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> | |
21 | #include <linux/slab.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/vmalloc.h> | |
25 | #include <linux/bio.h> | |
26 | #include <linux/sysctl.h> | |
27 | #include <linux/proc_fs.h> | |
28 | #include <linux/workqueue.h> | |
29 | #include <linux/percpu.h> | |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/hash.h> | |
4df08c52 | 32 | #include <linux/kthread.h> |
b20a3503 | 33 | #include <linux/migrate.h> |
3fcfab16 | 34 | #include <linux/backing-dev.h> |
7dfb7103 | 35 | #include <linux/freezer.h> |
1da177e4 | 36 | |
7989cb8e | 37 | static kmem_zone_t *xfs_buf_zone; |
a6867a68 | 38 | STATIC int xfsbufd(void *); |
27496a8c | 39 | STATIC int xfsbufd_wakeup(int, gfp_t); |
ce8e922c | 40 | STATIC void xfs_buf_delwri_queue(xfs_buf_t *, int); |
8e1f936b RR |
41 | static struct shrinker xfs_buf_shake = { |
42 | .shrink = xfsbufd_wakeup, | |
43 | .seeks = DEFAULT_SEEKS, | |
44 | }; | |
23ea4032 | 45 | |
7989cb8e | 46 | static struct workqueue_struct *xfslogd_workqueue; |
0829c360 | 47 | struct workqueue_struct *xfsdatad_workqueue; |
1da177e4 | 48 | |
ce8e922c | 49 | #ifdef XFS_BUF_TRACE |
1da177e4 | 50 | void |
ce8e922c NS |
51 | xfs_buf_trace( |
52 | xfs_buf_t *bp, | |
1da177e4 LT |
53 | char *id, |
54 | void *data, | |
55 | void *ra) | |
56 | { | |
ce8e922c NS |
57 | ktrace_enter(xfs_buf_trace_buf, |
58 | bp, id, | |
59 | (void *)(unsigned long)bp->b_flags, | |
60 | (void *)(unsigned long)bp->b_hold.counter, | |
d63f154a | 61 | (void *)(unsigned long)bp->b_sema.count, |
1da177e4 LT |
62 | (void *)current, |
63 | data, ra, | |
ce8e922c NS |
64 | (void *)(unsigned long)((bp->b_file_offset>>32) & 0xffffffff), |
65 | (void *)(unsigned long)(bp->b_file_offset & 0xffffffff), | |
66 | (void *)(unsigned long)bp->b_buffer_length, | |
1da177e4 LT |
67 | NULL, NULL, NULL, NULL, NULL); |
68 | } | |
ce8e922c NS |
69 | ktrace_t *xfs_buf_trace_buf; |
70 | #define XFS_BUF_TRACE_SIZE 4096 | |
71 | #define XB_TRACE(bp, id, data) \ | |
72 | xfs_buf_trace(bp, id, (void *)data, (void *)__builtin_return_address(0)) | |
1da177e4 | 73 | #else |
ce8e922c | 74 | #define XB_TRACE(bp, id, data) do { } while (0) |
1da177e4 LT |
75 | #endif |
76 | ||
ce8e922c NS |
77 | #ifdef XFS_BUF_LOCK_TRACKING |
78 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
79 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
80 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 81 | #else |
ce8e922c NS |
82 | # define XB_SET_OWNER(bp) do { } while (0) |
83 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
84 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
85 | #endif |
86 | ||
ce8e922c NS |
87 | #define xb_to_gfp(flags) \ |
88 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \ | |
89 | ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN) | |
1da177e4 | 90 | |
ce8e922c NS |
91 | #define xb_to_km(flags) \ |
92 | (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP) | |
1da177e4 | 93 | |
ce8e922c NS |
94 | #define xfs_buf_allocate(flags) \ |
95 | kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags)) | |
96 | #define xfs_buf_deallocate(bp) \ | |
97 | kmem_zone_free(xfs_buf_zone, (bp)); | |
1da177e4 LT |
98 | |
99 | /* | |
ce8e922c | 100 | * Page Region interfaces. |
1da177e4 | 101 | * |
ce8e922c NS |
102 | * For pages in filesystems where the blocksize is smaller than the |
103 | * pagesize, we use the page->private field (long) to hold a bitmap | |
104 | * of uptodate regions within the page. | |
1da177e4 | 105 | * |
ce8e922c | 106 | * Each such region is "bytes per page / bits per long" bytes long. |
1da177e4 | 107 | * |
ce8e922c NS |
108 | * NBPPR == number-of-bytes-per-page-region |
109 | * BTOPR == bytes-to-page-region (rounded up) | |
110 | * BTOPRT == bytes-to-page-region-truncated (rounded down) | |
1da177e4 LT |
111 | */ |
112 | #if (BITS_PER_LONG == 32) | |
113 | #define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */ | |
114 | #elif (BITS_PER_LONG == 64) | |
115 | #define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */ | |
116 | #else | |
117 | #error BITS_PER_LONG must be 32 or 64 | |
118 | #endif | |
119 | #define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG) | |
120 | #define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT) | |
121 | #define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT)) | |
122 | ||
123 | STATIC unsigned long | |
124 | page_region_mask( | |
125 | size_t offset, | |
126 | size_t length) | |
127 | { | |
128 | unsigned long mask; | |
129 | int first, final; | |
130 | ||
131 | first = BTOPR(offset); | |
132 | final = BTOPRT(offset + length - 1); | |
133 | first = min(first, final); | |
134 | ||
135 | mask = ~0UL; | |
136 | mask <<= BITS_PER_LONG - (final - first); | |
137 | mask >>= BITS_PER_LONG - (final); | |
138 | ||
139 | ASSERT(offset + length <= PAGE_CACHE_SIZE); | |
140 | ASSERT((final - first) < BITS_PER_LONG && (final - first) >= 0); | |
141 | ||
142 | return mask; | |
143 | } | |
144 | ||
7989cb8e | 145 | STATIC_INLINE void |
1da177e4 LT |
146 | set_page_region( |
147 | struct page *page, | |
148 | size_t offset, | |
149 | size_t length) | |
150 | { | |
4c21e2f2 HD |
151 | set_page_private(page, |
152 | page_private(page) | page_region_mask(offset, length)); | |
153 | if (page_private(page) == ~0UL) | |
1da177e4 LT |
154 | SetPageUptodate(page); |
155 | } | |
156 | ||
7989cb8e | 157 | STATIC_INLINE int |
1da177e4 LT |
158 | test_page_region( |
159 | struct page *page, | |
160 | size_t offset, | |
161 | size_t length) | |
162 | { | |
163 | unsigned long mask = page_region_mask(offset, length); | |
164 | ||
4c21e2f2 | 165 | return (mask && (page_private(page) & mask) == mask); |
1da177e4 LT |
166 | } |
167 | ||
1da177e4 | 168 | /* |
ce8e922c | 169 | * Internal xfs_buf_t object manipulation |
1da177e4 LT |
170 | */ |
171 | ||
172 | STATIC void | |
ce8e922c NS |
173 | _xfs_buf_initialize( |
174 | xfs_buf_t *bp, | |
1da177e4 | 175 | xfs_buftarg_t *target, |
204ab25f | 176 | xfs_off_t range_base, |
1da177e4 | 177 | size_t range_length, |
ce8e922c | 178 | xfs_buf_flags_t flags) |
1da177e4 LT |
179 | { |
180 | /* | |
ce8e922c | 181 | * We don't want certain flags to appear in b_flags. |
1da177e4 | 182 | */ |
ce8e922c NS |
183 | flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD); |
184 | ||
185 | memset(bp, 0, sizeof(xfs_buf_t)); | |
186 | atomic_set(&bp->b_hold, 1); | |
b4dd330b | 187 | init_completion(&bp->b_iowait); |
ce8e922c NS |
188 | INIT_LIST_HEAD(&bp->b_list); |
189 | INIT_LIST_HEAD(&bp->b_hash_list); | |
190 | init_MUTEX_LOCKED(&bp->b_sema); /* held, no waiters */ | |
191 | XB_SET_OWNER(bp); | |
192 | bp->b_target = target; | |
193 | bp->b_file_offset = range_base; | |
1da177e4 LT |
194 | /* |
195 | * Set buffer_length and count_desired to the same value initially. | |
196 | * I/O routines should use count_desired, which will be the same in | |
197 | * most cases but may be reset (e.g. XFS recovery). | |
198 | */ | |
ce8e922c NS |
