]>
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" | |
40 | #include "xfs_ag.h" | |
41 | #include "xfs_dmapi.h" | |
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 *); |
27496a8c | 47 | STATIC int xfsbufd_wakeup(int, gfp_t); |
ce8e922c | 48 | STATIC void xfs_buf_delwri_queue(xfs_buf_t *, int); |
8e1f936b RR |
49 | static struct shrinker xfs_buf_shake = { |
50 | .shrink = xfsbufd_wakeup, | |
51 | .seeks = DEFAULT_SEEKS, | |
52 | }; | |
23ea4032 | 53 | |
7989cb8e | 54 | static struct workqueue_struct *xfslogd_workqueue; |
0829c360 | 55 | struct workqueue_struct *xfsdatad_workqueue; |
c626d174 | 56 | struct workqueue_struct *xfsconvertd_workqueue; |
1da177e4 | 57 | |
ce8e922c NS |
58 | #ifdef XFS_BUF_LOCK_TRACKING |
59 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
60 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
61 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 62 | #else |
ce8e922c NS |
63 | # define XB_SET_OWNER(bp) do { } while (0) |
64 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
65 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
66 | #endif |
67 | ||
ce8e922c NS |
68 | #define xb_to_gfp(flags) \ |
69 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \ | |
70 | ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN) | |
1da177e4 | 71 | |
ce8e922c NS |
72 | #define xb_to_km(flags) \ |
73 | (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP) | |
1da177e4 | 74 | |
ce8e922c NS |
75 | #define xfs_buf_allocate(flags) \ |
76 | kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags)) | |
77 | #define xfs_buf_deallocate(bp) \ | |
78 | kmem_zone_free(xfs_buf_zone, (bp)); | |
1da177e4 | 79 | |
73c77e2c JB |
80 | static inline int |
81 | xfs_buf_is_vmapped( | |
82 | struct xfs_buf *bp) | |
83 | { | |
84 | /* | |
85 | * Return true if the buffer is vmapped. | |
86 | * | |
87 | * The XBF_MAPPED flag is set if the buffer should be mapped, but the | |
88 | * code is clever enough to know it doesn't have to map a single page, | |
89 | * so the check has to be both for XBF_MAPPED and bp->b_page_count > 1. | |
90 | */ | |
91 | return (bp->b_flags & XBF_MAPPED) && bp->b_page_count > 1; | |
92 | } | |
93 | ||
94 | static inline int | |
95 | xfs_buf_vmap_len( | |
96 | struct xfs_buf *bp) | |
97 | { | |
98 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
99 | } | |
100 | ||
1da177e4 | 101 | /* |
ce8e922c | 102 | * Page Region interfaces. |
1da177e4 | 103 | * |
ce8e922c NS |
104 | * For pages in filesystems where the blocksize is smaller than the |
105 | * pagesize, we use the page->private field (long) to hold a bitmap | |
106 | * of uptodate regions within the page. | |
1da177e4 | 107 | * |
ce8e922c | 108 | * Each such region is "bytes per page / bits per long" bytes long. |
1da177e4 | 109 | * |
ce8e922c NS |
110 | * NBPPR == number-of-bytes-per-page-region |
111 | * BTOPR == bytes-to-page-region (rounded up) | |
112 | * BTOPRT == bytes-to-page-region-truncated (rounded down) | |
1da177e4 LT |
113 | */ |
114 | #if (BITS_PER_LONG == 32) | |
115 | #define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */ | |
116 | #elif (BITS_PER_LONG == 64) | |
117 | #define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */ | |
118 | #else | |
119 | #error BITS_PER_LONG must be 32 or 64 | |
120 | #endif | |
121 | #define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG) | |
122 | #define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT) | |
123 | #define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT)) | |
124 | ||
125 | STATIC unsigned long | |
126 | page_region_mask( | |
127 | size_t offset, | |
128 | size_t length) | |
129 | { | |
130 | unsigned long mask; | |
131 | int first, final; | |
132 | ||
133 | first = BTOPR(offset); | |
134 | final = BTOPRT(offset + length - 1); | |
135 | first = min(first, final); | |
136 | ||
137 | mask = ~0UL; | |
138 | mask <<= BITS_PER_LONG - (final - first); | |
139 | mask >>= BITS_PER_LONG - (final); | |
140 | ||
141 | ASSERT(offset + length <= PAGE_CACHE_SIZE); | |
142 | ASSERT((final - first) < BITS_PER_LONG && (final - first) >= 0); | |
143 | ||
144 | return mask; | |
145 | } | |
146 | ||
b8f82a4a | 147 | STATIC void |
1da177e4 LT |
148 | set_page_region( |
149 | struct page *page, | |
150 | size_t offset, | |
151 | size_t length) | |
152 | { | |
4c21e2f2 HD |
153 | set_page_private(page, |
154 | page_private(page) | page_region_mask(offset, length)); | |
155 | if (page_private(page) == ~0UL) | |
1da177e4 LT |
156 | SetPageUptodate(page); |
157 | } | |
158 | ||
b8f82a4a | 159 | STATIC int |
1da177e4 LT |
160 | test_page_region( |
161 | struct page *page, | |
162 | size_t offset, | |
163 | size_t length) | |
164 | { | |
165 | unsigned long mask = page_region_mask(offset, length); | |
166 | ||
4c21e2f2 | 167 | return (mask && (page_private(page) & mask) == mask); |
1da177e4 LT |
168 | } |
169 | ||
1da177e4 | 170 | /* |
ce8e922c | 171 | * Internal xfs_buf_t object manipulation |
1da177e4 LT |
172 | */ |
173 | ||
174 | STATIC void | |
ce8e922c NS |
175 | _xfs_buf_initialize( |
176 | xfs_buf_t *bp, | |
1da177e4 | 177 | xfs_buftarg_t *target, |
204ab25f | 178 | xfs_off_t range_base, |
1da177e4 | 179 | size_t range_length, |
ce8e922c | 180 | xfs_buf_flags_t flags) |
1da177e4 LT |
181 | { |
182 | /* | |
ce8e922c | 183 | * We don't want certain flags to appear in b_flags. |
1da177e4 | 184 | */ |
ce8e922c NS |
185 | flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD); |
186 | ||
187 | memset(bp, 0, sizeof(xfs_buf_t)); | |
188 | atomic_set(&bp->b_hold, 1); | |
b4dd330b | 189 | init_completion(&bp->b_iowait); |
ce8e922c NS |
190 | INIT_LIST_HEAD(&bp->b_list); |
191 | INIT_LIST_HEAD(&bp->b_hash_list); | |
192 | init_MUTEX_LOCKED(&bp->b_sema); /* held, no waiters */ | |
193 | XB_SET_OWNER(bp); | |
194 | bp->b_target = target; | |
195 | bp->b_file_offset = range_base; | |
1da177e4 LT |
196 | /* |
197 | * Set buffer_length and count_desired to the same value initially. | |
198 | * I/O routines should use count_desired, which will be the same in | |
199 | * most cases but may be reset (e.g. XFS recovery). | |
200 | */ | |
ce8e922c NS |
201 | bp->b_buffer_length = bp->b_count_desired = range_length; |
202 | bp->b_flags = flags; | |
203 | bp->b_bn = XFS_BUF_DADDR_NULL; | |
204 | atomic_set(&bp->b_pin_count, 0); | |
205 | init_waitqueue_head(&bp->b_waiters); | |
206 | ||
207 | XFS_STATS_INC(xb_create); | |
0b1b213f CH |
208 | |
209 | trace_xfs_buf_init(bp, _RET_IP_); | |
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 NS |
223 | if (bp->b_pages == NULL) { |
224 | bp->b_offset = xfs_buf_poff(bp->b_file_offset); | |
225 | bp->b_page_count = page_count; | |
226 | if (page_count <= XB_PAGES) { | |
227 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 228 | } else { |
ce8e922c NS |
229 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
230 | page_count, xb_to_km(flags)); | |
231 | if (bp->b_pages == NULL) | |
1da177e4 LT |
232 | return -ENOMEM; |
233 | } | |
ce8e922c | 234 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
235 | } |
236 | return 0; | |
237 | } | |
238 | ||
239 | /* | |
ce8e922c | 240 | * Frees b_pages if it was allocated. |
1da177e4 LT |
241 | */ |
242 | STATIC void | |
ce8e922c | 243 | _xfs_buf_free_pages( |
1da177e4 LT |
244 | xfs_buf_t *bp) |
245 | { | |
ce8e922c | 246 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 247 | kmem_free(bp->b_pages); |
3fc98b1a | 248 | bp->b_pages = NULL; |
1da177e4 LT |
249 | } |
250 | } | |
251 | ||
252 | /* | |
253 | * Releases the specified buffer. | |
254 | * | |
255 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 256 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
257 | * hashed and refcounted buffers |
258 | */ | |
259 | void | |
ce8e922c | 260 | xfs_buf_free( |
1da177e4 LT |
261 | xfs_buf_t *bp) |
262 | { | |
0b1b213f | 263 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 264 | |
ce8e922c | 265 | ASSERT(list_empty(&bp->b_hash_list)); |
1da177e4 | 266 | |
1fa40b01 | 267 | if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) { |
1da177e4 LT |
268 | uint i; |
269 | ||
73c77e2c | 270 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
271 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
272 | bp->b_page_count); | |
1da177e4 | 273 | |
948ecdb4 NS |
274 | for (i = 0; i < bp->b_page_count; i++) { |
275 | struct page *page = bp->b_pages[i]; | |
276 | ||
1fa40b01 CH |
277 | if (bp->b_flags & _XBF_PAGE_CACHE) |
278 | ASSERT(!PagePrivate(page)); | |
948ecdb4 NS |
279 | page_cache_release(page); |
280 | } | |
1da177e4 | 281 | } |
3fc98b1a | 282 | _xfs_buf_free_pages(bp); |
ce8e922c | 283 | xfs_buf_deallocate(bp); |
1da177e4 LT |
284 | } |
285 | ||
286 | /* | |
287 | * Finds all pages for buffer in question and builds it's page list. | |
288 | */ | |
