ops) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 1039 | if (schedule) { |
c4028958 | 1040 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 1041 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 1042 | } else { |
c4028958 | 1043 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
1044 | } |
1045 | } else { | |
1813dd64 | 1046 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
b4dd330b | 1047 | complete(&bp->b_iowait); |
1da177e4 LT |
1048 | } |
1049 | } | |
1050 | ||
1da177e4 | 1051 | void |
ce8e922c NS |
1052 | xfs_buf_ioerror( |
1053 | xfs_buf_t *bp, | |
1054 | int error) | |
1da177e4 LT |
1055 | { |
1056 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c | 1057 | bp->b_error = (unsigned short)error; |
0b1b213f | 1058 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1059 | } |
1060 | ||
901796af CH |
1061 | void |
1062 | xfs_buf_ioerror_alert( | |
1063 | struct xfs_buf *bp, | |
1064 | const char *func) | |
1065 | { | |
1066 | xfs_alert(bp->b_target->bt_mount, | |
aa0e8833 DC |
1067 | "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d", |
1068 | (__uint64_t)XFS_BUF_ADDR(bp), func, bp->b_error, bp->b_length); | |
901796af CH |
1069 | } |
1070 | ||
4e23471a CH |
1071 | /* |
1072 | * Called when we want to stop a buffer from getting written or read. | |
1a1a3e97 | 1073 | * We attach the EIO error, muck with its flags, and call xfs_buf_ioend |
4e23471a CH |
1074 | * so that the proper iodone callbacks get called. |
1075 | */ | |
1076 | STATIC int | |
1077 | xfs_bioerror( | |
1078 | xfs_buf_t *bp) | |
1079 | { | |
1080 | #ifdef XFSERRORDEBUG | |
1081 | ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone); | |
1082 | #endif | |
1083 | ||
1084 | /* | |
1085 | * No need to wait until the buffer is unpinned, we aren't flushing it. | |
1086 | */ | |
5a52c2a5 | 1087 | xfs_buf_ioerror(bp, EIO); |
4e23471a CH |
1088 | |
1089 | /* | |
1a1a3e97 | 1090 | * We're calling xfs_buf_ioend, so delete XBF_DONE flag. |
4e23471a CH |
1091 | */ |
1092 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1093 | XFS_BUF_UNDONE(bp); |
c867cb61 | 1094 | xfs_buf_stale(bp); |
4e23471a | 1095 | |
1a1a3e97 | 1096 | xfs_buf_ioend(bp, 0); |
4e23471a CH |
1097 | |
1098 | return EIO; | |
1099 | } | |
1100 | ||
1101 | /* | |
1102 | * Same as xfs_bioerror, except that we are releasing the buffer | |
1a1a3e97 | 1103 | * here ourselves, and avoiding the xfs_buf_ioend call. |
4e23471a CH |
1104 | * This is meant for userdata errors; metadata bufs come with |
1105 | * iodone functions attached, so that we can track down errors. | |
1106 | */ | |
83a0adc3 | 1107 | int |
4e23471a CH |
1108 | xfs_bioerror_relse( |
1109 | struct xfs_buf *bp) | |
1110 | { | |
ed43233b | 1111 | int64_t fl = bp->b_flags; |
4e23471a CH |
1112 | /* |
1113 | * No need to wait until the buffer is unpinned. | |
1114 | * We aren't flushing it. | |
1115 | * | |
1116 | * chunkhold expects B_DONE to be set, whether | |
1117 | * we actually finish the I/O or not. We don't want to | |
1118 | * change that interface. | |
1119 | */ | |
1120 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1121 | XFS_BUF_DONE(bp); |
c867cb61 | 1122 | xfs_buf_stale(bp); |
cb669ca5 | 1123 | bp->b_iodone = NULL; |
0cadda1c | 1124 | if (!(fl & XBF_ASYNC)) { |
4e23471a CH |
1125 | /* |
1126 | * Mark b_error and B_ERROR _both_. | |
1127 | * Lot's of chunkcache code assumes that. | |
1128 | * There's no reason to mark error for | |
1129 | * ASYNC buffers. | |
1130 | */ | |
5a52c2a5 | 1131 | xfs_buf_ioerror(bp, EIO); |
5fde0326 | 1132 | complete(&bp->b_iowait); |
4e23471a CH |
1133 | } else { |
1134 | xfs_buf_relse(bp); | |
1135 | } | |
1136 | ||
1137 | return EIO; | |
1138 | } | |
1139 | ||
a2dcf5df | 1140 | STATIC int |