199 | bp->b_buffer_length = bp->b_count_desired = range_length; |
200 | bp->b_flags = flags; | |
201 | bp->b_bn = XFS_BUF_DADDR_NULL; | |
202 | atomic_set(&bp->b_pin_count, 0); | |
203 | init_waitqueue_head(&bp->b_waiters); | |
204 | ||
205 | XFS_STATS_INC(xb_create); | |
206 | XB_TRACE(bp, "initialize", target); | |
1da177e4 LT |
207 | } |
208 | ||
209 | /* | |
ce8e922c NS |
210 | * Allocate a page array capable of holding a specified number |
211 | * of pages, and point the page buf at it. | |
1da177e4 LT |
212 | */ |
213 | STATIC int | |
ce8e922c NS |
214 | _xfs_buf_get_pages( |
215 | xfs_buf_t *bp, | |
1da177e4 | 216 | int page_count, |
ce8e922c | 217 | xfs_buf_flags_t flags) |
1da177e4 LT |
218 | { |
219 | /* Make sure that we have a page list */ | |
ce8e922c NS |
220 | if (bp->b_pages == NULL) { |
221 | bp->b_offset = xfs_buf_poff(bp->b_file_offset); | |
222 | bp->b_page_count = page_count; | |
223 | if (page_count <= XB_PAGES) { | |
224 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 225 | } else { |
ce8e922c NS |
226 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
227 | page_count, xb_to_km(flags)); | |
228 | if (bp->b_pages == NULL) | |
1da177e4 LT |
229 | return -ENOMEM; |
230 | } | |
ce8e922c | 231 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
232 | } |
233 | return 0; | |
234 | } | |
235 | ||
236 | /* | |
ce8e922c | 237 | * Frees b_pages if it was allocated. |
1da177e4 LT |
238 | */ |
239 | STATIC void | |
ce8e922c | 240 | _xfs_buf_free_pages( |
1da177e4 LT |
241 | xfs_buf_t *bp) |
242 | { | |
ce8e922c | 243 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 244 | kmem_free(bp->b_pages); |
1da177e4 LT |
245 | } |
246 | } | |
247 | ||
248 | /* | |
249 | * Releases the specified buffer. | |
250 | * | |
251 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 252 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
253 | * hashed and refcounted buffers |
254 | */ | |
255 | void | |
ce8e922c | 256 | xfs_buf_free( |
1da177e4 LT |
257 | xfs_buf_t *bp) |
258 | { | |
ce8e922c | 259 | XB_TRACE(bp, "free", 0); |
1da177e4 | 260 | |
ce8e922c | 261 | ASSERT(list_empty(&bp->b_hash_list)); |
1da177e4 | 262 | |
1fa40b01 | 263 | if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) { |
1da177e4 LT |
264 | uint i; |
265 | ||
ce8e922c | 266 | if ((bp->b_flags & XBF_MAPPED) && (bp->b_page_count > 1)) |
0087167c | 267 | vm_unmap_ram(bp->b_addr - bp->b_offset, bp->b_page_count); |
1da177e4 | 268 | |
948ecdb4 NS |
269 | for (i = 0; i < bp->b_page_count; i++) { |
270 | struct page *page = bp->b_pages[i]; | |
271 | ||
1fa40b01 CH |
272 | if (bp->b_flags & _XBF_PAGE_CACHE) |
273 | ASSERT(!PagePrivate(page)); | |
948ecdb4 NS |
274 | page_cache_release(page); |
275 | } | |
ce8e922c | 276 | _xfs_buf_free_pages(bp); |
1da177e4 LT |
277 | } |
278 | ||
ce8e922c | 279 | xfs_buf_deallocate(bp); |
1da177e4 LT |
280 | } |
281 | ||
282 | /* | |
283 | * Finds all pages for buffer in question and builds it's page list. | |
284 | */ | |
285 | STATIC int | |
ce8e922c | 286 | _xfs_buf_lookup_pages( |
1da177e4 LT |
287 | xfs_buf_t *bp, |
288 | uint flags) | |
289 | { | |
ce8e922c NS |
290 | struct address_space *mapping = bp->b_target->bt_mapping; |
291 | size_t blocksize = bp->b_target->bt_bsize; | |
292 | size_t size = bp->b_count_desired; | |
1da177e4 | 293 | size_t nbytes, offset; |
ce8e922c | 294 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 LT |
295 | unsigned short page_count, i; |
296 | pgoff_t first; | |
204ab25f | 297 | xfs_off_t end; |
1da177e4 LT |
298 | int error; |
299 | ||
ce8e922c NS |
300 | end = bp->b_file_offset + bp->b_buffer_length; |
301 | page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset); | |
1da177e4 | 302 | |
ce8e922c | 303 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
304 | if (unlikely(error)) |
305 | return error; | |
ce8e922c | 306 | bp->b_flags |= _XBF_PAGE_CACHE; |
1da177e4 | 307 | |
ce8e922c NS |
308 | offset = bp->b_offset; |
309 | first = bp->b_file_offset >> PAGE_CACHE_SHIFT; | |
1da177e4 | 310 | |
ce8e922c | 311 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
312 | struct page *page; |
313 | uint retries = 0; | |
314 | ||
315 | retry: | |
316 | page = find_or_create_page(mapping, first + i, gfp_mask); | |
317 | if (unlikely(page == NULL)) { | |
ce8e922c NS |
318 | if (flags & XBF_READ_AHEAD) { |
319 | bp->b_page_count = i; | |
6ab455ee CH |
320 | for (i = 0; i < bp->b_page_count; i++) |
321 | unlock_page(bp->b_pages[i]); | |
1da177e4 LT |
322 | return -ENOMEM; |
323 | } | |
324 | ||
325 | /* | |
326 | * This could deadlock. | |
327 | * | |
328 | * But until all the XFS lowlevel code is revamped to | |
329 | * handle buffer allocation failures we can't do much. | |
330 | */ | |
331 | if (!(++retries % 100)) | |
332 | printk(KERN_ERR | |
333 | "XFS: possible memory allocation " | |
334 | "deadlock in %s (mode:0x%x)\n", | |
34a622b2 | 335 | __func__, gfp_mask); |
1da177e4 | 336 | |
ce8e922c | 337 | XFS_STATS_INC(xb_page_retries); |
23ea4032 | 338 | xfsbufd_wakeup(0, gfp_mask); |
3fcfab16 | 339 | congestion_wait(WRITE, HZ/50); |
1da177e4 LT |
340 | goto retry; |
341 | } | |
342 | ||
ce8e922c | 343 | XFS_STATS_INC(xb_page_found); |
1da177e4 LT |
344 | |
345 | nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset); | |
346 | size -= nbytes; | |
347 | ||
948ecdb4 | 348 | ASSERT(!PagePrivate(page)); |
1da177e4 LT |
349 | if (!PageUptodate(page)) { |
350 | page_count--; | |
6ab455ee CH |
351 | if (blocksize >= PAGE_CACHE_SIZE) { |
352 | if (flags & XBF_READ) | |
353 | bp->b_flags |= _XBF_PAGE_LOCKED; | |
354 | } else if (!PagePrivate(page)) { | |
1da177e4 LT |
355 | if (test_page_region(page, offset, nbytes)) |
356 | page_count++; | |
357 | } | |
358 | } | |
359 | ||
ce8e922c | 360 | bp->b_pages[i] = page; |
1da177e4 LT |
361 | offset = 0; |
362 | } | |
363 | ||
6ab455ee CH |
364 | if (!(bp->b_flags & _XBF_PAGE_LOCKED)) { |
365 | for (i = 0; i < bp->b_page_count; i++) | |
366 | unlock_page(bp->b_pages[i]); | |
367 | } | |
368 | ||
ce8e922c NS |
369 | if (page_count == bp->b_page_count) |
370 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 371 | |
ce8e922c | 372 | XB_TRACE(bp, "lookup_pages", (long)page_count); |
1da177e4 LT |
373 | return error; |
374 | } | |
375 | ||
376 | /* | |
377 | * Map buffer into kernel address-space if nessecary. | |
378 | */ | |
379 | STATIC int | |
ce8e922c | 380 | _xfs_buf_map_pages( |
1da177e4 LT |
381 | xfs_buf_t *bp, |
382 | uint flags) | |
383 | { | |
384 | /* A single page buffer is always mappable */ | |
ce8e922c NS |
385 | if (bp->b_page_count == 1) { |
386 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; | |
387 | bp->b_flags |= XBF_MAPPED; | |
388 | } else if (flags & XBF_MAPPED) { | |
0087167c NP |
389 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, |
390 | -1, PAGE_KERNEL); | |
ce8e922c | 391 | if (unlikely(bp->b_addr == NULL)) |
1da177e4 | 392 | return -ENOMEM; |
ce8e922c NS |
393 | bp->b_addr += bp->b_offset; |
394 | bp->b_flags |= XBF_MAPPED; | |
