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 | |
b7963133 CH |
37 | #include "xfs_sb.h" |
38 | #include "xfs_inum.h" | |
39 | #include "xfs_ag.h" | |
40 | #include "xfs_dmapi.h" | |
41 | #include "xfs_mount.h" | |
0b1b213f | 42 | #include "xfs_trace.h" |
b7963133 | 43 | |
7989cb8e | 44 | static kmem_zone_t *xfs_buf_zone; |
a6867a68 | 45 | STATIC int xfsbufd(void *); |
27496a8c | 46 | STATIC int xfsbufd_wakeup(int, gfp_t); |
ce8e922c | 47 | STATIC void xfs_buf_delwri_queue(xfs_buf_t *, int); |
8e1f936b RR |
48 | static struct shrinker xfs_buf_shake = { |
49 | .shrink = xfsbufd_wakeup, | |
50 | .seeks = DEFAULT_SEEKS, | |
51 | }; | |
23ea4032 | 52 | |
7989cb8e | 53 | static struct workqueue_struct *xfslogd_workqueue; |
0829c360 | 54 | struct workqueue_struct *xfsdatad_workqueue; |
c626d174 | 55 | struct workqueue_struct *xfsconvertd_workqueue; |
1da177e4 | 56 | |
ce8e922c NS |
57 | #ifdef XFS_BUF_LOCK_TRACKING |
58 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
59 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
60 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 61 | #else |
ce8e922c NS |
62 | # define XB_SET_OWNER(bp) do { } while (0) |
63 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
64 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
65 | #endif |
66 | ||
ce8e922c NS |
67 | #define xb_to_gfp(flags) \ |
68 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \ | |
69 | ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN) | |
1da177e4 | 70 | |
ce8e922c NS |
71 | #define xb_to_km(flags) \ |
72 | (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP) | |
1da177e4 | 73 | |
ce8e922c NS |
74 | #define xfs_buf_allocate(flags) \ |
75 | kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags)) | |
76 | #define xfs_buf_deallocate(bp) \ | |
77 | kmem_zone_free(xfs_buf_zone, (bp)); | |
1da177e4 LT |
78 | |
79 | /* | |
ce8e922c | 80 | * Page Region interfaces. |
1da177e4 | 81 | * |
ce8e922c NS |
82 | * For pages in filesystems where the blocksize is smaller than the |
83 | * pagesize, we use the page->private field (long) to hold a bitmap | |
84 | * of uptodate regions within the page. | |
1da177e4 | 85 | * |
ce8e922c | 86 | * Each such region is "bytes per page / bits per long" bytes long. |
1da177e4 | 87 | * |
ce8e922c NS |
88 | * NBPPR == number-of-bytes-per-page-region |
89 | * BTOPR == bytes-to-page-region (rounded up) | |
90 | * BTOPRT == bytes-to-page-region-truncated (rounded down) | |
1da177e4 LT |
91 | */ |
92 | #if (BITS_PER_LONG == 32) | |
93 | #define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */ | |
94 | #elif (BITS_PER_LONG == 64) | |
95 | #define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */ | |
96 | #else | |
97 | #error BITS_PER_LONG must be 32 or 64 | |
98 | #endif | |
99 | #define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG) | |
100 | #define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT) | |
101 | #define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT)) | |
102 | ||
103 | STATIC unsigned long | |
104 | page_region_mask( | |
105 | size_t offset, | |
106 | size_t length) | |
107 | { | |
108 | unsigned long mask; | |
109 | int first, final; | |
110 | ||
111 | first = BTOPR(offset); | |
