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 | /* |
ce8e922c | 166 | * Internal xfs_buf_t object manipulation |
1da177e4 LT |
167 | */ |
168 | ||
169 | STATIC void | |
ce8e922c NS |
170 | _xfs_buf_initialize( |
171 | xfs_buf_t *bp, | |
1da177e4 | 172 | xfs_buftarg_t *target, |
204ab25f | 173 | xfs_off_t range_base, |
1da177e4 | 174 | size_t range_length, |
ce8e922c | 175 | xfs_buf_flags_t flags) |
1da177e4 LT |
176 | { |
177 | /* | |
ce8e922c | 178 | * We don't want certain flags to appear in b_flags. |
1da177e4 | 179 | */ |
ce8e922c NS |
180 | flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD); |
181 | ||
182 | memset(bp, 0, sizeof(xfs_buf_t)); | |
183 | atomic_set(&bp->b_hold, 1); | |
b4dd330b | 184 | init_completion(&bp->b_iowait); |
ce8e922c | 185 | INIT_LIST_HEAD(&bp->b_list); |
74f75a0c | 186 | RB_CLEAR_NODE(&bp->b_rbnode); |
a731cd11 | 187 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
ce8e922c NS |
188 | XB_SET_OWNER(bp); |
189 | bp->b_target = target; | |
190 | bp->b_file_offset = range_base; | |
1da177e4 LT |
191 | /* |
192 | * Set buffer_length and count_desired to the same value initially. | |
193 | * I/O routines should use count_desired, which will be the same in | |
194 | * most cases but may be reset (e.g. XFS recovery). | |
195 | */ | |
ce8e922c NS |
196 | bp->b_buffer_length = bp->b_count_desired = range_length; |
197 | bp->b_flags = flags; | |
198 | bp->b_bn = XFS_BUF_DADDR_NULL; | |
199 | atomic_set(&bp->b_pin_count, 0); | |
200 | init_waitqueue_head(&bp->b_waiters); | |
201 | ||
202 | XFS_STATS_INC(xb_create); | |
0b1b213f CH |
203 | |
204 | trace_xfs_buf_init(bp, _RET_IP_); | |
1da177e4 LT |
205 | } |
206 | ||
207 | /* | |
ce8e922c NS |
208 | * Allocate a page array capable of holding a specified number |
209 | * of pages, and point the page buf at it. | |
1da177e4 LT |
210 | */ |
211 | STATIC int | |
ce8e922c NS |
212 | _xfs_buf_get_pages( |
213 | xfs_buf_t *bp, | |
1da177e4 | 214 | int page_count, |
ce8e922c | 215 | xfs_buf_flags_t flags) |
1da177e4 LT |
216 | { |
217 | /* Make sure that we have a page list */ | |
ce8e922c NS |
218 | if (bp->b_pages == NULL) { |
219 | bp->b_offset = xfs_buf_poff(bp->b_file_offset); | |
220 | bp->b_page_count = page_count; | |
221 | if (page_count <= XB_PAGES) { | |
222 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 223 | } else { |
ce8e922c NS |
224 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
225 | page_count, xb_to_km(flags)); | |
226 | if (bp->b_pages == NULL) | |
1da177e4 LT |
227 | return -ENOMEM; |
228 | } | |
ce8e922c | 229 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
230 | } |
231 | return 0; | |
232 | } | |
233 | ||
234 | /* | |
ce8e922c | 235 | * Frees b_pages if it was allocated. |
1da177e4 LT |
236 | */ |
237 | STATIC void | |
ce8e922c | 238 | _xfs_buf_free_pages( |
1da177e4 LT |
239 | xfs_buf_t *bp) |
240 | { | |
ce8e922c | 241 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 242 | kmem_free(bp->b_pages); |
3fc98b1a | 243 | bp->b_pages = NULL; |
1da177e4 LT |
244 | } |
245 | } | |
246 | ||
247 | /* | |
248 | * Releases the specified buffer. | |
249 | * | |
250 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 251 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
252 | * hashed and refcounted buffers |
253 | */ | |
254 | void | |
ce8e922c | 255 | xfs_buf_free( |
1da177e4 LT |
256 | xfs_buf_t *bp) |
257 | { | |
0b1b213f | 258 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 259 | |
1fa40b01 | 260 | if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) { |
1da177e4 LT |
261 | uint i; |
262 | ||
73c77e2c | 263 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
264 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
265 | bp->b_page_count); | |
1da177e4 | 266 | |
948ecdb4 NS |
267 | for (i = 0; i < bp->b_page_count; i++) { |
268 | struct page *page = bp->b_pages[i]; | |
269 | ||
1fa40b01 CH |
270 | if (bp->b_flags & _XBF_PAGE_CACHE) |
271 | ASSERT(!PagePrivate(page)); | |
948ecdb4 NS |
272 | page_cache_release(page); |
273 | } | |
1da177e4 | 274 | } |
3fc98b1a | 275 | _xfs_buf_free_pages(bp); |
ce8e922c | 276 | xfs_buf_deallocate(bp); |
1da177e4 LT |
277 | } |
278 | ||
279 | /* | |
280 | * Finds all pages for buffer in question and builds it's page list. | |
281 | */ | |
282 | STATIC int | |
ce8e922c | 283 | _xfs_buf_lookup_pages( |
1da177e4 LT |
284 | xfs_buf_t *bp, |
285 | uint flags) | |
286 | { | |
ce8e922c NS |
287 | struct address_space *mapping = bp->b_target->bt_mapping; |
288 | size_t blocksize = bp->b_target->bt_bsize; | |
289 | size_t size = bp->b_count_desired; | |
1da177e4 | 290 | size_t nbytes, offset; |
ce8e922c | 291 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 LT |
292 | unsigned short page_count, i; |
293 | pgoff_t first; | |
204ab25f | 294 | xfs_off_t end; |
1da177e4 LT |
295 | int error; |
296 | ||
ce8e922c NS |
297 | end = bp->b_file_offset + bp->b_buffer_length; |
298 | page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset); | |
1da177e4 | 299 | |
ce8e922c | 300 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
301 | if (unlikely(error)) |
302 | return error; | |
ce8e922c | 303 | bp->b_flags |= _XBF_PAGE_CACHE; |
1da177e4 | 304 | |
ce8e922c NS |
305 | offset = bp->b_offset; |
306 | first = bp->b_file_offset >> PAGE_CACHE_SHIFT; | |
1da177e4 | 307 | |
ce8e922c | 308 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
309 | struct page *page; |
310 | uint retries = 0; | |
311 | ||
312 | retry: | |
313 | page = find_or_create_page(mapping, first + i, gfp_mask); | |
314 | if (unlikely(page == NULL)) { | |
ce8e922c NS |
315 | if (flags & XBF_READ_AHEAD) { |
316 | bp->b_page_count = i; | |
6ab455ee CH |
317 | for (i = 0; i < bp->b_page_count; i++) |
318 | unlock_page(bp->b_pages[i]); | |
1da177e4 LT |
319 | return -ENOMEM; |
320 | } | |
321 | ||
322 | /* | |
323 | * This could deadlock. | |
324 | * | |
325 | * But until all the XFS lowlevel code is revamped to | |
326 | * handle buffer allocation failures we can't do much. | |
327 | */ | |
328 | if (!(++retries % 100)) | |
329 | printk(KERN_ERR | |
330 | "XFS: possible memory allocation " | |
331 | "deadlock in %s (mode:0x%x)\n", | |
34a622b2 | 332 | __func__, gfp_mask); |
1da177e4 | 333 | |
ce8e922c | 334 | XFS_STATS_INC(xb_page_retries); |
8aa7e847 | 335 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
336 | goto retry; |
337 | } | |
338 | ||
ce8e922c | 339 | XFS_STATS_INC(xb_page_found); |
1da177e4 LT |
340 | |
341 | nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset); | |
342 | size -= nbytes; | |
343 | ||
948ecdb4 | 344 | ASSERT(!PagePrivate(page)); |
1da177e4 LT |
345 | if (!PageUptodate(page)) { |
346 | page_count--; | |
6ab455ee CH |
347 | if (blocksize >= PAGE_CACHE_SIZE) { |
348 | if (flags & XBF_READ) | |
349 | bp->b_flags |= _XBF_PAGE_LOCKED; | |
350 | } else if (!PagePrivate(page)) { | |
1da177e4 LT |
351 | if (test_page_region(page, offset, nbytes)) |
352 | page_count++; | |
353 | } | |
354 | } | |
355 | ||
ce8e922c | 356 | bp->b_pages[i] = page; |
1da177e4 LT |
357 | offset = 0; |
358 | } | |
359 | ||
6ab455ee CH |
360 | if (!(bp->b_flags & _XBF_PAGE_LOCKED)) { |
361 | for (i = 0; i < bp->b_page_count; i++) | |
362 | unlock_page(bp->b_pages[i]); | |
363 | } | |
364 | ||
ce8e922c NS |
365 | if (page_count == bp->b_page_count) |
366 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 367 | |
1da177e4 LT |
368 | return error; |
369 | } | |
370 | ||
371 | /* | |
372 | * Map buffer into kernel address-space if nessecary. | |
373 | */ | |
374 | STATIC int | |
ce8e922c | 375 | _xfs_buf_map_pages( |
1da177e4 LT |
376 | xfs_buf_t *bp, |
377 | uint flags) | |
378 | { | |
379 | /* A single page buffer is always mappable */ | |
ce8e922c NS |
380 | if (bp->b_page_count == 1) { |
381 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; | |
382 | bp->b_flags |= XBF_MAPPED; | |
383 | } else if (flags & XBF_MAPPED) { | |
8a262e57 AE |
384 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, |
385 | -1, PAGE_KERNEL); | |
ce8e922c | 386 | if (unlikely(bp->b_addr == NULL)) |
1da177e4 | 387 | return -ENOMEM; |
ce8e922c NS |
388 | bp->b_addr += bp->b_offset; |