1da177e4 LT |
395 | } |
396 | ||
397 | return 0; | |
398 | } | |
399 | ||
400 | /* | |
401 | * Finding and Reading Buffers | |
402 | */ | |
403 | ||
404 | /* | |
ce8e922c | 405 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 LT |
406 | * a given range of an inode. The buffer is returned |
407 | * locked. If other overlapping buffers exist, they are | |
408 | * released before the new buffer is created and locked, | |
409 | * which may imply that this call will block until those buffers | |
410 | * are unlocked. No I/O is implied by this call. | |
411 | */ | |
412 | xfs_buf_t * | |
ce8e922c | 413 | _xfs_buf_find( |
1da177e4 | 414 | xfs_buftarg_t *btp, /* block device target */ |
204ab25f | 415 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 416 | size_t isize, /* length of range */ |
ce8e922c NS |
417 | xfs_buf_flags_t flags, |
418 | xfs_buf_t *new_bp) | |
1da177e4 | 419 | { |
204ab25f | 420 | xfs_off_t range_base; |
1da177e4 LT |
421 | size_t range_length; |
422 | xfs_bufhash_t *hash; | |
ce8e922c | 423 | xfs_buf_t *bp, *n; |
1da177e4 LT |
424 | |
425 | range_base = (ioff << BBSHIFT); | |
426 | range_length = (isize << BBSHIFT); | |
427 | ||
428 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
ce8e922c | 429 | ASSERT(!(range_length < (1 << btp->bt_sshift))); |
204ab25f | 430 | ASSERT(!(range_base & (xfs_off_t)btp->bt_smask)); |
1da177e4 LT |
431 | |
432 | hash = &btp->bt_hash[hash_long((unsigned long)ioff, btp->bt_hashshift)]; | |
433 | ||
434 | spin_lock(&hash->bh_lock); | |
435 | ||
ce8e922c NS |
436 | list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) { |
437 | ASSERT(btp == bp->b_target); | |
438 | if (bp->b_file_offset == range_base && | |
439 | bp->b_buffer_length == range_length) { | |
1da177e4 | 440 | /* |
ce8e922c | 441 | * If we look at something, bring it to the |
1da177e4 LT |
442 | * front of the list for next time. |
443 | */ | |
ce8e922c NS |
444 | atomic_inc(&bp->b_hold); |
445 | list_move(&bp->b_hash_list, &hash->bh_list); | |
1da177e4 LT |
446 | goto found; |
447 | } | |
448 | } | |
449 | ||
450 | /* No match found */ | |
ce8e922c NS |
451 | if (new_bp) { |
452 | _xfs_buf_initialize(new_bp, btp, range_base, | |
1da177e4 | 453 | range_length, flags); |
ce8e922c NS |
454 | new_bp->b_hash = hash; |
455 | list_add(&new_bp->b_hash_list, &hash->bh_list); | |
1da177e4 | 456 | } else { |
ce8e922c | 457 | XFS_STATS_INC(xb_miss_locked); |
1da177e4 LT |
458 | } |
459 | ||
460 | spin_unlock(&hash->bh_lock); | |
ce8e922c | 461 | return new_bp; |
1da177e4 LT |
462 | |
463 | found: | |
464 | spin_unlock(&hash->bh_lock); | |
465 | ||
466 | /* Attempt to get the semaphore without sleeping, | |
467 | * if this does not work then we need to drop the | |
468 | * spinlock and do a hard attempt on the semaphore. | |
469 | */ | |
ce8e922c NS |
470 | if (down_trylock(&bp->b_sema)) { |
471 | if (!(flags & XBF_TRYLOCK)) { | |
1da177e4 | 472 | /* wait for buffer ownership */ |
ce8e922c NS |
473 | XB_TRACE(bp, "get_lock", 0); |
474 | xfs_buf_lock(bp); | |
475 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
476 | } else { |
477 | /* We asked for a trylock and failed, no need | |
478 | * to look at file offset and length here, we | |
ce8e922c NS |
479 | * know that this buffer at least overlaps our |
480 | * buffer and is locked, therefore our buffer | |
481 | * either does not exist, or is this buffer. | |
1da177e4 | 482 | */ |
ce8e922c NS |
483 | xfs_buf_rele(bp); |
484 | XFS_STATS_INC(xb_busy_locked); | |
485 | return NULL; | |
1da177e4 LT |
486 | } |
487 | } else { | |
488 | /* trylock worked */ | |
ce8e922c | 489 | XB_SET_OWNER(bp); |
1da177e4 LT |
490 | } |
491 | ||
ce8e922c NS |
492 | if (bp->b_flags & XBF_STALE) { |
493 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
494 | bp->b_flags &= XBF_MAPPED; | |
2f926587 | 495 | } |
ce8e922c NS |
496 | XB_TRACE(bp, "got_lock", 0); |
497 | XFS_STATS_INC(xb_get_locked); | |
498 | return bp; | |
1da177e4 LT |
499 | } |
500 | ||
501 | /* | |
ce8e922c | 502 | * Assembles a buffer covering the specified range. |
1da177e4 LT |
503 | * Storage in memory for all portions of the buffer will be allocated, |
504 | * although backing storage may not be. | |
505 | */ | |
506 | xfs_buf_t * | |
ce8e922c | 507 | xfs_buf_get_flags( |
1da177e4 | 508 | xfs_buftarg_t *target,/* target for buffer */ |
204ab25f | 509 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 510 | size_t isize, /* length of range */ |
ce8e922c | 511 | xfs_buf_flags_t flags) |
1da177e4 | 512 | { |
ce8e922c | 513 | xfs_buf_t *bp, *new_bp; |
1da177e4 LT |
514 | int error = 0, i; |
515 | ||
ce8e922c NS |
516 | new_bp = xfs_buf_allocate(flags); |
517 | if (unlikely(!new_bp)) | |
1da177e4 LT |
518 | return NULL; |
519 | ||
ce8e922c NS |
520 | bp = _xfs_buf_find(target, ioff, isize, flags, new_bp); |
521 | if (bp == new_bp) { | |
522 | error = _xfs_buf_lookup_pages(bp, flags); | |
1da177e4 LT |
523 | if (error) |
524 | goto no_buffer; | |
525 | } else { | |
ce8e922c NS |
526 | xfs_buf_deallocate(new_bp); |
527 | if (unlikely(bp == NULL)) | |
1da177e4 LT |
528 | return NULL; |
529 | } | |
530 | ||
ce8e922c NS |
531 | for (i = 0; i < bp->b_page_count; i++) |
532 | mark_page_accessed(bp->b_pages[i]); | |
1da177e4 | 533 | |
ce8e922c NS |
534 | if (!(bp->b_flags & XBF_MAPPED)) { |
535 | error = _xfs_buf_map_pages(bp, flags); | |
1da177e4 LT |
536 | if (unlikely(error)) { |
537 | printk(KERN_WARNING "%s: failed to map pages\n", | |
34a622b2 | 538 | __func__); |
1da177e4 LT |
539 | goto no_buffer; |
540 | } | |
541 | } | |
542 | ||
ce8e922c | 543 | XFS_STATS_INC(xb_get); |
1da177e4 LT |
544 | |
545 | /* | |
546 | * Always fill in the block number now, the mapped cases can do | |
547 | * their own overlay of this later. | |
548 | */ | |
ce8e922c NS |
549 | bp->b_bn = ioff; |
550 | bp->b_count_desired = bp->b_buffer_length; | |
1da177e4 | 551 | |
ce8e922c NS |
552 | XB_TRACE(bp, "get", (unsigned long)flags); |
553 | return bp; | |
1da177e4 LT |
554 | |
555 | no_buffer: | |
ce8e922c NS |
556 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
557 | xfs_buf_unlock(bp); | |
558 | xfs_buf_rele(bp); | |
1da177e4 LT |
559 | return NULL; |
560 | } | |
561 | ||
5d765b97 CH |
562 | STATIC int |
563 | _xfs_buf_read( | |
564 | xfs_buf_t *bp, | |
565 | xfs_buf_flags_t flags) | |
566 | { | |
567 | int status; | |
568 | ||
569 | XB_TRACE(bp, "_xfs_buf_read", (unsigned long)flags); | |
570 | ||
571 | ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE))); | |
572 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); | |
573 | ||
574 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \ | |
575 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
576 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \ | |
577 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
578 | ||
579 | status = xfs_buf_iorequest(bp); | |
580 | if (!status && !(flags & XBF_ASYNC)) | |
581 | status = xfs_buf_iowait(bp); | |
582 | return status; | |
583 | } | |
584 | ||
1da177e4 LT |