4e23471a CH |
1141 | xfs_bdstrat_cb( |
1142 | struct xfs_buf *bp) | |
1143 | { | |
ebad861b | 1144 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
4e23471a CH |
1145 | trace_xfs_bdstrat_shut(bp, _RET_IP_); |
1146 | /* | |
1147 | * Metadata write that didn't get logged but | |
1148 | * written delayed anyway. These aren't associated | |
1149 | * with a transaction, and can be ignored. | |
1150 | */ | |
1151 | if (!bp->b_iodone && !XFS_BUF_ISREAD(bp)) | |
1152 | return xfs_bioerror_relse(bp); | |
1153 | else | |
1154 | return xfs_bioerror(bp); | |
1155 | } | |
1156 | ||
1157 | xfs_buf_iorequest(bp); | |
1158 | return 0; | |
1159 | } | |
1160 | ||
a2dcf5df CH |
1161 | int |
1162 | xfs_bwrite( | |
1163 | struct xfs_buf *bp) | |
1164 | { | |
1165 | int error; | |
1166 | ||
1167 | ASSERT(xfs_buf_islocked(bp)); | |
1168 | ||
1169 | bp->b_flags |= XBF_WRITE; | |
ac8809f9 | 1170 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q | XBF_WRITE_FAIL); |
a2dcf5df CH |
1171 | |
1172 | xfs_bdstrat_cb(bp); | |
1173 | ||
1174 | error = xfs_buf_iowait(bp); | |
1175 | if (error) { | |
1176 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1177 | SHUTDOWN_META_IO_ERROR); | |
1178 | } | |
1179 | return error; | |
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 | |
37eb17e6 DC |
1198 | /* |
1199 | * don't overwrite existing errors - otherwise we can lose errors on | |
1200 | * buffers that require multiple bios to complete. | |
1201 | */ | |
1202 | if (!bp->b_error) | |
1203 | xfs_buf_ioerror(bp, -error); | |
1da177e4 | 1204 | |
37eb17e6 | 1205 | if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
73c77e2c JB |
1206 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); |
1207 | ||
ce8e922c | 1208 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1209 | bio_put(bio); |
1da177e4 LT |
1210 | } |
1211 | ||
3e85c868 DC |
1212 | static void |
1213 | xfs_buf_ioapply_map( | |
1214 | struct xfs_buf *bp, | |
1215 | int map, | |
1216 | int *buf_offset, | |
1217 | int *count, | |
1218 | int rw) | |
1da177e4 | 1219 | { |
3e85c868 DC |
1220 | int page_index; |
1221 | int total_nr_pages = bp->b_page_count; | |
1222 | int nr_pages; | |
1223 | struct bio *bio; | |
1224 | sector_t sector = bp->b_maps[map].bm_bn; | |
1225 | int size; | |
1226 | int offset; | |
1da177e4 | 1227 | |
ce8e922c | 1228 | total_nr_pages = bp->b_page_count; |
1da177e4 | 1229 | |
3e85c868 DC |
1230 | /* skip the pages in the buffer before the start offset */ |
1231 | page_index = 0; | |
1232 | offset = *buf_offset; | |
1233 | while (offset >= PAGE_SIZE) { | |
1234 | page_index++; | |
1235 | offset -= PAGE_SIZE; | |
f538d4da CH |
1236 | } |
1237 | ||
3e85c868 DC |
1238 | /* |
1239 | * Limit the IO size to the length of the current vector, and update the | |
1240 | * remaining IO count for the next time around. | |
1241 | */ | |
1242 | size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count); | |
1243 | *count -= size; | |
1244 | *buf_offset += size; | |
34951f5c | 1245 | |
1da177e4 | 1246 | next_chunk: |
ce8e922c | 1247 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1248 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1249 | if (nr_pages > total_nr_pages) | |
1250 | nr_pages = total_nr_pages; | |
1251 | ||
1252 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1253 | bio->bi_bdev = bp->b_target->bt_bdev; |
4f024f37 | 1254 | bio->bi_iter.bi_sector = sector; |
ce8e922c NS |
1255 | bio->bi_end_io = xfs_buf_bio_end_io; |
1256 | bio->bi_private = bp; | |
1da177e4 | 1257 | |
0e6e847f | 1258 | |
3e85c868 | 1259 | for (; size && nr_pages; nr_pages--, page_index++) { |
0e6e847f | 1260 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1261 | |
1262 | if (nbytes > size) | |