389 | bp->b_flags |= XBF_MAPPED; | |
1da177e4 LT |
390 | } |
391 | ||
392 | return 0; | |
393 | } | |
394 | ||
395 | /* | |
396 | * Finding and Reading Buffers | |
397 | */ | |
398 | ||
399 | /* | |
ce8e922c | 400 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 LT |
401 | * a given range of an inode. The buffer is returned |
402 | * locked. If other overlapping buffers exist, they are | |
403 | * released before the new buffer is created and locked, | |
404 | * which may imply that this call will block until those buffers | |
405 | * are unlocked. No I/O is implied by this call. | |
406 | */ | |
407 | xfs_buf_t * | |
ce8e922c | 408 | _xfs_buf_find( |
1da177e4 | 409 | xfs_buftarg_t *btp, /* block device target */ |
204ab25f | 410 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 411 | size_t isize, /* length of range */ |
ce8e922c NS |
412 | xfs_buf_flags_t flags, |
413 | xfs_buf_t *new_bp) | |
1da177e4 | 414 | { |
204ab25f | 415 | xfs_off_t range_base; |
1da177e4 | 416 | size_t range_length; |
74f75a0c DC |
417 | struct xfs_perag *pag; |
418 | struct rb_node **rbp; | |
419 | struct rb_node *parent; | |
420 | xfs_buf_t *bp; | |
1da177e4 LT |
421 | |
422 | range_base = (ioff << BBSHIFT); | |
423 | range_length = (isize << BBSHIFT); | |
424 | ||
425 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
ce8e922c | 426 | ASSERT(!(range_length < (1 << btp->bt_sshift))); |
204ab25f | 427 | ASSERT(!(range_base & (xfs_off_t)btp->bt_smask)); |
1da177e4 | 428 | |
74f75a0c DC |
429 | /* get tree root */ |
430 | pag = xfs_perag_get(btp->bt_mount, | |
431 | xfs_daddr_to_agno(btp->bt_mount, ioff)); | |
432 | ||
433 | /* walk tree */ | |
434 | spin_lock(&pag->pag_buf_lock); | |
435 | rbp = &pag->pag_buf_tree.rb_node; | |
436 | parent = NULL; | |
437 | bp = NULL; | |
438 | while (*rbp) { | |
439 | parent = *rbp; | |
440 | bp = rb_entry(parent, struct xfs_buf, b_rbnode); | |
441 | ||
442 | if (range_base < bp->b_file_offset) | |
443 | rbp = &(*rbp)->rb_left; | |
444 | else if (range_base > bp->b_file_offset) | |
445 | rbp = &(*rbp)->rb_right; | |
446 | else { | |
447 | /* | |
448 | * found a block offset match. If the range doesn't | |
449 | * match, the only way this is allowed is if the buffer | |
450 | * in the cache is stale and the transaction that made | |
451 | * it stale has not yet committed. i.e. we are | |
452 | * reallocating a busy extent. Skip this buffer and | |
453 | * continue searching to the right for an exact match. | |
454 | */ | |
455 | if (bp->b_buffer_length != range_length) { | |
456 | ASSERT(bp->b_flags & XBF_STALE); | |
457 | rbp = &(*rbp)->rb_right; | |
458 | continue; | |
459 | } | |
ce8e922c | 460 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
461 | goto found; |
462 | } | |
463 | } | |
464 | ||
465 | /* No match found */ | |
ce8e922c NS |
466 | if (new_bp) { |
467 | _xfs_buf_initialize(new_bp, btp, range_base, | |
1da177e4 | 468 | range_length, flags); |
74f75a0c DC |
469 | rb_link_node(&new_bp->b_rbnode, parent, rbp); |
470 | rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree); | |
471 | /* the buffer keeps the perag reference until it is freed */ | |
472 | new_bp->b_pag = pag; | |
473 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 474 | } else { |
ce8e922c | 475 | XFS_STATS_INC(xb_miss_locked); |
74f75a0c DC |
476 | spin_unlock(&pag->pag_buf_lock); |
477 | xfs_perag_put(pag); | |
1da177e4 | 478 | } |
ce8e922c | 479 | return new_bp; |
1da177e4 LT |
480 | |
481 | found: | |
74f75a0c DC |
482 | spin_unlock(&pag->pag_buf_lock); |
483 | xfs_perag_put(pag); | |
1da177e4 | 484 | |
90810b9e DC |
485 | if (xfs_buf_cond_lock(bp)) { |
486 | /* failed, so wait for the lock if requested. */ | |
ce8e922c | 487 | if (!(flags & XBF_TRYLOCK)) { |
ce8e922c NS |
488 | xfs_buf_lock(bp); |
489 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 | 490 | } else { |
ce8e922c NS |
491 | xfs_buf_rele(bp); |
492 | XFS_STATS_INC(xb_busy_locked); | |
493 | return NULL; | |
1da177e4 | 494 | } |
1da177e4 LT |
495 | } |
496 | ||
ce8e922c NS |
497 | if (bp->b_flags & XBF_STALE) { |
498 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
499 | bp->b_flags &= XBF_MAPPED; | |
2f926587 | 500 | } |
0b1b213f CH |
501 | |
502 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ce8e922c NS |
503 | XFS_STATS_INC(xb_get_locked); |
504 | return bp; | |
1da177e4 LT |
505 | } |
506 | ||
507 | /* | |
ce8e922c | 508 | * Assembles a buffer covering the specified range. |
1da177e4 LT |
509 | * Storage in memory for all portions of the buffer will be allocated, |
510 | * although backing storage may not be. | |
511 | */ | |
512 | xfs_buf_t * | |
6ad112bf | 513 | xfs_buf_get( |
1da177e4 | 514 | xfs_buftarg_t *target,/* target for buffer */ |
204ab25f | 515 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 516 | size_t isize, /* length of range */ |
ce8e922c | 517 | xfs_buf_flags_t flags) |
1da177e4 | 518 | { |
ce8e922c | 519 | xfs_buf_t *bp, *new_bp; |
1da177e4 LT |
520 | int error = 0, i; |
521 | ||
ce8e922c NS |
522 | new_bp = xfs_buf_allocate(flags); |
523 | if (unlikely(!new_bp)) | |
1da177e4 LT |
524 | return NULL; |
525 | ||
ce8e922c NS |
526 | bp = _xfs_buf_find(target, ioff, isize, flags, new_bp); |
527 | if (bp == new_bp) { | |
528 | error = _xfs_buf_lookup_pages(bp, flags); | |
1da177e4 LT |
529 | if (error) |
530 | goto no_buffer; | |
531 | } else { | |
ce8e922c NS |
532 | xfs_buf_deallocate(new_bp); |
533 | if (unlikely(bp == NULL)) | |
1da177e4 LT |
534 | return NULL; |
535 | } | |
536 | ||
ce8e922c NS |
537 | for (i = 0; i < bp->b_page_count; i++) |
538 | mark_page_accessed(bp->b_pages[i]); | |
1da177e4 | 539 | |
ce8e922c NS |
540 | if (!(bp->b_flags & XBF_MAPPED)) { |
541 | error = _xfs_buf_map_pages(bp, flags); | |
1da177e4 LT |
542 | if (unlikely(error)) { |
543 | printk(KERN_WARNING "%s: failed to map pages\n", | |
34a622b2 | 544 | __func__); |
1da177e4 LT |
545 | goto no_buffer; |
546 | } | |
547 | } | |
548 | ||
ce8e922c | 549 | XFS_STATS_INC(xb_get); |
1da177e4 LT |
550 | |
551 | /* | |
552 | * Always fill in the block number now, the mapped cases can do | |
553 | * their own overlay of this later. | |
554 | */ | |
ce8e922c NS |
555 | bp->b_bn = ioff; |
556 | bp->b_count_desired = bp->b_buffer_length; | |
1da177e4 | 557 | |
0b1b213f | 558 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 559 | return bp; |
1da177e4 LT |
560 | |
561 | no_buffer: | |
ce8e922c NS |
562 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
563 | xfs_buf_unlock(bp); | |
564 | xfs_buf_rele(bp); | |
1da177e4 LT |
565 | return NULL; |
566 | } | |
567 | ||
5d765b97 CH |
568 | STATIC int |
569 | _xfs_buf_read( | |
570 | xfs_buf_t *bp, | |
571 | xfs_buf_flags_t flags) | |
572 | { | |
573 | int status; | |
574 | ||
5d765b97 CH |
575 | ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE))); |
576 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); | |
577 | ||
578 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \ | |
579 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
580 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \ | |
581 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
582 | ||
583 | status = xfs_buf_iorequest(bp); | |
ec53d1db DC |
584 | if (status || XFS_BUF_ISERROR(bp) || (flags & XBF_ASYNC)) |
585 | return status; | |
586 | return xfs_buf_iowait(bp); | |
5d765b97 CH |
587 | } |
588 | ||
1da177e4 | 589 | xfs_buf_t * |
6ad112bf | 590 | xfs_buf_read( |
1da177e4 | 591 | xfs_buftarg_t *target, |
204ab25f | 592 | xfs_off_t ioff, |
1da177e4 | 593 | size_t isize, |
ce8e922c | 594 | xfs_buf_flags_t flags) |
1da177e4 | 595 | { |
ce8e922c NS |
596 | xfs_buf_t *bp; |
597 | ||
598 | flags |= XBF_READ; | |
599 | ||
6ad112bf | 600 | bp = xfs_buf_get(target, ioff, isize, flags); |
ce8e922c | 601 | if (bp) { |
0b1b213f CH |
602 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
603 | ||
ce8e922c | 604 | if (!XFS_BUF_ISDONE(bp)) { |
ce8e922c | 605 | XFS_STATS_INC(xb_get_read); |
5d765b97 | 606 | _xfs_buf_read(bp, flags); |
ce8e922c | 607 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
608 | /* |
609 | * Read ahead call which is already satisfied, | |