585 | xfs_buf_t * |
586 | xfs_buf_read_flags( | |
587 | xfs_buftarg_t *target, | |
204ab25f | 588 | xfs_off_t ioff, |
1da177e4 | 589 | size_t isize, |
ce8e922c | 590 | xfs_buf_flags_t flags) |
1da177e4 | 591 | { |
ce8e922c NS |
592 | xfs_buf_t *bp; |
593 | ||
594 | flags |= XBF_READ; | |
595 | ||
596 | bp = xfs_buf_get_flags(target, ioff, isize, flags); | |
597 | if (bp) { | |
598 | if (!XFS_BUF_ISDONE(bp)) { | |
599 | XB_TRACE(bp, "read", (unsigned long)flags); | |
600 | XFS_STATS_INC(xb_get_read); | |
5d765b97 | 601 | _xfs_buf_read(bp, flags); |
ce8e922c NS |
602 | } else if (flags & XBF_ASYNC) { |
603 | XB_TRACE(bp, "read_async", (unsigned long)flags); | |
1da177e4 LT |
604 | /* |
605 | * Read ahead call which is already satisfied, | |
606 | * drop the buffer | |
607 | */ | |
608 | goto no_buffer; | |
609 | } else { | |
ce8e922c | 610 | XB_TRACE(bp, "read_done", (unsigned long)flags); |
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); |
1da177e4 LT |
643 | xfs_buf_read_flags(target, ioff, isize, flags); |
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 | |
ce8e922c | 695 | rval = _xfs_buf_get_pages(bp, page_count, 0); |
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 | |
1fa40b01 | 748 | XB_TRACE(bp, "no_daddr", len); |
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 | { |
ce8e922c NS |
770 | atomic_inc(&bp->b_hold); |
771 | XB_TRACE(bp, "hold", 0); | |
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 | |
ce8e922c | 784 | XB_TRACE(bp, "rele", bp->b_relse); |
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; |
1da177e4 | 834 | if (locked) { |
ce8e922c | 835 | XB_SET_OWNER(bp); |
1da177e4 | 836 | } |
ce8e922c NS |
837 | XB_TRACE(bp, "cond_lock", (long)locked); |
838 | return locked ? 0 : -EBUSY; | |
1da177e4 LT |
839 | } |
840 | ||
841 | #if defined(DEBUG) || defined(XFS_BLI_TRACE) | |
1da177e4 | 842 | int |
ce8e922c NS |
843 | xfs_buf_lock_value( |
844 | xfs_buf_t *bp) | |
1da177e4 | 845 | { |
adaa693b | 846 | return bp->b_sema.count; |
1da177e4 LT |
847 | } |
848 | #endif | |
849 | ||
850 | /* | |
ce8e922c NS |
851 | * Locks a buffer object. |
852 | * Note that this in no way locks the underlying pages, so it is only | |
853 | * useful for synchronizing concurrent use of buffer objects, not for | |
854 | * synchronizing independent access to the underlying pages. | |
1da177e4 | 855 | */ |
ce8e922c NS |
856 | void |
857 | xfs_buf_lock( | |
858 | xfs_buf_t *bp) | |
1da177e4 | 859 | { |
ce8e922c NS |
860 | XB_TRACE(bp, "lock", 0); |
861 | if (atomic_read(&bp->b_io_remaining)) | |
862 | blk_run_address_space(bp->b_target->bt_mapping); | |
863 | down(&bp->b_sema); | |
864 | XB_SET_OWNER(bp); | |
865 | XB_TRACE(bp, "locked", 0); | |
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); | |
887 | XB_TRACE(bp, "unlock", 0); | |
1da177e4 LT |
888 | } |
889 | ||
890 | ||
891 | /* | |
892 | * Pinning Buffer Storage in Memory | |
ce8e922c | 893 | * Ensure that no attempt to force a buffer to disk will succeed. |
1da177e4 LT |
894 | */ |
895 | void | |
ce8e922c NS |
896 | xfs_buf_pin( |
897 | xfs_buf_t *bp) | |
1da177e4 | 898 | { |
ce8e922c NS |
899 | atomic_inc(&bp->b_pin_count); |
900 | XB_TRACE(bp, "pin", (long)bp->b_pin_count.counter); | |
1da177e4 LT |
901 | } |
902 | ||
1da177e4 | 903 | void |
ce8e922c NS |
904 | xfs_buf_unpin( |
905 | xfs_buf_t *bp) | |
1da177e4 | 906 | { |
ce8e922c NS |
907 | if (atomic_dec_and_test(&bp->b_pin_count)) |
908 | wake_up_all(&bp->b_waiters); | |
909 | XB_TRACE(bp, "unpin", (long)bp->b_pin_count.counter); | |
1da177e4 LT |
910 | } |
911 | ||
912 | int | |
ce8e922c NS |
913 | xfs_buf_ispin( |
914 | xfs_buf_t *bp) | |
1da177e4 | 915 | { |
ce8e922c | 916 | return atomic_read(&bp->b_pin_count); |
1da177e4 LT |
917 | } |
918 | ||
ce8e922c NS |
919 | STATIC void |
920 | xfs_buf_wait_unpin( | |
921 | xfs_buf_t *bp) | |
1da177e4 LT |
922 | { |
923 | DECLARE_WAITQUEUE (wait, current); | |
924 | ||
ce8e922c | 925 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
926 | return; |
927 | ||
ce8e922c | 928 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
929 | for (;;) { |
930 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 931 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 932 | break; |
ce8e922c NS |
933 | if (atomic_read(&bp->b_io_remaining)) |
934 | blk_run_address_space(bp->b_target->bt_mapping); | |
1da177e4 LT |
935 | schedule(); |
936 | } | |
ce8e922c | 937 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
938 | set_current_state(TASK_RUNNING); |
939 | } | |
940 | ||
941 | /* | |
942 | * Buffer Utility Routines | |
943 | */ | |
944 | ||
1da177e4 | 945 | STATIC void |
ce8e922c | 946 | xfs_buf_iodone_work( |
c4028958 | 947 | struct work_struct *work) |
1da177e4 | 948 | { |
c4028958 DH |
949 | xfs_buf_t *bp = |
950 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 951 | |
0bfefc46 DC |
952 | /* |
953 | * We can get an EOPNOTSUPP to ordered writes. Here we clear the | |
954 | * ordered flag and reissue them. Because we can't tell the higher | |
955 | * layers directly that they should not issue ordered I/O anymore, they | |
73f6aa4d | 956 | * need to check if the _XFS_BARRIER_FAILED flag was set during I/O completion. |
0bfefc46 DC |
957 | */ |
958 | if ((bp->b_error == EOPNOTSUPP) && | |
959 | (bp->b_flags & (XBF_ORDERED|XBF_ASYNC)) == (XBF_ORDERED|XBF_ASYNC)) { | |
960 | XB_TRACE(bp, "ordered_retry", bp->b_iodone); | |
961 | bp->b_flags &= ~XBF_ORDERED; | |
73f6aa4d | 962 | bp->b_flags |= _XFS_BARRIER_FAILED; |
0bfefc46 DC |
963 | xfs_buf_iorequest(bp); |
964 | } else if (bp->b_iodone) | |
ce8e922c NS |
965 | (*(bp->b_iodone))(bp); |
966 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
967 | xfs_buf_relse(bp); |
968 | } | |
969 | ||
970 | void | |
ce8e922c NS |
971 | xfs_buf_ioend( |
972 | xfs_buf_t *bp, | |
1da177e4 LT |
973 | int schedule) |
974 | { | |
77be55a5 | 975 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
ce8e922c NS |
976 | if (bp->b_error == 0) |
977 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 978 | |
ce8e922c | 979 | XB_TRACE(bp, "iodone", bp->b_iodone); |
1da177e4 | 980 | |
ce8e922c | 981 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 982 | if (schedule) { |
c4028958 | 983 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 984 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 985 | } else { |
c4028958 | 986 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
987 | } |
988 | } else { | |
b4dd330b | 989 | complete(&bp->b_iowait); |
1da177e4 LT |
990 | } |
991 | } | |
992 | ||
1da177e4 | 993 | void |
ce8e922c NS |
994 | xfs_buf_ioerror( |
995 | xfs_buf_t *bp, | |
996 | int error) | |
1da177e4 LT |
997 | { |
998 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c NS |
999 | bp->b_error = (unsigned short)error; |
1000 | XB_TRACE(bp, "ioerror", (unsigned long)error); | |
1da177e4 LT |
1001 | } |
1002 | ||
1da177e4 | 1003 | int |
5d765b97 CH |
1004 | xfs_bawrite( |
1005 | void *mp, | |
1006 | struct xfs_buf *bp) | |