1263 | nbytes = size; | |
1264 | ||
3e85c868 DC |
1265 | rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes, |
1266 | offset); | |
ce8e922c | 1267 | if (rbytes < nbytes) |
1da177e4 LT |
1268 | break; |
1269 | ||
1270 | offset = 0; | |
aa0e8833 | 1271 | sector += BTOBB(nbytes); |
1da177e4 LT |
1272 | size -= nbytes; |
1273 | total_nr_pages--; | |
1274 | } | |
1275 | ||
4f024f37 | 1276 | if (likely(bio->bi_iter.bi_size)) { |
73c77e2c JB |
1277 | if (xfs_buf_is_vmapped(bp)) { |
1278 | flush_kernel_vmap_range(bp->b_addr, | |
1279 | xfs_buf_vmap_len(bp)); | |
1280 | } | |
1da177e4 LT |
1281 | submit_bio(rw, bio); |
1282 | if (size) | |
1283 | goto next_chunk; | |
1284 | } else { | |
37eb17e6 DC |
1285 | /* |
1286 | * This is guaranteed not to be the last io reference count | |
1287 | * because the caller (xfs_buf_iorequest) holds a count itself. | |
1288 | */ | |
1289 | atomic_dec(&bp->b_io_remaining); | |
ce8e922c | 1290 | xfs_buf_ioerror(bp, EIO); |
ec53d1db | 1291 | bio_put(bio); |
1da177e4 | 1292 | } |
3e85c868 DC |
1293 | |
1294 | } | |
1295 | ||
1296 | STATIC void | |
1297 | _xfs_buf_ioapply( | |
1298 | struct xfs_buf *bp) | |
1299 | { | |
1300 | struct blk_plug plug; | |
1301 | int rw; | |
1302 | int offset; | |
1303 | int size; | |
1304 | int i; | |
1305 | ||
c163f9a1 DC |
1306 | /* |
1307 | * Make sure we capture only current IO errors rather than stale errors | |
1308 | * left over from previous use of the buffer (e.g. failed readahead). | |
1309 | */ | |
1310 | bp->b_error = 0; | |
1311 | ||
3e85c868 DC |
1312 | if (bp->b_flags & XBF_WRITE) { |
1313 | if (bp->b_flags & XBF_SYNCIO) | |
1314 | rw = WRITE_SYNC; | |
1315 | else | |
1316 | rw = WRITE; | |
1317 | if (bp->b_flags & XBF_FUA) | |
1318 | rw |= REQ_FUA; | |
1319 | if (bp->b_flags & XBF_FLUSH) | |
1320 | rw |= REQ_FLUSH; | |
1813dd64 DC |
1321 | |
1322 | /* | |
1323 | * Run the write verifier callback function if it exists. If | |
1324 | * this function fails it will mark the buffer with an error and | |
1325 | * the IO should not be dispatched. | |
1326 | */ | |
1327 | if (bp->b_ops) { | |
1328 | bp->b_ops->verify_write(bp); | |
1329 | if (bp->b_error) { | |
1330 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1331 | SHUTDOWN_CORRUPT_INCORE); | |
1332 | return; | |
1333 | } | |
1334 | } | |
3e85c868 DC |
1335 | } else if (bp->b_flags & XBF_READ_AHEAD) { |
1336 | rw = READA; | |
1337 | } else { | |
1338 | rw = READ; | |
1339 | } | |
1340 | ||
1341 | /* we only use the buffer cache for meta-data */ | |
1342 | rw |= REQ_META; | |
1343 | ||
1344 | /* | |
1345 | * Walk all the vectors issuing IO on them. Set up the initial offset | |
1346 | * into the buffer and the desired IO size before we start - | |
1347 | * _xfs_buf_ioapply_vec() will modify them appropriately for each | |
1348 | * subsequent call. | |
1349 | */ | |
1350 | offset = bp->b_offset; | |
1351 | size = BBTOB(bp->b_io_length); | |
1352 | blk_start_plug(&plug); | |
1353 | for (i = 0; i < bp->b_map_count; i++) { | |
1354 | xfs_buf_ioapply_map(bp, i, &offset, &size, rw); | |
1355 | if (bp->b_error) | |
1356 | break; | |
1357 | if (size <= 0) | |
1358 | break; /* all done */ | |
1359 | } | |
1360 | blk_finish_plug(&plug); | |
1da177e4 LT |
1361 | } |
1362 | ||
0e95f19a | 1363 | void |
ce8e922c NS |
1364 | xfs_buf_iorequest( |
1365 | xfs_buf_t *bp) | |
1da177e4 | 1366 | { |
0b1b213f | 1367 | trace_xfs_buf_iorequest(bp, _RET_IP_); |
1da177e4 | 1368 | |
43ff2122 | 1369 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
1da177e4 | 1370 | |
375ec69d | 1371 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1372 | xfs_buf_wait_unpin(bp); |
ce8e922c | 1373 | xfs_buf_hold(bp); |
1da177e4 | 1374 | |
8d6c1210 ES |
1375 | /* |