610 | * drop the buffer | |
611 | */ | |
612 | goto no_buffer; | |
613 | } else { | |
1da177e4 | 614 | /* We do not want read in the flags */ |
ce8e922c | 615 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
616 | } |
617 | } | |
618 | ||
ce8e922c | 619 | return bp; |
1da177e4 LT |
620 | |
621 | no_buffer: | |
ce8e922c NS |
622 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
623 | xfs_buf_unlock(bp); | |
624 | xfs_buf_rele(bp); | |
1da177e4 LT |
625 | return NULL; |
626 | } | |
627 | ||
1da177e4 | 628 | /* |
ce8e922c NS |
629 | * If we are not low on memory then do the readahead in a deadlock |
630 | * safe manner. | |
1da177e4 LT |
631 | */ |
632 | void | |
ce8e922c | 633 | xfs_buf_readahead( |
1da177e4 | 634 | xfs_buftarg_t *target, |
204ab25f | 635 | xfs_off_t ioff, |
1a1a3e97 | 636 | size_t isize) |
1da177e4 LT |
637 | { |
638 | struct backing_dev_info *bdi; | |
639 | ||
ce8e922c | 640 | bdi = target->bt_mapping->backing_dev_info; |
1da177e4 LT |
641 | if (bdi_read_congested(bdi)) |
642 | return; | |
643 | ||
1a1a3e97 CH |
644 | xfs_buf_read(target, ioff, isize, |
645 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD|XBF_DONT_BLOCK); | |
1da177e4 LT |
646 | } |
647 | ||
5adc94c2 DC |
648 | /* |
649 | * Read an uncached buffer from disk. Allocates and returns a locked | |
650 | * buffer containing the disk contents or nothing. | |
651 | */ | |
652 | struct xfs_buf * | |
653 | xfs_buf_read_uncached( | |
654 | struct xfs_mount *mp, | |
655 | struct xfs_buftarg *target, | |
656 | xfs_daddr_t daddr, | |
657 | size_t length, | |
658 | int flags) | |
659 | { | |
660 | xfs_buf_t *bp; | |
661 | int error; | |
662 | ||
663 | bp = xfs_buf_get_uncached(target, length, flags); | |
664 | if (!bp) | |
665 | return NULL; | |
666 | ||
667 | /* set up the buffer for a read IO */ | |
668 | xfs_buf_lock(bp); | |
669 | XFS_BUF_SET_ADDR(bp, daddr); | |
670 | XFS_BUF_READ(bp); | |
671 | XFS_BUF_BUSY(bp); | |
672 | ||
673 | xfsbdstrat(mp, bp); | |
1a1a3e97 | 674 | error = xfs_buf_iowait(bp); |
5adc94c2 DC |
675 | if (error || bp->b_error) { |
676 | xfs_buf_relse(bp); | |
677 | return NULL; | |
678 | } | |
679 | return bp; | |
1da177e4 LT |
680 | } |
681 | ||
682 | xfs_buf_t * | |
ce8e922c | 683 | xfs_buf_get_empty( |
1da177e4 LT |
684 | size_t len, |
685 | xfs_buftarg_t *target) | |
686 | { | |
ce8e922c | 687 | xfs_buf_t *bp; |
1da177e4 | 688 | |
ce8e922c NS |
689 | bp = xfs_buf_allocate(0); |
690 | if (bp) | |
691 | _xfs_buf_initialize(bp, target, 0, len, 0); | |
692 | return bp; | |
1da177e4 LT |
693 | } |
694 | ||
695 | static inline struct page * | |
696 | mem_to_page( | |
697 | void *addr) | |
698 | { | |
9e2779fa | 699 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
700 | return virt_to_page(addr); |
701 | } else { | |
702 | return vmalloc_to_page(addr); | |
703 | } | |
704 | } | |
705 | ||
706 | int | |
ce8e922c NS |
707 | xfs_buf_associate_memory( |
708 | xfs_buf_t *bp, | |
1da177e4 LT |
709 | void *mem, |
710 | size_t len) | |
711 | { | |
712 | int rval; | |
713 | int i = 0; | |
d1afb678 LM |
714 | unsigned long pageaddr; |
715 | unsigned long offset; | |
716 | size_t buflen; | |
1da177e4 LT |
717 | int page_count; |
718 | ||
d1afb678 LM |
719 | pageaddr = (unsigned long)mem & PAGE_CACHE_MASK; |
720 | offset = (unsigned long)mem - pageaddr; | |
721 | buflen = PAGE_CACHE_ALIGN(len + offset); | |
722 | page_count = buflen >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
723 | |
724 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
725 | if (bp->b_pages) |
726 | _xfs_buf_free_pages(bp); | |
1da177e4 | 727 | |
ce8e922c NS |
728 | bp->b_pages = NULL; |
729 | bp->b_addr = mem; | |
1da177e4 | 730 | |
36fae17a | 731 | rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK); |
1da177e4 LT |
732 | if (rval) |
733 | return rval; | |
734 | ||
ce8e922c | 735 | bp->b_offset = offset; |
d1afb678 LM |
736 | |
737 | for (i = 0; i < bp->b_page_count; i++) { | |
738 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
739 | pageaddr += PAGE_CACHE_SIZE; | |
1da177e4 | 740 | } |
1da177e4 | 741 | |
d1afb678 LM |
742 | bp->b_count_desired = len; |
743 | bp->b_buffer_length = buflen; | |
ce8e922c | 744 | bp->b_flags |= XBF_MAPPED; |
6ab455ee | 745 | bp->b_flags &= ~_XBF_PAGE_LOCKED; |
1da177e4 LT |
746 | |
747 | return 0; | |
748 | } | |
749 | ||
750 | xfs_buf_t * | |
686865f7 DC |
751 | xfs_buf_get_uncached( |
752 | struct xfs_buftarg *target, | |
1da177e4 | 753 | size_t len, |
686865f7 | 754 | int flags) |
1da177e4 | 755 | { |
1fa40b01 CH |
756 | unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT; |
757 | int error, i; | |
1da177e4 | 758 | xfs_buf_t *bp; |
1da177e4 | 759 | |
ce8e922c | 760 | bp = xfs_buf_allocate(0); |
1da177e4 LT |
761 | if (unlikely(bp == NULL)) |
762 | goto fail; | |
ce8e922c | 763 | _xfs_buf_initialize(bp, target, 0, len, 0); |
1da177e4 | 764 | |
1fa40b01 CH |
765 | error = _xfs_buf_get_pages(bp, page_count, 0); |
766 | if (error) | |
1da177e4 LT |
767 | goto fail_free_buf; |
768 | ||
1fa40b01 | 769 | for (i = 0; i < page_count; i++) { |
686865f7 | 770 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
771 | if (!bp->b_pages[i]) |
772 | goto fail_free_mem; | |
1da177e4 | 773 | } |
1fa40b01 | 774 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 775 | |
1fa40b01 CH |
776 | error = _xfs_buf_map_pages(bp, XBF_MAPPED); |
777 | if (unlikely(error)) { | |
778 | printk(KERN_WARNING "%s: failed to map pages\n", | |
34a622b2 | 779 | __func__); |
1da177e4 | 780 | goto fail_free_mem; |
1fa40b01 | 781 | } |
1da177e4 | 782 | |
ce8e922c | 783 | xfs_buf_unlock(bp); |
1da177e4 | 784 | |
686865f7 | 785 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 786 | return bp; |
1fa40b01 | 787 | |
1da177e4 | 788 | fail_free_mem: |
1fa40b01 CH |
789 | while (--i >= 0) |
790 | __free_page(bp->b_pages[i]); | |
ca165b88 | 791 | _xfs_buf_free_pages(bp); |
1da177e4 | 792 | fail_free_buf: |
ca165b88 | 793 | xfs_buf_deallocate(bp); |
1da177e4 LT |
794 | fail: |
795 | return NULL; | |
796 | } | |
797 | ||
798 | /* | |
1da177e4 LT |
799 | * Increment reference count on buffer, to hold the buffer concurrently |
800 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
801 | * Must hold the buffer already to call this function. |
802 | */ | |
803 | void | |
ce8e922c NS |
804 | xfs_buf_hold( |
805 | xfs_buf_t *bp) | |
1da177e4 | 806 | { |
0b1b213f | 807 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 808 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
809 | } |
810 | ||
811 | /* | |
ce8e922c NS |
812 | * Releases a hold on the specified buffer. If the |
813 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
814 | */ |
815 | void | |
ce8e922c NS |
816 | xfs_buf_rele( |
817 | xfs_buf_t *bp) | |
1da177e4 | 818 | { |
74f75a0c | 819 | struct xfs_perag *pag = bp->b_pag; |
1da177e4 | 820 | |
0b1b213f | 821 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 822 | |
74f75a0c | 823 | if (!pag) { |
fad3aa1e | 824 | ASSERT(!bp->b_relse); |
74f75a0c | 825 | ASSERT(RB_EMPTY_NODE(&bp->b_rbnode)); |
fad3aa1e NS |
826 | if (atomic_dec_and_test(&bp->b_hold)) |
827 | xfs_buf_free(bp); | |
828 | return; | |
829 | } | |
830 | ||
74f75a0c | 831 | ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode)); |
3790689f | 832 | ASSERT(atomic_read(&bp->b_hold) > 0); |
74f75a0c | 833 | if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) { |
ce8e922c NS |
834 | if (bp->b_relse) { |
835 | atomic_inc(&bp->b_hold); | |
74f75a0c DC |
836 | spin_unlock(&pag->pag_buf_lock); |
837 | bp->b_relse(bp); | |
1da177e4 | 838 | } else { |
ce8e922c | 839 | ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q))); |
74f75a0c DC |
840 | rb_erase(&bp->b_rbnode, &pag->pag_buf_tree); |
841 | spin_unlock(&pag->pag_buf_lock); | |
842 | xfs_perag_put(pag); | |
ce8e922c | 843 | xfs_buf_free(bp); |
1da177e4 LT |
844 | } |
845 | } | |
846 | } | |
847 | ||
848 | ||
849 | /* | |
850 | * Mutual exclusion on buffers. Locking model: | |
851 | * | |
852 | * Buffers associated with inodes for which buffer locking | |
853 | * is not enabled are not protected by semaphores, and are | |