289 | STATIC int | |
ce8e922c | 290 | _xfs_buf_lookup_pages( |
1da177e4 LT |
291 | xfs_buf_t *bp, |
292 | uint flags) | |
293 | { | |
ce8e922c NS |
294 | struct address_space *mapping = bp->b_target->bt_mapping; |
295 | size_t blocksize = bp->b_target->bt_bsize; | |
296 | size_t size = bp->b_count_desired; | |
1da177e4 | 297 | size_t nbytes, offset; |
ce8e922c | 298 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 LT |
299 | unsigned short page_count, i; |
300 | pgoff_t first; | |
204ab25f | 301 | xfs_off_t end; |
1da177e4 LT |
302 | int error; |
303 | ||
ce8e922c NS |
304 | end = bp->b_file_offset + bp->b_buffer_length; |
305 | page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset); | |
1da177e4 | 306 | |
ce8e922c | 307 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
308 | if (unlikely(error)) |
309 | return error; | |
ce8e922c | 310 | bp->b_flags |= _XBF_PAGE_CACHE; |
1da177e4 | 311 | |
ce8e922c NS |
312 | offset = bp->b_offset; |
313 | first = bp->b_file_offset >> PAGE_CACHE_SHIFT; | |
1da177e4 | 314 | |
ce8e922c | 315 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
316 | struct page *page; |
317 | uint retries = 0; | |
318 | ||
319 | retry: | |
320 | page = find_or_create_page(mapping, first + i, gfp_mask); | |
321 | if (unlikely(page == NULL)) { | |
ce8e922c NS |
322 | if (flags & XBF_READ_AHEAD) { |
323 | bp->b_page_count = i; | |
6ab455ee CH |
324 | for (i = 0; i < bp->b_page_count; i++) |
325 | unlock_page(bp->b_pages[i]); | |
1da177e4 LT |
326 | return -ENOMEM; |
327 | } | |
328 | ||
329 | /* | |
330 | * This could deadlock. | |
331 | * | |
332 | * But until all the XFS lowlevel code is revamped to | |
333 | * handle buffer allocation failures we can't do much. | |
334 | */ | |
335 | if (!(++retries % 100)) | |
336 | printk(KERN_ERR | |
337 | "XFS: possible memory allocation " | |
338 | "deadlock in %s (mode:0x%x)\n", | |
34a622b2 | 339 | __func__, gfp_mask); |
1da177e4 | 340 | |
ce8e922c | 341 | XFS_STATS_INC(xb_page_retries); |
23ea4032 | 342 | xfsbufd_wakeup(0, gfp_mask); |
8aa7e847 | 343 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
344 | goto retry; |
345 | } | |
346 | ||
ce8e922c | 347 | XFS_STATS_INC(xb_page_found); |
1da177e4 LT |
348 | |
349 | nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset); | |
350 | size -= nbytes; | |
351 | ||
948ecdb4 | 352 | ASSERT(!PagePrivate(page)); |
1da177e4 LT |
353 | if (!PageUptodate(page)) { |
354 | page_count--; | |
6ab455ee CH |
355 | if (blocksize >= PAGE_CACHE_SIZE) { |
356 | if (flags & XBF_READ) | |
357 | bp->b_flags |= _XBF_PAGE_LOCKED; | |
358 | } else if (!PagePrivate(page)) { | |
1da177e4 LT |
359 | if (test_page_region(page, offset, nbytes)) |
360 | page_count++; | |
361 | } | |
362 | } | |
363 | ||
ce8e922c | 364 | bp->b_pages[i] = page; |
1da177e4 LT |
365 | offset = 0; |
366 | } | |
367 | ||
6ab455ee CH |
368 | if (!(bp->b_flags & _XBF_PAGE_LOCKED)) { |
369 | for (i = 0; i < bp->b_page_count; i++) | |
370 | unlock_page(bp->b_pages[i]); | |
371 | } | |
372 | ||
ce8e922c NS |
373 | if (page_count == bp->b_page_count) |
374 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 375 | |
1da177e4 LT |
376 | return error; |
377 | } | |
378 | ||
379 | /* | |
380 | * Map buffer into kernel address-space if nessecary. | |
381 | */ | |
382 | STATIC int | |
ce8e922c | 383 | _xfs_buf_map_pages( |
1da177e4 LT |
384 | xfs_buf_t *bp, |
385 | uint flags) | |
386 | { | |
387 | /* A single page buffer is always mappable */ | |
ce8e922c NS |
388 | if (bp->b_page_count == 1) { |
389 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; | |
390 | bp->b_flags |= XBF_MAPPED; | |
391 | } else if (flags & XBF_MAPPED) { | |
8a262e57 AE |
392 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, |
393 | -1, PAGE_KERNEL); | |
ce8e922c | 394 | if (unlikely(bp->b_addr == NULL)) |
1da177e4 | 395 | return -ENOMEM; |
ce8e922c NS |
396 | bp->b_addr += bp->b_offset; |
397 | bp->b_flags |= XBF_MAPPED; | |
1da177e4 LT |
398 | } |
399 | ||
400 | return 0; | |
401 | } | |
402 | ||
403 | /* | |
404 | * Finding and Reading Buffers | |
405 | */ | |
406 | ||
407 | /* | |
ce8e922c | 408 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 LT |
409 | * a given range of an inode. The buffer is returned |
410 | * locked. If other overlapping buffers exist, they are | |
411 | * released before the new buffer is created and locked, | |
412 | * which may imply that this call will block until those buffers | |
413 | * are unlocked. No I/O is implied by this call. | |
414 | */ | |
415 | xfs_buf_t * | |
ce8e922c | 416 | _xfs_buf_find( |
1da177e4 | 417 | xfs_buftarg_t *btp, /* block device target */ |
204ab25f | 418 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 419 | size_t isize, /* length of range */ |
ce8e922c NS |
420 | xfs_buf_flags_t flags, |
421 | xfs_buf_t *new_bp) | |
1da177e4 | 422 | { |
204ab25f | 423 | xfs_off_t range_base; |
1da177e4 LT |
424 | size_t range_length; |
425 | xfs_bufhash_t *hash; | |
ce8e922c | 426 | xfs_buf_t *bp, *n; |
1da177e4 LT |
427 | |
428 | range_base = (ioff << BBSHIFT); | |
429 | range_length = (isize << BBSHIFT); | |
430 | ||
431 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
ce8e922c | 432 | ASSERT(!(range_length < (1 << btp->bt_sshift))); |
204ab25f | 433 | ASSERT(!(range_base & (xfs_off_t)btp->bt_smask)); |
1da177e4 LT |
434 | |
435 | hash = &btp->bt_hash[hash_long((unsigned long)ioff, btp->bt_hashshift)]; | |
436 | ||
437 | spin_lock(&hash->bh_lock); | |
438 | ||
ce8e922c NS |
439 | list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) { |
440 | ASSERT(btp == bp->b_target); | |
441 | if (bp->b_file_offset == range_base && | |
442 | bp->b_buffer_length == range_length) { | |
1da177e4 | 443 | /* |
ce8e922c | 444 | * If we look at something, bring it to the |
1da177e4 LT |
445 | * front of the list for next time. |
446 | */ | |
ce8e922c NS |
447 | atomic_inc(&bp->b_hold); |
448 | list_move(&bp->b_hash_list, &hash->bh_list); | |
1da177e4 LT |
449 | goto found; |
450 | } | |
451 | } | |
452 | ||
453 | /* No match found */ | |
ce8e922c NS |
454 | if (new_bp) { |
455 | _xfs_buf_initialize(new_bp, btp, range_base, | |
1da177e4 | 456 | range_length, flags); |
ce8e922c NS |
457 | new_bp->b_hash = hash; |
458 | list_add(&new_bp->b_hash_list, &hash->bh_list); | |
1da177e4 | 459 | } else { |
ce8e922c | 460 | XFS_STATS_INC(xb_miss_locked); |
1da177e4 LT |
461 | } |
462 | ||
463 | spin_unlock(&hash->bh_lock); | |
ce8e922c | 464 | return new_bp; |
1da177e4 LT |
465 | |
466 | found: | |
467 | spin_unlock(&hash->bh_lock); | |
468 | ||
469 | /* Attempt to get the semaphore without sleeping, | |
470 | * if this does not work then we need to drop the | |
471 | * spinlock and do a hard attempt on the semaphore. | |
472 | */ | |
ce8e922c NS |
473 | if (down_trylock(&bp->b_sema)) { |
474 | if (!(flags & XBF_TRYLOCK)) { | |
1da177e4 | 475 | /* wait for buffer ownership */ |
ce8e922c NS |
476 | xfs_buf_lock(bp); |
477 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
478 | } else { |
479 | /* We asked for a trylock and failed, no need | |
480 | * to look at file offset and length here, we | |
ce8e922c NS |
481 | * know that this buffer at least overlaps our |
482 | * buffer and is locked, therefore our buffer | |
483 | * either does not exist, or is this buffer. | |
1da177e4 | 484 | */ |
ce8e922c NS |
485 | xfs_buf_rele(bp); |
486 | XFS_STATS_INC(xb_busy_locked); | |
487 | return NULL; | |
1da177e4 LT |
488 | } |
489 | } else { | |
490 | /* trylock worked */ | |
ce8e922c | 491 | XB_SET_OWNER(bp); |
1da177e4 LT |
492 | } |
493 | ||
ce8e922c NS |
494 | if (bp->b_flags & XBF_STALE) { |
495 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
496 | bp->b_flags &= XBF_MAPPED; | |
2f926587 | 497 | } |
0b1b213f CH |
498 | |
499 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ce8e922c NS |
500 | XFS_STATS_INC(xb_get_locked); |
501 | return bp; | |
1da177e4 LT |
502 | } |
503 | ||
504 | /* | |
ce8e922c | 505 | * Assembles a buffer covering the specified range. |
1da177e4 LT |
506 | * Storage in memory for all portions of the buffer will be allocated, |
507 | * although backing storage may not be. | |
508 | */ | |
509 | xfs_buf_t * | |
6ad112bf | 510 | xfs_buf_get( |
1da177e4 | 511 | xfs_buftarg_t *target,/* target for buffer */ |
204ab25f | 512 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 513 | size_t isize, /* length of range */ |
ce8e922c | 514 | xfs_buf_flags_t flags) |
1da177e4 | 515 | { |
ce8e922c | 516 | xfs_buf_t *bp, *new_bp; |
1da177e4 LT |
517 | int error = 0, i; |
518 | ||
ce8e922c NS |
519 | new_bp = xfs_buf_allocate(flags); |
520 | if (unlikely(!new_bp)) | |
1da177e4 LT |
521 | return NULL; |
522 | ||
ce8e922c NS |
523 | bp = _xfs_buf_find(target, ioff, isize, flags, new_bp); |
524 | if (bp == new_bp) { | |
