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> |
089716aa | 36 | #include <linux/list_sort.h> |
1da177e4 | 37 | |
b7963133 CH |
38 | #include "xfs_sb.h" |
39 | #include "xfs_inum.h" | |
ed3b4d6c | 40 | #include "xfs_log.h" |
b7963133 | 41 | #include "xfs_ag.h" |
b7963133 | 42 | #include "xfs_mount.h" |
0b1b213f | 43 | #include "xfs_trace.h" |
b7963133 | 44 | |
7989cb8e | 45 | static kmem_zone_t *xfs_buf_zone; |
a6867a68 | 46 | STATIC int xfsbufd(void *); |
ce8e922c | 47 | STATIC void xfs_buf_delwri_queue(xfs_buf_t *, int); |
23ea4032 | 48 | |
7989cb8e | 49 | static struct workqueue_struct *xfslogd_workqueue; |
0829c360 | 50 | struct workqueue_struct *xfsdatad_workqueue; |
c626d174 | 51 | struct workqueue_struct *xfsconvertd_workqueue; |
1da177e4 | 52 | |
ce8e922c NS |
53 | #ifdef XFS_BUF_LOCK_TRACKING |
54 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
55 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
56 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 57 | #else |
ce8e922c NS |
58 | # define XB_SET_OWNER(bp) do { } while (0) |
59 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
60 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
61 | #endif |
62 | ||
ce8e922c NS |
63 | #define xb_to_gfp(flags) \ |
64 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \ | |
65 | ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN) | |
1da177e4 | 66 | |
ce8e922c NS |
67 | #define xb_to_km(flags) \ |
68 | (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP) | |
1da177e4 | 69 | |
ce8e922c NS |
70 | #define xfs_buf_allocate(flags) \ |
71 | kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags)) | |
72 | #define xfs_buf_deallocate(bp) \ | |
73 | kmem_zone_free(xfs_buf_zone, (bp)); | |
1da177e4 | 74 | |
73c77e2c JB |
75 | static inline int |
76 | xfs_buf_is_vmapped( | |
77 | struct xfs_buf *bp) | |
78 | { | |
79 | /* | |
80 | * Return true if the buffer is vmapped. | |
81 | * | |
82 | * The XBF_MAPPED flag is set if the buffer should be mapped, but the | |
83 | * code is clever enough to know it doesn't have to map a single page, | |
84 | * so the check has to be both for XBF_MAPPED and bp->b_page_count > 1. | |
85 | */ | |
86 | return (bp->b_flags & XBF_MAPPED) && bp->b_page_count > 1; | |
87 | } | |
88 | ||
89 | static inline int | |
90 | xfs_buf_vmap_len( | |
91 | struct xfs_buf *bp) | |
92 | { | |
93 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
94 | } | |
95 | ||
1da177e4 | 96 | /* |
ce8e922c | 97 | * Page Region interfaces. |
1da177e4 | 98 | * |
ce8e922c NS |
99 | * For pages in filesystems where the blocksize is smaller than the |
100 | * pagesize, we use the page->private field (long) to hold a bitmap | |
101 | * of uptodate regions within the page. | |
1da177e4 | 102 | * |
ce8e922c | 103 | * Each such region is "bytes per page / bits per long" bytes long. |
1da177e4 | 104 | * |
ce8e922c NS |
105 | * NBPPR == number-of-bytes-per-page-region |
106 | * BTOPR == bytes-to-page-region (rounded up) | |
107 | * BTOPRT == bytes-to-page-region-truncated (rounded down) | |
1da177e4 LT |
108 | */ |
109 | #if (BITS_PER_LONG == 32) | |
110 | #define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */ | |
111 | #elif (BITS_PER_LONG == 64) | |
112 | #define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */ | |
113 | #else | |
114 | #error BITS_PER_LONG must be 32 or 64 | |
115 | #endif | |
116 | #define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG) | |
117 | #define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT) | |
118 | #define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT)) | |
119 | ||
120 | STATIC unsigned long | |
121 | page_region_mask( | |
122 | size_t offset, | |
123 | size_t length) | |
124 | { | |
125 | unsigned long mask; | |
126 | int first, final; | |
127 | ||
128 | first = BTOPR(offset); | |
129 | final = BTOPRT(offset + length - 1); | |
130 | first = min(first, final); | |
131 | ||
132 | mask = ~0UL; | |
133 | mask <<= BITS_PER_LONG - (final - first); | |
134 | mask >>= BITS_PER_LONG - (final); | |
135 | ||
136 | ASSERT(offset + length <= PAGE_CACHE_SIZE); | |
137 | ASSERT((final - first) < BITS_PER_LONG && (final - first) >= 0); | |
138 | ||
139 | return mask; | |
140 | } | |
141 | ||
b8f82a4a | 142 | STATIC void |
1da177e4 LT |
143 | set_page_region( |
144 | struct page *page, | |
145 | size_t offset, | |
146 | size_t length) | |
147 | { | |
4c21e2f2 HD |
148 | set_page_private(page, |
149 | page_private(page) | page_region_mask(offset, length)); | |
150 | if (page_private(page) == ~0UL) | |
1da177e4 LT |
151 | SetPageUptodate(page); |
152 | } | |
153 | ||
b8f82a4a | 154 | STATIC int |
1da177e4 LT |
155 | test_page_region( |
156 | struct page *page, | |
157 | size_t offset, | |
158 | size_t length) | |
159 | { | |
160 | unsigned long mask = page_region_mask(offset, length); | |
161 | ||
4c21e2f2 | 162 | return (mask && (page_private(page) & mask) == mask); |
1da177e4 LT |
163 | } |
164 | ||
1da177e4 | 165 | /* |
430cbeb8 DC |
166 | * xfs_buf_lru_add - add a buffer to the LRU. |
167 | * | |
168 | * The LRU takes a new reference to the buffer so that it will only be freed | |
169 | * once the shrinker takes the buffer off the LRU. | |
170 | */ | |
171 | STATIC void | |
172 | xfs_buf_lru_add( | |
173 | struct xfs_buf *bp) | |
174 | { | |
175 | struct xfs_buftarg *btp = bp->b_target; | |
176 | ||
177 | spin_lock(&btp->bt_lru_lock); | |
178 | if (list_empty(&bp->b_lru)) { | |
179 | atomic_inc(&bp->b_hold); | |
180 | list_add_tail(&bp->b_lru, &btp->bt_lru); | |
181 | btp->bt_lru_nr++; | |
182 | } | |
183 | spin_unlock(&btp->bt_lru_lock); | |
184 | } | |
185 | ||
186 | /* | |
187 | * xfs_buf_lru_del - remove a buffer from the LRU | |
188 | * | |
189 | * The unlocked check is safe here because it only occurs when there are not | |
190 | * b_lru_ref counts left on the inode under the pag->pag_buf_lock. it is there | |
191 | * to optimise the shrinker removing the buffer from the LRU and calling | |
192 | * xfs_buf_free(). i.e. it removes an unneccessary round trip on the | |
193 | * bt_lru_lock. | |
1da177e4 | 194 | */ |
430cbeb8 DC |
195 | STATIC void |
196 | xfs_buf_lru_del( | |
197 | struct xfs_buf *bp) | |
198 | { | |
199 | struct xfs_buftarg *btp = bp->b_target; | |
200 | ||
201 | if (list_empty(&bp->b_lru)) | |
202 | return; | |
203 | ||
204 | spin_lock(&btp->bt_lru_lock); | |
205 | if (!list_empty(&bp->b_lru)) { | |
206 | list_del_init(&bp->b_lru); | |
207 | btp->bt_lru_nr--; | |
208 | } | |
209 | spin_unlock(&btp->bt_lru_lock); | |
210 | } | |
211 | ||
212 | /* | |
213 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
214 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
215 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
216 | * to remove the reference that LRU holds on the buffer. | |
217 | * | |
218 | * This prevents build-up of stale buffers on the LRU. | |
219 | */ | |
220 | void | |
221 | xfs_buf_stale( | |
222 | struct xfs_buf *bp) | |
223 | { | |
224 | bp->b_flags |= XBF_STALE; | |
225 | atomic_set(&(bp)->b_lru_ref, 0); | |
226 | if (!list_empty(&bp->b_lru)) { | |
227 | struct xfs_buftarg *btp = bp->b_target; | |
228 | ||
229 | spin_lock(&btp->bt_lru_lock); | |
230 | if (!list_empty(&bp->b_lru)) { | |
231 | list_del_init(&bp->b_lru); | |
232 | btp->bt_lru_nr--; | |
233 | atomic_dec(&bp->b_hold); | |
234 | } | |
235 | spin_unlock(&btp->bt_lru_lock); | |
236 | } | |
237 | ASSERT(atomic_read(&bp->b_hold) >= 1); | |
238 | } | |
1da177e4 LT |
239 | |
240 | STATIC void | |
ce8e922c NS |
241 | _xfs_buf_initialize( |
242 | xfs_buf_t *bp, | |
1da177e4 | 243 | xfs_buftarg_t *target, |
204ab25f | 244 | xfs_off_t range_base, |
1da177e4 | 245 | size_t range_length, |
ce8e922c | 246 | xfs_buf_flags_t flags) |
1da177e4 LT |
247 | { |
248 | /* | |
ce8e922c | 249 | * We don't want certain flags to appear in b_flags. |
1da177e4 | 250 | */ |
ce8e922c NS |
251 | flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD); |
252 | ||
253 | memset(bp, 0, sizeof(xfs_buf_t)); | |
254 | atomic_set(&bp->b_hold, 1); | |
430cbeb8 | 255 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 256 | init_completion(&bp->b_iowait); |
430cbeb8 | 257 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 258 | INIT_LIST_HEAD(&bp->b_list); |
74f75a0c | 259 | RB_CLEAR_NODE(&bp->b_rbnode); |
a731cd11 | 260 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
ce8e922c NS |
261 | XB_SET_OWNER(bp); |
262 | bp->b_target = target; | |
263 | bp->b_file_offset = range_base; | |
1da177e4 LT |
264 | /* |
265 | * Set buffer_length and count_desired to the same value initially. | |
266 | * I/O routines should use count_desired, which will be the same in | |
267 | * most cases but may be reset (e.g. XFS recovery). | |
268 | */ | |
ce8e922c NS |
269 | bp->b_buffer_length = bp->b_count_desired = range_length; |
270 | bp->b_flags = flags; | |
271 | bp->b_bn = XFS_BUF_DADDR_NULL; | |
272 | atomic_set(&bp->b_pin_count, 0); | |
273 | init_waitqueue_head(&bp->b_waiters); | |
274 | ||
275 | XFS_STATS_INC(xb_create); | |
0b1b213f CH |
276 | |
277 | trace_xfs_buf_init(bp, _RET_IP_); | |
1da177e4 LT |
278 | } |
279 | ||
280 | /* | |
ce8e922c NS |
281 | * Allocate a page array capable of holding a specified number |
282 | * of pages, and point the page buf at it. | |