854 | * assumed to be exclusively owned by the caller. There is a | |
855 | * spinlock in the buffer, used by the caller when concurrent | |
856 | * access is possible. | |
857 | */ | |
858 | ||
859 | /* | |
90810b9e DC |
860 | * Locks a buffer object, if it is not already locked. Note that this in |
861 | * no way locks the underlying pages, so it is only useful for | |
862 | * synchronizing concurrent use of buffer objects, not for synchronizing | |
863 | * independent access to the underlying pages. | |
864 | * | |
865 | * If we come across a stale, pinned, locked buffer, we know that we are | |
866 | * being asked to lock a buffer that has been reallocated. Because it is | |
867 | * pinned, we know that the log has not been pushed to disk and hence it | |
868 | * will still be locked. Rather than continuing to have trylock attempts | |
869 | * fail until someone else pushes the log, push it ourselves before | |
870 | * returning. This means that the xfsaild will not get stuck trying | |
871 | * to push on stale inode buffers. | |
1da177e4 LT |
872 | */ |
873 | int | |
ce8e922c NS |
874 | xfs_buf_cond_lock( |
875 | xfs_buf_t *bp) | |
1da177e4 LT |
876 | { |
877 | int locked; | |
878 | ||
ce8e922c | 879 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 880 | if (locked) |
ce8e922c | 881 | XB_SET_OWNER(bp); |
90810b9e DC |
882 | else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
883 | xfs_log_force(bp->b_target->bt_mount, 0); | |
0b1b213f CH |
884 | |
885 | trace_xfs_buf_cond_lock(bp, _RET_IP_); | |
ce8e922c | 886 | return locked ? 0 : -EBUSY; |
1da177e4 LT |
887 | } |
888 | ||
1da177e4 | 889 | int |
ce8e922c NS |
890 | xfs_buf_lock_value( |
891 | xfs_buf_t *bp) | |
1da177e4 | 892 | { |
adaa693b | 893 | return bp->b_sema.count; |
1da177e4 | 894 | } |
1da177e4 LT |
895 | |
896 | /* | |
ce8e922c NS |
897 | * Locks a buffer object. |
898 | * Note that this in no way locks the underlying pages, so it is only | |
899 | * useful for synchronizing concurrent use of buffer objects, not for | |
900 | * synchronizing independent access to the underlying pages. | |
ed3b4d6c DC |
901 | * |
902 | * If we come across a stale, pinned, locked buffer, we know that we | |
903 | * are being asked to lock a buffer that has been reallocated. Because | |
904 | * it is pinned, we know that the log has not been pushed to disk and | |
905 | * hence it will still be locked. Rather than sleeping until someone | |
906 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 907 | */ |
ce8e922c NS |
908 | void |
909 | xfs_buf_lock( | |
910 | xfs_buf_t *bp) | |
1da177e4 | 911 | { |
0b1b213f CH |
912 | trace_xfs_buf_lock(bp, _RET_IP_); |
913 | ||
ed3b4d6c | 914 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 915 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
916 | if (atomic_read(&bp->b_io_remaining)) |
917 | blk_run_address_space(bp->b_target->bt_mapping); | |
918 | down(&bp->b_sema); | |
919 | XB_SET_OWNER(bp); | |
0b1b213f CH |
920 | |
921 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
922 | } |
923 | ||
924 | /* | |
ce8e922c | 925 | * Releases the lock on the buffer object. |
2f926587 | 926 | * If the buffer is marked delwri but is not queued, do so before we |
ce8e922c | 927 | * unlock the buffer as we need to set flags correctly. We also need to |
2f926587 DC |
928 | * take a reference for the delwri queue because the unlocker is going to |
929 | * drop their's and they don't know we just queued it. | |
1da177e4 LT |
930 | */ |
931 | void | |
ce8e922c NS |
932 | xfs_buf_unlock( |
933 | xfs_buf_t *bp) | |
1da177e4 | 934 | { |
ce8e922c NS |
935 | if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) { |
936 | atomic_inc(&bp->b_hold); | |
937 | bp->b_flags |= XBF_ASYNC; | |
938 | xfs_buf_delwri_queue(bp, 0); | |
2f926587 DC |
939 | } |
940 | ||
ce8e922c NS |
941 | XB_CLEAR_OWNER(bp); |
942 | up(&bp->b_sema); | |
0b1b213f CH |
943 | |
944 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
945 | } |
946 | ||
ce8e922c NS |
947 | STATIC void |
948 | xfs_buf_wait_unpin( | |
949 | xfs_buf_t *bp) | |
1da177e4 LT |
950 | { |
951 | DECLARE_WAITQUEUE (wait, current); | |
952 | ||
ce8e922c | 953 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
954 | return; |
955 | ||
ce8e922c | 956 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
957 | for (;;) { |
958 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 959 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 960 | break; |
ce8e922c NS |
961 | if (atomic_read(&bp->b_io_remaining)) |
962 | blk_run_address_space(bp->b_target->bt_mapping); | |
1da177e4 LT |
963 | schedule(); |
964 | } | |
ce8e922c | 965 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
966 | set_current_state(TASK_RUNNING); |
967 | } | |
968 | ||
969 | /* | |
970 | * Buffer Utility Routines | |
971 | */ | |
972 | ||
1da177e4 | 973 | STATIC void |
ce8e922c | 974 | xfs_buf_iodone_work( |
c4028958 | 975 | struct work_struct *work) |
1da177e4 | 976 | { |
c4028958 DH |
977 | xfs_buf_t *bp = |
978 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 979 | |
80f6c29d | 980 | if (bp->b_iodone) |
ce8e922c NS |
981 | (*(bp->b_iodone))(bp); |
982 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
983 | xfs_buf_relse(bp); |
984 | } | |
985 | ||
986 | void | |
ce8e922c NS |
987 | xfs_buf_ioend( |
988 | xfs_buf_t *bp, | |
1da177e4 LT |
989 | int schedule) |
990 | { | |
0b1b213f CH |
991 | trace_xfs_buf_iodone(bp, _RET_IP_); |
992 | ||
77be55a5 | 993 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
ce8e922c NS |
994 | if (bp->b_error == 0) |
995 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 996 | |
ce8e922c | 997 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 998 | if (schedule) { |
c4028958 | 999 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 1000 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 1001 | } else { |
c4028958 | 1002 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
1003 | } |
1004 | } else { | |
b4dd330b | 1005 | complete(&bp->b_iowait); |
1da177e4 LT |
1006 | } |
1007 | } | |
1008 | ||
1da177e4 | 1009 | void |
ce8e922c NS |
1010 | xfs_buf_ioerror( |
1011 | xfs_buf_t *bp, | |
1012 | int error) | |
1da177e4 LT |
1013 | { |
1014 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c | 1015 | bp->b_error = (unsigned short)error; |
0b1b213f | 1016 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1017 | } |
1018 | ||
1da177e4 | 1019 | int |
64e0bc7d CH |
1020 | xfs_bwrite( |
1021 | struct xfs_mount *mp, | |
5d765b97 | 1022 | struct xfs_buf *bp) |
1da177e4 | 1023 | { |
8c38366f | 1024 | int error; |
1da177e4 | 1025 | |
64e0bc7d | 1026 | bp->b_flags |= XBF_WRITE; |
8c38366f | 1027 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ); |
1da177e4 | 1028 | |
5d765b97 | 1029 | xfs_buf_delwri_dequeue(bp); |
939d723b | 1030 | xfs_bdstrat_cb(bp); |
1da177e4 | 1031 | |
8c38366f CH |
1032 | error = xfs_buf_iowait(bp); |
1033 | if (error) | |
1034 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1035 | xfs_buf_relse(bp); | |
64e0bc7d | 1036 | return error; |
5d765b97 | 1037 | } |
1da177e4 | 1038 | |
5d765b97 CH |
1039 | void |
1040 | xfs_bdwrite( | |
1041 | void *mp, | |
1042 | struct xfs_buf *bp) | |
1043 | { | |
0b1b213f | 1044 | trace_xfs_buf_bdwrite(bp, _RET_IP_); |
1da177e4 | 1045 | |
5d765b97 CH |
1046 | bp->b_flags &= ~XBF_READ; |
1047 | bp->b_flags |= (XBF_DELWRI | XBF_ASYNC); | |
1048 | ||
1049 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1050 | } |
1051 | ||
4e23471a CH |
1052 | /* |
1053 | * Called when we want to stop a buffer from getting written or read. | |
1a1a3e97 | 1054 | * We attach the EIO error, muck with its flags, and call xfs_buf_ioend |
4e23471a CH |
1055 | * so that the proper iodone callbacks get called. |
1056 | */ | |
1057 | STATIC int | |
1058 | xfs_bioerror( | |
1059 | xfs_buf_t *bp) | |
1060 | { | |
1061 | #ifdef XFSERRORDEBUG | |
1062 | ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone); | |
1063 | #endif | |
1064 | ||
1065 | /* | |
1066 | * No need to wait until the buffer is unpinned, we aren't flushing it. | |
1067 | */ | |
1068 | XFS_BUF_ERROR(bp, EIO); | |
1069 | ||
1070 | /* | |