1376 | * Set the count to 1 initially, this will stop an I/O | |
1da177e4 | 1377 | * completion callout which happens before we have started |
ce8e922c | 1378 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1379 | */ |
ce8e922c NS |
1380 | atomic_set(&bp->b_io_remaining, 1); |
1381 | _xfs_buf_ioapply(bp); | |
8d6c1210 ES |
1382 | /* |
1383 | * If _xfs_buf_ioapply failed, we'll get back here with | |
1384 | * only the reference we took above. _xfs_buf_ioend will | |
1385 | * drop it to zero, so we'd better not queue it for later, | |
1386 | * or we'll free it before it's done. | |
1387 | */ | |
1388 | _xfs_buf_ioend(bp, bp->b_error ? 0 : 1); | |
1da177e4 | 1389 | |
ce8e922c | 1390 | xfs_buf_rele(bp); |
1da177e4 LT |
1391 | } |
1392 | ||
1393 | /* | |
0e95f19a | 1394 | * Waits for I/O to complete on the buffer supplied. It returns immediately if |
8d6c1210 ES |
1395 | * no I/O is pending or there is already a pending error on the buffer, in which |
1396 | * case nothing will ever complete. It returns the I/O error code, if any, or | |
1397 | * 0 if there was no error. | |
1da177e4 LT |
1398 | */ |
1399 | int | |
ce8e922c NS |
1400 | xfs_buf_iowait( |
1401 | xfs_buf_t *bp) | |
1da177e4 | 1402 | { |
0b1b213f CH |
1403 | trace_xfs_buf_iowait(bp, _RET_IP_); |
1404 | ||
0e95f19a DC |
1405 | if (!bp->b_error) |
1406 | wait_for_completion(&bp->b_iowait); | |
0b1b213f CH |
1407 | |
1408 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
ce8e922c | 1409 | return bp->b_error; |
1da177e4 LT |
1410 | } |
1411 | ||
ce8e922c NS |
1412 | xfs_caddr_t |
1413 | xfs_buf_offset( | |
1414 | xfs_buf_t *bp, | |
1da177e4 LT |
1415 | size_t offset) |
1416 | { | |
1417 | struct page *page; | |
1418 | ||
611c9946 | 1419 | if (bp->b_addr) |
62926044 | 1420 | return bp->b_addr + offset; |
1da177e4 | 1421 | |
ce8e922c | 1422 | offset += bp->b_offset; |
0e6e847f DC |
1423 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
1424 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_SIZE-1)); | |
1da177e4 LT |
1425 | } |
1426 | ||
1427 | /* | |
1da177e4 LT |
1428 | * Move data into or out of a buffer. |
1429 | */ | |
1430 | void | |
ce8e922c NS |
1431 | xfs_buf_iomove( |
1432 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1433 | size_t boff, /* starting buffer offset */ |
1434 | size_t bsize, /* length to copy */ | |
b9c48649 | 1435 | void *data, /* data address */ |
ce8e922c | 1436 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 | 1437 | { |
795cac72 | 1438 | size_t bend; |
1da177e4 LT |
1439 | |
1440 | bend = boff + bsize; | |
1441 | while (boff < bend) { | |
795cac72 DC |
1442 | struct page *page; |
1443 | int page_index, page_offset, csize; | |
1444 | ||
1445 | page_index = (boff + bp->b_offset) >> PAGE_SHIFT; | |
1446 | page_offset = (boff + bp->b_offset) & ~PAGE_MASK; | |
1447 | page = bp->b_pages[page_index]; | |
1448 | csize = min_t(size_t, PAGE_SIZE - page_offset, | |
1449 | BBTOB(bp->b_io_length) - boff); | |
1da177e4 | 1450 | |
795cac72 | 1451 | ASSERT((csize + page_offset) <= PAGE_SIZE); |
1da177e4 LT |
1452 | |
1453 | switch (mode) { | |
ce8e922c | 1454 | case XBRW_ZERO: |
795cac72 | 1455 | memset(page_address(page) + page_offset, 0, csize); |
1da177e4 | 1456 | break; |
ce8e922c | 1457 | case XBRW_READ: |
795cac72 | 1458 | memcpy(data, page_address(page) + page_offset, csize); |
1da177e4 | 1459 | break; |
ce8e922c | 1460 | case XBRW_WRITE: |
795cac72 | 1461 | memcpy(page_address(page) + page_offset, data, csize); |
1da177e4 LT |
1462 | } |
1463 | ||
1464 | boff += csize; | |
1465 | data += csize; | |
1466 | } | |
1467 | } | |
1468 | ||
1469 | /* | |
ce8e922c | 1470 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1471 | */ |
1472 | ||
1473 | /* | |