525 | error = _xfs_buf_lookup_pages(bp, flags); | |
1da177e4 LT |
526 | if (error) |
527 | goto no_buffer; | |
528 | } else { | |
ce8e922c NS |
529 | xfs_buf_deallocate(new_bp); |
530 | if (unlikely(bp == NULL)) | |
1da177e4 LT |
531 | return NULL; |
532 | } | |
533 | ||
ce8e922c NS |
534 | for (i = 0; i < bp->b_page_count; i++) |
535 | mark_page_accessed(bp->b_pages[i]); | |
1da177e4 | 536 | |
ce8e922c NS |
537 | if (!(bp->b_flags & XBF_MAPPED)) { |
538 | error = _xfs_buf_map_pages(bp, flags); | |
1da177e4 LT |
539 | if (unlikely(error)) { |
540 | printk(KERN_WARNING "%s: failed to map pages\n", | |
34a622b2 | 541 | __func__); |
1da177e4 LT |
542 | goto no_buffer; |
543 | } | |
544 | } | |
545 | ||
ce8e922c | 546 | XFS_STATS_INC(xb_get); |
1da177e4 LT |
547 | |
548 | /* | |
549 | * Always fill in the block number now, the mapped cases can do | |
550 | * their own overlay of this later. | |
551 | */ | |
ce8e922c NS |
552 | bp->b_bn = ioff; |
553 | bp->b_count_desired = bp->b_buffer_length; | |
1da177e4 | 554 | |
0b1b213f | 555 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 556 | return bp; |
1da177e4 LT |
557 | |
558 | no_buffer: | |
ce8e922c NS |
559 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
560 | xfs_buf_unlock(bp); | |
561 | xfs_buf_rele(bp); | |
1da177e4 LT |
562 | return NULL; |
563 | } | |
564 | ||
5d765b97 CH |
565 | STATIC int |
566 | _xfs_buf_read( | |
567 | xfs_buf_t *bp, | |
568 | xfs_buf_flags_t flags) | |
569 | { | |
570 | int status; | |
571 | ||
5d765b97 CH |
572 | ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE))); |
573 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); | |
574 | ||
575 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \ | |
576 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
577 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \ | |
578 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
579 | ||
580 | status = xfs_buf_iorequest(bp); | |
581 | if (!status && !(flags & XBF_ASYNC)) | |
582 | status = xfs_buf_iowait(bp); | |
583 | return status; | |
584 | } | |
585 | ||
1da177e4 | 586 | xfs_buf_t * |
6ad112bf | 587 | xfs_buf_read( |
1da177e4 | 588 | xfs_buftarg_t *target, |
204ab25f | 589 | xfs_off_t ioff, |
1da177e4 | 590 | size_t isize, |
ce8e922c | 591 | xfs_buf_flags_t flags) |
1da177e4 | 592 | { |
ce8e922c NS |
593 | xfs_buf_t *bp; |
594 | ||
595 | flags |= XBF_READ; | |
596 | ||
6ad112bf | 597 | bp = xfs_buf_get(target, ioff, isize, flags); |
ce8e922c | 598 | if (bp) { |
0b1b213f CH |
599 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
600 | ||
ce8e922c | 601 | if (!XFS_BUF_ISDONE(bp)) { |
ce8e922c | 602 | XFS_STATS_INC(xb_get_read); |
5d765b97 | 603 | _xfs_buf_read(bp, flags); |
ce8e922c | 604 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
605 | /* |
606 | * Read ahead call which is already satisfied, | |
607 | * drop the buffer | |
608 | */ | |
609 | goto no_buffer; | |
610 | } else { | |
1da177e4 | 611 | /* We do not want read in the flags */ |
ce8e922c | 612 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
613 | } |
614 | } | |
615 | ||
ce8e922c | 616 | return bp; |
1da177e4 LT |
617 | |
618 | no_buffer: | |
ce8e922c NS |
619 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
620 | xfs_buf_unlock(bp); | |
621 | xfs_buf_rele(bp); | |
1da177e4 LT |
622 | return NULL; |
623 | } | |
624 | ||
1da177e4 | 625 | /* |
ce8e922c NS |
626 | * If we are not low on memory then do the readahead in a deadlock |
627 | * safe manner. | |
1da177e4 LT |
628 | */ |
629 | void | |
ce8e922c | 630 | xfs_buf_readahead( |
1da177e4 | 631 | xfs_buftarg_t *target, |
204ab25f | 632 | xfs_off_t ioff, |
1da177e4 | 633 | size_t isize, |
ce8e922c | 634 | xfs_buf_flags_t flags) |
1da177e4 LT |
635 | { |
636 | struct backing_dev_info *bdi; | |
637 | ||
ce8e922c | 638 | bdi = target->bt_mapping->backing_dev_info; |
1da177e4 LT |
639 | if (bdi_read_congested(bdi)) |
640 | return; | |
641 | ||
ce8e922c | 642 | flags |= (XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD); |
6ad112bf | 643 | xfs_buf_read(target, ioff, isize, flags); |
1da177e4 LT |
644 | } |
645 | ||
646 | xfs_buf_t * | |
ce8e922c | 647 | xfs_buf_get_empty( |
1da177e4 LT |
648 | size_t len, |
649 | xfs_buftarg_t *target) | |
650 | { | |
ce8e922c | 651 | xfs_buf_t *bp; |
1da177e4 | 652 | |
ce8e922c NS |
653 | bp = xfs_buf_allocate(0); |
654 | if (bp) | |
655 | _xfs_buf_initialize(bp, target, 0, len, 0); | |
656 | return bp; | |
1da177e4 LT |
657 | } |
658 | ||
659 | static inline struct page * | |
660 | mem_to_page( | |
661 | void *addr) | |
662 | { | |
9e2779fa | 663 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
664 | return virt_to_page(addr); |
665 | } else { | |
666 | return vmalloc_to_page(addr); | |
667 | } | |
668 | } | |
669 | ||
670 | int | |
ce8e922c NS |
671 | xfs_buf_associate_memory( |
672 | xfs_buf_t *bp, | |
1da177e4 LT |
673 | void *mem, |
674 | size_t len) | |
675 | { | |
676 | int rval; | |
677 | int i = 0; | |
d1afb678 LM |
678 | unsigned long pageaddr; |
679 | unsigned long offset; | |
680 | size_t buflen; | |
1da177e4 LT |
681 | int page_count; |
682 | ||
d1afb678 LM |
683 | pageaddr = (unsigned long)mem & PAGE_CACHE_MASK; |
684 | offset = (unsigned long)mem - pageaddr; | |
685 | buflen = PAGE_CACHE_ALIGN(len + offset); | |
686 | page_count = buflen >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
687 | |
688 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
689 | if (bp->b_pages) |
690 | _xfs_buf_free_pages(bp); | |
1da177e4 | 691 | |
ce8e922c NS |
692 | bp->b_pages = NULL; |
693 | bp->b_addr = mem; | |
1da177e4 | 694 | |
36fae17a | 695 | rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK); |
1da177e4 LT |
696 | if (rval) |
697 | return rval; | |
698 | ||
ce8e922c | 699 | bp->b_offset = offset; |
d1afb678 LM |
700 | |
701 | for (i = 0; i < bp->b_page_count; i++) { | |
702 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
703 | pageaddr += PAGE_CACHE_SIZE; | |
1da177e4 | 704 | } |
1da177e4 | 705 | |
d1afb678 LM |
706 | bp->b_count_desired = len; |
707 | bp->b_buffer_length = buflen; | |
ce8e922c | 708 | bp->b_flags |= XBF_MAPPED; |
6ab455ee | 709 | bp->b_flags &= ~_XBF_PAGE_LOCKED; |
1da177e4 LT |
710 | |
711 | return 0; | |
712 | } | |
713 | ||
714 | xfs_buf_t * | |
ce8e922c | 715 | xfs_buf_get_noaddr( |
1da177e4 LT |
716 | size_t len, |
717 | xfs_buftarg_t *target) | |
718 | { | |
1fa40b01 CH |
719 | unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT; |
720 | int error, i; | |
1da177e4 | 721 | xfs_buf_t *bp; |
1da177e4 | 722 | |
ce8e922c | 723 | bp = xfs_buf_allocate(0); |
1da177e4 LT |
724 | if (unlikely(bp == NULL)) |
725 | goto fail; | |
ce8e922c | 726 | _xfs_buf_initialize(bp, target, 0, len, 0); |
1da177e4 | 727 | |
1fa40b01 CH |
728 | error = _xfs_buf_get_pages(bp, page_count, 0); |
729 | if (error) | |
1da177e4 LT |
730 | goto fail_free_buf; |
731 | ||
1fa40b01 CH |
732 | for (i = 0; i < page_count; i++) { |
733 | bp->b_pages[i] = alloc_page(GFP_KERNEL); | |
734 | if (!bp->b_pages[i]) | |
735 | goto fail_free_mem; | |
1da177e4 | 736 | } |
1fa40b01 | 737 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 738 | |
1fa40b01 CH |
739 | error = _xfs_buf_map_pages(bp, XBF_MAPPED); |
740 | if (unlikely(error)) { | |
741 | printk(KERN_WARNING "%s: failed to map pages\n", | |
34a622b2 | 742 | __func__); |
1da177e4 | 743 | goto fail_free_mem; |
1fa40b01 | 744 | } |
1da177e4 | 745 | |
ce8e922c | 746 | xfs_buf_unlock(bp); |
1da177e4 | 747 | |
0b1b213f | 748 | trace_xfs_buf_get_noaddr(bp, _RET_IP_); |
1da177e4 | 749 | return bp; |
1fa40b01 | 750 | |
1da177e4 | 751 | fail_free_mem: |
1fa40b01 CH |
752 | while (--i >= 0) |
753 | __free_page(bp->b_pages[i]); | |
ca165b88 | 754 | _xfs_buf_free_pages(bp); |
1da177e4 | 755 | fail_free_buf: |
ca165b88 | 756 | xfs_buf_deallocate(bp); |
1da177e4 LT |
757 | fail: |
758 | return NULL; | |
759 | } | |
760 | ||
761 | /* | |
1da177e4 LT |
762 | * Increment reference count on buffer, to hold the buffer concurrently |
763 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
764 | * Must hold the buffer already to call this function. |
765 | */ | |
766 | void | |
ce8e922c NS |
767 | xfs_buf_hold( |
768 | xfs_buf_t *bp) | |
1da177e4 | 769 | { |
0b1b213f | 770 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 771 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
772 | } |
773 | ||
774 | /* | |
ce8e922c NS |
775 | * Releases a hold on the specified buffer. If the |
776 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
777 | */ |
778 | void | |
ce8e922c NS |
779 | xfs_buf_rele( |
780 | xfs_buf_t *bp) | |
1da177e4 | 781 | { |
ce8e922c | 782 | xfs_bufhash_t *hash = bp->b_hash; |
1da177e4 | 783 | |
0b1b213f | 784 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 785 | |
fad3aa1e NS |
786 | if (unlikely(!hash)) { |