1a1a3e97 | 1071 | * We're calling xfs_buf_ioend, so delete XBF_DONE flag. |
4e23471a CH |
1072 | */ |
1073 | XFS_BUF_UNREAD(bp); | |
1074 | XFS_BUF_UNDELAYWRITE(bp); | |
1075 | XFS_BUF_UNDONE(bp); | |
1076 | XFS_BUF_STALE(bp); | |
1077 | ||
1a1a3e97 | 1078 | xfs_buf_ioend(bp, 0); |
4e23471a CH |
1079 | |
1080 | return EIO; | |
1081 | } | |
1082 | ||
1083 | /* | |
1084 | * Same as xfs_bioerror, except that we are releasing the buffer | |
1a1a3e97 | 1085 | * here ourselves, and avoiding the xfs_buf_ioend call. |
4e23471a CH |
1086 | * This is meant for userdata errors; metadata bufs come with |
1087 | * iodone functions attached, so that we can track down errors. | |
1088 | */ | |
1089 | STATIC int | |
1090 | xfs_bioerror_relse( | |
1091 | struct xfs_buf *bp) | |
1092 | { | |
1093 | int64_t fl = XFS_BUF_BFLAGS(bp); | |
1094 | /* | |
1095 | * No need to wait until the buffer is unpinned. | |
1096 | * We aren't flushing it. | |
1097 | * | |
1098 | * chunkhold expects B_DONE to be set, whether | |
1099 | * we actually finish the I/O or not. We don't want to | |
1100 | * change that interface. | |
1101 | */ | |
1102 | XFS_BUF_UNREAD(bp); | |
1103 | XFS_BUF_UNDELAYWRITE(bp); | |
1104 | XFS_BUF_DONE(bp); | |
1105 | XFS_BUF_STALE(bp); | |
1106 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
0cadda1c | 1107 | if (!(fl & XBF_ASYNC)) { |
4e23471a CH |
1108 | /* |
1109 | * Mark b_error and B_ERROR _both_. | |
1110 | * Lot's of chunkcache code assumes that. | |
1111 | * There's no reason to mark error for | |
1112 | * ASYNC buffers. | |
1113 | */ | |
1114 | XFS_BUF_ERROR(bp, EIO); | |
1115 | XFS_BUF_FINISH_IOWAIT(bp); | |
1116 | } else { | |
1117 | xfs_buf_relse(bp); | |
1118 | } | |
1119 | ||
1120 | return EIO; | |
1121 | } | |
1122 | ||
1123 | ||
1124 | /* | |
1125 | * All xfs metadata buffers except log state machine buffers | |
1126 | * get this attached as their b_bdstrat callback function. | |
1127 | * This is so that we can catch a buffer | |
1128 | * after prematurely unpinning it to forcibly shutdown the filesystem. | |
1129 | */ | |
1130 | int | |
1131 | xfs_bdstrat_cb( | |
1132 | struct xfs_buf *bp) | |
1133 | { | |
ebad861b | 1134 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
4e23471a CH |
1135 | trace_xfs_bdstrat_shut(bp, _RET_IP_); |
1136 | /* | |
1137 | * Metadata write that didn't get logged but | |
1138 | * written delayed anyway. These aren't associated | |
1139 | * with a transaction, and can be ignored. | |
1140 | */ | |
1141 | if (!bp->b_iodone && !XFS_BUF_ISREAD(bp)) | |
1142 | return xfs_bioerror_relse(bp); | |
1143 | else | |
1144 | return xfs_bioerror(bp); | |
1145 | } | |
1146 | ||
1147 | xfs_buf_iorequest(bp); | |
1148 | return 0; | |
1149 | } | |
1150 | ||
1151 | /* | |
1152 | * Wrapper around bdstrat so that we can stop data from going to disk in case | |
1153 | * we are shutting down the filesystem. Typically user data goes thru this | |
1154 | * path; one of the exceptions is the superblock. | |
1155 | */ | |
1156 | void | |
1157 | xfsbdstrat( | |
1158 | struct xfs_mount *mp, | |
1159 | struct xfs_buf *bp) | |
1160 | { | |
1161 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1162 | trace_xfs_bdstrat_shut(bp, _RET_IP_); | |
1163 | xfs_bioerror_relse(bp); | |
1164 | return; | |
1165 | } | |
1166 | ||
1167 | xfs_buf_iorequest(bp); | |
1168 | } | |
1169 | ||
b8f82a4a | 1170 | STATIC void |
ce8e922c NS |
1171 | _xfs_buf_ioend( |
1172 | xfs_buf_t *bp, | |
1da177e4 LT |
1173 | int schedule) |
1174 | { | |
6ab455ee CH |
1175 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
1176 | bp->b_flags &= ~_XBF_PAGE_LOCKED; | |
ce8e922c | 1177 | xfs_buf_ioend(bp, schedule); |
6ab455ee | 1178 | } |
1da177e4 LT |
1179 | } |
1180 | ||
782e3b3b | 1181 | STATIC void |
ce8e922c | 1182 | xfs_buf_bio_end_io( |
1da177e4 | 1183 | struct bio *bio, |
1da177e4 LT |
1184 | int error) |
1185 | { | |
ce8e922c NS |
1186 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1187 | unsigned int blocksize = bp->b_target->bt_bsize; | |
eedb5530 | 1188 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1da177e4 | 1189 | |
cfbe5267 | 1190 | xfs_buf_ioerror(bp, -error); |
1da177e4 | 1191 | |
73c77e2c JB |
1192 | if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
1193 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); | |
1194 | ||
eedb5530 | 1195 | do { |
1da177e4 LT |
1196 | struct page *page = bvec->bv_page; |
1197 | ||
948ecdb4 | 1198 | ASSERT(!PagePrivate(page)); |
ce8e922c NS |
1199 | if (unlikely(bp->b_error)) { |
1200 | if (bp->b_flags & XBF_READ) | |
eedb5530 | 1201 | ClearPageUptodate(page); |
ce8e922c | 1202 | } else if (blocksize >= PAGE_CACHE_SIZE) { |
1da177e4 LT |
1203 | SetPageUptodate(page); |
1204 | } else if (!PagePrivate(page) && | |
ce8e922c | 1205 | (bp->b_flags & _XBF_PAGE_CACHE)) { |
1da177e4 LT |
1206 | set_page_region(page, bvec->bv_offset, bvec->bv_len); |
1207 | } | |
1208 | ||
eedb5530 NS |
1209 | if (--bvec >= bio->bi_io_vec) |
1210 | prefetchw(&bvec->bv_page->flags); | |
6ab455ee CH |
1211 | |
1212 | if (bp->b_flags & _XBF_PAGE_LOCKED) | |
1213 | unlock_page(page); | |
eedb5530 | 1214 | } while (bvec >= bio->bi_io_vec); |
1da177e4 | 1215 | |
ce8e922c | 1216 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1217 | bio_put(bio); |
1da177e4 LT |
1218 | } |
1219 | ||
1220 | STATIC void | |
ce8e922c NS |
1221 | _xfs_buf_ioapply( |
1222 | xfs_buf_t *bp) | |
1da177e4 | 1223 | { |
a9759f2d | 1224 | int rw, map_i, total_nr_pages, nr_pages; |
1da177e4 | 1225 | struct bio *bio; |
ce8e922c NS |
1226 | int offset = bp->b_offset; |
1227 | int size = bp->b_count_desired; | |
1228 | sector_t sector = bp->b_bn; | |
1229 | unsigned int blocksize = bp->b_target->bt_bsize; | |
1da177e4 | 1230 | |
ce8e922c | 1231 | total_nr_pages = bp->b_page_count; |
1da177e4 LT |
1232 | map_i = 0; |
1233 | ||
ce8e922c NS |
1234 | if (bp->b_flags & XBF_ORDERED) { |
1235 | ASSERT(!(bp->b_flags & XBF_READ)); | |
80f6c29d | 1236 | rw = WRITE_FLUSH_FUA; |
2ee1abad | 1237 | } else if (bp->b_flags & XBF_LOG_BUFFER) { |
51bdd706 NS |
1238 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); |
1239 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1240 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC; | |
2ee1abad DC |
1241 | } else if (bp->b_flags & _XBF_RUN_QUEUES) { |
1242 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); | |
1243 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1244 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META; | |
51bdd706 NS |
1245 | } else { |
1246 | rw = (bp->b_flags & XBF_WRITE) ? WRITE : | |
1247 | (bp->b_flags & XBF_READ_AHEAD) ? READA : READ; | |
f538d4da CH |
1248 | } |
1249 | ||
ce8e922c | 1250 | /* Special code path for reading a sub page size buffer in -- |
1da177e4 LT |
1251 | * we populate up the whole page, and hence the other metadata |
1252 | * in the same page. This optimization is only valid when the | |
ce8e922c | 1253 | * filesystem block size is not smaller than the page size. |
1da177e4 | 1254 | */ |
ce8e922c | 1255 | if ((bp->b_buffer_length < PAGE_CACHE_SIZE) && |
6ab455ee CH |
1256 | ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) == |
1257 | (XBF_READ|_XBF_PAGE_LOCKED)) && | |
ce8e922c | 1258 | (blocksize >= PAGE_CACHE_SIZE)) { |
1da177e4 LT |
1259 | bio = bio_alloc(GFP_NOIO, 1); |
1260 | ||
ce8e922c | 1261 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1262 | bio->bi_sector = sector - (offset >> BBSHIFT); |
ce8e922c NS |
1263 | bio->bi_end_io = xfs_buf_bio_end_io; |
1264 | bio->bi_private = bp; | |
1da177e4 | 1265 | |
ce8e922c | 1266 | bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0); |
1da177e4 LT |
1267 | size = 0; |
1268 | ||
ce8e922c | 1269 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1270 | |
1271 | goto submit_io; | |
1272 | } | |
1273 | ||
1da177e4 | 1274 | next_chunk: |
ce8e922c | 1275 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1276 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1277 | if (nr_pages > total_nr_pages) | |
1278 | nr_pages = total_nr_pages; | |
1279 | ||
1280 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1281 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1282 | bio->bi_sector = sector; |