1da177e4 LT |
283 | */ |
284 | STATIC int | |
ce8e922c NS |
285 | _xfs_buf_get_pages( |
286 | xfs_buf_t *bp, | |
1da177e4 | 287 | int page_count, |
ce8e922c | 288 | xfs_buf_flags_t flags) |
1da177e4 LT |
289 | { |
290 | /* Make sure that we have a page list */ | |
ce8e922c NS |
291 | if (bp->b_pages == NULL) { |
292 | bp->b_offset = xfs_buf_poff(bp->b_file_offset); | |
293 | bp->b_page_count = page_count; | |
294 | if (page_count <= XB_PAGES) { | |
295 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 296 | } else { |
ce8e922c NS |
297 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
298 | page_count, xb_to_km(flags)); | |
299 | if (bp->b_pages == NULL) | |
1da177e4 LT |
300 | return -ENOMEM; |
301 | } | |
ce8e922c | 302 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
303 | } |
304 | return 0; | |
305 | } | |
306 | ||
307 | /* | |
ce8e922c | 308 | * Frees b_pages if it was allocated. |
1da177e4 LT |
309 | */ |
310 | STATIC void | |
ce8e922c | 311 | _xfs_buf_free_pages( |
1da177e4 LT |
312 | xfs_buf_t *bp) |
313 | { | |
ce8e922c | 314 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 315 | kmem_free(bp->b_pages); |
3fc98b1a | 316 | bp->b_pages = NULL; |
1da177e4 LT |
317 | } |
318 | } | |
319 | ||
320 | /* | |
321 | * Releases the specified buffer. | |
322 | * | |
323 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 324 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
325 | * hashed and refcounted buffers |
326 | */ | |
327 | void | |
ce8e922c | 328 | xfs_buf_free( |
1da177e4 LT |
329 | xfs_buf_t *bp) |
330 | { | |
0b1b213f | 331 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 332 | |
430cbeb8 DC |
333 | ASSERT(list_empty(&bp->b_lru)); |
334 | ||
1fa40b01 | 335 | if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) { |
1da177e4 LT |
336 | uint i; |
337 | ||
73c77e2c | 338 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
339 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
340 | bp->b_page_count); | |
1da177e4 | 341 | |
948ecdb4 NS |
342 | for (i = 0; i < bp->b_page_count; i++) { |
343 | struct page *page = bp->b_pages[i]; | |
344 | ||
1fa40b01 CH |
345 | if (bp->b_flags & _XBF_PAGE_CACHE) |
346 | ASSERT(!PagePrivate(page)); | |
948ecdb4 NS |
347 | page_cache_release(page); |
348 | } | |
1da177e4 | 349 | } |
3fc98b1a | 350 | _xfs_buf_free_pages(bp); |
ce8e922c | 351 | xfs_buf_deallocate(bp); |
1da177e4 LT |
352 | } |
353 | ||
354 | /* | |
355 | * Finds all pages for buffer in question and builds it's page list. | |
356 | */ | |
357 | STATIC int | |
ce8e922c | 358 | _xfs_buf_lookup_pages( |
1da177e4 LT |
359 | xfs_buf_t *bp, |
360 | uint flags) | |
361 | { | |
ce8e922c NS |
362 | struct address_space *mapping = bp->b_target->bt_mapping; |
363 | size_t blocksize = bp->b_target->bt_bsize; | |
364 | size_t size = bp->b_count_desired; | |
1da177e4 | 365 | size_t nbytes, offset; |
ce8e922c | 366 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 LT |
367 | unsigned short page_count, i; |
368 | pgoff_t first; | |
204ab25f | 369 | xfs_off_t end; |
1da177e4 LT |
370 | int error; |
371 | ||
ce8e922c NS |
372 | end = bp->b_file_offset + bp->b_buffer_length; |
373 | page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset); | |
1da177e4 | 374 | |
ce8e922c | 375 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
376 | if (unlikely(error)) |
377 | return error; | |
ce8e922c | 378 | bp->b_flags |= _XBF_PAGE_CACHE; |
1da177e4 | 379 | |
ce8e922c NS |
380 | offset = bp->b_offset; |
381 | first = bp->b_file_offset >> PAGE_CACHE_SHIFT; | |
1da177e4 | 382 | |
ce8e922c | 383 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
384 | struct page *page; |
385 | uint retries = 0; | |
386 | ||
387 | retry: | |
388 | page = find_or_create_page(mapping, first + i, gfp_mask); | |
389 | if (unlikely(page == NULL)) { | |
ce8e922c NS |
390 | if (flags & XBF_READ_AHEAD) { |
391 | bp->b_page_count = i; | |
6ab455ee CH |
392 | for (i = 0; i < bp->b_page_count; i++) |
393 | unlock_page(bp->b_pages[i]); | |
1da177e4 LT |
394 | return -ENOMEM; |
395 | } | |
396 | ||
397 | /* | |
398 | * This could deadlock. | |
399 | * | |
400 | * But until all the XFS lowlevel code is revamped to | |
401 | * handle buffer allocation failures we can't do much. | |
402 | */ | |
403 | if (!(++retries % 100)) | |
404 | printk(KERN_ERR | |
405 | "XFS: possible memory allocation " | |
406 | "deadlock in %s (mode:0x%x)\n", | |
34a622b2 | 407 | __func__, gfp_mask); |
1da177e4 | 408 | |
ce8e922c | 409 | XFS_STATS_INC(xb_page_retries); |
8aa7e847 | 410 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
411 | goto retry; |
412 | } | |
413 | ||
ce8e922c | 414 | XFS_STATS_INC(xb_page_found); |
1da177e4 LT |
415 | |
416 | nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset); | |
417 | size -= nbytes; | |
418 | ||
948ecdb4 | 419 | ASSERT(!PagePrivate(page)); |
1da177e4 LT |
420 | if (!PageUptodate(page)) { |
421 | page_count--; | |
6ab455ee CH |
422 | if (blocksize >= PAGE_CACHE_SIZE) { |
423 | if (flags & XBF_READ) | |
424 | bp->b_flags |= _XBF_PAGE_LOCKED; | |
425 | } else if (!PagePrivate(page)) { | |
1da177e4 LT |
426 | if (test_page_region(page, offset, nbytes)) |
427 | page_count++; | |
428 | } | |
429 | } | |
430 | ||
ce8e922c | 431 | bp->b_pages[i] = page; |
1da177e4 LT |
432 | offset = 0; |
433 | } | |
434 | ||
6ab455ee CH |
435 | if (!(bp->b_flags & _XBF_PAGE_LOCKED)) { |
436 | for (i = 0; i < bp->b_page_count; i++) | |
437 | unlock_page(bp->b_pages[i]); | |
438 | } | |
439 | ||
ce8e922c NS |
440 | if (page_count == bp->b_page_count) |
441 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 442 | |
1da177e4 LT |
443 | return error; |
444 | } | |
445 | ||
446 | /* | |
447 | * Map buffer into kernel address-space if nessecary. | |
448 | */ | |
449 | STATIC int | |
ce8e922c | 450 | _xfs_buf_map_pages( |
1da177e4 LT |
451 | xfs_buf_t *bp, |
452 | uint flags) | |
453 | { | |
454 | /* A single page buffer is always mappable */ | |
ce8e922c NS |
455 | if (bp->b_page_count == 1) { |
456 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; | |
457 | bp->b_flags |= XBF_MAPPED; | |
458 | } else if (flags & XBF_MAPPED) { | |
8a262e57 AE |
459 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, |
460 | -1, PAGE_KERNEL); | |
ce8e922c | 461 | if (unlikely(bp->b_addr == NULL)) |
1da177e4 | 462 | return -ENOMEM; |
ce8e922c NS |
463 | bp->b_addr += bp->b_offset; |
464 | bp->b_flags |= XBF_MAPPED; | |
1da177e4 LT |
465 | } |
466 | ||
467 | return 0; | |
468 | } | |
469 | ||
470 | /* | |
471 | * Finding and Reading Buffers | |
472 | */ | |
473 | ||
474 | /* | |
ce8e922c | 475 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 LT |
476 | * a given range of an inode. The buffer is returned |
477 | * locked. If other overlapping buffers exist, they are | |
478 | * released before the new buffer is created and locked, | |
479 | * which may imply that this call will block until those buffers | |
480 | * are unlocked. No I/O is implied by this call. | |
481 | */ | |
482 | xfs_buf_t * | |
ce8e922c | 483 | _xfs_buf_find( |
1da177e4 | 484 | xfs_buftarg_t *btp, /* block device target */ |
204ab25f | 485 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 486 | size_t isize, /* length of range */ |
ce8e922c NS |
487 | xfs_buf_flags_t flags, |
488 | xfs_buf_t *new_bp) | |
1da177e4 | 489 | { |
204ab25f | 490 | xfs_off_t range_base; |
1da177e4 | 491 | size_t range_length; |
74f75a0c DC |
492 | struct xfs_perag *pag; |
493 | struct rb_node **rbp; | |
494 | struct rb_node *parent; | |
495 | xfs_buf_t *bp; | |
1da177e4 LT |
496 | |
497 | range_base = (ioff << BBSHIFT); | |
498 | range_length = (isize << BBSHIFT); | |
499 | ||
500 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
ce8e922c | 501 | ASSERT(!(range_length < (1 << btp->bt_sshift))); |
204ab25f | 502 | ASSERT(!(range_base & (xfs_off_t)btp->bt_smask)); |
1da177e4 | 503 | |
74f75a0c DC |
504 | /* get tree root */ |
505 | pag = xfs_perag_get(btp->bt_mount, | |
506 | xfs_daddr_to_agno(btp->bt_mount, ioff)); | |
507 | ||
508 | /* walk tree */ | |
509 | spin_lock(&pag->pag_buf_lock); | |
510 | rbp = &pag->pag_buf_tree.rb_node; | |
511 | parent = NULL; | |
512 | bp = NULL; | |
513 | while (*rbp) { | |
514 | parent = *rbp; | |
515 | bp = rb_entry(parent, struct xfs_buf, b_rbnode); | |
516 | ||
517 | if (range_base < bp->b_file_offset) | |
518 | rbp = &(*rbp)->rb_left; | |
519 | else if (range_base > bp->b_file_offset) | |
520 | rbp = &(*rbp)->rb_right; | |
521 | else { | |
522 | /* | |
523 | * found a block offset match. If the range doesn't | |
524 | * match, the only way this is allowed is if the buffer | |
525 | * in the cache is stale and the transaction that made | |
526 | * it stale has not yet committed. i.e. we are | |
527 | * reallocating a busy extent. Skip this buffer and | |
528 | * continue searching to the right for an exact match. | |
529 | */ | |
530 | if (bp->b_buffer_length != range_length) { | |
531 | ASSERT(bp->b_flags & XBF_STALE); | |
532 | rbp = &(*rbp)->rb_right; | |
533 | continue; | |
534 | } | |
ce8e922c | 535 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
536 | goto found; |
537 | } | |
538 | } | |
539 | ||
540 | /* No match found */ | |
ce8e922c NS |
541 | if (new_bp) { |
542 | _xfs_buf_initialize(new_bp, btp, range_base, | |