1da177e4 | 1007 | { |
5d765b97 | 1008 | XB_TRACE(bp, "bawrite", 0); |
1da177e4 | 1009 | |
5d765b97 | 1010 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); |
1da177e4 | 1011 | |
5d765b97 | 1012 | xfs_buf_delwri_dequeue(bp); |
1da177e4 | 1013 | |
5d765b97 CH |
1014 | bp->b_flags &= ~(XBF_READ | XBF_DELWRI | XBF_READ_AHEAD); |
1015 | bp->b_flags |= (XBF_WRITE | XBF_ASYNC | _XBF_RUN_QUEUES); | |
1da177e4 | 1016 | |
15ac08a8 | 1017 | bp->b_mount = mp; |
5d765b97 CH |
1018 | bp->b_strat = xfs_bdstrat_cb; |
1019 | return xfs_bdstrat_cb(bp); | |
1020 | } | |
1da177e4 | 1021 | |
5d765b97 CH |
1022 | void |
1023 | xfs_bdwrite( | |
1024 | void *mp, | |
1025 | struct xfs_buf *bp) | |
1026 | { | |
1027 | XB_TRACE(bp, "bdwrite", 0); | |
1da177e4 | 1028 | |
5d765b97 | 1029 | bp->b_strat = xfs_bdstrat_cb; |
15ac08a8 | 1030 | bp->b_mount = mp; |
1da177e4 | 1031 | |
5d765b97 CH |
1032 | bp->b_flags &= ~XBF_READ; |
1033 | bp->b_flags |= (XBF_DELWRI | XBF_ASYNC); | |
1034 | ||
1035 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1036 | } |
1037 | ||
7989cb8e | 1038 | STATIC_INLINE void |
ce8e922c NS |
1039 | _xfs_buf_ioend( |
1040 | xfs_buf_t *bp, | |
1da177e4 LT |
1041 | int schedule) |
1042 | { | |
6ab455ee CH |
1043 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
1044 | bp->b_flags &= ~_XBF_PAGE_LOCKED; | |
ce8e922c | 1045 | xfs_buf_ioend(bp, schedule); |
6ab455ee | 1046 | } |
1da177e4 LT |
1047 | } |
1048 | ||
782e3b3b | 1049 | STATIC void |
ce8e922c | 1050 | xfs_buf_bio_end_io( |
1da177e4 | 1051 | struct bio *bio, |
1da177e4 LT |
1052 | int error) |
1053 | { | |
ce8e922c NS |
1054 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1055 | unsigned int blocksize = bp->b_target->bt_bsize; | |
eedb5530 | 1056 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1da177e4 | 1057 | |
cfbe5267 | 1058 | xfs_buf_ioerror(bp, -error); |
1da177e4 | 1059 | |
eedb5530 | 1060 | do { |
1da177e4 LT |
1061 | struct page *page = bvec->bv_page; |
1062 | ||
948ecdb4 | 1063 | ASSERT(!PagePrivate(page)); |
ce8e922c NS |
1064 | if (unlikely(bp->b_error)) { |
1065 | if (bp->b_flags & XBF_READ) | |
eedb5530 | 1066 | ClearPageUptodate(page); |
ce8e922c | 1067 | } else if (blocksize >= PAGE_CACHE_SIZE) { |
1da177e4 LT |
1068 | SetPageUptodate(page); |
1069 | } else if (!PagePrivate(page) && | |
ce8e922c | 1070 | (bp->b_flags & _XBF_PAGE_CACHE)) { |
1da177e4 LT |
1071 | set_page_region(page, bvec->bv_offset, bvec->bv_len); |
1072 | } | |
1073 | ||
eedb5530 NS |
1074 | if (--bvec >= bio->bi_io_vec) |
1075 | prefetchw(&bvec->bv_page->flags); | |
6ab455ee CH |
1076 | |
1077 | if (bp->b_flags & _XBF_PAGE_LOCKED) | |
1078 | unlock_page(page); | |
eedb5530 | 1079 | } while (bvec >= bio->bi_io_vec); |
1da177e4 | 1080 | |
ce8e922c | 1081 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1082 | bio_put(bio); |
1da177e4 LT |
1083 | } |
1084 | ||
1085 | STATIC void | |
ce8e922c NS |
1086 | _xfs_buf_ioapply( |
1087 | xfs_buf_t *bp) | |
1da177e4 | 1088 | { |
a9759f2d | 1089 | int rw, map_i, total_nr_pages, nr_pages; |
1da177e4 | 1090 | struct bio *bio; |
ce8e922c NS |
1091 | int offset = bp->b_offset; |
1092 | int size = bp->b_count_desired; | |
1093 | sector_t sector = bp->b_bn; | |
1094 | unsigned int blocksize = bp->b_target->bt_bsize; | |
1da177e4 | 1095 | |
ce8e922c | 1096 | total_nr_pages = bp->b_page_count; |
1da177e4 LT |
1097 | map_i = 0; |
1098 | ||
ce8e922c NS |
1099 | if (bp->b_flags & XBF_ORDERED) { |
1100 | ASSERT(!(bp->b_flags & XBF_READ)); | |
f538d4da | 1101 | rw = WRITE_BARRIER; |
51bdd706 NS |
1102 | } else if (bp->b_flags & _XBF_RUN_QUEUES) { |
1103 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); | |
1104 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1105 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC; | |
1106 | } else { | |
1107 | rw = (bp->b_flags & XBF_WRITE) ? WRITE : | |
1108 | (bp->b_flags & XBF_READ_AHEAD) ? READA : READ; | |
f538d4da CH |
1109 | } |
1110 | ||
ce8e922c | 1111 | /* Special code path for reading a sub page size buffer in -- |
1da177e4 LT |
1112 | * we populate up the whole page, and hence the other metadata |
1113 | * in the same page. This optimization is only valid when the | |
ce8e922c | 1114 | * filesystem block size is not smaller than the page size. |
1da177e4 | 1115 | */ |
ce8e922c | 1116 | if ((bp->b_buffer_length < PAGE_CACHE_SIZE) && |
6ab455ee CH |
1117 | ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) == |
1118 | (XBF_READ|_XBF_PAGE_LOCKED)) && | |
ce8e922c | 1119 | (blocksize >= PAGE_CACHE_SIZE)) { |
1da177e4 LT |
1120 | bio = bio_alloc(GFP_NOIO, 1); |
1121 | ||
ce8e922c | 1122 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1123 | bio->bi_sector = sector - (offset >> BBSHIFT); |
ce8e922c NS |
1124 | bio->bi_end_io = xfs_buf_bio_end_io; |
1125 | bio->bi_private = bp; | |
1da177e4 | 1126 | |
ce8e922c | 1127 | bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0); |
1da177e4 LT |
1128 | size = 0; |
1129 | ||
ce8e922c | 1130 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1131 | |
1132 | goto submit_io; | |
1133 | } | |
1134 | ||
1da177e4 | 1135 | next_chunk: |
ce8e922c | 1136 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1137 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1138 | if (nr_pages > total_nr_pages) | |
1139 | nr_pages = total_nr_pages; | |
1140 | ||
1141 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1142 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1143 | bio->bi_sector = sector; |
ce8e922c NS |
1144 | bio->bi_end_io = xfs_buf_bio_end_io; |
1145 | bio->bi_private = bp; | |
1da177e4 LT |
1146 | |
1147 | for (; size && nr_pages; nr_pages--, map_i++) { | |
ce8e922c | 1148 | int rbytes, nbytes = PAGE_CACHE_SIZE - offset; |
1da177e4 LT |
1149 | |
1150 | if (nbytes > size) | |
1151 | nbytes = size; | |
1152 | ||
ce8e922c NS |
1153 | rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset); |
1154 | if (rbytes < nbytes) | |
1da177e4 LT |
1155 | break; |
1156 | ||
1157 | offset = 0; | |
1158 | sector += nbytes >> BBSHIFT; | |
1159 | size -= nbytes; | |
1160 | total_nr_pages--; | |
1161 | } | |
1162 | ||
1163 | submit_io: | |
1164 | if (likely(bio->bi_size)) { | |
1165 | submit_bio(rw, bio); | |
1166 | if (size) | |
1167 | goto next_chunk; | |
1168 | } else { | |
1169 | bio_put(bio); | |
ce8e922c | 1170 | xfs_buf_ioerror(bp, EIO); |
1da177e4 LT |
1171 | } |
1172 | } | |
1173 | ||
1da177e4 | 1174 | int |
ce8e922c NS |
1175 | xfs_buf_iorequest( |
1176 | xfs_buf_t *bp) | |
1da177e4 | 1177 | { |
ce8e922c | 1178 | XB_TRACE(bp, "iorequest", 0); |
1da177e4 | 1179 | |
ce8e922c NS |
1180 | if (bp->b_flags & XBF_DELWRI) { |
1181 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1182 | return 0; |
1183 | } | |
1184 | ||
ce8e922c NS |
1185 | if (bp->b_flags & XBF_WRITE) { |
1186 | xfs_buf_wait_unpin(bp); | |
1da177e4 LT |
1187 | } |
1188 | ||
ce8e922c | 1189 | xfs_buf_hold(bp); |
1da177e4 LT |
1190 | |
1191 | /* Set the count to 1 initially, this will stop an I/O | |