787 | ASSERT(!bp->b_relse); | |
788 | if (atomic_dec_and_test(&bp->b_hold)) | |
789 | xfs_buf_free(bp); | |
790 | return; | |
791 | } | |
792 | ||
3790689f | 793 | ASSERT(atomic_read(&bp->b_hold) > 0); |
ce8e922c NS |
794 | if (atomic_dec_and_lock(&bp->b_hold, &hash->bh_lock)) { |
795 | if (bp->b_relse) { | |
796 | atomic_inc(&bp->b_hold); | |
1da177e4 | 797 | spin_unlock(&hash->bh_lock); |
ce8e922c NS |
798 | (*(bp->b_relse)) (bp); |
799 | } else if (bp->b_flags & XBF_FS_MANAGED) { | |
1da177e4 | 800 | spin_unlock(&hash->bh_lock); |
1da177e4 | 801 | } else { |
ce8e922c NS |
802 | ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q))); |
803 | list_del_init(&bp->b_hash_list); | |
1da177e4 | 804 | spin_unlock(&hash->bh_lock); |
ce8e922c | 805 | xfs_buf_free(bp); |
1da177e4 LT |
806 | } |
807 | } | |
808 | } | |
809 | ||
810 | ||
811 | /* | |
812 | * Mutual exclusion on buffers. Locking model: | |
813 | * | |
814 | * Buffers associated with inodes for which buffer locking | |
815 | * is not enabled are not protected by semaphores, and are | |
816 | * assumed to be exclusively owned by the caller. There is a | |
817 | * spinlock in the buffer, used by the caller when concurrent | |
818 | * access is possible. | |
819 | */ | |
820 | ||
821 | /* | |
ce8e922c NS |
822 | * Locks a buffer object, if it is not already locked. |
823 | * Note that this in no way locks the underlying pages, so it is only | |
824 | * useful for synchronizing concurrent use of buffer objects, not for | |
825 | * synchronizing independent access to the underlying pages. | |
1da177e4 LT |
826 | */ |
827 | int | |
ce8e922c NS |
828 | xfs_buf_cond_lock( |
829 | xfs_buf_t *bp) | |
1da177e4 LT |
830 | { |
831 | int locked; | |
832 | ||
ce8e922c | 833 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 834 | if (locked) |
ce8e922c | 835 | XB_SET_OWNER(bp); |
0b1b213f CH |
836 | |
837 | trace_xfs_buf_cond_lock(bp, _RET_IP_); | |
ce8e922c | 838 | return locked ? 0 : -EBUSY; |
1da177e4 LT |
839 | } |
840 | ||
1da177e4 | 841 | int |
ce8e922c NS |
842 | xfs_buf_lock_value( |
843 | xfs_buf_t *bp) | |
1da177e4 | 844 | { |
adaa693b | 845 | return bp->b_sema.count; |
1da177e4 | 846 | } |
1da177e4 LT |
847 | |
848 | /* | |
ce8e922c NS |
849 | * Locks a buffer object. |
850 | * Note that this in no way locks the underlying pages, so it is only | |
851 | * useful for synchronizing concurrent use of buffer objects, not for | |
852 | * synchronizing independent access to the underlying pages. | |
1da177e4 | 853 | */ |
ce8e922c NS |
854 | void |
855 | xfs_buf_lock( | |
856 | xfs_buf_t *bp) | |
1da177e4 | 857 | { |
0b1b213f CH |
858 | trace_xfs_buf_lock(bp, _RET_IP_); |
859 | ||
ce8e922c NS |
860 | if (atomic_read(&bp->b_io_remaining)) |
861 | blk_run_address_space(bp->b_target->bt_mapping); | |
862 | down(&bp->b_sema); | |
863 | XB_SET_OWNER(bp); | |
0b1b213f CH |
864 | |
865 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
866 | } |
867 | ||
868 | /* | |
ce8e922c | 869 | * Releases the lock on the buffer object. |
2f926587 | 870 | * If the buffer is marked delwri but is not queued, do so before we |
ce8e922c | 871 | * unlock the buffer as we need to set flags correctly. We also need to |
2f926587 DC |
872 | * take a reference for the delwri queue because the unlocker is going to |
873 | * drop their's and they don't know we just queued it. | |
1da177e4 LT |
874 | */ |
875 | void | |
ce8e922c NS |
876 | xfs_buf_unlock( |
877 | xfs_buf_t *bp) | |
1da177e4 | 878 | { |
ce8e922c NS |
879 | if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) { |
880 | atomic_inc(&bp->b_hold); | |
881 | bp->b_flags |= XBF_ASYNC; | |
882 | xfs_buf_delwri_queue(bp, 0); | |
2f926587 DC |
883 | } |
884 | ||
ce8e922c NS |
885 | XB_CLEAR_OWNER(bp); |
886 | up(&bp->b_sema); | |
0b1b213f CH |
887 | |
888 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
889 | } |
890 | ||
891 | ||
892 | /* | |
893 | * Pinning Buffer Storage in Memory | |
ce8e922c | 894 | * Ensure that no attempt to force a buffer to disk will succeed. |
1da177e4 LT |
895 | */ |
896 | void | |
ce8e922c NS |
897 | xfs_buf_pin( |
898 | xfs_buf_t *bp) | |
1da177e4 | 899 | { |
0b1b213f | 900 | trace_xfs_buf_pin(bp, _RET_IP_); |
ce8e922c | 901 | atomic_inc(&bp->b_pin_count); |
1da177e4 LT |
902 | } |
903 | ||
1da177e4 | 904 | void |
ce8e922c NS |
905 | xfs_buf_unpin( |
906 | xfs_buf_t *bp) | |
1da177e4 | 907 | { |
0b1b213f CH |
908 | trace_xfs_buf_unpin(bp, _RET_IP_); |
909 | ||
ce8e922c NS |
910 | if (atomic_dec_and_test(&bp->b_pin_count)) |
911 | wake_up_all(&bp->b_waiters); | |
1da177e4 LT |
912 | } |
913 | ||
914 | int | |
ce8e922c NS |
915 | xfs_buf_ispin( |
916 | xfs_buf_t *bp) | |
1da177e4 | 917 | { |
ce8e922c | 918 | return atomic_read(&bp->b_pin_count); |
1da177e4 LT |
919 | } |
920 | ||
ce8e922c NS |
921 | STATIC void |
922 | xfs_buf_wait_unpin( | |
923 | xfs_buf_t *bp) | |
1da177e4 LT |
924 | { |
925 | DECLARE_WAITQUEUE (wait, current); | |
926 | ||
ce8e922c | 927 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
928 | return; |
929 | ||
ce8e922c | 930 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
931 | for (;;) { |
932 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 933 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 934 | break; |
ce8e922c NS |
935 | if (atomic_read(&bp->b_io_remaining)) |
936 | blk_run_address_space(bp->b_target->bt_mapping); | |
1da177e4 LT |
937 | schedule(); |
938 | } | |
ce8e922c | 939 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
940 | set_current_state(TASK_RUNNING); |
941 | } | |
942 | ||
943 | /* | |
944 | * Buffer Utility Routines | |
945 | */ | |
946 | ||
1da177e4 | 947 | STATIC void |
ce8e922c | 948 | xfs_buf_iodone_work( |
c4028958 | 949 | struct work_struct *work) |
1da177e4 | 950 | { |
c4028958 DH |
951 | xfs_buf_t *bp = |
952 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 953 | |
0bfefc46 DC |
954 | /* |
955 | * We can get an EOPNOTSUPP to ordered writes. Here we clear the | |
956 | * ordered flag and reissue them. Because we can't tell the higher | |
957 | * layers directly that they should not issue ordered I/O anymore, they | |
73f6aa4d | 958 | * need to check if the _XFS_BARRIER_FAILED flag was set during I/O completion. |
0bfefc46 DC |
959 | */ |
960 | if ((bp->b_error == EOPNOTSUPP) && | |
961 | (bp->b_flags & (XBF_ORDERED|XBF_ASYNC)) == (XBF_ORDERED|XBF_ASYNC)) { | |
0b1b213f | 962 | trace_xfs_buf_ordered_retry(bp, _RET_IP_); |
0bfefc46 | 963 | bp->b_flags &= ~XBF_ORDERED; |
73f6aa4d | 964 | bp->b_flags |= _XFS_BARRIER_FAILED; |
0bfefc46 DC |
965 | xfs_buf_iorequest(bp); |
966 | } else if (bp->b_iodone) | |
ce8e922c NS |
967 | (*(bp->b_iodone))(bp); |
968 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
969 | xfs_buf_relse(bp); |
970 | } | |
971 | ||
972 | void | |
ce8e922c NS |
973 | xfs_buf_ioend( |
974 | xfs_buf_t *bp, | |
1da177e4 LT |
975 | int schedule) |
976 | { | |
0b1b213f CH |
977 | trace_xfs_buf_iodone(bp, _RET_IP_); |
978 | ||
77be55a5 | 979 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
ce8e922c NS |
980 | if (bp->b_error == 0) |
981 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 982 | |
ce8e922c | 983 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 984 | if (schedule) { |
c4028958 | 985 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 986 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 987 | } else { |
c4028958 | 988 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
989 | } |
990 | } else { | |
b4dd330b | 991 | complete(&bp->b_iowait); |
1da177e4 LT |
992 | } |
993 | } | |
994 | ||
1da177e4 | 995 | void |
ce8e922c NS |
996 | xfs_buf_ioerror( |
997 | xfs_buf_t *bp, | |
998 | int error) | |
1da177e4 LT |
999 | { |
1000 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c | 1001 | bp->b_error = (unsigned short)error; |
0b1b213f | 1002 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1003 | } |
1004 | ||
1da177e4 | 1005 | int |
64e0bc7d CH |
1006 | xfs_bwrite( |
1007 | struct xfs_mount *mp, | |
5d765b97 | 1008 | struct xfs_buf *bp) |
1da177e4 | 1009 | { |
64e0bc7d CH |
1010 | int iowait = (bp->b_flags & XBF_ASYNC) == 0; |
1011 | int error = 0; | |
1da177e4 | 1012 | |
64e0bc7d CH |
1013 | bp->b_strat = xfs_bdstrat_cb; |
1014 | bp->b_mount = mp; | |
1015 | bp->b_flags |= XBF_WRITE; | |
1016 | if (!iowait) | |
1017 | bp->b_flags |= _XBF_RUN_QUEUES; | |
1da177e4 | 1018 | |
5d765b97 | 1019 | xfs_buf_delwri_dequeue(bp); |
64e0bc7d | 1020 | xfs_buf_iostrategy(bp); |
1da177e4 | 1021 | |
64e0bc7d CH |
1022 | if (iowait) { |
1023 | error = xfs_buf_iowait(bp); | |
1024 | if (error) | |
1025 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1026 | xfs_buf_relse(bp); | |
1027 | } | |
1da177e4 | 1028 | |
64e0bc7d | 1029 | return error; |
5d765b97 | 1030 | } |
1da177e4 | 1031 | |
5d765b97 CH |