1da177e4 | 543 | range_length, flags); |
74f75a0c DC |
544 | rb_link_node(&new_bp->b_rbnode, parent, rbp); |
545 | rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree); | |
546 | /* the buffer keeps the perag reference until it is freed */ | |
547 | new_bp->b_pag = pag; | |
548 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 549 | } else { |
ce8e922c | 550 | XFS_STATS_INC(xb_miss_locked); |
74f75a0c DC |
551 | spin_unlock(&pag->pag_buf_lock); |
552 | xfs_perag_put(pag); | |
1da177e4 | 553 | } |
ce8e922c | 554 | return new_bp; |
1da177e4 LT |
555 | |
556 | found: | |
74f75a0c DC |
557 | spin_unlock(&pag->pag_buf_lock); |
558 | xfs_perag_put(pag); | |
1da177e4 | 559 | |
90810b9e DC |
560 | if (xfs_buf_cond_lock(bp)) { |
561 | /* failed, so wait for the lock if requested. */ | |
ce8e922c | 562 | if (!(flags & XBF_TRYLOCK)) { |
ce8e922c NS |
563 | xfs_buf_lock(bp); |
564 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 | 565 | } else { |
ce8e922c NS |
566 | xfs_buf_rele(bp); |
567 | XFS_STATS_INC(xb_busy_locked); | |
568 | return NULL; | |
1da177e4 | 569 | } |
1da177e4 LT |
570 | } |
571 | ||
ce8e922c NS |
572 | if (bp->b_flags & XBF_STALE) { |
573 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
574 | bp->b_flags &= XBF_MAPPED; | |
2f926587 | 575 | } |
0b1b213f CH |
576 | |
577 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ce8e922c NS |
578 | XFS_STATS_INC(xb_get_locked); |
579 | return bp; | |
1da177e4 LT |
580 | } |
581 | ||
582 | /* | |
ce8e922c | 583 | * Assembles a buffer covering the specified range. |
1da177e4 LT |
584 | * Storage in memory for all portions of the buffer will be allocated, |
585 | * although backing storage may not be. | |
586 | */ | |
587 | xfs_buf_t * | |
6ad112bf | 588 | xfs_buf_get( |
1da177e4 | 589 | xfs_buftarg_t *target,/* target for buffer */ |
204ab25f | 590 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 591 | size_t isize, /* length of range */ |
ce8e922c | 592 | xfs_buf_flags_t flags) |
1da177e4 | 593 | { |
ce8e922c | 594 | xfs_buf_t *bp, *new_bp; |
1da177e4 LT |
595 | int error = 0, i; |
596 | ||
ce8e922c NS |
597 | new_bp = xfs_buf_allocate(flags); |
598 | if (unlikely(!new_bp)) | |
1da177e4 LT |
599 | return NULL; |
600 | ||
ce8e922c NS |
601 | bp = _xfs_buf_find(target, ioff, isize, flags, new_bp); |
602 | if (bp == new_bp) { | |
603 | error = _xfs_buf_lookup_pages(bp, flags); | |
1da177e4 LT |
604 | if (error) |
605 | goto no_buffer; | |
606 | } else { | |
ce8e922c NS |
607 | xfs_buf_deallocate(new_bp); |
608 | if (unlikely(bp == NULL)) | |
1da177e4 LT |
609 | return NULL; |
610 | } | |
611 | ||
ce8e922c NS |
612 | for (i = 0; i < bp->b_page_count; i++) |
613 | mark_page_accessed(bp->b_pages[i]); | |
1da177e4 | 614 | |
ce8e922c NS |
615 | if (!(bp->b_flags & XBF_MAPPED)) { |
616 | error = _xfs_buf_map_pages(bp, flags); | |
1da177e4 LT |
617 | if (unlikely(error)) { |
618 | printk(KERN_WARNING "%s: failed to map pages\n", | |
34a622b2 | 619 | __func__); |
1da177e4 LT |
620 | goto no_buffer; |
621 | } | |
622 | } | |
623 | ||
ce8e922c | 624 | XFS_STATS_INC(xb_get); |
1da177e4 LT |
625 | |
626 | /* | |
627 | * Always fill in the block number now, the mapped cases can do | |
628 | * their own overlay of this later. | |
629 | */ | |
ce8e922c NS |
630 | bp->b_bn = ioff; |
631 | bp->b_count_desired = bp->b_buffer_length; | |
1da177e4 | 632 | |
0b1b213f | 633 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 634 | return bp; |
1da177e4 LT |
635 | |
636 | no_buffer: | |
ce8e922c NS |
637 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
638 | xfs_buf_unlock(bp); | |
639 | xfs_buf_rele(bp); | |
1da177e4 LT |
640 | return NULL; |
641 | } | |
642 | ||
5d765b97 CH |
643 | STATIC int |
644 | _xfs_buf_read( | |
645 | xfs_buf_t *bp, | |
646 | xfs_buf_flags_t flags) | |
647 | { | |
648 | int status; | |
649 | ||
5d765b97 CH |
650 | ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE))); |
651 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); | |
652 | ||
653 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \ | |
654 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
655 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \ | |
656 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
657 | ||
658 | status = xfs_buf_iorequest(bp); | |
ec53d1db DC |
659 | if (status || XFS_BUF_ISERROR(bp) || (flags & XBF_ASYNC)) |
660 | return status; | |
661 | return xfs_buf_iowait(bp); | |
5d765b97 CH |
662 | } |
663 | ||
1da177e4 | 664 | xfs_buf_t * |
6ad112bf | 665 | xfs_buf_read( |
1da177e4 | 666 | xfs_buftarg_t *target, |
204ab25f | 667 | xfs_off_t ioff, |
1da177e4 | 668 | size_t isize, |
ce8e922c | 669 | xfs_buf_flags_t flags) |
1da177e4 | 670 | { |
ce8e922c NS |
671 | xfs_buf_t *bp; |
672 | ||
673 | flags |= XBF_READ; | |
674 | ||
6ad112bf | 675 | bp = xfs_buf_get(target, ioff, isize, flags); |
ce8e922c | 676 | if (bp) { |
0b1b213f CH |
677 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
678 | ||
ce8e922c | 679 | if (!XFS_BUF_ISDONE(bp)) { |
ce8e922c | 680 | XFS_STATS_INC(xb_get_read); |
5d765b97 | 681 | _xfs_buf_read(bp, flags); |
ce8e922c | 682 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
683 | /* |
684 | * Read ahead call which is already satisfied, | |
685 | * drop the buffer | |
686 | */ | |
687 | goto no_buffer; | |
688 | } else { | |
1da177e4 | 689 | /* We do not want read in the flags */ |
ce8e922c | 690 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
691 | } |
692 | } | |
693 | ||
ce8e922c | 694 | return bp; |
1da177e4 LT |
695 | |
696 | no_buffer: | |
ce8e922c NS |
697 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
698 | xfs_buf_unlock(bp); | |
699 | xfs_buf_rele(bp); | |
1da177e4 LT |
700 | return NULL; |
701 | } | |
702 | ||
1da177e4 | 703 | /* |
ce8e922c NS |
704 | * If we are not low on memory then do the readahead in a deadlock |
705 | * safe manner. | |
1da177e4 LT |
706 | */ |
707 | void | |
ce8e922c | 708 | xfs_buf_readahead( |
1da177e4 | 709 | xfs_buftarg_t *target, |
204ab25f | 710 | xfs_off_t ioff, |
1a1a3e97 | 711 | size_t isize) |
1da177e4 LT |
712 | { |
713 | struct backing_dev_info *bdi; | |
714 | ||
ce8e922c | 715 | bdi = target->bt_mapping->backing_dev_info; |
1da177e4 LT |
716 | if (bdi_read_congested(bdi)) |
717 | return; | |
718 | ||
1a1a3e97 CH |
719 | xfs_buf_read(target, ioff, isize, |
720 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD|XBF_DONT_BLOCK); | |
1da177e4 LT |
721 | } |
722 | ||
5adc94c2 DC |
723 | /* |
724 | * Read an uncached buffer from disk. Allocates and returns a locked | |
725 | * buffer containing the disk contents or nothing. | |
726 | */ | |
727 | struct xfs_buf * | |
728 | xfs_buf_read_uncached( | |
729 | struct xfs_mount *mp, | |
730 | struct xfs_buftarg *target, | |
731 | xfs_daddr_t daddr, | |
732 | size_t length, | |
733 | int flags) | |
734 | { | |
735 | xfs_buf_t *bp; | |
736 | int error; | |
737 | ||
738 | bp = xfs_buf_get_uncached(target, length, flags); | |
739 | if (!bp) | |
740 | return NULL; | |
741 | ||
742 | /* set up the buffer for a read IO */ | |
743 | xfs_buf_lock(bp); | |
744 | XFS_BUF_SET_ADDR(bp, daddr); | |
745 | XFS_BUF_READ(bp); | |
746 | XFS_BUF_BUSY(bp); | |
747 | ||
748 | xfsbdstrat(mp, bp); | |
1a1a3e97 | 749 | error = xfs_buf_iowait(bp); |
5adc94c2 DC |
750 | if (error || bp->b_error) { |
751 | xfs_buf_relse(bp); | |
752 | return NULL; | |
753 | } | |
754 | return bp; | |
1da177e4 LT |
755 | } |
756 | ||
757 | xfs_buf_t * | |
ce8e922c | 758 | xfs_buf_get_empty( |
1da177e4 LT |
759 | size_t len, |
760 | xfs_buftarg_t *target) | |
761 | { | |
ce8e922c | 762 | xfs_buf_t *bp; |
1da177e4 | 763 | |
ce8e922c NS |
764 | bp = xfs_buf_allocate(0); |
765 | if (bp) | |
766 | _xfs_buf_initialize(bp, target, 0, len, 0); | |
767 | return bp; | |
1da177e4 LT |
768 | } |
769 | ||
770 | static inline struct page * | |
771 | mem_to_page( | |
772 | void *addr) | |
773 | { | |
9e2779fa | 774 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
775 | return virt_to_page(addr); |
776 | } else { | |
777 | return vmalloc_to_page(addr); | |
778 | } | |
779 | } | |
780 | ||
781 | int | |
ce8e922c NS |
782 | xfs_buf_associate_memory( |
783 | xfs_buf_t *bp, | |
1da177e4 LT |
784 | void *mem, |
785 | size_t len) | |
786 | { | |
787 | int rval; | |
788 | int i = 0; | |
d1afb678 LM |
789 | unsigned long pageaddr; |
790 | unsigned long offset; | |
791 | size_t buflen; | |
1da177e4 LT |
792 | int page_count; |
793 | ||
d1afb678 LM |
794 | pageaddr = (unsigned long)mem & PAGE_CACHE_MASK; |
795 | offset = (unsigned long)mem - pageaddr; | |
796 | buflen = PAGE_CACHE_ALIGN(len + offset); | |
797 | page_count = buflen >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
798 | |
799 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
800 | if (bp->b_pages) |
801 | _xfs_buf_free_pages(bp); | |
1da177e4 | 802 | |
ce8e922c NS |
803 | bp->b_pages = NULL; |
804 | bp->b_addr = mem; | |
1da177e4 | 805 | |
36fae17a | 806 | rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK); |
1da177e4 LT |
807 | if (rval) |
808 | return rval; | |
809 | ||
ce8e922c | 810 | bp->b_offset = offset; |
d1afb678 LM |
811 | |
812 | for (i = 0; i < bp->b_page_count; i++) { | |
813 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
814 | pageaddr += PAGE_CACHE_SIZE; | |
1da177e4 | 815 | } |