1192 | * completion callout which happens before we have started | |
ce8e922c | 1193 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1194 | */ |
ce8e922c NS |
1195 | atomic_set(&bp->b_io_remaining, 1); |
1196 | _xfs_buf_ioapply(bp); | |
1197 | _xfs_buf_ioend(bp, 0); | |
1da177e4 | 1198 | |
ce8e922c | 1199 | xfs_buf_rele(bp); |
1da177e4 LT |
1200 | return 0; |
1201 | } | |
1202 | ||
1203 | /* | |
ce8e922c NS |
1204 | * Waits for I/O to complete on the buffer supplied. |
1205 | * It returns immediately if no I/O is pending. | |
1206 | * It returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1207 | */ |
1208 | int | |
ce8e922c NS |
1209 | xfs_buf_iowait( |
1210 | xfs_buf_t *bp) | |
1da177e4 | 1211 | { |
ce8e922c NS |
1212 | XB_TRACE(bp, "iowait", 0); |
1213 | if (atomic_read(&bp->b_io_remaining)) | |
1214 | blk_run_address_space(bp->b_target->bt_mapping); | |
b4dd330b | 1215 | wait_for_completion(&bp->b_iowait); |
ce8e922c NS |
1216 | XB_TRACE(bp, "iowaited", (long)bp->b_error); |
1217 | return bp->b_error; | |
1da177e4 LT |
1218 | } |
1219 | ||
ce8e922c NS |
1220 | xfs_caddr_t |
1221 | xfs_buf_offset( | |
1222 | xfs_buf_t *bp, | |
1da177e4 LT |
1223 | size_t offset) |
1224 | { | |
1225 | struct page *page; | |
1226 | ||
ce8e922c NS |
1227 | if (bp->b_flags & XBF_MAPPED) |
1228 | return XFS_BUF_PTR(bp) + offset; | |
1da177e4 | 1229 | |
ce8e922c NS |
1230 | offset += bp->b_offset; |
1231 | page = bp->b_pages[offset >> PAGE_CACHE_SHIFT]; | |
1232 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1)); | |
1da177e4 LT |
1233 | } |
1234 | ||
1235 | /* | |
1da177e4 LT |
1236 | * Move data into or out of a buffer. |
1237 | */ | |
1238 | void | |
ce8e922c NS |
1239 | xfs_buf_iomove( |
1240 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1241 | size_t boff, /* starting buffer offset */ |
1242 | size_t bsize, /* length to copy */ | |
1243 | caddr_t data, /* data address */ | |
ce8e922c | 1244 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 LT |
1245 | { |
1246 | size_t bend, cpoff, csize; | |
1247 | struct page *page; | |
1248 | ||
1249 | bend = boff + bsize; | |
1250 | while (boff < bend) { | |
ce8e922c NS |
1251 | page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)]; |
1252 | cpoff = xfs_buf_poff(boff + bp->b_offset); | |
1da177e4 | 1253 | csize = min_t(size_t, |
ce8e922c | 1254 | PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff); |
1da177e4 LT |
1255 | |
1256 | ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE)); | |
1257 | ||
1258 | switch (mode) { | |
ce8e922c | 1259 | case XBRW_ZERO: |
1da177e4 LT |
1260 | memset(page_address(page) + cpoff, 0, csize); |
1261 | break; | |
ce8e922c | 1262 | case XBRW_READ: |
1da177e4 LT |
1263 | memcpy(data, page_address(page) + cpoff, csize); |
1264 | break; | |
ce8e922c | 1265 | case XBRW_WRITE: |
1da177e4 LT |
1266 | memcpy(page_address(page) + cpoff, data, csize); |
1267 | } | |
1268 | ||
1269 | boff += csize; | |
1270 | data += csize; | |
1271 | } | |
1272 | } | |
1273 | ||
1274 | /* | |
ce8e922c | 1275 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1276 | */ |
1277 | ||
1278 | /* | |
ce8e922c NS |
1279 | * Wait for any bufs with callbacks that have been submitted but |
1280 | * have not yet returned... walk the hash list for the target. | |
1da177e4 LT |
1281 | */ |
1282 | void | |
1283 | xfs_wait_buftarg( | |
1284 | xfs_buftarg_t *btp) | |
1285 | { | |
1286 | xfs_buf_t *bp, *n; | |
1287 | xfs_bufhash_t *hash; | |
1288 | uint i; | |
1289 | ||
1290 | for (i = 0; i < (1 << btp->bt_hashshift); i++) { | |
1291 | hash = &btp->bt_hash[i]; | |
1292 | again: | |
1293 | spin_lock(&hash->bh_lock); | |
ce8e922c NS |
1294 | list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) { |
1295 | ASSERT(btp == bp->b_target); | |
1296 | if (!(bp->b_flags & XBF_FS_MANAGED)) { | |
1da177e4 | 1297 | spin_unlock(&hash->bh_lock); |
2f926587 DC |
1298 | /* |
1299 | * Catch superblock reference count leaks | |
1300 | * immediately | |
1301 | */ | |
ce8e922c | 1302 | BUG_ON(bp->b_bn == 0); |
1da177e4 LT |
1303 | delay(100); |
1304 | goto again; | |
1305 | } | |
1306 | } | |
1307 | spin_unlock(&hash->bh_lock); | |
1308 | } | |
1309 | } | |
1310 | ||
1311 | /* | |
ce8e922c NS |
1312 | * Allocate buffer hash table for a given target. |
1313 | * For devices containing metadata (i.e. not the log/realtime devices) | |
1314 | * we need to allocate a much larger hash table. | |
1da177e4 LT |
1315 | */ |
1316 | STATIC void | |
1317 | xfs_alloc_bufhash( | |
1318 | xfs_buftarg_t *btp, | |
1319 | int external) | |
1320 | { | |
1321 | unsigned int i; | |
1322 | ||
1323 | btp->bt_hashshift = external ? 3 : 8; /* 8 or 256 buckets */ | |
1324 | btp->bt_hashmask = (1 << btp->bt_hashshift) - 1; | |
1325 | btp->bt_hash = kmem_zalloc((1 << btp->bt_hashshift) * | |
93c189c1 | 1326 | sizeof(xfs_bufhash_t), KM_SLEEP | KM_LARGE); |
1da177e4 LT |
1327 | for (i = 0; i < (1 << btp->bt_hashshift); i++) { |
1328 | spin_lock_init(&btp->bt_hash[i].bh_lock); | |
1329 | INIT_LIST_HEAD(&btp->bt_hash[i].bh_list); | |
1330 | } | |
1331 | } | |
1332 | ||
1333 | STATIC void | |
1334 | xfs_free_bufhash( | |
1335 | xfs_buftarg_t *btp) | |
1336 | { | |
f0e2d93c | 1337 | kmem_free(btp->bt_hash); |
1da177e4 LT |
1338 | btp->bt_hash = NULL; |
1339 | } | |
1340 | ||
a6867a68 | 1341 | /* |
ce8e922c | 1342 | * buftarg list for delwrite queue processing |
a6867a68 | 1343 | */ |
e6a0e9cd | 1344 | static LIST_HEAD(xfs_buftarg_list); |
7989cb8e | 1345 | static DEFINE_SPINLOCK(xfs_buftarg_lock); |
a6867a68 DC |
1346 | |
1347 | STATIC void | |
1348 | xfs_register_buftarg( | |
1349 | xfs_buftarg_t *btp) | |
1350 | { | |
1351 | spin_lock(&xfs_buftarg_lock); | |
1352 | list_add(&btp->bt_list, &xfs_buftarg_list); | |
1353 | spin_unlock(&xfs_buftarg_lock); | |
1354 | } | |
1355 | ||
1356 | STATIC void | |
1357 | xfs_unregister_buftarg( | |
1358 | xfs_buftarg_t *btp) | |
1359 | { | |
1360 | spin_lock(&xfs_buftarg_lock); | |
1361 | list_del(&btp->bt_list); | |
1362 | spin_unlock(&xfs_buftarg_lock); | |
1363 | } | |
1364 | ||
1da177e4 LT |
1365 | void |
1366 | xfs_free_buftarg( | |
19f354d4 | 1367 | xfs_buftarg_t *btp) |
1da177e4 LT |
1368 | { |
1369 | xfs_flush_buftarg(btp, 1); | |
f4a9f28a | 1370 | xfs_blkdev_issue_flush(btp); |
1da177e4 | 1371 | xfs_free_bufhash(btp); |
ce8e922c | 1372 | iput(btp->bt_mapping->host); |
a6867a68 | 1373 | |
ce8e922c NS |
1374 | /* Unregister the buftarg first so that we don't get a |
1375 | * wakeup finding a non-existent task | |
1376 | */ | |
a6867a68 DC |
1377 | xfs_unregister_buftarg(btp); |
1378 | kthread_stop(btp->bt_task); | |
1379 | ||
f0e2d93c | 1380 | kmem_free(btp); |
1da177e4 LT |
1381 | } |
1382 | ||
1da177e4 LT |
1383 | STATIC int |
1384 | xfs_setsize_buftarg_flags( | |
1385 | xfs_buftarg_t *btp, | |
1386 | unsigned int blocksize, | |
1387 | unsigned int sectorsize, | |
1388 | int verbose) | |
1389 | { | |
ce8e922c NS |
1390 | btp->bt_bsize = blocksize; |
1391 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1392 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1393 | |