1032 | void |
1033 | xfs_bdwrite( | |
1034 | void *mp, | |
1035 | struct xfs_buf *bp) | |
1036 | { | |
0b1b213f | 1037 | trace_xfs_buf_bdwrite(bp, _RET_IP_); |
1da177e4 | 1038 | |
5d765b97 | 1039 | bp->b_strat = xfs_bdstrat_cb; |
15ac08a8 | 1040 | bp->b_mount = mp; |
1da177e4 | 1041 | |
5d765b97 CH |
1042 | bp->b_flags &= ~XBF_READ; |
1043 | bp->b_flags |= (XBF_DELWRI | XBF_ASYNC); | |
1044 | ||
1045 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1046 | } |
1047 | ||
4e23471a CH |
1048 | /* |
1049 | * Called when we want to stop a buffer from getting written or read. | |
1050 | * We attach the EIO error, muck with its flags, and call biodone | |
1051 | * so that the proper iodone callbacks get called. | |
1052 | */ | |
1053 | STATIC int | |
1054 | xfs_bioerror( | |
1055 | xfs_buf_t *bp) | |
1056 | { | |
1057 | #ifdef XFSERRORDEBUG | |
1058 | ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone); | |
1059 | #endif | |
1060 | ||
1061 | /* | |
1062 | * No need to wait until the buffer is unpinned, we aren't flushing it. | |
1063 | */ | |
1064 | XFS_BUF_ERROR(bp, EIO); | |
1065 | ||
1066 | /* | |
1067 | * We're calling biodone, so delete XBF_DONE flag. | |
1068 | */ | |
1069 | XFS_BUF_UNREAD(bp); | |
1070 | XFS_BUF_UNDELAYWRITE(bp); | |
1071 | XFS_BUF_UNDONE(bp); | |
1072 | XFS_BUF_STALE(bp); | |
1073 | ||
1074 | XFS_BUF_CLR_BDSTRAT_FUNC(bp); | |
1075 | xfs_biodone(bp); | |
1076 | ||
1077 | return EIO; | |
1078 | } | |
1079 | ||
1080 | /* | |
1081 | * Same as xfs_bioerror, except that we are releasing the buffer | |
1082 | * here ourselves, and avoiding the biodone call. | |
1083 | * This is meant for userdata errors; metadata bufs come with | |
1084 | * iodone functions attached, so that we can track down errors. | |
1085 | */ | |
1086 | STATIC int | |
1087 | xfs_bioerror_relse( | |
1088 | struct xfs_buf *bp) | |
1089 | { | |
1090 | int64_t fl = XFS_BUF_BFLAGS(bp); | |
1091 | /* | |
1092 | * No need to wait until the buffer is unpinned. | |
1093 | * We aren't flushing it. | |
1094 | * | |
1095 | * chunkhold expects B_DONE to be set, whether | |
1096 | * we actually finish the I/O or not. We don't want to | |
1097 | * change that interface. | |
1098 | */ | |
1099 | XFS_BUF_UNREAD(bp); | |
1100 | XFS_BUF_UNDELAYWRITE(bp); | |
1101 | XFS_BUF_DONE(bp); | |
1102 | XFS_BUF_STALE(bp); | |
1103 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
1104 | XFS_BUF_CLR_BDSTRAT_FUNC(bp); | |
0cadda1c | 1105 | if (!(fl & XBF_ASYNC)) { |
4e23471a CH |
1106 | /* |
1107 | * Mark b_error and B_ERROR _both_. | |
1108 | * Lot's of chunkcache code assumes that. | |
1109 | * There's no reason to mark error for | |
1110 | * ASYNC buffers. | |
1111 | */ | |
1112 | XFS_BUF_ERROR(bp, EIO); | |
1113 | XFS_BUF_FINISH_IOWAIT(bp); | |
1114 | } else { | |
1115 | xfs_buf_relse(bp); | |
1116 | } | |
1117 | ||
1118 | return EIO; | |
1119 | } | |
1120 | ||
1121 | ||
1122 | /* | |
1123 | * All xfs metadata buffers except log state machine buffers | |
1124 | * get this attached as their b_bdstrat callback function. | |
1125 | * This is so that we can catch a buffer | |
1126 | * after prematurely unpinning it to forcibly shutdown the filesystem. | |
1127 | */ | |
1128 | int | |
1129 | xfs_bdstrat_cb( | |
1130 | struct xfs_buf *bp) | |
1131 | { | |
1132 | if (XFS_FORCED_SHUTDOWN(bp->b_mount)) { | |
1133 | trace_xfs_bdstrat_shut(bp, _RET_IP_); | |
1134 | /* | |
1135 | * Metadata write that didn't get logged but | |
1136 | * written delayed anyway. These aren't associated | |
1137 | * with a transaction, and can be ignored. | |
1138 | */ | |
1139 | if (!bp->b_iodone && !XFS_BUF_ISREAD(bp)) | |
1140 | return xfs_bioerror_relse(bp); | |
1141 | else | |
1142 | return xfs_bioerror(bp); | |
1143 | } | |
1144 | ||
1145 | xfs_buf_iorequest(bp); | |
1146 | return 0; | |
1147 | } | |
1148 | ||
1149 | /* | |
1150 | * Wrapper around bdstrat so that we can stop data from going to disk in case | |
1151 | * we are shutting down the filesystem. Typically user data goes thru this | |
1152 | * path; one of the exceptions is the superblock. | |
1153 | */ | |
1154 | void | |
1155 | xfsbdstrat( | |
1156 | struct xfs_mount *mp, | |
1157 | struct xfs_buf *bp) | |
1158 | { | |
1159 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1160 | trace_xfs_bdstrat_shut(bp, _RET_IP_); | |
1161 | xfs_bioerror_relse(bp); | |
1162 | return; | |
1163 | } | |
1164 | ||
1165 | xfs_buf_iorequest(bp); | |
1166 | } | |
1167 | ||
b8f82a4a | 1168 | STATIC void |
ce8e922c NS |
1169 | _xfs_buf_ioend( |
1170 | xfs_buf_t *bp, | |
1da177e4 LT |
1171 | int schedule) |
1172 | { | |
6ab455ee CH |
1173 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
1174 | bp->b_flags &= ~_XBF_PAGE_LOCKED; | |
ce8e922c | 1175 | xfs_buf_ioend(bp, schedule); |
6ab455ee | 1176 | } |
1da177e4 LT |
1177 | } |
1178 | ||
782e3b3b | 1179 | STATIC void |
ce8e922c | 1180 | xfs_buf_bio_end_io( |
1da177e4 | 1181 | struct bio *bio, |
1da177e4 LT |
1182 | int error) |
1183 | { | |
ce8e922c NS |
1184 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1185 | unsigned int blocksize = bp->b_target->bt_bsize; | |
eedb5530 | 1186 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1da177e4 | 1187 | |
cfbe5267 | 1188 | xfs_buf_ioerror(bp, -error); |
1da177e4 | 1189 | |
73c77e2c JB |
1190 | if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
1191 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); | |
1192 | ||
eedb5530 | 1193 | do { |
1da177e4 LT |
1194 | struct page *page = bvec->bv_page; |
1195 | ||
948ecdb4 | 1196 | ASSERT(!PagePrivate(page)); |
ce8e922c NS |
1197 | if (unlikely(bp->b_error)) { |
1198 | if (bp->b_flags & XBF_READ) | |
eedb5530 | 1199 | ClearPageUptodate(page); |
ce8e922c | 1200 | } else if (blocksize >= PAGE_CACHE_SIZE) { |
1da177e4 LT |
1201 | SetPageUptodate(page); |
1202 | } else if (!PagePrivate(page) && | |
ce8e922c | 1203 | (bp->b_flags & _XBF_PAGE_CACHE)) { |
1da177e4 LT |
1204 | set_page_region(page, bvec->bv_offset, bvec->bv_len); |
1205 | } | |
1206 | ||
eedb5530 NS |
1207 | if (--bvec >= bio->bi_io_vec) |
1208 | prefetchw(&bvec->bv_page->flags); | |
6ab455ee CH |
1209 | |
1210 | if (bp->b_flags & _XBF_PAGE_LOCKED) | |
1211 | unlock_page(page); | |
eedb5530 | 1212 | } while (bvec >= bio->bi_io_vec); |
1da177e4 | 1213 | |
ce8e922c | 1214 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1215 | bio_put(bio); |
1da177e4 LT |
1216 | } |
1217 | ||
1218 | STATIC void | |
ce8e922c NS |
1219 | _xfs_buf_ioapply( |
1220 | xfs_buf_t *bp) | |
1da177e4 | 1221 | { |
a9759f2d | 1222 | int rw, map_i, total_nr_pages, nr_pages; |
1da177e4 | 1223 | struct bio *bio; |
ce8e922c NS |
1224 | int offset = bp->b_offset; |
1225 | int size = bp->b_count_desired; | |
1226 | sector_t sector = bp->b_bn; | |
1227 | unsigned int blocksize = bp->b_target->bt_bsize; | |
1da177e4 | 1228 | |
ce8e922c | 1229 | total_nr_pages = bp->b_page_count; |
1da177e4 LT |
1230 | map_i = 0; |
1231 | ||
ce8e922c NS |
1232 | if (bp->b_flags & XBF_ORDERED) { |
1233 | ASSERT(!(bp->b_flags & XBF_READ)); | |
f538d4da | 1234 | rw = WRITE_BARRIER; |
2ee1abad | 1235 | } else if (bp->b_flags & XBF_LOG_BUFFER) { |
51bdd706 NS |
1236 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); |
1237 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1238 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC; | |
2ee1abad DC |
1239 | } else if (bp->b_flags & _XBF_RUN_QUEUES) { |
1240 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); | |
1241 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1242 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META; | |
51bdd706 NS |
1243 | } else { |
1244 | rw = (bp->b_flags & XBF_WRITE) ? WRITE : | |
1245 | (bp->b_flags & XBF_READ_AHEAD) ? READA : READ; | |
f538d4da CH |
1246 | } |
1247 | ||
ce8e922c | 1248 | /* Special code path for reading a sub page size buffer in -- |
1da177e4 LT |
1249 | * we populate up the whole page, and hence the other metadata |
1250 | * in the same page. This optimization is only valid when the | |
ce8e922c | 1251 | * filesystem block size is not smaller than the page size. |
1da177e4 | 1252 | */ |
ce8e922c | 1253 | if ((bp->b_buffer_length < PAGE_CACHE_SIZE) && |
6ab455ee CH |
1254 | ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) == |
1255 | (XBF_READ|_XBF_PAGE_LOCKED)) && | |
ce8e922c | 1256 | (blocksize >= PAGE_CACHE_SIZE)) { |
1da177e4 LT |
1257 | bio = bio_alloc(GFP_NOIO, 1); |
1258 | ||
ce8e922c | 1259 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1260 | bio->bi_sector = sector - (offset >> BBSHIFT); |
ce8e922c NS |
1261 | bio->bi_end_io = xfs_buf_bio_end_io; |
1262 | bio->bi_private = bp; | |
1da177e4 | 1263 | |
ce8e922c | 1264 | bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0); |
1da177e4 LT |
1265 | size = 0; |
1266 | ||