ce8e922c NS |
1283 | bio->bi_end_io = xfs_buf_bio_end_io; |
1284 | bio->bi_private = bp; | |
1da177e4 LT |
1285 | |
1286 | for (; size && nr_pages; nr_pages--, map_i++) { | |
ce8e922c | 1287 | int rbytes, nbytes = PAGE_CACHE_SIZE - offset; |
1da177e4 LT |
1288 | |
1289 | if (nbytes > size) | |
1290 | nbytes = size; | |
1291 | ||
ce8e922c NS |
1292 | rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset); |
1293 | if (rbytes < nbytes) | |
1da177e4 LT |
1294 | break; |
1295 | ||
1296 | offset = 0; | |
1297 | sector += nbytes >> BBSHIFT; | |
1298 | size -= nbytes; | |
1299 | total_nr_pages--; | |
1300 | } | |
1301 | ||
1302 | submit_io: | |
1303 | if (likely(bio->bi_size)) { | |
73c77e2c JB |
1304 | if (xfs_buf_is_vmapped(bp)) { |
1305 | flush_kernel_vmap_range(bp->b_addr, | |
1306 | xfs_buf_vmap_len(bp)); | |
1307 | } | |
1da177e4 LT |
1308 | submit_bio(rw, bio); |
1309 | if (size) | |
1310 | goto next_chunk; | |
1311 | } else { | |
ec53d1db DC |
1312 | /* |
1313 | * if we get here, no pages were added to the bio. However, | |
1314 | * we can't just error out here - if the pages are locked then | |
1315 | * we have to unlock them otherwise we can hang on a later | |
1316 | * access to the page. | |
1317 | */ | |
ce8e922c | 1318 | xfs_buf_ioerror(bp, EIO); |
ec53d1db DC |
1319 | if (bp->b_flags & _XBF_PAGE_LOCKED) { |
1320 | int i; | |
1321 | for (i = 0; i < bp->b_page_count; i++) | |
1322 | unlock_page(bp->b_pages[i]); | |
1323 | } | |
1324 | bio_put(bio); | |
1da177e4 LT |
1325 | } |
1326 | } | |
1327 | ||
1da177e4 | 1328 | int |
ce8e922c NS |
1329 | xfs_buf_iorequest( |
1330 | xfs_buf_t *bp) | |
1da177e4 | 1331 | { |
0b1b213f | 1332 | trace_xfs_buf_iorequest(bp, _RET_IP_); |
1da177e4 | 1333 | |
ce8e922c NS |
1334 | if (bp->b_flags & XBF_DELWRI) { |
1335 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1336 | return 0; |
1337 | } | |
1338 | ||
ce8e922c NS |
1339 | if (bp->b_flags & XBF_WRITE) { |
1340 | xfs_buf_wait_unpin(bp); | |
1da177e4 LT |
1341 | } |
1342 | ||
ce8e922c | 1343 | xfs_buf_hold(bp); |
1da177e4 LT |
1344 | |
1345 | /* Set the count to 1 initially, this will stop an I/O | |
1346 | * completion callout which happens before we have started | |
ce8e922c | 1347 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1348 | */ |
ce8e922c NS |
1349 | atomic_set(&bp->b_io_remaining, 1); |
1350 | _xfs_buf_ioapply(bp); | |
1351 | _xfs_buf_ioend(bp, 0); | |
1da177e4 | 1352 | |
ce8e922c | 1353 | xfs_buf_rele(bp); |
1da177e4 LT |
1354 | return 0; |
1355 | } | |
1356 | ||
1357 | /* | |
ce8e922c NS |
1358 | * Waits for I/O to complete on the buffer supplied. |
1359 | * It returns immediately if no I/O is pending. | |
1360 | * It returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1361 | */ |
1362 | int | |
ce8e922c NS |
1363 | xfs_buf_iowait( |
1364 | xfs_buf_t *bp) | |
1da177e4 | 1365 | { |
0b1b213f CH |
1366 | trace_xfs_buf_iowait(bp, _RET_IP_); |
1367 | ||
ce8e922c NS |
1368 | if (atomic_read(&bp->b_io_remaining)) |
1369 | blk_run_address_space(bp->b_target->bt_mapping); | |
b4dd330b | 1370 | wait_for_completion(&bp->b_iowait); |
0b1b213f CH |
1371 | |
1372 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
ce8e922c | 1373 | return bp->b_error; |
1da177e4 LT |
1374 | } |
1375 | ||
ce8e922c NS |
1376 | xfs_caddr_t |
1377 | xfs_buf_offset( | |
1378 | xfs_buf_t *bp, | |
1da177e4 LT |
1379 | size_t offset) |
1380 | { | |
1381 | struct page *page; | |
1382 | ||
ce8e922c NS |
1383 | if (bp->b_flags & XBF_MAPPED) |
1384 | return XFS_BUF_PTR(bp) + offset; | |
1da177e4 | 1385 | |
ce8e922c NS |
1386 | offset += bp->b_offset; |
1387 | page = bp->b_pages[offset >> PAGE_CACHE_SHIFT]; | |
1388 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1)); | |
1da177e4 LT |
1389 | } |
1390 | ||
1391 | /* | |
1da177e4 LT |
1392 | * Move data into or out of a buffer. |
1393 | */ | |
1394 | void | |
ce8e922c NS |
1395 | xfs_buf_iomove( |
1396 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1397 | size_t boff, /* starting buffer offset */ |
1398 | size_t bsize, /* length to copy */ | |
b9c48649 | 1399 | void *data, /* data address */ |
ce8e922c | 1400 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 LT |
1401 | { |
1402 | size_t bend, cpoff, csize; | |
1403 | struct page *page; | |
1404 | ||
1405 | bend = boff + bsize; | |
1406 | while (boff < bend) { | |
ce8e922c NS |
1407 | page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)]; |
1408 | cpoff = xfs_buf_poff(boff + bp->b_offset); | |
1da177e4 | 1409 | csize = min_t(size_t, |
ce8e922c | 1410 | PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff); |
1da177e4 LT |
1411 | |
1412 | ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE)); | |
1413 | ||
1414 | switch (mode) { | |
ce8e922c | 1415 | case XBRW_ZERO: |
1da177e4 LT |
1416 | memset(page_address(page) + cpoff, 0, csize); |
1417 | break; | |
ce8e922c | 1418 | case XBRW_READ: |
1da177e4 LT |
1419 | memcpy(data, page_address(page) + cpoff, csize); |
1420 | break; | |
ce8e922c | 1421 | case XBRW_WRITE: |
1da177e4 LT |
1422 | memcpy(page_address(page) + cpoff, data, csize); |
1423 | } | |
1424 | ||
1425 | boff += csize; | |
1426 | data += csize; | |
1427 | } | |
1428 | } | |
1429 | ||
1430 | /* | |
ce8e922c | 1431 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1432 | */ |
1433 | ||
1434 | /* | |
ce8e922c NS |
1435 | * Wait for any bufs with callbacks that have been submitted but |
1436 | * have not yet returned... walk the hash list for the target. | |
1da177e4 LT |
1437 | */ |
1438 | void | |
1439 | xfs_wait_buftarg( | |
74f75a0c | 1440 | struct xfs_buftarg *btp) |
1da177e4 | 1441 | { |
74f75a0c DC |
1442 | struct xfs_perag *pag; |
1443 | uint i; | |
1da177e4 | 1444 | |
74f75a0c DC |
1445 | for (i = 0; i < btp->bt_mount->m_sb.sb_agcount; i++) { |
1446 | pag = xfs_perag_get(btp->bt_mount, i); | |
1447 | spin_lock(&pag->pag_buf_lock); | |
1448 | while (rb_first(&pag->pag_buf_tree)) { | |
1449 | spin_unlock(&pag->pag_buf_lock); | |
26af6552 | 1450 | delay(100); |
74f75a0c | 1451 | spin_lock(&pag->pag_buf_lock); |
1da177e4 | 1452 | } |
74f75a0c DC |
1453 | spin_unlock(&pag->pag_buf_lock); |
1454 | xfs_perag_put(pag); | |
1da177e4 LT |
1455 | } |
1456 | } | |
1457 | ||
ff57ab21 DC |
1458 | int |
1459 | xfs_buftarg_shrink( | |
1460 | struct shrinker *shrink, | |
1461 | int nr_to_scan, | |
1462 | gfp_t mask) | |
a6867a68 | 1463 | { |
ff57ab21 DC |
1464 | struct xfs_buftarg *btp = container_of(shrink, |
1465 | struct xfs_buftarg, bt_shrinker); | |
1466 | if (nr_to_scan) { | |
1467 | if (test_bit(XBT_FORCE_SLEEP, &btp->bt_flags)) | |
1468 | return -1; | |
1469 | if (list_empty(&btp->bt_delwrite_queue)) | |
1470 | return -1; | |
1471 | set_bit(XBT_FORCE_FLUSH, &btp->bt_flags); | |
1472 | wake_up_process(btp->bt_task); | |
1473 | } | |
1474 | return list_empty(&btp->bt_delwrite_queue) ? -1 : 1; | |
a6867a68 DC |
1475 | } |
1476 | ||
1da177e4 LT |
1477 | void |
1478 | xfs_free_buftarg( | |
b7963133 CH |
1479 | struct xfs_mount *mp, |
1480 | struct xfs_buftarg *btp) | |
1da177e4 | 1481 | { |
ff57ab21 DC |
1482 | unregister_shrinker(&btp->bt_shrinker); |
1483 | ||
1da177e4 | 1484 | xfs_flush_buftarg(btp, 1); |
b7963133 CH |
1485 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1486 | xfs_blkdev_issue_flush(btp); | |
ce8e922c | 1487 | iput(btp->bt_mapping->host); |
a6867a68 | 1488 | |
a6867a68 | 1489 | kthread_stop(btp->bt_task); |
f0e2d93c | 1490 | kmem_free(btp); |
1da177e4 LT |
1491 | } |
1492 | ||
1da177e4 LT |
1493 | STATIC int |
1494 | xfs_setsize_buftarg_flags( | |
1495 | xfs_buftarg_t *btp, | |
1496 | unsigned int blocksize, | |
1497 | unsigned int sectorsize, | |
1498 | int verbose) | |
1499 | { | |
ce8e922c NS |
1500 | btp->bt_bsize = blocksize; |
1501 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1502 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1503 | |
ce8e922c | 1504 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
1da177e4 LT |
1505 | printk(KERN_WARNING |
1506 | "XFS: Cannot set_blocksize to %u on device %s\n", | |
1507 | sectorsize, XFS_BUFTARG_NAME(btp)); | |
1508 | return EINVAL; | |
1509 | } | |
1510 | ||