ce8e922c | 1394 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
1da177e4 LT |
1395 | printk(KERN_WARNING |
1396 | "XFS: Cannot set_blocksize to %u on device %s\n", | |
1397 | sectorsize, XFS_BUFTARG_NAME(btp)); | |
1398 | return EINVAL; | |
1399 | } | |
1400 | ||
1401 | if (verbose && | |
1402 | (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) { | |
1403 | printk(KERN_WARNING | |
1404 | "XFS: %u byte sectors in use on device %s. " | |
1405 | "This is suboptimal; %u or greater is ideal.\n", | |
1406 | sectorsize, XFS_BUFTARG_NAME(btp), | |
1407 | (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG); | |
1408 | } | |
1409 | ||
1410 | return 0; | |
1411 | } | |
1412 | ||
1413 | /* | |
ce8e922c NS |
1414 | * When allocating the initial buffer target we have not yet |
1415 | * read in the superblock, so don't know what sized sectors | |
1416 | * are being used is at this early stage. Play safe. | |
1417 | */ | |
1da177e4 LT |
1418 | STATIC int |
1419 | xfs_setsize_buftarg_early( | |
1420 | xfs_buftarg_t *btp, | |
1421 | struct block_device *bdev) | |
1422 | { | |
1423 | return xfs_setsize_buftarg_flags(btp, | |
1424 | PAGE_CACHE_SIZE, bdev_hardsect_size(bdev), 0); | |
1425 | } | |
1426 | ||
1427 | int | |
1428 | xfs_setsize_buftarg( | |
1429 | xfs_buftarg_t *btp, | |
1430 | unsigned int blocksize, | |
1431 | unsigned int sectorsize) | |
1432 | { | |
1433 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1434 | } | |
1435 | ||
1436 | STATIC int | |
1437 | xfs_mapping_buftarg( | |
1438 | xfs_buftarg_t *btp, | |
1439 | struct block_device *bdev) | |
1440 | { | |
1441 | struct backing_dev_info *bdi; | |
1442 | struct inode *inode; | |
1443 | struct address_space *mapping; | |
f5e54d6e | 1444 | static const struct address_space_operations mapping_aops = { |
1da177e4 | 1445 | .sync_page = block_sync_page, |
e965f963 | 1446 | .migratepage = fail_migrate_page, |
1da177e4 LT |
1447 | }; |
1448 | ||
1449 | inode = new_inode(bdev->bd_inode->i_sb); | |
1450 | if (!inode) { | |
1451 | printk(KERN_WARNING | |
1452 | "XFS: Cannot allocate mapping inode for device %s\n", | |
1453 | XFS_BUFTARG_NAME(btp)); | |
1454 | return ENOMEM; | |
1455 | } | |
1456 | inode->i_mode = S_IFBLK; | |
1457 | inode->i_bdev = bdev; | |
1458 | inode->i_rdev = bdev->bd_dev; | |
1459 | bdi = blk_get_backing_dev_info(bdev); | |
1460 | if (!bdi) | |
1461 | bdi = &default_backing_dev_info; | |
1462 | mapping = &inode->i_data; | |
1463 | mapping->a_ops = &mapping_aops; | |
1464 | mapping->backing_dev_info = bdi; | |
1465 | mapping_set_gfp_mask(mapping, GFP_NOFS); | |
ce8e922c | 1466 | btp->bt_mapping = mapping; |
1da177e4 LT |
1467 | return 0; |
1468 | } | |
1469 | ||
a6867a68 DC |
1470 | STATIC int |
1471 | xfs_alloc_delwrite_queue( | |
1472 | xfs_buftarg_t *btp) | |
1473 | { | |
1474 | int error = 0; | |
1475 | ||
1476 | INIT_LIST_HEAD(&btp->bt_list); | |
1477 | INIT_LIST_HEAD(&btp->bt_delwrite_queue); | |
007c61c6 | 1478 | spin_lock_init(&btp->bt_delwrite_lock); |
a6867a68 DC |
1479 | btp->bt_flags = 0; |
1480 | btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd"); | |
1481 | if (IS_ERR(btp->bt_task)) { | |
1482 | error = PTR_ERR(btp->bt_task); | |
1483 | goto out_error; | |
1484 | } | |
1485 | xfs_register_buftarg(btp); | |
1486 | out_error: | |
1487 | return error; | |
1488 | } | |
1489 | ||
1da177e4 LT |
1490 | xfs_buftarg_t * |
1491 | xfs_alloc_buftarg( | |
1492 | struct block_device *bdev, | |
1493 | int external) | |
1494 | { | |
1495 | xfs_buftarg_t *btp; | |
1496 | ||
1497 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1498 | ||
ce8e922c NS |
1499 | btp->bt_dev = bdev->bd_dev; |
1500 | btp->bt_bdev = bdev; | |
1da177e4 LT |
1501 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1502 | goto error; | |
1503 | if (xfs_mapping_buftarg(btp, bdev)) | |
1504 | goto error; | |
a6867a68 DC |
1505 | if (xfs_alloc_delwrite_queue(btp)) |
1506 | goto error; | |
1da177e4 LT |
1507 | xfs_alloc_bufhash(btp, external); |
1508 | return btp; | |
1509 | ||
1510 | error: | |
f0e2d93c | 1511 | kmem_free(btp); |
1da177e4 LT |
1512 | return NULL; |
1513 | } | |
1514 | ||
1515 | ||
1516 | /* | |
ce8e922c | 1517 | * Delayed write buffer handling |
1da177e4 | 1518 | */ |
1da177e4 | 1519 | STATIC void |
ce8e922c NS |
1520 | xfs_buf_delwri_queue( |
1521 | xfs_buf_t *bp, | |
1da177e4 LT |
1522 | int unlock) |
1523 | { | |
ce8e922c NS |
1524 | struct list_head *dwq = &bp->b_target->bt_delwrite_queue; |
1525 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; | |
a6867a68 | 1526 | |
ce8e922c NS |
1527 | XB_TRACE(bp, "delwri_q", (long)unlock); |
1528 | ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC)); | |
1da177e4 | 1529 | |
a6867a68 | 1530 | spin_lock(dwlk); |
1da177e4 | 1531 | /* If already in the queue, dequeue and place at tail */ |
ce8e922c NS |
1532 | if (!list_empty(&bp->b_list)) { |
1533 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1534 | if (unlock) | |
1535 | atomic_dec(&bp->b_hold); | |
1536 | list_del(&bp->b_list); | |
1da177e4 LT |
1537 | } |
1538 | ||
ce8e922c NS |
1539 | bp->b_flags |= _XBF_DELWRI_Q; |
1540 | list_add_tail(&bp->b_list, dwq); | |
1541 | bp->b_queuetime = jiffies; | |
a6867a68 | 1542 | spin_unlock(dwlk); |
1da177e4 LT |
1543 | |
1544 | if (unlock) | |
ce8e922c | 1545 | xfs_buf_unlock(bp); |
1da177e4 LT |
1546 | } |
1547 | ||
1548 | void | |
ce8e922c NS |
1549 | xfs_buf_delwri_dequeue( |
1550 | xfs_buf_t *bp) | |
1da177e4 | 1551 | { |
ce8e922c | 1552 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; |
1da177e4 LT |
1553 | int dequeued = 0; |
1554 | ||
a6867a68 | 1555 | spin_lock(dwlk); |
ce8e922c NS |
1556 | if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) { |
1557 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1558 | list_del_init(&bp->b_list); | |
1da177e4 LT |
1559 | dequeued = 1; |
1560 | } | |
ce8e922c | 1561 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q); |
a6867a68 | 1562 | spin_unlock(dwlk); |
1da177e4 LT |
1563 | |
1564 | if (dequeued) | |
ce8e922c | 1565 | xfs_buf_rele(bp); |
1da177e4 | 1566 | |
ce8e922c | 1567 | XB_TRACE(bp, "delwri_dq", (long)dequeued); |
1da177e4 LT |
1568 | } |
1569 | ||
1570 | STATIC void | |
ce8e922c | 1571 | xfs_buf_runall_queues( |
1da177e4 LT |
1572 | struct workqueue_struct *queue) |
1573 | { | |
1574 | flush_workqueue(queue); | |
1575 | } | |
1576 | ||
1da177e4 | 1577 | STATIC int |
23ea4032 | 1578 | xfsbufd_wakeup( |
15c84a47 NS |
1579 | int priority, |
1580 | gfp_t mask) | |
1da177e4 | 1581 | { |
da7f93e9 | 1582 | xfs_buftarg_t *btp; |
a6867a68 DC |
1583 | |
1584 | spin_lock(&xfs_buftarg_lock); | |
da7f93e9 | 1585 | list_for_each_entry(btp, &xfs_buftarg_list, bt_list) { |
ce8e922c | 1586 | if (test_bit(XBT_FORCE_SLEEP, &btp->bt_flags)) |
a6867a68 | 1587 | continue; |
ce8e922c | 1588 | set_bit(XBT_FORCE_FLUSH, &btp->bt_flags); |
a6867a68 DC |
1589 | wake_up_process(btp->bt_task); |
1590 | } | |
1591 | spin_unlock(&xfs_buftarg_lock); | |
1da177e4 LT |