1da177e4 | 816 | |
d1afb678 LM |
817 | bp->b_count_desired = len; |
818 | bp->b_buffer_length = buflen; | |
ce8e922c | 819 | bp->b_flags |= XBF_MAPPED; |
6ab455ee | 820 | bp->b_flags &= ~_XBF_PAGE_LOCKED; |
1da177e4 LT |
821 | |
822 | return 0; | |
823 | } | |
824 | ||
825 | xfs_buf_t * | |
686865f7 DC |
826 | xfs_buf_get_uncached( |
827 | struct xfs_buftarg *target, | |
1da177e4 | 828 | size_t len, |
686865f7 | 829 | int flags) |
1da177e4 | 830 | { |
1fa40b01 CH |
831 | unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT; |
832 | int error, i; | |
1da177e4 | 833 | xfs_buf_t *bp; |
1da177e4 | 834 | |
ce8e922c | 835 | bp = xfs_buf_allocate(0); |
1da177e4 LT |
836 | if (unlikely(bp == NULL)) |
837 | goto fail; | |
ce8e922c | 838 | _xfs_buf_initialize(bp, target, 0, len, 0); |
1da177e4 | 839 | |
1fa40b01 CH |
840 | error = _xfs_buf_get_pages(bp, page_count, 0); |
841 | if (error) | |
1da177e4 LT |
842 | goto fail_free_buf; |
843 | ||
1fa40b01 | 844 | for (i = 0; i < page_count; i++) { |
686865f7 | 845 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
846 | if (!bp->b_pages[i]) |
847 | goto fail_free_mem; | |
1da177e4 | 848 | } |
1fa40b01 | 849 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 850 | |
1fa40b01 CH |
851 | error = _xfs_buf_map_pages(bp, XBF_MAPPED); |
852 | if (unlikely(error)) { | |
853 | printk(KERN_WARNING "%s: failed to map pages\n", | |
34a622b2 | 854 | __func__); |
1da177e4 | 855 | goto fail_free_mem; |
1fa40b01 | 856 | } |
1da177e4 | 857 | |
ce8e922c | 858 | xfs_buf_unlock(bp); |
1da177e4 | 859 | |
686865f7 | 860 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 861 | return bp; |
1fa40b01 | 862 | |
1da177e4 | 863 | fail_free_mem: |
1fa40b01 CH |
864 | while (--i >= 0) |
865 | __free_page(bp->b_pages[i]); | |
ca165b88 | 866 | _xfs_buf_free_pages(bp); |
1da177e4 | 867 | fail_free_buf: |
ca165b88 | 868 | xfs_buf_deallocate(bp); |
1da177e4 LT |
869 | fail: |
870 | return NULL; | |
871 | } | |
872 | ||
873 | /* | |
1da177e4 LT |
874 | * Increment reference count on buffer, to hold the buffer concurrently |
875 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
876 | * Must hold the buffer already to call this function. |
877 | */ | |
878 | void | |
ce8e922c NS |
879 | xfs_buf_hold( |
880 | xfs_buf_t *bp) | |
1da177e4 | 881 | { |
0b1b213f | 882 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 883 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
884 | } |
885 | ||
886 | /* | |
ce8e922c NS |
887 | * Releases a hold on the specified buffer. If the |
888 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
889 | */ |
890 | void | |
ce8e922c NS |
891 | xfs_buf_rele( |
892 | xfs_buf_t *bp) | |
1da177e4 | 893 | { |
74f75a0c | 894 | struct xfs_perag *pag = bp->b_pag; |
1da177e4 | 895 | |
0b1b213f | 896 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 897 | |
74f75a0c | 898 | if (!pag) { |
fad3aa1e | 899 | ASSERT(!bp->b_relse); |
430cbeb8 | 900 | ASSERT(list_empty(&bp->b_lru)); |
74f75a0c | 901 | ASSERT(RB_EMPTY_NODE(&bp->b_rbnode)); |
fad3aa1e NS |
902 | if (atomic_dec_and_test(&bp->b_hold)) |
903 | xfs_buf_free(bp); | |
904 | return; | |
905 | } | |
906 | ||
74f75a0c | 907 | ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode)); |
430cbeb8 | 908 | |
3790689f | 909 | ASSERT(atomic_read(&bp->b_hold) > 0); |
74f75a0c | 910 | if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) { |
ce8e922c NS |
911 | if (bp->b_relse) { |
912 | atomic_inc(&bp->b_hold); | |
74f75a0c DC |
913 | spin_unlock(&pag->pag_buf_lock); |
914 | bp->b_relse(bp); | |
430cbeb8 DC |
915 | } else if (!(bp->b_flags & XBF_STALE) && |
916 | atomic_read(&bp->b_lru_ref)) { | |
917 | xfs_buf_lru_add(bp); | |
918 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 919 | } else { |
430cbeb8 | 920 | xfs_buf_lru_del(bp); |
ce8e922c | 921 | ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q))); |
74f75a0c DC |
922 | rb_erase(&bp->b_rbnode, &pag->pag_buf_tree); |
923 | spin_unlock(&pag->pag_buf_lock); | |
924 | xfs_perag_put(pag); | |
ce8e922c | 925 | xfs_buf_free(bp); |
1da177e4 LT |
926 | } |
927 | } | |
928 | } | |
929 | ||
930 | ||
931 | /* | |
932 | * Mutual exclusion on buffers. Locking model: | |
933 | * | |
934 | * Buffers associated with inodes for which buffer locking | |
935 | * is not enabled are not protected by semaphores, and are | |
936 | * assumed to be exclusively owned by the caller. There is a | |
937 | * spinlock in the buffer, used by the caller when concurrent | |
938 | * access is possible. | |
939 | */ | |
940 | ||
941 | /* | |
90810b9e DC |
942 | * Locks a buffer object, if it is not already locked. Note that this in |
943 | * no way locks the underlying pages, so it is only useful for | |
944 | * synchronizing concurrent use of buffer objects, not for synchronizing | |
945 | * independent access to the underlying pages. | |
946 | * | |
947 | * If we come across a stale, pinned, locked buffer, we know that we are | |
948 | * being asked to lock a buffer that has been reallocated. Because it is | |
949 | * pinned, we know that the log has not been pushed to disk and hence it | |
950 | * will still be locked. Rather than continuing to have trylock attempts | |
951 | * fail until someone else pushes the log, push it ourselves before | |
952 | * returning. This means that the xfsaild will not get stuck trying | |
953 | * to push on stale inode buffers. | |
1da177e4 LT |
954 | */ |
955 | int | |
ce8e922c NS |
956 | xfs_buf_cond_lock( |
957 | xfs_buf_t *bp) | |
1da177e4 LT |
958 | { |
959 | int locked; | |
960 | ||
ce8e922c | 961 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 962 | if (locked) |
ce8e922c | 963 | XB_SET_OWNER(bp); |
90810b9e DC |
964 | else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
965 | xfs_log_force(bp->b_target->bt_mount, 0); | |
0b1b213f CH |
966 | |
967 | trace_xfs_buf_cond_lock(bp, _RET_IP_); | |
ce8e922c | 968 | return locked ? 0 : -EBUSY; |
1da177e4 LT |
969 | } |
970 | ||
1da177e4 | 971 | int |
ce8e922c NS |
972 | xfs_buf_lock_value( |
973 | xfs_buf_t *bp) | |
1da177e4 | 974 | { |
adaa693b | 975 | return bp->b_sema.count; |
1da177e4 | 976 | } |
1da177e4 LT |
977 | |
978 | /* | |
ce8e922c NS |
979 | * Locks a buffer object. |
980 | * Note that this in no way locks the underlying pages, so it is only | |
981 | * useful for synchronizing concurrent use of buffer objects, not for | |
982 | * synchronizing independent access to the underlying pages. | |
ed3b4d6c DC |
983 | * |
984 | * If we come across a stale, pinned, locked buffer, we know that we | |
985 | * are being asked to lock a buffer that has been reallocated. Because | |
986 | * it is pinned, we know that the log has not been pushed to disk and | |
987 | * hence it will still be locked. Rather than sleeping until someone | |
988 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 989 | */ |
ce8e922c NS |
990 | void |
991 | xfs_buf_lock( | |
992 | xfs_buf_t *bp) | |
1da177e4 | 993 | { |
0b1b213f CH |
994 | trace_xfs_buf_lock(bp, _RET_IP_); |
995 | ||
ed3b4d6c | 996 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 997 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
998 | if (atomic_read(&bp->b_io_remaining)) |
999 | blk_run_address_space(bp->b_target->bt_mapping); | |
1000 | down(&bp->b_sema); | |
1001 | XB_SET_OWNER(bp); | |
0b1b213f CH |
1002 | |
1003 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
1004 | } |
1005 | ||
1006 | /* | |
ce8e922c | 1007 | * Releases the lock on the buffer object. |
2f926587 | 1008 | * If the buffer is marked delwri but is not queued, do so before we |
ce8e922c | 1009 | * unlock the buffer as we need to set flags correctly. We also need to |
2f926587 DC |
1010 | * take a reference for the delwri queue because the unlocker is going to |
1011 | * drop their's and they don't know we just queued it. | |
1da177e4 LT |
1012 | */ |
1013 | void | |
ce8e922c NS |
1014 | xfs_buf_unlock( |
1015 | xfs_buf_t *bp) | |
1da177e4 | 1016 | { |
ce8e922c NS |
1017 | if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) { |
1018 | atomic_inc(&bp->b_hold); | |
1019 | bp->b_flags |= XBF_ASYNC; | |
1020 | xfs_buf_delwri_queue(bp, 0); | |
2f926587 DC |
1021 | } |
1022 | ||
ce8e922c NS |
1023 | XB_CLEAR_OWNER(bp); |
1024 | up(&bp->b_sema); | |
0b1b213f CH |
1025 | |
1026 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
1027 | } |
1028 | ||
ce8e922c NS |
1029 | STATIC void |
1030 | xfs_buf_wait_unpin( | |
1031 | xfs_buf_t *bp) | |
1da177e4 LT |
1032 | { |
1033 | DECLARE_WAITQUEUE (wait, current); | |
1034 | ||
ce8e922c | 1035 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
1036 | return; |
1037 | ||
ce8e922c | 1038 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1039 | for (;;) { |
1040 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 1041 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 1042 | break; |
ce8e922c NS |
1043 | if (atomic_read(&bp->b_io_remaining)) |
1044 | blk_run_address_space(bp->b_target->bt_mapping); | |
1da177e4 LT |
1045 | schedule(); |
1046 | } | |
ce8e922c | 1047 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1048 | set_current_state(TASK_RUNNING); |
1049 | } | |
1050 | ||
1051 | /* | |
1052 | * Buffer Utility Routines | |
1053 | */ | |
1054 | ||
1da177e4 | 1055 | STATIC void |
ce8e922c | 1056 | xfs_buf_iodone_work( |