430cbeb8 DC |
1474 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1475 | * returned. These buffers will have an elevated hold count, so wait on those | |
1476 | * while freeing all the buffers only held by the LRU. | |
1da177e4 | 1477 | */ |
e80dfa19 DC |
1478 | static enum lru_status |
1479 | xfs_buftarg_wait_rele( | |
1480 | struct list_head *item, | |
1481 | spinlock_t *lru_lock, | |
1482 | void *arg) | |
1483 | ||
1da177e4 | 1484 | { |
e80dfa19 | 1485 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); |
a4082357 | 1486 | struct list_head *dispose = arg; |
430cbeb8 | 1487 | |
e80dfa19 | 1488 | if (atomic_read(&bp->b_hold) > 1) { |
a4082357 | 1489 | /* need to wait, so skip it this pass */ |
e80dfa19 | 1490 | trace_xfs_buf_wait_buftarg(bp, _RET_IP_); |
a4082357 | 1491 | return LRU_SKIP; |
1da177e4 | 1492 | } |
a4082357 DC |
1493 | if (!spin_trylock(&bp->b_lock)) |
1494 | return LRU_SKIP; | |
e80dfa19 | 1495 | |
a4082357 DC |
1496 | /* |
1497 | * clear the LRU reference count so the buffer doesn't get | |
1498 | * ignored in xfs_buf_rele(). | |
1499 | */ | |
1500 | atomic_set(&bp->b_lru_ref, 0); | |
1501 | bp->b_state |= XFS_BSTATE_DISPOSE; | |
1502 | list_move(item, dispose); | |
1503 | spin_unlock(&bp->b_lock); | |
1504 | return LRU_REMOVED; | |
1da177e4 LT |
1505 | } |
1506 | ||
e80dfa19 DC |
1507 | void |
1508 | xfs_wait_buftarg( | |
1509 | struct xfs_buftarg *btp) | |
1510 | { | |
a4082357 DC |
1511 | LIST_HEAD(dispose); |
1512 | int loop = 0; | |
1513 | ||
1514 | /* loop until there is nothing left on the lru list. */ | |
1515 | while (list_lru_count(&btp->bt_lru)) { | |
e80dfa19 | 1516 | list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele, |
a4082357 DC |
1517 | &dispose, LONG_MAX); |
1518 | ||
1519 | while (!list_empty(&dispose)) { | |
1520 | struct xfs_buf *bp; | |
1521 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1522 | list_del_init(&bp->b_lru); | |
ac8809f9 DC |
1523 | if (bp->b_flags & XBF_WRITE_FAIL) { |
1524 | xfs_alert(btp->bt_mount, | |
1525 | "Corruption Alert: Buffer at block 0x%llx had permanent write failures!\n" | |
1526 | "Please run xfs_repair to determine the extent of the problem.", | |
1527 | (long long)bp->b_bn); | |
1528 | } | |
a4082357 DC |
1529 | xfs_buf_rele(bp); |
1530 | } | |
1531 | if (loop++ != 0) | |
1532 | delay(100); | |
1533 | } | |
e80dfa19 DC |
1534 | } |
1535 | ||
1536 | static enum lru_status | |
1537 | xfs_buftarg_isolate( | |
1538 | struct list_head *item, | |
1539 | spinlock_t *lru_lock, | |
1540 | void *arg) | |
1541 | { | |
1542 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); | |
1543 | struct list_head *dispose = arg; | |
1544 | ||
a4082357 DC |
1545 | /* |
1546 | * we are inverting the lru lock/bp->b_lock here, so use a trylock. | |
1547 | * If we fail to get the lock, just skip it. | |
1548 | */ | |
1549 | if (!spin_trylock(&bp->b_lock)) | |
1550 | return LRU_SKIP; | |
e80dfa19 DC |
1551 | /* |
1552 | * Decrement the b_lru_ref count unless the value is already | |
1553 | * zero. If the value is already zero, we need to reclaim the | |
1554 | * buffer, otherwise it gets another trip through the LRU. | |
1555 | */ | |
a4082357 DC |
1556 | if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) { |
1557 | spin_unlock(&bp->b_lock); | |
e80dfa19 | 1558 | return LRU_ROTATE; |
a4082357 | 1559 | } |
e80dfa19 | 1560 | |
a4082357 | 1561 | bp->b_state |= XFS_BSTATE_DISPOSE; |
e80dfa19 | 1562 | list_move(item, dispose); |
a4082357 | 1563 | spin_unlock(&bp->b_lock); |
e80dfa19 DC |
1564 | return LRU_REMOVED; |
1565 | } | |
1566 | ||
addbda40 | 1567 | static unsigned long |
e80dfa19 | 1568 | xfs_buftarg_shrink_scan( |
ff57ab21 | 1569 | struct shrinker *shrink, |