1592 | return 0; |
1593 | } | |
1594 | ||
585e6d88 DC |
1595 | /* |
1596 | * Move as many buffers as specified to the supplied list | |
1597 | * idicating if we skipped any buffers to prevent deadlocks. | |
1598 | */ | |
1599 | STATIC int | |
1600 | xfs_buf_delwri_split( | |
1601 | xfs_buftarg_t *target, | |
1602 | struct list_head *list, | |
5e6a07df | 1603 | unsigned long age) |
585e6d88 DC |
1604 | { |
1605 | xfs_buf_t *bp, *n; | |
1606 | struct list_head *dwq = &target->bt_delwrite_queue; | |
1607 | spinlock_t *dwlk = &target->bt_delwrite_lock; | |
1608 | int skipped = 0; | |
5e6a07df | 1609 | int force; |
585e6d88 | 1610 | |
5e6a07df | 1611 | force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
585e6d88 DC |
1612 | INIT_LIST_HEAD(list); |
1613 | spin_lock(dwlk); | |
1614 | list_for_each_entry_safe(bp, n, dwq, b_list) { | |
1615 | XB_TRACE(bp, "walkq1", (long)xfs_buf_ispin(bp)); | |
1616 | ASSERT(bp->b_flags & XBF_DELWRI); | |
1617 | ||
1618 | if (!xfs_buf_ispin(bp) && !xfs_buf_cond_lock(bp)) { | |
5e6a07df | 1619 | if (!force && |
585e6d88 DC |
1620 | time_before(jiffies, bp->b_queuetime + age)) { |
1621 | xfs_buf_unlock(bp); | |
1622 | break; | |
1623 | } | |
1624 | ||
1625 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q| | |
1626 | _XBF_RUN_QUEUES); | |
1627 | bp->b_flags |= XBF_WRITE; | |
1628 | list_move_tail(&bp->b_list, list); | |
1629 | } else | |
1630 | skipped++; | |
1631 | } | |
1632 | spin_unlock(dwlk); | |
1633 | ||
1634 | return skipped; | |
1635 | ||
1636 | } | |
1637 | ||
1da177e4 | 1638 | STATIC int |
23ea4032 | 1639 | xfsbufd( |
585e6d88 | 1640 | void *data) |
1da177e4 | 1641 | { |
585e6d88 DC |
1642 | struct list_head tmp; |
1643 | xfs_buftarg_t *target = (xfs_buftarg_t *)data; | |
1644 | int count; | |
1645 | xfs_buf_t *bp; | |
1da177e4 | 1646 | |
1da177e4 LT |
1647 | current->flags |= PF_MEMALLOC; |
1648 | ||
978c7b2f RW |
1649 | set_freezable(); |
1650 | ||
1da177e4 | 1651 | do { |
3e1d1d28 | 1652 | if (unlikely(freezing(current))) { |
ce8e922c | 1653 | set_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
3e1d1d28 | 1654 | refrigerator(); |
abd0cf7a | 1655 | } else { |
ce8e922c | 1656 | clear_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
abd0cf7a | 1657 | } |
1da177e4 | 1658 | |
15c84a47 NS |
1659 | schedule_timeout_interruptible( |
1660 | xfs_buf_timer_centisecs * msecs_to_jiffies(10)); | |
1da177e4 | 1661 | |
585e6d88 | 1662 | xfs_buf_delwri_split(target, &tmp, |
5e6a07df | 1663 | xfs_buf_age_centisecs * msecs_to_jiffies(10)); |
1da177e4 | 1664 | |
585e6d88 | 1665 | count = 0; |
1da177e4 | 1666 | while (!list_empty(&tmp)) { |
ce8e922c NS |
1667 | bp = list_entry(tmp.next, xfs_buf_t, b_list); |
1668 | ASSERT(target == bp->b_target); | |
1da177e4 | 1669 | |
ce8e922c NS |
1670 | list_del_init(&bp->b_list); |
1671 | xfs_buf_iostrategy(bp); | |
585e6d88 | 1672 | count++; |
1da177e4 LT |
1673 | } |
1674 | ||
f07c2250 NS |
1675 | if (count) |
1676 | blk_run_address_space(target->bt_mapping); | |
1da177e4 | 1677 | |
4df08c52 | 1678 | } while (!kthread_should_stop()); |
1da177e4 | 1679 | |
4df08c52 | 1680 | return 0; |
1da177e4 LT |
1681 | } |
1682 | ||
1683 | /* | |
ce8e922c NS |
1684 | * Go through all incore buffers, and release buffers if they belong to |
1685 | * the given device. This is used in filesystem error handling to | |
1686 | * preserve the consistency of its metadata. | |
1da177e4 LT |
1687 | */ |
1688 | int | |
1689 | xfs_flush_buftarg( | |
585e6d88 DC |
1690 | xfs_buftarg_t *target, |
1691 | int wait) | |
1da177e4 | 1692 | { |
585e6d88 DC |
1693 | struct list_head tmp; |
1694 | xfs_buf_t *bp, *n; | |
1695 | int pincount = 0; | |
1da177e4 | 1696 | |
ce8e922c NS |
1697 | xfs_buf_runall_queues(xfsdatad_workqueue); |
1698 | xfs_buf_runall_queues(xfslogd_workqueue); | |
1da177e4 | 1699 | |
5e6a07df DC |
1700 | set_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
1701 | pincount = xfs_buf_delwri_split(target, &tmp, 0); | |
1da177e4 LT |
1702 | |
1703 | /* | |
1704 | * Dropped the delayed write list lock, now walk the temporary list | |
1705 | */ | |
ce8e922c | 1706 | list_for_each_entry_safe(bp, n, &tmp, b_list) { |
585e6d88 | 1707 | ASSERT(target == bp->b_target); |
1da177e4 | 1708 | if (wait) |
ce8e922c | 1709 | bp->b_flags &= ~XBF_ASYNC; |
1da177e4 | 1710 | else |
ce8e922c | 1711 | list_del_init(&bp->b_list); |
1da177e4 | 1712 | |
ce8e922c | 1713 | xfs_buf_iostrategy(bp); |
1da177e4 LT |
1714 | } |
1715 | ||
f07c2250 NS |
1716 | if (wait) |
1717 | blk_run_address_space(target->bt_mapping); | |
1718 | ||
1da177e4 LT |
1719 | /* |
1720 | * Remaining list items must be flushed before returning | |
1721 | */ | |
1722 | while (!list_empty(&tmp)) { | |
ce8e922c | 1723 | bp = list_entry(tmp.next, xfs_buf_t, b_list); |
1da177e4 | 1724 | |
ce8e922c NS |
1725 | list_del_init(&bp->b_list); |
1726 | xfs_iowait(bp); | |
1727 | xfs_buf_relse(bp); | |
1da177e4 LT |
1728 | } |
1729 | ||
1da177e4 LT |
1730 | return pincount; |
1731 | } | |
1732 | ||
04d8b284 | 1733 | int __init |
ce8e922c | 1734 | xfs_buf_init(void) |
1da177e4 | 1735 | { |
ce8e922c | 1736 | #ifdef XFS_BUF_TRACE |
5695ef46 | 1737 | xfs_buf_trace_buf = ktrace_alloc(XFS_BUF_TRACE_SIZE, KM_NOFS); |
04d8b284 CH |
1738 | #endif |
1739 | ||
8758280f NS |
1740 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1741 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1742 | if (!xfs_buf_zone) |
04d8b284 CH |
1743 | goto out_free_trace_buf; |
1744 | ||
b4337692 | 1745 | xfslogd_workqueue = create_workqueue("xfslogd"); |
23ea4032 | 1746 | if (!xfslogd_workqueue) |
04d8b284 | 1747 | goto out_free_buf_zone; |
1da177e4 | 1748 | |
b4337692 | 1749 | xfsdatad_workqueue = create_workqueue("xfsdatad"); |
23ea4032 CH |
1750 | if (!xfsdatad_workqueue) |
1751 | goto out_destroy_xfslogd_workqueue; | |
1da177e4 | 1752 | |
8e1f936b | 1753 | register_shrinker(&xfs_buf_shake); |
23ea4032 | 1754 | return 0; |
1da177e4 | 1755 | |
23ea4032 CH |
1756 | out_destroy_xfslogd_workqueue: |
1757 | destroy_workqueue(xfslogd_workqueue); | |
23ea4032 | 1758 | out_free_buf_zone: |
ce8e922c | 1759 | kmem_zone_destroy(xfs_buf_zone); |
04d8b284 | 1760 | out_free_trace_buf: |
ce8e922c NS |
1761 | #ifdef XFS_BUF_TRACE |
1762 | ktrace_free(xfs_buf_trace_buf); | |
23ea4032 | 1763 | #endif |
8758280f | 1764 | return -ENOMEM; |
1da177e4 LT |
1765 | } |
1766 | ||
1da177e4 | 1767 | void |
ce8e922c | 1768 | xfs_buf_terminate(void) |
1da177e4 | 1769 | { |
8e1f936b | 1770 | unregister_shrinker(&xfs_buf_shake); |
04d8b284 CH |
1771 | destroy_workqueue(xfsdatad_workqueue); |
1772 | destroy_workqueue(xfslogd_workqueue); | |
ce8e922c NS |
1773 | kmem_zone_destroy(xfs_buf_zone); |
1774 | #ifdef XFS_BUF_TRACE | |
1775 | ktrace_free(xfs_buf_trace_buf); | |
1da177e4 | 1776 | #endif |
1da177e4 | 1777 | } |
e6a0e9cd TS |
1778 | |
1779 | #ifdef CONFIG_KDB_MODULES | |
1780 | struct list_head * | |
1781 | xfs_get_buftarg_list(void) | |
1782 | { | |
1783 | return &xfs_buftarg_list; | |
1784 | } | |
1785 | #endif |