c4028958 | 1057 | struct work_struct *work) |
1da177e4 | 1058 | { |
c4028958 DH |
1059 | xfs_buf_t *bp = |
1060 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 1061 | |
80f6c29d | 1062 | if (bp->b_iodone) |
ce8e922c NS |
1063 | (*(bp->b_iodone))(bp); |
1064 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
1065 | xfs_buf_relse(bp); |
1066 | } | |
1067 | ||
1068 | void | |
ce8e922c NS |
1069 | xfs_buf_ioend( |
1070 | xfs_buf_t *bp, | |
1da177e4 LT |
1071 | int schedule) |
1072 | { | |
0b1b213f CH |
1073 | trace_xfs_buf_iodone(bp, _RET_IP_); |
1074 | ||
77be55a5 | 1075 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
ce8e922c NS |
1076 | if (bp->b_error == 0) |
1077 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1078 | |
ce8e922c | 1079 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 1080 | if (schedule) { |
c4028958 | 1081 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 1082 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 1083 | } else { |
c4028958 | 1084 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
1085 | } |
1086 | } else { | |
b4dd330b | 1087 | complete(&bp->b_iowait); |
1da177e4 LT |
1088 | } |
1089 | } | |
1090 | ||
1da177e4 | 1091 | void |
ce8e922c NS |
1092 | xfs_buf_ioerror( |
1093 | xfs_buf_t *bp, | |
1094 | int error) | |
1da177e4 LT |
1095 | { |
1096 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c | 1097 | bp->b_error = (unsigned short)error; |
0b1b213f | 1098 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1099 | } |
1100 | ||
1da177e4 | 1101 | int |
64e0bc7d CH |
1102 | xfs_bwrite( |
1103 | struct xfs_mount *mp, | |
5d765b97 | 1104 | struct xfs_buf *bp) |
1da177e4 | 1105 | { |
8c38366f | 1106 | int error; |
1da177e4 | 1107 | |
64e0bc7d | 1108 | bp->b_flags |= XBF_WRITE; |
8c38366f | 1109 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ); |
1da177e4 | 1110 | |
5d765b97 | 1111 | xfs_buf_delwri_dequeue(bp); |
939d723b | 1112 | xfs_bdstrat_cb(bp); |
1da177e4 | 1113 | |
8c38366f CH |
1114 | error = xfs_buf_iowait(bp); |
1115 | if (error) | |
1116 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1117 | xfs_buf_relse(bp); | |
64e0bc7d | 1118 | return error; |
5d765b97 | 1119 | } |
1da177e4 | 1120 | |
5d765b97 CH |
1121 | void |
1122 | xfs_bdwrite( | |
1123 | void *mp, | |
1124 | struct xfs_buf *bp) | |
1125 | { | |
0b1b213f | 1126 | trace_xfs_buf_bdwrite(bp, _RET_IP_); |
1da177e4 | 1127 | |
5d765b97 CH |
1128 | bp->b_flags &= ~XBF_READ; |
1129 | bp->b_flags |= (XBF_DELWRI | XBF_ASYNC); | |
1130 | ||
1131 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1132 | } |
1133 | ||
4e23471a CH |
1134 | /* |
1135 | * Called when we want to stop a buffer from getting written or read. | |
1a1a3e97 | 1136 | * We attach the EIO error, muck with its flags, and call xfs_buf_ioend |
4e23471a CH |
1137 | * so that the proper iodone callbacks get called. |
1138 | */ | |
1139 | STATIC int | |
1140 | xfs_bioerror( | |
1141 | xfs_buf_t *bp) | |
1142 | { | |
1143 | #ifdef XFSERRORDEBUG | |
1144 | ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone); | |
1145 | #endif | |
1146 | ||
1147 | /* | |
1148 | * No need to wait until the buffer is unpinned, we aren't flushing it. | |
1149 | */ | |
1150 | XFS_BUF_ERROR(bp, EIO); | |
1151 | ||
1152 | /* | |
1a1a3e97 | 1153 | * We're calling xfs_buf_ioend, so delete XBF_DONE flag. |
4e23471a CH |
1154 | */ |
1155 | XFS_BUF_UNREAD(bp); | |
1156 | XFS_BUF_UNDELAYWRITE(bp); | |
1157 | XFS_BUF_UNDONE(bp); | |
1158 | XFS_BUF_STALE(bp); | |
1159 | ||
1a1a3e97 | 1160 | xfs_buf_ioend(bp, 0); |
4e23471a CH |
1161 | |
1162 | return EIO; | |
1163 | } | |
1164 | ||
1165 | /* | |
1166 | * Same as xfs_bioerror, except that we are releasing the buffer | |
1a1a3e97 | 1167 | * here ourselves, and avoiding the xfs_buf_ioend call. |
4e23471a CH |
1168 | * This is meant for userdata errors; metadata bufs come with |
1169 | * iodone functions attached, so that we can track down errors. | |
1170 | */ | |
1171 | STATIC int | |
1172 | xfs_bioerror_relse( | |
1173 | struct xfs_buf *bp) | |
1174 | { | |
1175 | int64_t fl = XFS_BUF_BFLAGS(bp); | |
1176 | /* | |
1177 | * No need to wait until the buffer is unpinned. | |
1178 | * We aren't flushing it. | |
1179 | * | |
1180 | * chunkhold expects B_DONE to be set, whether | |
1181 | * we actually finish the I/O or not. We don't want to | |
1182 | * change that interface. | |
1183 | */ | |
1184 | XFS_BUF_UNREAD(bp); | |
1185 | XFS_BUF_UNDELAYWRITE(bp); | |
1186 | XFS_BUF_DONE(bp); | |
1187 | XFS_BUF_STALE(bp); | |
1188 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
0cadda1c | 1189 | if (!(fl & XBF_ASYNC)) { |
4e23471a CH |
1190 | /* |
1191 | * Mark b_error and B_ERROR _both_. | |
1192 | * Lot's of chunkcache code assumes that. | |
1193 | * There's no reason to mark error for | |
1194 | * ASYNC buffers. | |
1195 | */ | |
1196 | XFS_BUF_ERROR(bp, EIO); | |
1197 | XFS_BUF_FINISH_IOWAIT(bp); | |
1198 | } else { | |
1199 | xfs_buf_relse(bp); | |
1200 | } | |
1201 | ||
1202 | return EIO; | |
1203 | } | |
1204 | ||
1205 | ||
1206 | /* | |
1207 | * All xfs metadata buffers except log state machine buffers | |
1208 | * get this attached as their b_bdstrat callback function. | |
1209 | * This is so that we can catch a buffer | |
1210 | * after prematurely unpinning it to forcibly shutdown the filesystem. | |
1211 | */ | |
1212 | int | |
1213 | xfs_bdstrat_cb( | |
1214 | struct xfs_buf *bp) | |
1215 | { | |
ebad861b | 1216 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
4e23471a CH |
1217 | trace_xfs_bdstrat_shut(bp, _RET_IP_); |
1218 | /* | |
1219 | * Metadata write that didn't get logged but | |
1220 | * written delayed anyway. These aren't associated | |
1221 | * with a transaction, and can be ignored. | |
1222 | */ | |
1223 | if (!bp->b_iodone && !XFS_BUF_ISREAD(bp)) | |
1224 | return xfs_bioerror_relse(bp); | |
1225 | else | |
1226 | return xfs_bioerror(bp); | |
1227 | } | |
1228 | ||
1229 | xfs_buf_iorequest(bp); | |
1230 | return 0; | |
1231 | } | |
1232 | ||
1233 | /* | |
1234 | * Wrapper around bdstrat so that we can stop data from going to disk in case | |
1235 | * we are shutting down the filesystem. Typically user data goes thru this | |
1236 | * path; one of the exceptions is the superblock. | |
1237 | */ | |
1238 | void | |
1239 | xfsbdstrat( | |
1240 | struct xfs_mount *mp, | |
1241 | struct xfs_buf *bp) | |
1242 | { | |
1243 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1244 | trace_xfs_bdstrat_shut(bp, _RET_IP_); | |
1245 | xfs_bioerror_relse(bp); | |
1246 | return; | |
1247 | } | |
1248 | ||
1249 | xfs_buf_iorequest(bp); | |
1250 | } | |
1251 | ||
b8f82a4a | 1252 | STATIC void |
ce8e922c NS |
1253 | _xfs_buf_ioend( |
1254 | xfs_buf_t *bp, | |
1da177e4 LT |
1255 | int schedule) |
1256 | { | |
6ab455ee CH |
1257 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
1258 | bp->b_flags &= ~_XBF_PAGE_LOCKED; | |
ce8e922c | 1259 | xfs_buf_ioend(bp, schedule); |
6ab455ee | 1260 | } |
1da177e4 LT |
1261 | } |
1262 | ||
782e3b3b | 1263 | STATIC void |
ce8e922c | 1264 | xfs_buf_bio_end_io( |
1da177e4 | 1265 | struct bio *bio, |
1da177e4 LT |
1266 | int error) |
1267 | { | |
ce8e922c NS |
1268 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1269 | unsigned int blocksize = bp->b_target->bt_bsize; | |
eedb5530 | 1270 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1da177e4 | 1271 | |
cfbe5267 | 1272 | xfs_buf_ioerror(bp, -error); |
1da177e4 | 1273 | |
73c77e2c JB |
1274 | if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
1275 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); | |
1276 | ||
eedb5530 | 1277 | do { |
1da177e4 LT |
1278 | struct page *page = bvec->bv_page; |
1279 | ||
948ecdb4 | 1280 | ASSERT(!PagePrivate(page)); |
ce8e922c NS |
1281 | if (unlikely(bp->b_error)) { |
1282 | if (bp->b_flags & XBF_READ) | |
eedb5530 | 1283 | ClearPageUptodate(page); |
ce8e922c | 1284 | } else if (blocksize >= PAGE_CACHE_SIZE) { |
1da177e4 LT |
1285 | SetPageUptodate(page); |
1286 | } else if (!PagePrivate(page) && | |
ce8e922c | 1287 | (bp->b_flags & _XBF_PAGE_CACHE)) { |
1da177e4 LT |
1288 | set_page_region(page, bvec->bv_offset, bvec->bv_len); |
1289 | } | |
1290 | ||
eedb5530 NS |
1291 | if (--bvec >= bio->bi_io_vec) |
1292 | prefetchw(&bvec->bv_page->flags); | |
6ab455ee CH |
1293 | |
1294 | if (bp->b_flags & _XBF_PAGE_LOCKED) | |
1295 | unlock_page(page); | |
eedb5530 | 1296 | } while (bvec >= bio->bi_io_vec); |
1da177e4 | 1297 | |
ce8e922c | 1298 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1299 | bio_put(bio); |
1da177e4 LT |
1300 | } |
1301 | ||
1302 | STATIC void | |
ce8e922c NS |
1303 | _xfs_buf_ioapply( |
1304 | xfs_buf_t *bp) | |
1da177e4 | 1305 | { |
a9759f2d | 1306 | int rw, map_i, total_nr_pages, nr_pages; |
1da177e4 | 1307 | struct bio *bio; |
ce8e922c NS |
1308 | int offset = bp->b_offset; |
1309 | int size = bp->b_count_desired; | |
1310 | sector_t sector = bp->b_bn; | |
1311 | unsigned int blocksize = bp->b_target->bt_bsize; | |
1da177e4 | 1312 | |
ce8e922c | 1313 | total_nr_pages = bp->b_page_count; |
1da177e4 LT |
1314 | map_i = 0; |
1315 | ||
ce8e922c NS |
1316 | if (bp->b_flags & XBF_ORDERED) { |
1317 | ASSERT(!(bp->b_flags & XBF_READ)); | |
80f6c29d | 1318 | rw = WRITE_FLUSH_FUA; |
2ee1abad | 1319 | } else if (bp->b_flags & XBF_LOG_BUFFER) { |