ce8e922c | 1267 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1268 | |
1269 | goto submit_io; | |
1270 | } | |
1271 | ||
1da177e4 | 1272 | next_chunk: |
ce8e922c | 1273 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1274 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1275 | if (nr_pages > total_nr_pages) | |
1276 | nr_pages = total_nr_pages; | |
1277 | ||
1278 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1279 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1280 | bio->bi_sector = sector; |
ce8e922c NS |
1281 | bio->bi_end_io = xfs_buf_bio_end_io; |
1282 | bio->bi_private = bp; | |
1da177e4 LT |
1283 | |
1284 | for (; size && nr_pages; nr_pages--, map_i++) { | |
ce8e922c | 1285 | int rbytes, nbytes = PAGE_CACHE_SIZE - offset; |
1da177e4 LT |
1286 | |
1287 | if (nbytes > size) | |
1288 | nbytes = size; | |
1289 | ||
ce8e922c NS |
1290 | rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset); |
1291 | if (rbytes < nbytes) | |
1da177e4 LT |
1292 | break; |
1293 | ||
1294 | offset = 0; | |
1295 | sector += nbytes >> BBSHIFT; | |
1296 | size -= nbytes; | |
1297 | total_nr_pages--; | |
1298 | } | |
1299 | ||
1300 | submit_io: | |
1301 | if (likely(bio->bi_size)) { | |
73c77e2c JB |
1302 | if (xfs_buf_is_vmapped(bp)) { |
1303 | flush_kernel_vmap_range(bp->b_addr, | |
1304 | xfs_buf_vmap_len(bp)); | |
1305 | } | |
1da177e4 LT |
1306 | submit_bio(rw, bio); |
1307 | if (size) | |
1308 | goto next_chunk; | |
1309 | } else { | |
1310 | bio_put(bio); | |
ce8e922c | 1311 | xfs_buf_ioerror(bp, EIO); |
1da177e4 LT |
1312 | } |
1313 | } | |
1314 | ||
1da177e4 | 1315 | int |
ce8e922c NS |
1316 | xfs_buf_iorequest( |
1317 | xfs_buf_t *bp) | |
1da177e4 | 1318 | { |
0b1b213f | 1319 | trace_xfs_buf_iorequest(bp, _RET_IP_); |
1da177e4 | 1320 | |
ce8e922c NS |
1321 | if (bp->b_flags & XBF_DELWRI) { |
1322 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1323 | return 0; |
1324 | } | |
1325 | ||
ce8e922c NS |
1326 | if (bp->b_flags & XBF_WRITE) { |
1327 | xfs_buf_wait_unpin(bp); | |
1da177e4 LT |
1328 | } |
1329 | ||
ce8e922c | 1330 | xfs_buf_hold(bp); |
1da177e4 LT |
1331 | |
1332 | /* Set the count to 1 initially, this will stop an I/O | |
1333 | * completion callout which happens before we have started | |
ce8e922c | 1334 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1335 | */ |
ce8e922c NS |
1336 | atomic_set(&bp->b_io_remaining, 1); |
1337 | _xfs_buf_ioapply(bp); | |
1338 | _xfs_buf_ioend(bp, 0); | |
1da177e4 | 1339 | |
ce8e922c | 1340 | xfs_buf_rele(bp); |
1da177e4 LT |
1341 | return 0; |
1342 | } | |
1343 | ||
1344 | /* | |
ce8e922c NS |
1345 | * Waits for I/O to complete on the buffer supplied. |
1346 | * It returns immediately if no I/O is pending. | |
1347 | * It returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1348 | */ |
1349 | int | |
ce8e922c NS |
1350 | xfs_buf_iowait( |
1351 | xfs_buf_t *bp) | |
1da177e4 | 1352 | { |
0b1b213f CH |
1353 | trace_xfs_buf_iowait(bp, _RET_IP_); |
1354 | ||
ce8e922c NS |
1355 | if (atomic_read(&bp->b_io_remaining)) |
1356 | blk_run_address_space(bp->b_target->bt_mapping); | |
b4dd330b | 1357 | wait_for_completion(&bp->b_iowait); |
0b1b213f CH |
1358 | |
1359 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
ce8e922c | 1360 | return bp->b_error; |
1da177e4 LT |
1361 | } |
1362 | ||
ce8e922c NS |
1363 | xfs_caddr_t |
1364 | xfs_buf_offset( | |
1365 | xfs_buf_t *bp, | |
1da177e4 LT |
1366 | size_t offset) |
1367 | { | |
1368 | struct page *page; | |
1369 | ||
ce8e922c NS |
1370 | if (bp->b_flags & XBF_MAPPED) |
1371 | return XFS_BUF_PTR(bp) + offset; | |
1da177e4 | 1372 | |
ce8e922c NS |
1373 | offset += bp->b_offset; |
1374 | page = bp->b_pages[offset >> PAGE_CACHE_SHIFT]; | |
1375 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1)); | |
1da177e4 LT |
1376 | } |
1377 | ||
1378 | /* | |
1da177e4 LT |
1379 | * Move data into or out of a buffer. |
1380 | */ | |
1381 | void | |
ce8e922c NS |
1382 | xfs_buf_iomove( |
1383 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1384 | size_t boff, /* starting buffer offset */ |
1385 | size_t bsize, /* length to copy */ | |
b9c48649 | 1386 | void *data, /* data address */ |
ce8e922c | 1387 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 LT |
1388 | { |
1389 | size_t bend, cpoff, csize; | |
1390 | struct page *page; | |
1391 | ||
1392 | bend = boff + bsize; | |
1393 | while (boff < bend) { | |
ce8e922c NS |
1394 | page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)]; |
1395 | cpoff = xfs_buf_poff(boff + bp->b_offset); | |
1da177e4 | 1396 | csize = min_t(size_t, |
ce8e922c | 1397 | PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff); |
1da177e4 LT |
1398 | |
1399 | ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE)); | |
1400 | ||
1401 | switch (mode) { | |
ce8e922c | 1402 | case XBRW_ZERO: |
1da177e4 LT |
1403 | memset(page_address(page) + cpoff, 0, csize); |
1404 | break; | |
ce8e922c | 1405 | case XBRW_READ: |
1da177e4 LT |
1406 | memcpy(data, page_address(page) + cpoff, csize); |
1407 | break; | |
ce8e922c | 1408 | case XBRW_WRITE: |
1da177e4 LT |
1409 | memcpy(page_address(page) + cpoff, data, csize); |
1410 | } | |
1411 | ||
1412 | boff += csize; | |
1413 | data += csize; | |
1414 | } | |
1415 | } | |
1416 | ||
1417 | /* | |
ce8e922c | 1418 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1419 | */ |
1420 | ||
1421 | /* | |
ce8e922c NS |
1422 | * Wait for any bufs with callbacks that have been submitted but |
1423 | * have not yet returned... walk the hash list for the target. | |
1da177e4 LT |
1424 | */ |
1425 | void | |
1426 | xfs_wait_buftarg( | |
1427 | xfs_buftarg_t *btp) | |
1428 | { | |
1429 | xfs_buf_t *bp, *n; | |
1430 | xfs_bufhash_t *hash; | |
1431 | uint i; | |
1432 | ||
1433 | for (i = 0; i < (1 << btp->bt_hashshift); i++) { | |
1434 | hash = &btp->bt_hash[i]; | |
1435 | again: | |
1436 | spin_lock(&hash->bh_lock); | |
ce8e922c NS |
1437 | list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) { |
1438 | ASSERT(btp == bp->b_target); | |
1439 | if (!(bp->b_flags & XBF_FS_MANAGED)) { | |
1da177e4 | 1440 | spin_unlock(&hash->bh_lock); |
2f926587 DC |
1441 | /* |
1442 | * Catch superblock reference count leaks | |
1443 | * immediately | |
1444 | */ | |
ce8e922c | 1445 | BUG_ON(bp->b_bn == 0); |
1da177e4 LT |
1446 | delay(100); |
1447 | goto again; | |
1448 | } | |
1449 | } | |
1450 | spin_unlock(&hash->bh_lock); | |
1451 | } | |
1452 | } | |
1453 | ||
1454 | /* | |
ce8e922c NS |
1455 | * Allocate buffer hash table for a given target. |
1456 | * For devices containing metadata (i.e. not the log/realtime devices) | |
1457 | * we need to allocate a much larger hash table. | |
1da177e4 LT |
1458 | */ |
1459 | STATIC void | |
1460 | xfs_alloc_bufhash( | |
1461 | xfs_buftarg_t *btp, | |
1462 | int external) | |
1463 | { | |
1464 | unsigned int i; | |
1465 | ||
1466 | btp->bt_hashshift = external ? 3 : 8; /* 8 or 256 buckets */ | |
1467 | btp->bt_hashmask = (1 << btp->bt_hashshift) - 1; | |
bdfb0430 CH |
1468 | btp->bt_hash = kmem_zalloc_large((1 << btp->bt_hashshift) * |
1469 | sizeof(xfs_bufhash_t)); | |
1da177e4 LT |
1470 | for (i = 0; i < (1 << btp->bt_hashshift); i++) { |
1471 | spin_lock_init(&btp->bt_hash[i].bh_lock); | |
1472 | INIT_LIST_HEAD(&btp->bt_hash[i].bh_list); | |
1473 | } | |
1474 | } | |
1475 | ||
1476 | STATIC void | |
1477 | xfs_free_bufhash( | |
1478 | xfs_buftarg_t *btp) | |
1479 | { | |
bdfb0430 | 1480 | kmem_free_large(btp->bt_hash); |
1da177e4 LT |
1481 | btp->bt_hash = NULL; |
1482 | } | |
1483 | ||
a6867a68 | 1484 | /* |
ce8e922c | 1485 | * buftarg list for delwrite queue processing |
a6867a68 | 1486 | */ |
e6a0e9cd | 1487 | static LIST_HEAD(xfs_buftarg_list); |
7989cb8e | 1488 | static DEFINE_SPINLOCK(xfs_buftarg_lock); |
a6867a68 DC |
1489 | |
1490 | STATIC void | |
1491 | xfs_register_buftarg( | |
1492 | xfs_buftarg_t *btp) | |
1493 | { | |
1494 | spin_lock(&xfs_buftarg_lock); | |
1495 | list_add(&btp->bt_list, &xfs_buftarg_list); | |
1496 | spin_unlock(&xfs_buftarg_lock); | |
1497 | } | |
1498 | ||
1499 | STATIC void | |
1500 | xfs_unregister_buftarg( | |
1501 | xfs_buftarg_t *btp) | |
1502 | { | |
1503 | spin_lock(&xfs_buftarg_lock); | |
1504 | list_del(&btp->bt_list); | |
1505 | spin_unlock(&xfs_buftarg_lock); | |
1506 | } | |
1507 | ||
1da177e4 LT |
1508 | void |
1509 | xfs_free_buftarg( | |
b7963133 CH |
1510 | struct xfs_mount *mp, |
1511 | struct xfs_buftarg *btp) | |
1da177e4 LT |
1512 | { |
1513 | xfs_flush_buftarg(btp, 1); | |
b7963133 CH |
1514 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1515 | xfs_blkdev_issue_flush(btp); | |
1da177e4 | 1516 | xfs_free_bufhash(btp); |
ce8e922c | 1517 | iput(btp->bt_mapping->host); |
a6867a68 | 1518 | |
ce8e922c NS |
1519 | /* Unregister the buftarg first so that we don't get a |
1520 | * wakeup finding a non-existent task | |