1495f230 | 1570 | struct shrink_control *sc) |
a6867a68 | 1571 | { |
ff57ab21 DC |
1572 | struct xfs_buftarg *btp = container_of(shrink, |
1573 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1574 | LIST_HEAD(dispose); |
addbda40 | 1575 | unsigned long freed; |
e80dfa19 | 1576 | unsigned long nr_to_scan = sc->nr_to_scan; |
430cbeb8 | 1577 | |
e80dfa19 DC |
1578 | freed = list_lru_walk_node(&btp->bt_lru, sc->nid, xfs_buftarg_isolate, |
1579 | &dispose, &nr_to_scan); | |
430cbeb8 DC |
1580 | |
1581 | while (!list_empty(&dispose)) { | |
e80dfa19 | 1582 | struct xfs_buf *bp; |
430cbeb8 DC |
1583 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); |
1584 | list_del_init(&bp->b_lru); | |
1585 | xfs_buf_rele(bp); | |
1586 | } | |
1587 | ||
e80dfa19 DC |
1588 | return freed; |
1589 | } | |
1590 | ||
addbda40 | 1591 | static unsigned long |
e80dfa19 DC |
1592 | xfs_buftarg_shrink_count( |
1593 | struct shrinker *shrink, | |
1594 | struct shrink_control *sc) | |
1595 | { | |
1596 | struct xfs_buftarg *btp = container_of(shrink, | |
1597 | struct xfs_buftarg, bt_shrinker); | |
1598 | return list_lru_count_node(&btp->bt_lru, sc->nid); | |
a6867a68 DC |
1599 | } |
1600 | ||
1da177e4 LT |
1601 | void |
1602 | xfs_free_buftarg( | |
b7963133 CH |
1603 | struct xfs_mount *mp, |
1604 | struct xfs_buftarg *btp) | |
1da177e4 | 1605 | { |
ff57ab21 | 1606 | unregister_shrinker(&btp->bt_shrinker); |
f5e1dd34 | 1607 | list_lru_destroy(&btp->bt_lru); |
ff57ab21 | 1608 | |
b7963133 CH |
1609 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1610 | xfs_blkdev_issue_flush(btp); | |
a6867a68 | 1611 | |
f0e2d93c | 1612 | kmem_free(btp); |
1da177e4 LT |
1613 | } |
1614 | ||
3fefdeee ES |
1615 | int |
1616 | xfs_setsize_buftarg( | |
1da177e4 LT |
1617 | xfs_buftarg_t *btp, |
1618 | unsigned int blocksize, | |
3fefdeee | 1619 | unsigned int sectorsize) |
1da177e4 | 1620 | { |
7c71ee78 | 1621 | /* Set up metadata sector size info */ |
6da54179 ES |
1622 | btp->bt_meta_sectorsize = sectorsize; |
1623 | btp->bt_meta_sectormask = sectorsize - 1; | |
1da177e4 | 1624 | |
ce8e922c | 1625 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
02b102df CH |
1626 | char name[BDEVNAME_SIZE]; |
1627 | ||
1628 | bdevname(btp->bt_bdev, name); | |
1629 | ||
4f10700a | 1630 | xfs_warn(btp->bt_mount, |
08e96e1a | 1631 | "Cannot set_blocksize to %u on device %s", |
02b102df | 1632 | sectorsize, name); |
1da177e4 LT |
1633 | return EINVAL; |
1634 | } | |
1635 | ||
7c71ee78 ES |
1636 | /* Set up device logical sector size mask */ |
1637 | btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev); | |
1638 | btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1; | |
1639 | ||
1da177e4 LT |
1640 | return 0; |
1641 | } | |
1642 | ||
1643 | /* | |
3fefdeee ES |
1644 | * When allocating the initial buffer target we have not yet |
1645 | * read in the superblock, so don't know what sized sectors | |
1646 | * are being used at this early stage. Play safe. | |
ce8e922c | 1647 | */ |
1da177e4 LT |
1648 | STATIC int |
1649 | xfs_setsize_buftarg_early( | |
1650 | xfs_buftarg_t *btp, | |
1651 | struct block_device *bdev) | |
1652 | { | |
3fefdeee ES |
1653 | return xfs_setsize_buftarg(btp, PAGE_SIZE, |
1654 | bdev_logical_block_size(bdev)); | |
1da177e4 LT |
1655 | } |
1656 | ||
1da177e4 LT |
1657 | xfs_buftarg_t * |
1658 | xfs_alloc_buftarg( | |
ebad861b | 1659 | struct xfs_mount *mp, |
1da177e4 | 1660 | struct block_device *bdev, |
e2a07812 JE |
1661 | int external, |
1662 | const char *fsname) | |
1da177e4 LT |
1663 | { |
1664 | xfs_buftarg_t *btp; | |
1665 | ||
b17cb364 | 1666 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP | KM_NOFS); |
1da177e4 | 1667 | |
ebad861b | 1668 | btp->bt_mount = mp; |