51bdd706 NS |
1320 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); |
1321 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1322 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC; | |
2ee1abad DC |
1323 | } else if (bp->b_flags & _XBF_RUN_QUEUES) { |
1324 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); | |
1325 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1326 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META; | |
51bdd706 NS |
1327 | } else { |
1328 | rw = (bp->b_flags & XBF_WRITE) ? WRITE : | |
1329 | (bp->b_flags & XBF_READ_AHEAD) ? READA : READ; | |
f538d4da CH |
1330 | } |
1331 | ||
ce8e922c | 1332 | /* Special code path for reading a sub page size buffer in -- |
1da177e4 LT |
1333 | * we populate up the whole page, and hence the other metadata |
1334 | * in the same page. This optimization is only valid when the | |
ce8e922c | 1335 | * filesystem block size is not smaller than the page size. |
1da177e4 | 1336 | */ |
ce8e922c | 1337 | if ((bp->b_buffer_length < PAGE_CACHE_SIZE) && |
6ab455ee CH |
1338 | ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) == |
1339 | (XBF_READ|_XBF_PAGE_LOCKED)) && | |
ce8e922c | 1340 | (blocksize >= PAGE_CACHE_SIZE)) { |
1da177e4 LT |
1341 | bio = bio_alloc(GFP_NOIO, 1); |
1342 | ||
ce8e922c | 1343 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1344 | bio->bi_sector = sector - (offset >> BBSHIFT); |
ce8e922c NS |
1345 | bio->bi_end_io = xfs_buf_bio_end_io; |
1346 | bio->bi_private = bp; | |
1da177e4 | 1347 | |
ce8e922c | 1348 | bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0); |
1da177e4 LT |
1349 | size = 0; |
1350 | ||
ce8e922c | 1351 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1352 | |
1353 | goto submit_io; | |
1354 | } | |
1355 | ||
1da177e4 | 1356 | next_chunk: |
ce8e922c | 1357 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1358 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1359 | if (nr_pages > total_nr_pages) | |
1360 | nr_pages = total_nr_pages; | |
1361 | ||
1362 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1363 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1364 | bio->bi_sector = sector; |
ce8e922c NS |
1365 | bio->bi_end_io = xfs_buf_bio_end_io; |
1366 | bio->bi_private = bp; | |
1da177e4 LT |
1367 | |
1368 | for (; size && nr_pages; nr_pages--, map_i++) { | |
ce8e922c | 1369 | int rbytes, nbytes = PAGE_CACHE_SIZE - offset; |
1da177e4 LT |
1370 | |
1371 | if (nbytes > size) | |
1372 | nbytes = size; | |
1373 | ||
ce8e922c NS |
1374 | rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset); |
1375 | if (rbytes < nbytes) | |
1da177e4 LT |
1376 | break; |
1377 | ||
1378 | offset = 0; | |
1379 | sector += nbytes >> BBSHIFT; | |
1380 | size -= nbytes; | |
1381 | total_nr_pages--; | |
1382 | } | |
1383 | ||
1384 | submit_io: | |
1385 | if (likely(bio->bi_size)) { | |
73c77e2c JB |
1386 | if (xfs_buf_is_vmapped(bp)) { |
1387 | flush_kernel_vmap_range(bp->b_addr, | |
1388 | xfs_buf_vmap_len(bp)); | |
1389 | } | |
1da177e4 LT |
1390 | submit_bio(rw, bio); |
1391 | if (size) | |
1392 | goto next_chunk; | |
1393 | } else { | |
ec53d1db DC |
1394 | /* |
1395 | * if we get here, no pages were added to the bio. However, | |
1396 | * we can't just error out here - if the pages are locked then | |
1397 | * we have to unlock them otherwise we can hang on a later | |
1398 | * access to the page. | |
1399 | */ | |
ce8e922c | 1400 | xfs_buf_ioerror(bp, EIO); |
ec53d1db DC |
1401 | if (bp->b_flags & _XBF_PAGE_LOCKED) { |
1402 | int i; | |
1403 | for (i = 0; i < bp->b_page_count; i++) | |
1404 | unlock_page(bp->b_pages[i]); | |
1405 | } | |
1406 | bio_put(bio); | |
1da177e4 LT |
1407 | } |
1408 | } | |
1409 | ||
1da177e4 | 1410 | int |
ce8e922c NS |
1411 | xfs_buf_iorequest( |
1412 | xfs_buf_t *bp) | |
1da177e4 | 1413 | { |
0b1b213f | 1414 | trace_xfs_buf_iorequest(bp, _RET_IP_); |
1da177e4 | 1415 | |
ce8e922c NS |
1416 | if (bp->b_flags & XBF_DELWRI) { |
1417 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1418 | return 0; |
1419 | } | |
1420 | ||
ce8e922c NS |
1421 | if (bp->b_flags & XBF_WRITE) { |
1422 | xfs_buf_wait_unpin(bp); | |
1da177e4 LT |
1423 | } |
1424 | ||
ce8e922c | 1425 | xfs_buf_hold(bp); |
1da177e4 LT |
1426 | |
1427 | /* Set the count to 1 initially, this will stop an I/O | |
1428 | * completion callout which happens before we have started | |
ce8e922c | 1429 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1430 | */ |
ce8e922c NS |
1431 | atomic_set(&bp->b_io_remaining, 1); |
1432 | _xfs_buf_ioapply(bp); | |
1433 | _xfs_buf_ioend(bp, 0); | |
1da177e4 | 1434 | |
ce8e922c | 1435 | xfs_buf_rele(bp); |
1da177e4 LT |
1436 | return 0; |
1437 | } | |
1438 | ||
1439 | /* | |
ce8e922c NS |
1440 | * Waits for I/O to complete on the buffer supplied. |
1441 | * It returns immediately if no I/O is pending. | |
1442 | * It returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1443 | */ |
1444 | int | |
ce8e922c NS |
1445 | xfs_buf_iowait( |
1446 | xfs_buf_t *bp) | |
1da177e4 | 1447 | { |
0b1b213f CH |
1448 | trace_xfs_buf_iowait(bp, _RET_IP_); |
1449 | ||
ce8e922c NS |
1450 | if (atomic_read(&bp->b_io_remaining)) |
1451 | blk_run_address_space(bp->b_target->bt_mapping); | |
b4dd330b | 1452 | wait_for_completion(&bp->b_iowait); |
0b1b213f CH |
1453 | |
1454 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
ce8e922c | 1455 | return bp->b_error; |
1da177e4 LT |
1456 | } |
1457 | ||
ce8e922c NS |
1458 | xfs_caddr_t |
1459 | xfs_buf_offset( | |
1460 | xfs_buf_t *bp, | |
1da177e4 LT |
1461 | size_t offset) |
1462 | { | |
1463 | struct page *page; | |
1464 | ||
ce8e922c NS |
1465 | if (bp->b_flags & XBF_MAPPED) |
1466 | return XFS_BUF_PTR(bp) + offset; | |
1da177e4 | 1467 | |
ce8e922c NS |
1468 | offset += bp->b_offset; |
1469 | page = bp->b_pages[offset >> PAGE_CACHE_SHIFT]; | |
1470 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1)); | |
1da177e4 LT |
1471 | } |
1472 | ||
1473 | /* | |
1da177e4 LT |
1474 | * Move data into or out of a buffer. |
1475 | */ | |
1476 | void | |
ce8e922c NS |
1477 | xfs_buf_iomove( |
1478 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1479 | size_t boff, /* starting buffer offset */ |
1480 | size_t bsize, /* length to copy */ | |
b9c48649 | 1481 | void *data, /* data address */ |
ce8e922c | 1482 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 LT |
1483 | { |
1484 | size_t bend, cpoff, csize; | |
1485 | struct page *page; | |
1486 | ||
1487 | bend = boff + bsize; | |
1488 | while (boff < bend) { | |
ce8e922c NS |
1489 | page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)]; |
1490 | cpoff = xfs_buf_poff(boff + bp->b_offset); | |
1da177e4 | 1491 | csize = min_t(size_t, |
ce8e922c | 1492 | PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff); |
1da177e4 LT |
1493 | |
1494 | ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE)); | |
1495 | ||
1496 | switch (mode) { | |
ce8e922c | 1497 | case XBRW_ZERO: |
1da177e4 LT |
1498 | memset(page_address(page) + cpoff, 0, csize); |
1499 | break; | |
ce8e922c | 1500 | case XBRW_READ: |
1da177e4 LT |
1501 | memcpy(data, page_address(page) + cpoff, csize); |
1502 | break; | |
ce8e922c | 1503 | case XBRW_WRITE: |
1da177e4 LT |
1504 | memcpy(page_address(page) + cpoff, data, csize); |
1505 | } | |
1506 | ||
1507 | boff += csize; | |
1508 | data += csize; | |
1509 | } | |
1510 | } | |
1511 | ||
1512 | /* | |
ce8e922c | 1513 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1514 | */ |
1515 | ||
1516 | /* | |
430cbeb8 DC |
1517 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1518 | * returned. These buffers will have an elevated hold count, so wait on those | |
1519 | * while freeing all the buffers only held by the LRU. | |
1da177e4 LT |
1520 | */ |
1521 | void | |
1522 | xfs_wait_buftarg( | |
74f75a0c | 1523 | struct xfs_buftarg *btp) |
1da177e4 | 1524 | { |
430cbeb8 DC |
1525 | struct xfs_buf *bp; |
1526 | ||
1527 | restart: | |
1528 | spin_lock(&btp->bt_lru_lock); | |
1529 | while (!list_empty(&btp->bt_lru)) { | |
1530 | bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru); | |
1531 | if (atomic_read(&bp->b_hold) > 1) { | |
1532 | spin_unlock(&btp->bt_lru_lock); | |
26af6552 | 1533 | delay(100); |
430cbeb8 | 1534 | goto restart; |
1da177e4 | 1535 | } |
430cbeb8 DC |
1536 | /* |
1537 | * clear the LRU reference count so the bufer doesn't get | |
1538 | * ignored in xfs_buf_rele(). | |
1539 | */ | |
1540 | atomic_set(&bp->b_lru_ref, 0); | |
1541 | spin_unlock(&btp->bt_lru_lock); | |
1542 | xfs_buf_rele(bp); | |
1543 | spin_lock(&btp->bt_lru_lock); | |
1da177e4 | 1544 | } |
430cbeb8 | 1545 | spin_unlock(&btp->bt_lru_lock); |
1da177e4 LT |
1546 | } |
1547 | ||
ff57ab21 DC |
1548 | int |
1549 | xfs_buftarg_shrink( | |
1550 | struct shrinker *shrink, | |
1551 | int nr_to_scan, | |
1552 | gfp_t mask) | |
a6867a68 | 1553 | { |
ff57ab21 DC |
1554 | struct xfs_buftarg *btp = container_of(shrink, |
1555 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 DC |
1556 | struct xfs_buf *bp; |
1557 | LIST_HEAD(dispose); | |
1558 | ||
1559 | if (!nr_to_scan) | |
1560 | return btp->bt_lru_nr; | |
1561 | ||
1562 | spin_lock(&btp->bt_lru_lock); | |
1563 | while (!list_empty(&btp->bt_lru)) { | |
1564 | if (nr_to_scan-- <= 0) | |
1565 | break; | |
1566 | ||