1521 | */ | |
a6867a68 DC |
1522 | xfs_unregister_buftarg(btp); |
1523 | kthread_stop(btp->bt_task); | |
1524 | ||
f0e2d93c | 1525 | kmem_free(btp); |
1da177e4 LT |
1526 | } |
1527 | ||
1da177e4 LT |
1528 | STATIC int |
1529 | xfs_setsize_buftarg_flags( | |
1530 | xfs_buftarg_t *btp, | |
1531 | unsigned int blocksize, | |
1532 | unsigned int sectorsize, | |
1533 | int verbose) | |
1534 | { | |
ce8e922c NS |
1535 | btp->bt_bsize = blocksize; |
1536 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1537 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1538 | |
ce8e922c | 1539 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
1da177e4 LT |
1540 | printk(KERN_WARNING |
1541 | "XFS: Cannot set_blocksize to %u on device %s\n", | |
1542 | sectorsize, XFS_BUFTARG_NAME(btp)); | |
1543 | return EINVAL; | |
1544 | } | |
1545 | ||
1546 | if (verbose && | |
1547 | (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) { | |
1548 | printk(KERN_WARNING | |
1549 | "XFS: %u byte sectors in use on device %s. " | |
1550 | "This is suboptimal; %u or greater is ideal.\n", | |
1551 | sectorsize, XFS_BUFTARG_NAME(btp), | |
1552 | (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG); | |
1553 | } | |
1554 | ||
1555 | return 0; | |
1556 | } | |
1557 | ||
1558 | /* | |
ce8e922c NS |
1559 | * When allocating the initial buffer target we have not yet |
1560 | * read in the superblock, so don't know what sized sectors | |
1561 | * are being used is at this early stage. Play safe. | |
1562 | */ | |
1da177e4 LT |
1563 | STATIC int |
1564 | xfs_setsize_buftarg_early( | |
1565 | xfs_buftarg_t *btp, | |
1566 | struct block_device *bdev) | |
1567 | { | |
1568 | return xfs_setsize_buftarg_flags(btp, | |
e1defc4f | 1569 | PAGE_CACHE_SIZE, bdev_logical_block_size(bdev), 0); |
1da177e4 LT |
1570 | } |
1571 | ||
1572 | int | |
1573 | xfs_setsize_buftarg( | |
1574 | xfs_buftarg_t *btp, | |
1575 | unsigned int blocksize, | |
1576 | unsigned int sectorsize) | |
1577 | { | |
1578 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1579 | } | |
1580 | ||
1581 | STATIC int | |
1582 | xfs_mapping_buftarg( | |
1583 | xfs_buftarg_t *btp, | |
1584 | struct block_device *bdev) | |
1585 | { | |
1586 | struct backing_dev_info *bdi; | |
1587 | struct inode *inode; | |
1588 | struct address_space *mapping; | |
f5e54d6e | 1589 | static const struct address_space_operations mapping_aops = { |
1da177e4 | 1590 | .sync_page = block_sync_page, |
e965f963 | 1591 | .migratepage = fail_migrate_page, |
1da177e4 LT |
1592 | }; |
1593 | ||
1594 | inode = new_inode(bdev->bd_inode->i_sb); | |
1595 | if (!inode) { | |
1596 | printk(KERN_WARNING | |
1597 | "XFS: Cannot allocate mapping inode for device %s\n", | |
1598 | XFS_BUFTARG_NAME(btp)); | |
1599 | return ENOMEM; | |
1600 | } | |
1601 | inode->i_mode = S_IFBLK; | |
1602 | inode->i_bdev = bdev; | |
1603 | inode->i_rdev = bdev->bd_dev; | |
1604 | bdi = blk_get_backing_dev_info(bdev); | |
1605 | if (!bdi) | |
1606 | bdi = &default_backing_dev_info; | |
1607 | mapping = &inode->i_data; | |
1608 | mapping->a_ops = &mapping_aops; | |
1609 | mapping->backing_dev_info = bdi; | |
1610 | mapping_set_gfp_mask(mapping, GFP_NOFS); | |
ce8e922c | 1611 | btp->bt_mapping = mapping; |
1da177e4 LT |
1612 | return 0; |
1613 | } | |
1614 | ||
a6867a68 DC |
1615 | STATIC int |
1616 | xfs_alloc_delwrite_queue( | |
1617 | xfs_buftarg_t *btp) | |
1618 | { | |
1619 | int error = 0; | |
1620 | ||
1621 | INIT_LIST_HEAD(&btp->bt_list); | |
1622 | INIT_LIST_HEAD(&btp->bt_delwrite_queue); | |
007c61c6 | 1623 | spin_lock_init(&btp->bt_delwrite_lock); |
a6867a68 DC |
1624 | btp->bt_flags = 0; |
1625 | btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd"); | |
1626 | if (IS_ERR(btp->bt_task)) { | |
1627 | error = PTR_ERR(btp->bt_task); | |
1628 | goto out_error; | |
1629 | } | |
1630 | xfs_register_buftarg(btp); | |
1631 | out_error: | |
1632 | return error; | |
1633 | } | |
1634 | ||
1da177e4 LT |
1635 | xfs_buftarg_t * |
1636 | xfs_alloc_buftarg( | |
1637 | struct block_device *bdev, | |
1638 | int external) | |
1639 | { | |
1640 | xfs_buftarg_t *btp; | |
1641 | ||
1642 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1643 | ||
ce8e922c NS |
1644 | btp->bt_dev = bdev->bd_dev; |
1645 | btp->bt_bdev = bdev; | |
1da177e4 LT |
1646 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1647 | goto error; | |
1648 | if (xfs_mapping_buftarg(btp, bdev)) | |
1649 | goto error; | |
a6867a68 DC |
1650 | if (xfs_alloc_delwrite_queue(btp)) |
1651 | goto error; | |
1da177e4 LT |
1652 | xfs_alloc_bufhash(btp, external); |
1653 | return btp; | |
1654 | ||
1655 | error: | |
f0e2d93c | 1656 | kmem_free(btp); |
1da177e4 LT |
1657 | return NULL; |
1658 | } | |
1659 | ||
1660 | ||
1661 | /* | |
ce8e922c | 1662 | * Delayed write buffer handling |
1da177e4 | 1663 | */ |
1da177e4 | 1664 | STATIC void |
ce8e922c NS |
1665 | xfs_buf_delwri_queue( |
1666 | xfs_buf_t *bp, | |
1da177e4 LT |
1667 | int unlock) |
1668 | { | |
ce8e922c NS |
1669 | struct list_head *dwq = &bp->b_target->bt_delwrite_queue; |
1670 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; | |
a6867a68 | 1671 | |
0b1b213f CH |
1672 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
1673 | ||
ce8e922c | 1674 | ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC)); |
1da177e4 | 1675 | |
a6867a68 | 1676 | spin_lock(dwlk); |
1da177e4 | 1677 | /* If already in the queue, dequeue and place at tail */ |
ce8e922c NS |
1678 | if (!list_empty(&bp->b_list)) { |
1679 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1680 | if (unlock) | |
1681 | atomic_dec(&bp->b_hold); | |
1682 | list_del(&bp->b_list); | |
1da177e4 LT |
1683 | } |
1684 | ||
c9c12971 DC |
1685 | if (list_empty(dwq)) { |
1686 | /* start xfsbufd as it is about to have something to do */ | |
1687 | wake_up_process(bp->b_target->bt_task); | |
1688 | } | |
1689 | ||
ce8e922c NS |
1690 | bp->b_flags |= _XBF_DELWRI_Q; |
1691 | list_add_tail(&bp->b_list, dwq); | |
1692 | bp->b_queuetime = jiffies; | |
a6867a68 | 1693 | spin_unlock(dwlk); |
1da177e4 LT |
1694 | |
1695 | if (unlock) | |
ce8e922c | 1696 | xfs_buf_unlock(bp); |
1da177e4 LT |
1697 | } |
1698 | ||
1699 | void | |
ce8e922c NS |
1700 | xfs_buf_delwri_dequeue( |
1701 | xfs_buf_t *bp) | |
1da177e4 | 1702 | { |
ce8e922c | 1703 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; |
1da177e4 LT |
1704 | int dequeued = 0; |
1705 | ||
a6867a68 | 1706 | spin_lock(dwlk); |
ce8e922c NS |
1707 | if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) { |
1708 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1709 | list_del_init(&bp->b_list); | |
1da177e4 LT |
1710 | dequeued = 1; |
1711 | } | |
ce8e922c | 1712 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q); |
a6867a68 | 1713 | spin_unlock(dwlk); |
1da177e4 LT |
1714 | |
1715 | if (dequeued) | |
ce8e922c | 1716 | xfs_buf_rele(bp); |
1da177e4 | 1717 | |
0b1b213f | 1718 | trace_xfs_buf_delwri_dequeue(bp, _RET_IP_); |
1da177e4 LT |
1719 | } |
1720 | ||
d808f617 DC |
1721 | /* |
1722 | * If a delwri buffer needs to be pushed before it has aged out, then promote | |
1723 | * it to the head of the delwri queue so that it will be flushed on the next | |
1724 | * xfsbufd run. We do this by resetting the queuetime of the buffer to be older | |
1725 | * than the age currently needed to flush the buffer. Hence the next time the | |
1726 | * xfsbufd sees it is guaranteed to be considered old enough to flush. | |
1727 | */ | |
1728 | void | |
1729 | xfs_buf_delwri_promote( | |
1730 | struct xfs_buf *bp) | |
1731 | { | |
1732 | struct xfs_buftarg *btp = bp->b_target; | |
1733 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1; | |
1734 | ||
1735 | ASSERT(bp->b_flags & XBF_DELWRI); | |
1736 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1737 | ||
1738 | /* | |
1739 | * Check the buffer age before locking the delayed write queue as we | |
1740 | * don't need to promote buffers that are already past the flush age. | |
1741 | */ | |
1742 | if (bp->b_queuetime < jiffies - age) | |
1743 | return; | |
1744 | bp->b_queuetime = jiffies - age; | |
1745 | spin_lock(&btp->bt_delwrite_lock); | |
1746 | list_move(&bp->b_list, &btp->bt_delwrite_queue); | |
1747 | spin_unlock(&btp->bt_delwrite_lock); | |
1748 | } | |
1749 | ||
1da177e4 | 1750 | STATIC void |
ce8e922c | 1751 | xfs_buf_runall_queues( |
1da177e4 LT |
1752 | struct workqueue_struct *queue) |
1753 | { | |
1754 | flush_workqueue(queue); | |
1755 | } | |
1756 | ||
1da177e4 | 1757 | STATIC int |
23ea4032 | 1758 | xfsbufd_wakeup( |
15c84a47 NS |
1759 | int priority, |
1760 | gfp_t mask) | |
1da177e4 | 1761 | { |
da7f93e9 | 1762 | xfs_buftarg_t *btp; |
a6867a68 DC |
1763 | |
1764 | spin_lock(&xfs_buftarg_lock); | |
da7f93e9 | 1765 | list_for_each_entry(btp, &xfs_buftarg_list, bt_list) { |