ce8e922c NS |
1669 | btp->bt_dev = bdev->bd_dev; |
1670 | btp->bt_bdev = bdev; | |
0e6e847f DC |
1671 | btp->bt_bdi = blk_get_backing_dev_info(bdev); |
1672 | if (!btp->bt_bdi) | |
1673 | goto error; | |
1674 | ||
1da177e4 LT |
1675 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1676 | goto error; | |
5ca302c8 GC |
1677 | |
1678 | if (list_lru_init(&btp->bt_lru)) | |
1679 | goto error; | |
1680 | ||
e80dfa19 DC |
1681 | btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count; |
1682 | btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan; | |
ff57ab21 | 1683 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; |
e80dfa19 | 1684 | btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE; |
ff57ab21 | 1685 | register_shrinker(&btp->bt_shrinker); |
1da177e4 LT |
1686 | return btp; |
1687 | ||
1688 | error: | |
f0e2d93c | 1689 | kmem_free(btp); |
1da177e4 LT |
1690 | return NULL; |
1691 | } | |
1692 | ||
1da177e4 | 1693 | /* |
43ff2122 CH |
1694 | * Add a buffer to the delayed write list. |
1695 | * | |
1696 | * This queues a buffer for writeout if it hasn't already been. Note that | |
1697 | * neither this routine nor the buffer list submission functions perform | |
1698 | * any internal synchronization. It is expected that the lists are thread-local | |
1699 | * to the callers. | |
1700 | * | |
1701 | * Returns true if we queued up the buffer, or false if it already had | |
1702 | * been on the buffer list. | |
1da177e4 | 1703 | */ |
43ff2122 | 1704 | bool |
ce8e922c | 1705 | xfs_buf_delwri_queue( |
43ff2122 CH |
1706 | struct xfs_buf *bp, |
1707 | struct list_head *list) | |
1da177e4 | 1708 | { |
43ff2122 | 1709 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 1710 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1711 | |
43ff2122 CH |
1712 | /* |
1713 | * If the buffer is already marked delwri it already is queued up | |
1714 | * by someone else for imediate writeout. Just ignore it in that | |
1715 | * case. | |
1716 | */ | |
1717 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
1718 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
1719 | return false; | |
1da177e4 | 1720 | } |
1da177e4 | 1721 | |
43ff2122 | 1722 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
1723 | |
1724 | /* | |
43ff2122 CH |
1725 | * If a buffer gets written out synchronously or marked stale while it |
1726 | * is on a delwri list we lazily remove it. To do this, the other party | |
1727 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
1728 | * It remains referenced and on the list. In a rare corner case it | |
1729 | * might get readded to a delwri list after the synchronous writeout, in | |
1730 | * which case we need just need to re-add the flag here. | |
d808f617 | 1731 | */ |
43ff2122 CH |
1732 | bp->b_flags |= _XBF_DELWRI_Q; |
1733 | if (list_empty(&bp->b_list)) { | |
1734 | atomic_inc(&bp->b_hold); | |
1735 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 1736 | } |
585e6d88 | 1737 | |
43ff2122 | 1738 | return true; |
585e6d88 DC |
1739 | } |
1740 | ||
089716aa DC |
1741 | /* |
1742 | * Compare function is more complex than it needs to be because | |
1743 | * the return value is only 32 bits and we are doing comparisons | |
1744 | * on 64 bit values | |
1745 | */ | |
1746 | static int | |
1747 | xfs_buf_cmp( | |
1748 | void *priv, | |
1749 | struct list_head *a, | |
1750 | struct list_head *b) | |
1751 | { | |
1752 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1753 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1754 | xfs_daddr_t diff; | |
1755 | ||
f4b42421 | 1756 | diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn; |
089716aa DC |
1757 | if (diff < 0) |
1758 | return -1; | |
1759 | if (diff > 0) | |