1567 | bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru); | |
1568 | ||
1569 | /* | |
1570 | * Decrement the b_lru_ref count unless the value is already | |
1571 | * zero. If the value is already zero, we need to reclaim the | |
1572 | * buffer, otherwise it gets another trip through the LRU. | |
1573 | */ | |
1574 | if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) { | |
1575 | list_move_tail(&bp->b_lru, &btp->bt_lru); | |
1576 | continue; | |
1577 | } | |
1578 | ||
1579 | /* | |
1580 | * remove the buffer from the LRU now to avoid needing another | |
1581 | * lock round trip inside xfs_buf_rele(). | |
1582 | */ | |
1583 | list_move(&bp->b_lru, &dispose); | |
1584 | btp->bt_lru_nr--; | |
ff57ab21 | 1585 | } |
430cbeb8 DC |
1586 | spin_unlock(&btp->bt_lru_lock); |
1587 | ||
1588 | while (!list_empty(&dispose)) { | |
1589 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1590 | list_del_init(&bp->b_lru); | |
1591 | xfs_buf_rele(bp); | |
1592 | } | |
1593 | ||
1594 | return btp->bt_lru_nr; | |
a6867a68 DC |
1595 | } |
1596 | ||
1da177e4 LT |
1597 | void |
1598 | xfs_free_buftarg( | |
b7963133 CH |
1599 | struct xfs_mount *mp, |
1600 | struct xfs_buftarg *btp) | |
1da177e4 | 1601 | { |
ff57ab21 DC |
1602 | unregister_shrinker(&btp->bt_shrinker); |
1603 | ||
1da177e4 | 1604 | xfs_flush_buftarg(btp, 1); |
b7963133 CH |
1605 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1606 | xfs_blkdev_issue_flush(btp); | |
ce8e922c | 1607 | iput(btp->bt_mapping->host); |
a6867a68 | 1608 | |
a6867a68 | 1609 | kthread_stop(btp->bt_task); |
f0e2d93c | 1610 | kmem_free(btp); |
1da177e4 LT |
1611 | } |
1612 | ||
1da177e4 LT |
1613 | STATIC int |
1614 | xfs_setsize_buftarg_flags( | |
1615 | xfs_buftarg_t *btp, | |
1616 | unsigned int blocksize, | |
1617 | unsigned int sectorsize, | |
1618 | int verbose) | |
1619 | { | |
ce8e922c NS |
1620 | btp->bt_bsize = blocksize; |
1621 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1622 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1623 | |
ce8e922c | 1624 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
1da177e4 LT |
1625 | printk(KERN_WARNING |
1626 | "XFS: Cannot set_blocksize to %u on device %s\n", | |
1627 | sectorsize, XFS_BUFTARG_NAME(btp)); | |
1628 | return EINVAL; | |
1629 | } | |
1630 | ||
1631 | if (verbose && | |
1632 | (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) { | |
1633 | printk(KERN_WARNING | |
1634 | "XFS: %u byte sectors in use on device %s. " | |
1635 | "This is suboptimal; %u or greater is ideal.\n", | |
1636 | sectorsize, XFS_BUFTARG_NAME(btp), | |
1637 | (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG); | |
1638 | } | |
1639 | ||
1640 | return 0; | |
1641 | } | |
1642 | ||
1643 | /* | |
ce8e922c NS |
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 is at this early stage. Play safe. | |
1647 | */ | |
1da177e4 LT |
1648 | STATIC int |
1649 | xfs_setsize_buftarg_early( | |
1650 | xfs_buftarg_t *btp, | |
1651 | struct block_device *bdev) | |
1652 | { | |
1653 | return xfs_setsize_buftarg_flags(btp, | |
e1defc4f | 1654 | PAGE_CACHE_SIZE, bdev_logical_block_size(bdev), 0); |
1da177e4 LT |
1655 | } |
1656 | ||
1657 | int | |
1658 | xfs_setsize_buftarg( | |
1659 | xfs_buftarg_t *btp, | |
1660 | unsigned int blocksize, | |
1661 | unsigned int sectorsize) | |
1662 | { | |
1663 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1664 | } | |
1665 | ||
1666 | STATIC int | |
1667 | xfs_mapping_buftarg( | |
1668 | xfs_buftarg_t *btp, | |
1669 | struct block_device *bdev) | |
1670 | { | |
1671 | struct backing_dev_info *bdi; | |
1672 | struct inode *inode; | |
1673 | struct address_space *mapping; | |
f5e54d6e | 1674 | static const struct address_space_operations mapping_aops = { |
1da177e4 | 1675 | .sync_page = block_sync_page, |
e965f963 | 1676 | .migratepage = fail_migrate_page, |
1da177e4 LT |
1677 | }; |
1678 | ||
1679 | inode = new_inode(bdev->bd_inode->i_sb); | |
1680 | if (!inode) { | |
1681 | printk(KERN_WARNING | |
1682 | "XFS: Cannot allocate mapping inode for device %s\n", | |
1683 | XFS_BUFTARG_NAME(btp)); | |
1684 | return ENOMEM; | |
1685 | } | |
85fe4025 | 1686 | inode->i_ino = get_next_ino(); |
1da177e4 LT |
1687 | inode->i_mode = S_IFBLK; |
1688 | inode->i_bdev = bdev; | |
1689 | inode->i_rdev = bdev->bd_dev; | |
1690 | bdi = blk_get_backing_dev_info(bdev); | |
1691 | if (!bdi) | |
1692 | bdi = &default_backing_dev_info; | |
1693 | mapping = &inode->i_data; | |
1694 | mapping->a_ops = &mapping_aops; | |
1695 | mapping->backing_dev_info = bdi; | |
1696 | mapping_set_gfp_mask(mapping, GFP_NOFS); | |
ce8e922c | 1697 | btp->bt_mapping = mapping; |
1da177e4 LT |
1698 | return 0; |
1699 | } | |
1700 | ||
a6867a68 DC |
1701 | STATIC int |
1702 | xfs_alloc_delwrite_queue( | |
e2a07812 JE |
1703 | xfs_buftarg_t *btp, |
1704 | const char *fsname) | |
a6867a68 | 1705 | { |
a6867a68 | 1706 | INIT_LIST_HEAD(&btp->bt_delwrite_queue); |
007c61c6 | 1707 | spin_lock_init(&btp->bt_delwrite_lock); |
a6867a68 | 1708 | btp->bt_flags = 0; |
e2a07812 | 1709 | btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd/%s", fsname); |
ff57ab21 DC |
1710 | if (IS_ERR(btp->bt_task)) |
1711 | return PTR_ERR(btp->bt_task); | |
1712 | return 0; | |
a6867a68 DC |
1713 | } |
1714 | ||
1da177e4 LT |
1715 | xfs_buftarg_t * |
1716 | xfs_alloc_buftarg( | |
ebad861b | 1717 | struct xfs_mount *mp, |
1da177e4 | 1718 | struct block_device *bdev, |
e2a07812 JE |
1719 | int external, |
1720 | const char *fsname) | |
1da177e4 LT |
1721 | { |
1722 | xfs_buftarg_t *btp; | |
1723 | ||
1724 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1725 | ||
ebad861b | 1726 | btp->bt_mount = mp; |
ce8e922c NS |
1727 | btp->bt_dev = bdev->bd_dev; |
1728 | btp->bt_bdev = bdev; | |
430cbeb8 DC |
1729 | INIT_LIST_HEAD(&btp->bt_lru); |
1730 | spin_lock_init(&btp->bt_lru_lock); | |
1da177e4 LT |
1731 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1732 | goto error; | |
1733 | if (xfs_mapping_buftarg(btp, bdev)) | |
1734 | goto error; | |
e2a07812 | 1735 | if (xfs_alloc_delwrite_queue(btp, fsname)) |
a6867a68 | 1736 | goto error; |
ff57ab21 DC |
1737 | btp->bt_shrinker.shrink = xfs_buftarg_shrink; |
1738 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; | |
1739 | register_shrinker(&btp->bt_shrinker); | |
1da177e4 LT |
1740 | return btp; |
1741 | ||
1742 | error: | |
f0e2d93c | 1743 | kmem_free(btp); |
1da177e4 LT |
1744 | return NULL; |
1745 | } | |
1746 | ||
1747 | ||
1748 | /* | |
ce8e922c | 1749 | * Delayed write buffer handling |
1da177e4 | 1750 | */ |
1da177e4 | 1751 | STATIC void |
ce8e922c NS |
1752 | xfs_buf_delwri_queue( |
1753 | xfs_buf_t *bp, | |
1da177e4 LT |
1754 | int unlock) |
1755 | { | |
ce8e922c NS |
1756 | struct list_head *dwq = &bp->b_target->bt_delwrite_queue; |
1757 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; | |
a6867a68 | 1758 | |
0b1b213f CH |
1759 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
1760 | ||
ce8e922c | 1761 | ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC)); |
1da177e4 | 1762 | |
a6867a68 | 1763 | spin_lock(dwlk); |
1da177e4 | 1764 | /* If already in the queue, dequeue and place at tail */ |
ce8e922c NS |
1765 | if (!list_empty(&bp->b_list)) { |
1766 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1767 | if (unlock) | |
1768 | atomic_dec(&bp->b_hold); | |
1769 | list_del(&bp->b_list); | |
1da177e4 LT |
1770 | } |
1771 | ||
c9c12971 DC |
1772 | if (list_empty(dwq)) { |
1773 | /* start xfsbufd as it is about to have something to do */ | |
1774 | wake_up_process(bp->b_target->bt_task); | |
1775 | } | |
1776 | ||
ce8e922c NS |
1777 | bp->b_flags |= _XBF_DELWRI_Q; |
1778 | list_add_tail(&bp->b_list, dwq); | |
1779 | bp->b_queuetime = jiffies; | |
a6867a68 | 1780 | spin_unlock(dwlk); |
1da177e4 LT |
1781 | |
1782 | if (unlock) | |
ce8e922c | 1783 | xfs_buf_unlock(bp); |
1da177e4 LT |
1784 | } |
1785 | ||
1786 | void | |
ce8e922c NS |
1787 | xfs_buf_delwri_dequeue( |
1788 | xfs_buf_t *bp) | |
1da177e4 | 1789 | { |
ce8e922c | 1790 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; |
1da177e4 LT |
1791 | int dequeued = 0; |
1792 | ||
a6867a68 | 1793 | spin_lock(dwlk); |
ce8e922c NS |
1794 | if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) { |
1795 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1796 | list_del_init(&bp->b_list); | |
1da177e4 LT |
1797 | dequeued = 1; |
1798 | } | |
ce8e922c | 1799 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q); |
a6867a68 | 1800 | spin_unlock(dwlk); |
1da177e4 LT |
1801 | |
1802 | if (dequeued) | |
ce8e922c | 1803 | xfs_buf_rele(bp); |
1da177e4 | 1804 | |
0b1b213f | 1805 | trace_xfs_buf_delwri_dequeue(bp, _RET_IP_); |
1da177e4 LT |
1806 | } |
1807 | ||
d808f617 DC |
1808 | /* |
1809 | * If a delwri buffer needs to be pushed before it has aged out, then promote | |
1810 | * it to the head of the delwri queue so that it will be flushed on the next | |
1811 | * xfsbufd run. We do this by resetting the queuetime of the buffer to be older | |
1812 | * than the age currently needed to flush the buffer. Hence the next time the | |
1813 | * xfsbufd sees it is guaranteed to be considered old enough to flush. | |
1814 | */ | |
1815 | void | |
1816 | xfs_buf_delwri_promote( | |
1817 | struct xfs_buf *bp) | |
1818 | { | |