1511 | if (verbose && | |
1512 | (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) { | |
1513 | printk(KERN_WARNING | |
1514 | "XFS: %u byte sectors in use on device %s. " | |
1515 | "This is suboptimal; %u or greater is ideal.\n", | |
1516 | sectorsize, XFS_BUFTARG_NAME(btp), | |
1517 | (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG); | |
1518 | } | |
1519 | ||
1520 | return 0; | |
1521 | } | |
1522 | ||
1523 | /* | |
ce8e922c NS |
1524 | * When allocating the initial buffer target we have not yet |
1525 | * read in the superblock, so don't know what sized sectors | |
1526 | * are being used is at this early stage. Play safe. | |
1527 | */ | |
1da177e4 LT |
1528 | STATIC int |
1529 | xfs_setsize_buftarg_early( | |
1530 | xfs_buftarg_t *btp, | |
1531 | struct block_device *bdev) | |
1532 | { | |
1533 | return xfs_setsize_buftarg_flags(btp, | |
e1defc4f | 1534 | PAGE_CACHE_SIZE, bdev_logical_block_size(bdev), 0); |
1da177e4 LT |
1535 | } |
1536 | ||
1537 | int | |
1538 | xfs_setsize_buftarg( | |
1539 | xfs_buftarg_t *btp, | |
1540 | unsigned int blocksize, | |
1541 | unsigned int sectorsize) | |
1542 | { | |
1543 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1544 | } | |
1545 | ||
1546 | STATIC int | |
1547 | xfs_mapping_buftarg( | |
1548 | xfs_buftarg_t *btp, | |
1549 | struct block_device *bdev) | |
1550 | { | |
1551 | struct backing_dev_info *bdi; | |
1552 | struct inode *inode; | |
1553 | struct address_space *mapping; | |
f5e54d6e | 1554 | static const struct address_space_operations mapping_aops = { |
1da177e4 | 1555 | .sync_page = block_sync_page, |
e965f963 | 1556 | .migratepage = fail_migrate_page, |
1da177e4 LT |
1557 | }; |
1558 | ||
1559 | inode = new_inode(bdev->bd_inode->i_sb); | |
1560 | if (!inode) { | |
1561 | printk(KERN_WARNING | |
1562 | "XFS: Cannot allocate mapping inode for device %s\n", | |
1563 | XFS_BUFTARG_NAME(btp)); | |
1564 | return ENOMEM; | |
1565 | } | |
85fe4025 | 1566 | inode->i_ino = get_next_ino(); |
1da177e4 LT |
1567 | inode->i_mode = S_IFBLK; |
1568 | inode->i_bdev = bdev; | |
1569 | inode->i_rdev = bdev->bd_dev; | |
1570 | bdi = blk_get_backing_dev_info(bdev); | |
1571 | if (!bdi) | |
1572 | bdi = &default_backing_dev_info; | |
1573 | mapping = &inode->i_data; | |
1574 | mapping->a_ops = &mapping_aops; | |
1575 | mapping->backing_dev_info = bdi; | |
1576 | mapping_set_gfp_mask(mapping, GFP_NOFS); | |
ce8e922c | 1577 | btp->bt_mapping = mapping; |
1da177e4 LT |
1578 | return 0; |
1579 | } | |
1580 | ||
a6867a68 DC |
1581 | STATIC int |
1582 | xfs_alloc_delwrite_queue( | |
e2a07812 JE |
1583 | xfs_buftarg_t *btp, |
1584 | const char *fsname) | |
a6867a68 | 1585 | { |
a6867a68 | 1586 | INIT_LIST_HEAD(&btp->bt_delwrite_queue); |
007c61c6 | 1587 | spin_lock_init(&btp->bt_delwrite_lock); |
a6867a68 | 1588 | btp->bt_flags = 0; |
e2a07812 | 1589 | btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd/%s", fsname); |
ff57ab21 DC |
1590 | if (IS_ERR(btp->bt_task)) |
1591 | return PTR_ERR(btp->bt_task); | |
1592 | return 0; | |
a6867a68 DC |
1593 | } |
1594 | ||
1da177e4 LT |
1595 | xfs_buftarg_t * |
1596 | xfs_alloc_buftarg( | |
ebad861b | 1597 | struct xfs_mount *mp, |
1da177e4 | 1598 | struct block_device *bdev, |
e2a07812 JE |
1599 | int external, |
1600 | const char *fsname) | |
1da177e4 LT |
1601 | { |
1602 | xfs_buftarg_t *btp; | |
1603 | ||
1604 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1605 | ||
ebad861b | 1606 | btp->bt_mount = mp; |
ce8e922c NS |
1607 | btp->bt_dev = bdev->bd_dev; |
1608 | btp->bt_bdev = bdev; | |
1da177e4 LT |
1609 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1610 | goto error; | |
1611 | if (xfs_mapping_buftarg(btp, bdev)) | |
1612 | goto error; | |
e2a07812 | 1613 | if (xfs_alloc_delwrite_queue(btp, fsname)) |
a6867a68 | 1614 | goto error; |
ff57ab21 DC |
1615 | btp->bt_shrinker.shrink = xfs_buftarg_shrink; |
1616 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; | |
1617 | register_shrinker(&btp->bt_shrinker); | |
1da177e4 LT |
1618 | return btp; |
1619 | ||
1620 | error: | |
f0e2d93c | 1621 | kmem_free(btp); |
1da177e4 LT |
1622 | return NULL; |
1623 | } | |
1624 | ||
1625 | ||
1626 | /* | |
ce8e922c | 1627 | * Delayed write buffer handling |
1da177e4 | 1628 | */ |
1da177e4 | 1629 | STATIC void |
ce8e922c NS |
1630 | xfs_buf_delwri_queue( |
1631 | xfs_buf_t *bp, | |
1da177e4 LT |
1632 | int unlock) |
1633 | { | |
ce8e922c NS |
1634 | struct list_head *dwq = &bp->b_target->bt_delwrite_queue; |
1635 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; | |
a6867a68 | 1636 | |
0b1b213f CH |
1637 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
1638 | ||
ce8e922c | 1639 | ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC)); |
1da177e4 | 1640 | |
a6867a68 | 1641 | spin_lock(dwlk); |
1da177e4 | 1642 | /* If already in the queue, dequeue and place at tail */ |
ce8e922c NS |
1643 | if (!list_empty(&bp->b_list)) { |
1644 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1645 | if (unlock) | |
1646 | atomic_dec(&bp->b_hold); | |
1647 | list_del(&bp->b_list); | |
1da177e4 LT |
1648 | } |
1649 | ||
c9c12971 DC |
1650 | if (list_empty(dwq)) { |
1651 | /* start xfsbufd as it is about to have something to do */ | |
1652 | wake_up_process(bp->b_target->bt_task); | |
1653 | } | |
1654 | ||
ce8e922c NS |
1655 | bp->b_flags |= _XBF_DELWRI_Q; |
1656 | list_add_tail(&bp->b_list, dwq); | |
1657 | bp->b_queuetime = jiffies; | |
a6867a68 | 1658 | spin_unlock(dwlk); |
1da177e4 LT |
1659 | |
1660 | if (unlock) | |
ce8e922c | 1661 | xfs_buf_unlock(bp); |
1da177e4 LT |
1662 | } |
1663 | ||
1664 | void | |
ce8e922c NS |
1665 | xfs_buf_delwri_dequeue( |
1666 | xfs_buf_t *bp) | |
1da177e4 | 1667 | { |
ce8e922c | 1668 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; |
1da177e4 LT |
1669 | int dequeued = 0; |
1670 | ||
a6867a68 | 1671 | spin_lock(dwlk); |
ce8e922c NS |
1672 | if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) { |
1673 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1674 | list_del_init(&bp->b_list); | |
1da177e4 LT |
1675 | dequeued = 1; |
1676 | } | |
ce8e922c | 1677 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q); |
a6867a68 | 1678 | spin_unlock(dwlk); |
1da177e4 LT |
1679 | |
1680 | if (dequeued) | |
ce8e922c | 1681 | xfs_buf_rele(bp); |
1da177e4 | 1682 | |
0b1b213f | 1683 | trace_xfs_buf_delwri_dequeue(bp, _RET_IP_); |
1da177e4 LT |
1684 | } |
1685 | ||
d808f617 DC |
1686 | /* |
1687 | * If a delwri buffer needs to be pushed before it has aged out, then promote | |
1688 | * it to the head of the delwri queue so that it will be flushed on the next | |
1689 | * xfsbufd run. We do this by resetting the queuetime of the buffer to be older | |
1690 | * than the age currently needed to flush the buffer. Hence the next time the | |
1691 | * xfsbufd sees it is guaranteed to be considered old enough to flush. | |
1692 | */ | |
1693 | void | |
1694 | xfs_buf_delwri_promote( | |
1695 | struct xfs_buf *bp) | |
1696 | { | |
1697 | struct xfs_buftarg *btp = bp->b_target; | |
1698 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1; | |
1699 | ||
1700 | ASSERT(bp->b_flags & XBF_DELWRI); | |
1701 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1702 | ||
1703 | /* | |
1704 | * Check the buffer age before locking the delayed write queue as we | |
1705 | * don't need to promote buffers that are already past the flush age. | |
1706 | */ | |
1707 | if (bp->b_queuetime < jiffies - age) | |
1708 | return; | |
1709 | bp->b_queuetime = jiffies - age; | |
1710 | spin_lock(&btp->bt_delwrite_lock); | |
1711 | list_move(&bp->b_list, &btp->bt_delwrite_queue); | |
1712 | spin_unlock(&btp->bt_delwrite_lock); | |
1713 | } | |
1714 | ||
1da177e4 | 1715 | STATIC void |
ce8e922c | 1716 | xfs_buf_runall_queues( |
1da177e4 LT |
1717 | struct workqueue_struct *queue) |
1718 | { | |
1719 | flush_workqueue(queue); | |
1720 | } | |
1721 | ||
585e6d88 DC |
1722 | /* |
1723 | * Move as many buffers as specified to the supplied list | |
1724 | * idicating if we skipped any buffers to prevent deadlocks. | |
1725 | */ | |
1726 | STATIC int | |
1727 | xfs_buf_delwri_split( | |