1760 | return 1; | |
1761 | return 0; | |
1762 | } | |
1763 | ||
43ff2122 CH |
1764 | static int |
1765 | __xfs_buf_delwri_submit( | |
1766 | struct list_head *buffer_list, | |
1767 | struct list_head *io_list, | |
1768 | bool wait) | |
1da177e4 | 1769 | { |
43ff2122 CH |
1770 | struct blk_plug plug; |
1771 | struct xfs_buf *bp, *n; | |
1772 | int pinned = 0; | |
1773 | ||
1774 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { | |
1775 | if (!wait) { | |
1776 | if (xfs_buf_ispinned(bp)) { | |
1777 | pinned++; | |
1778 | continue; | |
1779 | } | |
1780 | if (!xfs_buf_trylock(bp)) | |
1781 | continue; | |
1782 | } else { | |
1783 | xfs_buf_lock(bp); | |
1784 | } | |
978c7b2f | 1785 | |
43ff2122 CH |
1786 | /* |
1787 | * Someone else might have written the buffer synchronously or | |
1788 | * marked it stale in the meantime. In that case only the | |
1789 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
1790 | * reference and remove it from the list here. | |
1791 | */ | |
1792 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
1793 | list_del_init(&bp->b_list); | |
1794 | xfs_buf_relse(bp); | |
1795 | continue; | |
1796 | } | |
c9c12971 | 1797 | |
43ff2122 CH |
1798 | list_move_tail(&bp->b_list, io_list); |
1799 | trace_xfs_buf_delwri_split(bp, _RET_IP_); | |
1800 | } | |
1da177e4 | 1801 | |
43ff2122 | 1802 | list_sort(NULL, io_list, xfs_buf_cmp); |
1da177e4 | 1803 | |
43ff2122 CH |
1804 | blk_start_plug(&plug); |
1805 | list_for_each_entry_safe(bp, n, io_list, b_list) { | |
ac8809f9 | 1806 | bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_ASYNC | XBF_WRITE_FAIL); |
43ff2122 | 1807 | bp->b_flags |= XBF_WRITE; |
a1b7ea5d | 1808 | |
43ff2122 CH |
1809 | if (!wait) { |
1810 | bp->b_flags |= XBF_ASYNC; | |
ce8e922c | 1811 | list_del_init(&bp->b_list); |
1da177e4 | 1812 | } |
43ff2122 CH |
1813 | xfs_bdstrat_cb(bp); |
1814 | } | |
1815 | blk_finish_plug(&plug); | |
1da177e4 | 1816 | |
43ff2122 | 1817 | return pinned; |
1da177e4 LT |
1818 | } |
1819 | ||
1820 | /* | |
43ff2122 CH |
1821 | * Write out a buffer list asynchronously. |
1822 | * | |
1823 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
1824 | * out and not wait for I/O completion on any of the buffers. This interface | |
1825 | * is only safely useable for callers that can track I/O completion by higher | |
1826 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
1827 | * function. | |
1da177e4 LT |
1828 | */ |
1829 | int | |
43ff2122 CH |
1830 | xfs_buf_delwri_submit_nowait( |
1831 | struct list_head *buffer_list) | |
1da177e4 | 1832 | { |
43ff2122 CH |
1833 | LIST_HEAD (io_list); |
1834 | return __xfs_buf_delwri_submit(buffer_list, &io_list, false); | |
1835 | } | |
1da177e4 | 1836 | |
43ff2122 CH |
1837 | /* |
1838 | * Write out a buffer list synchronously. | |
1839 | * | |
1840 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
1841 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
1842 | * so callers must have some other way of tracking buffers if they require such | |
1843 | * functionality. | |
1844 | */ | |
1845 | int | |
1846 | xfs_buf_delwri_submit( | |
1847 | struct list_head *buffer_list) | |
1848 | { | |
1849 | LIST_HEAD (io_list); | |
1850 | int error = 0, error2; | |
1851 | struct xfs_buf *bp; | |
1da177e4 | 1852 | |
43ff2122 | 1853 | __xfs_buf_delwri_submit(buffer_list, &io_list, true); |
1da177e4 | 1854 | |
43ff2122 CH |
1855 | /* Wait for IO to complete. */ |
1856 | while (!list_empty(&io_list)) { | |
1857 | bp = list_first_entry(&io_list, struct xfs_buf, b_list); | |
a1b7ea5d | 1858 | |
089716aa | 1859 | list_del_init(&bp->b_list); |
43ff2122 CH |
1860 | error2 = xfs_buf_iowait(bp); |
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 | } |