1819 | struct xfs_buftarg *btp = bp->b_target; | |
1820 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1; | |
1821 | ||
1822 | ASSERT(bp->b_flags & XBF_DELWRI); | |
1823 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1824 | ||
1825 | /* | |
1826 | * Check the buffer age before locking the delayed write queue as we | |
1827 | * don't need to promote buffers that are already past the flush age. | |
1828 | */ | |
1829 | if (bp->b_queuetime < jiffies - age) | |
1830 | return; | |
1831 | bp->b_queuetime = jiffies - age; | |
1832 | spin_lock(&btp->bt_delwrite_lock); | |
1833 | list_move(&bp->b_list, &btp->bt_delwrite_queue); | |
1834 | spin_unlock(&btp->bt_delwrite_lock); | |
1835 | } | |
1836 | ||
1da177e4 | 1837 | STATIC void |
ce8e922c | 1838 | xfs_buf_runall_queues( |
1da177e4 LT |
1839 | struct workqueue_struct *queue) |
1840 | { | |
1841 | flush_workqueue(queue); | |
1842 | } | |
1843 | ||
585e6d88 DC |
1844 | /* |
1845 | * Move as many buffers as specified to the supplied list | |
1846 | * idicating if we skipped any buffers to prevent deadlocks. | |
1847 | */ | |
1848 | STATIC int | |
1849 | xfs_buf_delwri_split( | |
1850 | xfs_buftarg_t *target, | |
1851 | struct list_head *list, | |
5e6a07df | 1852 | unsigned long age) |
585e6d88 DC |
1853 | { |
1854 | xfs_buf_t *bp, *n; | |
1855 | struct list_head *dwq = &target->bt_delwrite_queue; | |
1856 | spinlock_t *dwlk = &target->bt_delwrite_lock; | |
1857 | int skipped = 0; | |
5e6a07df | 1858 | int force; |
585e6d88 | 1859 | |
5e6a07df | 1860 | force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
585e6d88 DC |
1861 | INIT_LIST_HEAD(list); |
1862 | spin_lock(dwlk); | |
1863 | list_for_each_entry_safe(bp, n, dwq, b_list) { | |
585e6d88 DC |
1864 | ASSERT(bp->b_flags & XBF_DELWRI); |
1865 | ||
4d16e924 | 1866 | if (!XFS_BUF_ISPINNED(bp) && !xfs_buf_cond_lock(bp)) { |
5e6a07df | 1867 | if (!force && |
585e6d88 DC |
1868 | time_before(jiffies, bp->b_queuetime + age)) { |
1869 | xfs_buf_unlock(bp); | |
1870 | break; | |
1871 | } | |
1872 | ||
1873 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q| | |
1874 | _XBF_RUN_QUEUES); | |
1875 | bp->b_flags |= XBF_WRITE; | |
1876 | list_move_tail(&bp->b_list, list); | |
bfe27419 | 1877 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
585e6d88 DC |
1878 | } else |
1879 | skipped++; | |
1880 | } | |
1881 | spin_unlock(dwlk); | |
1882 | ||
1883 | return skipped; | |
1884 | ||
1885 | } | |
1886 | ||
089716aa DC |
1887 | /* |
1888 | * Compare function is more complex than it needs to be because | |
1889 | * the return value is only 32 bits and we are doing comparisons | |
1890 | * on 64 bit values | |
1891 | */ | |
1892 | static int | |
1893 | xfs_buf_cmp( | |
1894 | void *priv, | |
1895 | struct list_head *a, | |
1896 | struct list_head *b) | |
1897 | { | |
1898 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1899 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1900 | xfs_daddr_t diff; | |
1901 | ||
1902 | diff = ap->b_bn - bp->b_bn; | |
1903 | if (diff < 0) | |
1904 | return -1; | |
1905 | if (diff > 0) | |
1906 | return 1; | |
1907 | return 0; | |
1908 | } | |
1909 | ||
1910 | void | |
1911 | xfs_buf_delwri_sort( | |
1912 | xfs_buftarg_t *target, | |
1913 | struct list_head *list) | |
1914 | { | |
1915 | list_sort(NULL, list, xfs_buf_cmp); | |
1916 | } | |
1917 | ||
1da177e4 | 1918 | STATIC int |
23ea4032 | 1919 | xfsbufd( |
585e6d88 | 1920 | void *data) |
1da177e4 | 1921 | { |
089716aa | 1922 | xfs_buftarg_t *target = (xfs_buftarg_t *)data; |
1da177e4 | 1923 | |
1da177e4 LT |
1924 | current->flags |= PF_MEMALLOC; |
1925 | ||
978c7b2f RW |
1926 | set_freezable(); |
1927 | ||
1da177e4 | 1928 | do { |
c9c12971 DC |
1929 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10); |
1930 | long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10); | |
089716aa DC |
1931 | int count = 0; |
1932 | struct list_head tmp; | |
c9c12971 | 1933 | |
3e1d1d28 | 1934 | if (unlikely(freezing(current))) { |
ce8e922c | 1935 | set_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
3e1d1d28 | 1936 | refrigerator(); |
abd0cf7a | 1937 | } else { |
ce8e922c | 1938 | clear_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
abd0cf7a | 1939 | } |
1da177e4 | 1940 | |
c9c12971 DC |
1941 | /* sleep for a long time if there is nothing to do. */ |
1942 | if (list_empty(&target->bt_delwrite_queue)) | |
1943 | tout = MAX_SCHEDULE_TIMEOUT; | |
1944 | schedule_timeout_interruptible(tout); | |
1da177e4 | 1945 | |
c9c12971 | 1946 | xfs_buf_delwri_split(target, &tmp, age); |
089716aa | 1947 | list_sort(NULL, &tmp, xfs_buf_cmp); |
1da177e4 | 1948 | while (!list_empty(&tmp)) { |
089716aa DC |
1949 | struct xfs_buf *bp; |
1950 | bp = list_first_entry(&tmp, struct xfs_buf, b_list); | |
ce8e922c | 1951 | list_del_init(&bp->b_list); |
939d723b | 1952 | xfs_bdstrat_cb(bp); |
585e6d88 | 1953 | count++; |
1da177e4 | 1954 | } |
f07c2250 NS |
1955 | if (count) |
1956 | blk_run_address_space(target->bt_mapping); | |
1da177e4 | 1957 | |
4df08c52 | 1958 | } while (!kthread_should_stop()); |
1da177e4 | 1959 | |
4df08c52 | 1960 | return 0; |
1da177e4 LT |
1961 | } |
1962 | ||
1963 | /* | |
ce8e922c NS |
1964 | * Go through all incore buffers, and release buffers if they belong to |
1965 | * the given device. This is used in filesystem error handling to | |
1966 | * preserve the consistency of its metadata. | |
1da177e4 LT |
1967 | */ |
1968 | int | |
1969 | xfs_flush_buftarg( | |
585e6d88 DC |
1970 | xfs_buftarg_t *target, |
1971 | int wait) | |
1da177e4 | 1972 | { |
089716aa | 1973 | xfs_buf_t *bp; |
585e6d88 | 1974 | int pincount = 0; |
089716aa DC |
1975 | LIST_HEAD(tmp_list); |
1976 | LIST_HEAD(wait_list); | |
1da177e4 | 1977 | |
c626d174 | 1978 | xfs_buf_runall_queues(xfsconvertd_workqueue); |
ce8e922c NS |
1979 | xfs_buf_runall_queues(xfsdatad_workqueue); |
1980 | xfs_buf_runall_queues(xfslogd_workqueue); | |
1da177e4 | 1981 | |
5e6a07df | 1982 | set_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
089716aa | 1983 | pincount = xfs_buf_delwri_split(target, &tmp_list, 0); |
1da177e4 LT |
1984 | |
1985 | /* | |
089716aa DC |
1986 | * Dropped the delayed write list lock, now walk the temporary list. |
1987 | * All I/O is issued async and then if we need to wait for completion | |
1988 | * we do that after issuing all the IO. | |
1da177e4 | 1989 | */ |
089716aa DC |
1990 | list_sort(NULL, &tmp_list, xfs_buf_cmp); |
1991 | while (!list_empty(&tmp_list)) { | |
1992 | bp = list_first_entry(&tmp_list, struct xfs_buf, b_list); | |
585e6d88 | 1993 | ASSERT(target == bp->b_target); |
089716aa DC |
1994 | list_del_init(&bp->b_list); |
1995 | if (wait) { | |
ce8e922c | 1996 | bp->b_flags &= ~XBF_ASYNC; |
089716aa DC |
1997 | list_add(&bp->b_list, &wait_list); |
1998 | } | |
939d723b | 1999 | xfs_bdstrat_cb(bp); |
1da177e4 LT |
2000 | } |
2001 | ||
089716aa DC |
2002 | if (wait) { |
2003 | /* Expedite and wait for IO to complete. */ | |
f07c2250 | 2004 | blk_run_address_space(target->bt_mapping); |
089716aa DC |
2005 | while (!list_empty(&wait_list)) { |
2006 | bp = list_first_entry(&wait_list, struct xfs_buf, b_list); | |
f07c2250 | 2007 | |
089716aa | 2008 | list_del_init(&bp->b_list); |
1a1a3e97 | 2009 | xfs_buf_iowait(bp); |
089716aa DC |
2010 | xfs_buf_relse(bp); |
2011 | } | |
1da177e4 LT |
2012 | } |
2013 | ||
1da177e4 LT |
2014 | return pincount; |
2015 | } | |
2016 | ||
04d8b284 | 2017 | int __init |
ce8e922c | 2018 | xfs_buf_init(void) |
1da177e4 | 2019 | { |
8758280f NS |
2020 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
2021 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 2022 | if (!xfs_buf_zone) |
0b1b213f | 2023 | goto out; |
04d8b284 | 2024 | |
51749e47 | 2025 | xfslogd_workqueue = alloc_workqueue("xfslogd", |
6370a6ad | 2026 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 1); |
23ea4032 | 2027 | if (!xfslogd_workqueue) |
04d8b284 | 2028 | goto out_free_buf_zone; |
1da177e4 | 2029 | |
b4337692 | 2030 | xfsdatad_workqueue = create_workqueue("xfsdatad"); |
23ea4032 CH |
2031 | if (!xfsdatad_workqueue) |
2032 | goto out_destroy_xfslogd_workqueue; | |
1da177e4 | 2033 | |
c626d174 DC |
2034 | xfsconvertd_workqueue = create_workqueue("xfsconvertd"); |
2035 | if (!xfsconvertd_workqueue) | |
2036 | goto out_destroy_xfsdatad_workqueue; | |
2037 | ||
23ea4032 | 2038 | return 0; |
1da177e4 | 2039 | |
c626d174 DC |
2040 | out_destroy_xfsdatad_workqueue: |
2041 | destroy_workqueue(xfsdatad_workqueue); | |
23ea4032 CH |
2042 | out_destroy_xfslogd_workqueue: |
2043 | destroy_workqueue(xfslogd_workqueue); | |
23ea4032 | 2044 | out_free_buf_zone: |
ce8e922c | 2045 | kmem_zone_destroy(xfs_buf_zone); |
0b1b213f | 2046 | out: |
8758280f | 2047 | return -ENOMEM; |
1da177e4 LT |
2048 | } |
2049 | ||
1da177e4 | 2050 | void |
ce8e922c | 2051 | xfs_buf_terminate(void) |
1da177e4 | 2052 | { |
c626d174 | 2053 | destroy_workqueue(xfsconvertd_workqueue); |
04d8b284 CH |
2054 | destroy_workqueue(xfsdatad_workqueue); |
2055 | destroy_workqueue(xfslogd_workqueue); | |
ce8e922c | 2056 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 2057 | } |
e6a0e9cd TS |
2058 | |
2059 | #ifdef CONFIG_KDB_MODULES | |
2060 | struct list_head * | |
2061 | xfs_get_buftarg_list(void) | |
2062 | { | |
2063 | return &xfs_buftarg_list; | |
2064 | } | |
2065 | #endif |