1728 | xfs_buftarg_t *target, | |
1729 | struct list_head *list, | |
5e6a07df | 1730 | unsigned long age) |
585e6d88 DC |
1731 | { |
1732 | xfs_buf_t *bp, *n; | |
1733 | struct list_head *dwq = &target->bt_delwrite_queue; | |
1734 | spinlock_t *dwlk = &target->bt_delwrite_lock; | |
1735 | int skipped = 0; | |
5e6a07df | 1736 | int force; |
585e6d88 | 1737 | |
5e6a07df | 1738 | force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
585e6d88 DC |
1739 | INIT_LIST_HEAD(list); |
1740 | spin_lock(dwlk); | |
1741 | list_for_each_entry_safe(bp, n, dwq, b_list) { | |
585e6d88 DC |
1742 | ASSERT(bp->b_flags & XBF_DELWRI); |
1743 | ||
4d16e924 | 1744 | if (!XFS_BUF_ISPINNED(bp) && !xfs_buf_cond_lock(bp)) { |
5e6a07df | 1745 | if (!force && |
585e6d88 DC |
1746 | time_before(jiffies, bp->b_queuetime + age)) { |
1747 | xfs_buf_unlock(bp); | |
1748 | break; | |
1749 | } | |
1750 | ||
1751 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q| | |
1752 | _XBF_RUN_QUEUES); | |
1753 | bp->b_flags |= XBF_WRITE; | |
1754 | list_move_tail(&bp->b_list, list); | |
bfe27419 | 1755 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
585e6d88 DC |
1756 | } else |
1757 | skipped++; | |
1758 | } | |
1759 | spin_unlock(dwlk); | |
1760 | ||
1761 | return skipped; | |
1762 | ||
1763 | } | |
1764 | ||
089716aa DC |
1765 | /* |
1766 | * Compare function is more complex than it needs to be because | |
1767 | * the return value is only 32 bits and we are doing comparisons | |
1768 | * on 64 bit values | |
1769 | */ | |
1770 | static int | |
1771 | xfs_buf_cmp( | |
1772 | void *priv, | |
1773 | struct list_head *a, | |
1774 | struct list_head *b) | |
1775 | { | |
1776 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1777 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1778 | xfs_daddr_t diff; | |
1779 | ||
1780 | diff = ap->b_bn - bp->b_bn; | |
1781 | if (diff < 0) | |
1782 | return -1; | |
1783 | if (diff > 0) | |
1784 | return 1; | |
1785 | return 0; | |
1786 | } | |
1787 | ||
1788 | void | |
1789 | xfs_buf_delwri_sort( | |
1790 | xfs_buftarg_t *target, | |
1791 | struct list_head *list) | |
1792 | { | |
1793 | list_sort(NULL, list, xfs_buf_cmp); | |
1794 | } | |
1795 | ||
1da177e4 | 1796 | STATIC int |
23ea4032 | 1797 | xfsbufd( |
585e6d88 | 1798 | void *data) |
1da177e4 | 1799 | { |
089716aa | 1800 | xfs_buftarg_t *target = (xfs_buftarg_t *)data; |
1da177e4 | 1801 | |
1da177e4 LT |
1802 | current->flags |= PF_MEMALLOC; |
1803 | ||
978c7b2f RW |
1804 | set_freezable(); |
1805 | ||
1da177e4 | 1806 | do { |
c9c12971 DC |
1807 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10); |
1808 | long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10); | |
089716aa DC |
1809 | int count = 0; |
1810 | struct list_head tmp; | |
c9c12971 | 1811 | |
3e1d1d28 | 1812 | if (unlikely(freezing(current))) { |
ce8e922c | 1813 | set_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
3e1d1d28 | 1814 | refrigerator(); |
abd0cf7a | 1815 | } else { |
ce8e922c | 1816 | clear_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
abd0cf7a | 1817 | } |
1da177e4 | 1818 | |
c9c12971 DC |
1819 | /* sleep for a long time if there is nothing to do. */ |
1820 | if (list_empty(&target->bt_delwrite_queue)) | |
1821 | tout = MAX_SCHEDULE_TIMEOUT; | |
1822 | schedule_timeout_interruptible(tout); | |
1da177e4 | 1823 | |
c9c12971 | 1824 | xfs_buf_delwri_split(target, &tmp, age); |
089716aa | 1825 | list_sort(NULL, &tmp, xfs_buf_cmp); |
1da177e4 | 1826 | while (!list_empty(&tmp)) { |
089716aa DC |
1827 | struct xfs_buf *bp; |
1828 | bp = list_first_entry(&tmp, struct xfs_buf, b_list); | |
ce8e922c | 1829 | list_del_init(&bp->b_list); |
939d723b | 1830 | xfs_bdstrat_cb(bp); |
585e6d88 | 1831 | count++; |
1da177e4 | 1832 | } |
f07c2250 NS |
1833 | if (count) |
1834 | blk_run_address_space(target->bt_mapping); | |
1da177e4 | 1835 | |
4df08c52 | 1836 | } while (!kthread_should_stop()); |
1da177e4 | 1837 | |
4df08c52 | 1838 | return 0; |
1da177e4 LT |
1839 | } |
1840 | ||
1841 | /* | |
ce8e922c NS |
1842 | * Go through all incore buffers, and release buffers if they belong to |
1843 | * the given device. This is used in filesystem error handling to | |
1844 | * preserve the consistency of its metadata. | |
1da177e4 LT |
1845 | */ |
1846 | int | |
1847 | xfs_flush_buftarg( | |
585e6d88 DC |
1848 | xfs_buftarg_t *target, |
1849 | int wait) | |
1da177e4 | 1850 | { |
089716aa | 1851 | xfs_buf_t *bp; |
585e6d88 | 1852 | int pincount = 0; |
089716aa DC |
1853 | LIST_HEAD(tmp_list); |
1854 | LIST_HEAD(wait_list); | |
1da177e4 | 1855 | |
c626d174 | 1856 | xfs_buf_runall_queues(xfsconvertd_workqueue); |
ce8e922c NS |
1857 | xfs_buf_runall_queues(xfsdatad_workqueue); |
1858 | xfs_buf_runall_queues(xfslogd_workqueue); | |
1da177e4 | 1859 | |
5e6a07df | 1860 | set_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
089716aa | 1861 | pincount = xfs_buf_delwri_split(target, &tmp_list, 0); |
1da177e4 LT |
1862 | |
1863 | /* | |
089716aa DC |
1864 | * Dropped the delayed write list lock, now walk the temporary list. |
1865 | * All I/O is issued async and then if we need to wait for completion | |
1866 | * we do that after issuing all the IO. | |
1da177e4 | 1867 | */ |
089716aa DC |
1868 | list_sort(NULL, &tmp_list, xfs_buf_cmp); |
1869 | while (!list_empty(&tmp_list)) { | |
1870 | bp = list_first_entry(&tmp_list, struct xfs_buf, b_list); | |
585e6d88 | 1871 | ASSERT(target == bp->b_target); |
089716aa DC |
1872 | list_del_init(&bp->b_list); |
1873 | if (wait) { | |
ce8e922c | 1874 | bp->b_flags &= ~XBF_ASYNC; |
089716aa DC |
1875 | list_add(&bp->b_list, &wait_list); |
1876 | } | |
939d723b | 1877 | xfs_bdstrat_cb(bp); |
1da177e4 LT |
1878 | } |
1879 | ||
089716aa DC |
1880 | if (wait) { |
1881 | /* Expedite and wait for IO to complete. */ | |
f07c2250 | 1882 | blk_run_address_space(target->bt_mapping); |
089716aa DC |
1883 | while (!list_empty(&wait_list)) { |
1884 | bp = list_first_entry(&wait_list, struct xfs_buf, b_list); | |
f07c2250 | 1885 | |
089716aa | 1886 | list_del_init(&bp->b_list); |
1a1a3e97 | 1887 | xfs_buf_iowait(bp); |
089716aa DC |
1888 | xfs_buf_relse(bp); |
1889 | } | |
1da177e4 LT |
1890 | } |
1891 | ||
1da177e4 LT |
1892 | return pincount; |
1893 | } | |
1894 | ||
04d8b284 | 1895 | int __init |
ce8e922c | 1896 | xfs_buf_init(void) |
1da177e4 | 1897 | { |
8758280f NS |
1898 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1899 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1900 | if (!xfs_buf_zone) |
0b1b213f | 1901 | goto out; |
04d8b284 | 1902 | |
51749e47 | 1903 | xfslogd_workqueue = alloc_workqueue("xfslogd", |
6370a6ad | 1904 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 1); |
23ea4032 | 1905 | if (!xfslogd_workqueue) |
04d8b284 | 1906 | goto out_free_buf_zone; |
1da177e4 | 1907 | |
b4337692 | 1908 | xfsdatad_workqueue = create_workqueue("xfsdatad"); |
23ea4032 CH |
1909 | if (!xfsdatad_workqueue) |
1910 | goto out_destroy_xfslogd_workqueue; | |
1da177e4 | 1911 | |
c626d174 DC |
1912 | xfsconvertd_workqueue = create_workqueue("xfsconvertd"); |
1913 | if (!xfsconvertd_workqueue) | |
1914 | goto out_destroy_xfsdatad_workqueue; | |
1915 | ||
23ea4032 | 1916 | return 0; |
1da177e4 | 1917 | |
c626d174 DC |
1918 | out_destroy_xfsdatad_workqueue: |
1919 | destroy_workqueue(xfsdatad_workqueue); | |
23ea4032 CH |
1920 | out_destroy_xfslogd_workqueue: |
1921 | destroy_workqueue(xfslogd_workqueue); | |
23ea4032 | 1922 | out_free_buf_zone: |
ce8e922c | 1923 | kmem_zone_destroy(xfs_buf_zone); |
0b1b213f | 1924 | out: |
8758280f | 1925 | return -ENOMEM; |
1da177e4 LT |
1926 | } |
1927 | ||
1da177e4 | 1928 | void |
ce8e922c | 1929 | xfs_buf_terminate(void) |
1da177e4 | 1930 | { |
c626d174 | 1931 | destroy_workqueue(xfsconvertd_workqueue); |
04d8b284 CH |
1932 | destroy_workqueue(xfsdatad_workqueue); |
1933 | destroy_workqueue(xfslogd_workqueue); | |
ce8e922c | 1934 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 1935 | } |
e6a0e9cd TS |
1936 | |
1937 | #ifdef CONFIG_KDB_MODULES | |
1938 | struct list_head * | |
1939 | xfs_get_buftarg_list(void) | |
1940 | { | |
1941 | return &xfs_buftarg_list; | |
1942 | } | |
1943 | #endif |