ce8e922c | 1766 | if (test_bit(XBT_FORCE_SLEEP, &btp->bt_flags)) |
a6867a68 | 1767 | continue; |
c9c12971 DC |
1768 | if (list_empty(&btp->bt_delwrite_queue)) |
1769 | continue; | |
ce8e922c | 1770 | set_bit(XBT_FORCE_FLUSH, &btp->bt_flags); |
a6867a68 DC |
1771 | wake_up_process(btp->bt_task); |
1772 | } | |
1773 | spin_unlock(&xfs_buftarg_lock); | |
1da177e4 LT |
1774 | return 0; |
1775 | } | |
1776 | ||
585e6d88 DC |
1777 | /* |
1778 | * Move as many buffers as specified to the supplied list | |
1779 | * idicating if we skipped any buffers to prevent deadlocks. | |
1780 | */ | |
1781 | STATIC int | |
1782 | xfs_buf_delwri_split( | |
1783 | xfs_buftarg_t *target, | |
1784 | struct list_head *list, | |
5e6a07df | 1785 | unsigned long age) |
585e6d88 DC |
1786 | { |
1787 | xfs_buf_t *bp, *n; | |
1788 | struct list_head *dwq = &target->bt_delwrite_queue; | |
1789 | spinlock_t *dwlk = &target->bt_delwrite_lock; | |
1790 | int skipped = 0; | |
5e6a07df | 1791 | int force; |
585e6d88 | 1792 | |
5e6a07df | 1793 | force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
585e6d88 DC |
1794 | INIT_LIST_HEAD(list); |
1795 | spin_lock(dwlk); | |
1796 | list_for_each_entry_safe(bp, n, dwq, b_list) { | |
0b1b213f | 1797 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
585e6d88 DC |
1798 | ASSERT(bp->b_flags & XBF_DELWRI); |
1799 | ||
1800 | if (!xfs_buf_ispin(bp) && !xfs_buf_cond_lock(bp)) { | |
5e6a07df | 1801 | if (!force && |
585e6d88 DC |
1802 | time_before(jiffies, bp->b_queuetime + age)) { |
1803 | xfs_buf_unlock(bp); | |
1804 | break; | |
1805 | } | |
1806 | ||
1807 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q| | |
1808 | _XBF_RUN_QUEUES); | |
1809 | bp->b_flags |= XBF_WRITE; | |
1810 | list_move_tail(&bp->b_list, list); | |
1811 | } else | |
1812 | skipped++; | |
1813 | } | |
1814 | spin_unlock(dwlk); | |
1815 | ||
1816 | return skipped; | |
1817 | ||
1818 | } | |
1819 | ||
089716aa DC |
1820 | /* |
1821 | * Compare function is more complex than it needs to be because | |
1822 | * the return value is only 32 bits and we are doing comparisons | |
1823 | * on 64 bit values | |
1824 | */ | |
1825 | static int | |
1826 | xfs_buf_cmp( | |
1827 | void *priv, | |
1828 | struct list_head *a, | |
1829 | struct list_head *b) | |
1830 | { | |
1831 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1832 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1833 | xfs_daddr_t diff; | |
1834 | ||
1835 | diff = ap->b_bn - bp->b_bn; | |
1836 | if (diff < 0) | |
1837 | return -1; | |
1838 | if (diff > 0) | |
1839 | return 1; | |
1840 | return 0; | |
1841 | } | |
1842 | ||
1843 | void | |
1844 | xfs_buf_delwri_sort( | |
1845 | xfs_buftarg_t *target, | |
1846 | struct list_head *list) | |
1847 | { | |
1848 | list_sort(NULL, list, xfs_buf_cmp); | |
1849 | } | |
1850 | ||
1da177e4 | 1851 | STATIC int |
23ea4032 | 1852 | xfsbufd( |
585e6d88 | 1853 | void *data) |
1da177e4 | 1854 | { |
089716aa | 1855 | xfs_buftarg_t *target = (xfs_buftarg_t *)data; |
1da177e4 | 1856 | |
1da177e4 LT |
1857 | current->flags |= PF_MEMALLOC; |
1858 | ||
978c7b2f RW |
1859 | set_freezable(); |
1860 | ||
1da177e4 | 1861 | do { |
c9c12971 DC |
1862 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10); |
1863 | long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10); | |
089716aa DC |
1864 | int count = 0; |
1865 | struct list_head tmp; | |
c9c12971 | 1866 | |
3e1d1d28 | 1867 | if (unlikely(freezing(current))) { |
ce8e922c | 1868 | set_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
3e1d1d28 | 1869 | refrigerator(); |
abd0cf7a | 1870 | } else { |
ce8e922c | 1871 | clear_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
abd0cf7a | 1872 | } |
1da177e4 | 1873 | |
c9c12971 DC |
1874 | /* sleep for a long time if there is nothing to do. */ |
1875 | if (list_empty(&target->bt_delwrite_queue)) | |
1876 | tout = MAX_SCHEDULE_TIMEOUT; | |
1877 | schedule_timeout_interruptible(tout); | |
1da177e4 | 1878 | |
c9c12971 | 1879 | xfs_buf_delwri_split(target, &tmp, age); |
089716aa | 1880 | list_sort(NULL, &tmp, xfs_buf_cmp); |
1da177e4 | 1881 | while (!list_empty(&tmp)) { |
089716aa DC |
1882 | struct xfs_buf *bp; |
1883 | bp = list_first_entry(&tmp, struct xfs_buf, b_list); | |
ce8e922c NS |
1884 | list_del_init(&bp->b_list); |
1885 | xfs_buf_iostrategy(bp); | |
585e6d88 | 1886 | count++; |
1da177e4 | 1887 | } |
f07c2250 NS |
1888 | if (count) |
1889 | blk_run_address_space(target->bt_mapping); | |
1da177e4 | 1890 | |
4df08c52 | 1891 | } while (!kthread_should_stop()); |
1da177e4 | 1892 | |
4df08c52 | 1893 | return 0; |
1da177e4 LT |
1894 | } |
1895 | ||
1896 | /* | |
ce8e922c NS |
1897 | * Go through all incore buffers, and release buffers if they belong to |
1898 | * the given device. This is used in filesystem error handling to | |
1899 | * preserve the consistency of its metadata. | |
1da177e4 LT |
1900 | */ |
1901 | int | |
1902 | xfs_flush_buftarg( | |
585e6d88 DC |
1903 | xfs_buftarg_t *target, |
1904 | int wait) | |
1da177e4 | 1905 | { |
089716aa | 1906 | xfs_buf_t *bp; |
585e6d88 | 1907 | int pincount = 0; |
089716aa DC |
1908 | LIST_HEAD(tmp_list); |
1909 | LIST_HEAD(wait_list); | |
1da177e4 | 1910 | |
c626d174 | 1911 | xfs_buf_runall_queues(xfsconvertd_workqueue); |
ce8e922c NS |
1912 | xfs_buf_runall_queues(xfsdatad_workqueue); |
1913 | xfs_buf_runall_queues(xfslogd_workqueue); | |
1da177e4 | 1914 | |
5e6a07df | 1915 | set_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
089716aa | 1916 | pincount = xfs_buf_delwri_split(target, &tmp_list, 0); |
1da177e4 LT |
1917 | |
1918 | /* | |
089716aa DC |
1919 | * Dropped the delayed write list lock, now walk the temporary list. |
1920 | * All I/O is issued async and then if we need to wait for completion | |
1921 | * we do that after issuing all the IO. | |
1da177e4 | 1922 | */ |
089716aa DC |
1923 | list_sort(NULL, &tmp_list, xfs_buf_cmp); |
1924 | while (!list_empty(&tmp_list)) { | |
1925 | bp = list_first_entry(&tmp_list, struct xfs_buf, b_list); | |
585e6d88 | 1926 | ASSERT(target == bp->b_target); |
089716aa DC |
1927 | list_del_init(&bp->b_list); |
1928 | if (wait) { | |
ce8e922c | 1929 | bp->b_flags &= ~XBF_ASYNC; |
089716aa DC |
1930 | list_add(&bp->b_list, &wait_list); |
1931 | } | |
ce8e922c | 1932 | xfs_buf_iostrategy(bp); |
1da177e4 LT |
1933 | } |
1934 | ||
089716aa DC |
1935 | if (wait) { |
1936 | /* Expedite and wait for IO to complete. */ | |
f07c2250 | 1937 | blk_run_address_space(target->bt_mapping); |
089716aa DC |
1938 | while (!list_empty(&wait_list)) { |
1939 | bp = list_first_entry(&wait_list, struct xfs_buf, b_list); | |
f07c2250 | 1940 | |
089716aa DC |
1941 | list_del_init(&bp->b_list); |
1942 | xfs_iowait(bp); | |
1943 | xfs_buf_relse(bp); | |
1944 | } | |
1da177e4 LT |
1945 | } |
1946 | ||
1da177e4 LT |
1947 | return pincount; |
1948 | } | |
1949 | ||
04d8b284 | 1950 | int __init |
ce8e922c | 1951 | xfs_buf_init(void) |
1da177e4 | 1952 | { |
8758280f NS |
1953 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1954 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1955 | if (!xfs_buf_zone) |
0b1b213f | 1956 | goto out; |
04d8b284 | 1957 | |
b4337692 | 1958 | xfslogd_workqueue = create_workqueue("xfslogd"); |
23ea4032 | 1959 | if (!xfslogd_workqueue) |
04d8b284 | 1960 | goto out_free_buf_zone; |
1da177e4 | 1961 | |
b4337692 | 1962 | xfsdatad_workqueue = create_workqueue("xfsdatad"); |
23ea4032 CH |
1963 | if (!xfsdatad_workqueue) |
1964 | goto out_destroy_xfslogd_workqueue; | |
1da177e4 | 1965 | |
c626d174 DC |
1966 | xfsconvertd_workqueue = create_workqueue("xfsconvertd"); |
1967 | if (!xfsconvertd_workqueue) | |
1968 | goto out_destroy_xfsdatad_workqueue; | |
1969 | ||
8e1f936b | 1970 | register_shrinker(&xfs_buf_shake); |
23ea4032 | 1971 | return 0; |
1da177e4 | 1972 | |
c626d174 DC |
1973 | out_destroy_xfsdatad_workqueue: |
1974 | destroy_workqueue(xfsdatad_workqueue); | |
23ea4032 CH |
1975 | out_destroy_xfslogd_workqueue: |
1976 | destroy_workqueue(xfslogd_workqueue); | |
23ea4032 | 1977 | out_free_buf_zone: |
ce8e922c | 1978 | kmem_zone_destroy(xfs_buf_zone); |
0b1b213f | 1979 | out: |
8758280f | 1980 | return -ENOMEM; |
1da177e4 LT |
1981 | } |
1982 | ||
1da177e4 | 1983 | void |
ce8e922c | 1984 | xfs_buf_terminate(void) |
1da177e4 | 1985 | { |
8e1f936b | 1986 | unregister_shrinker(&xfs_buf_shake); |
c626d174 | 1987 | destroy_workqueue(xfsconvertd_workqueue); |
04d8b284 CH |
1988 | destroy_workqueue(xfsdatad_workqueue); |
1989 | destroy_workqueue(xfslogd_workqueue); | |
ce8e922c | 1990 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 1991 | } |
e6a0e9cd TS |
1992 | |
1993 | #ifdef CONFIG_KDB_MODULES | |
1994 | struct list_head * | |
1995 | xfs_get_buftarg_list(void) | |
1996 | { | |
1997 | return &xfs_buftarg_list; | |
1998 | } | |
1999 | #endif |