112 | final = BTOPRT(offset + length - 1); | |
113 | first = min(first, final); | |
114 | ||
115 | mask = ~0UL; | |
116 | mask <<= BITS_PER_LONG - (final - first); | |
117 | mask >>= BITS_PER_LONG - (final); | |
118 | ||
119 | ASSERT(offset + length <= PAGE_CACHE_SIZE); | |
120 | ASSERT((final - first) < BITS_PER_LONG && (final - first) >= 0); | |
121 | ||
122 | return mask; | |
123 | } | |
124 | ||
b8f82a4a | 125 | STATIC void |
1da177e4 LT |
126 | set_page_region( |
127 | struct page *page, | |
128 | size_t offset, | |
129 | size_t length) | |
130 | { | |
4c21e2f2 HD |
131 | set_page_private(page, |
132 | page_private(page) | page_region_mask(offset, length)); | |
133 | if (page_private(page) == ~0UL) | |
1da177e4 LT |
134 | SetPageUptodate(page); |
135 | } | |
136 | ||
b8f82a4a | 137 | STATIC int |
1da177e4 LT |
138 | test_page_region( |
139 | struct page *page, | |
140 | size_t offset, | |
141 | size_t length) | |
142 | { | |
143 | unsigned long mask = page_region_mask(offset, length); | |
144 | ||
4c21e2f2 | 145 | return (mask && (page_private(page) & mask) == mask); |
1da177e4 LT |
146 | } |
147 | ||
3a011a17 FB |
148 | /* |
149 | * Mapping of multi-page buffers into contiguous virtual space | |
150 | */ | |
151 | ||
152 | typedef struct a_list { | |
153 | void *vm_addr; | |
154 | struct a_list *next; | |
155 | } a_list_t; | |
156 | ||
157 | static a_list_t *as_free_head; | |
158 | static int as_list_len; | |
159 | static DEFINE_SPINLOCK(as_lock); | |
160 | ||
161 | /* | |
162 | * Try to batch vunmaps because they are costly. | |
163 | */ | |
164 | STATIC void | |
165 | free_address( | |
166 | void *addr) | |
167 | { | |
168 | a_list_t *aentry; | |
169 | ||
170 | #ifdef CONFIG_XEN | |
171 | /* | |
172 | * Xen needs to be able to make sure it can get an exclusive | |
173 | * RO mapping of pages it wants to turn into a pagetable. If | |
174 | * a newly allocated page is also still being vmap()ed by xfs, | |
175 | * it will cause pagetable construction to fail. This is a | |
176 | * quick workaround to always eagerly unmap pages so that Xen | |
177 | * is happy. | |
178 | */ | |
179 | vunmap(addr); | |
180 | return; | |
181 | #endif | |
182 | ||
183 | aentry = kmalloc(sizeof(a_list_t), GFP_NOWAIT); | |
184 | if (likely(aentry)) { | |
185 | spin_lock(&as_lock); | |
186 | aentry->next = as_free_head; | |
187 | aentry->vm_addr = addr; | |
188 | as_free_head = aentry; | |
189 | as_list_len++; | |
190 | spin_unlock(&as_lock); | |
191 | } else { | |
192 | vunmap(addr); | |
193 | } | |
194 | } | |
195 | ||
196 | STATIC void | |
197 | purge_addresses(void) | |
198 | { | |
199 | a_list_t *aentry, *old; | |
200 | ||
201 | if (as_free_head == NULL) | |
202 | return; | |
203 | ||
204 | spin_lock(&as_lock); | |
205 | aentry = as_free_head; | |
206 | as_free_head = NULL; | |
207 | as_list_len = 0; | |
208 | spin_unlock(&as_lock); | |
209 | ||
210 | while ((old = aentry) != NULL) { | |
211 | vunmap(aentry->vm_addr); | |
212 | aentry = aentry->next; | |
213 | kfree(old); | |
214 | } | |
215 | } | |
216 | ||
1da177e4 | 217 | /* |
ce8e922c | 218 | * Internal xfs_buf_t object manipulation |
1da177e4 LT |
219 | */ |
220 | ||
221 | STATIC void | |
ce8e922c NS |
222 | _xfs_buf_initialize( |
223 | xfs_buf_t *bp, | |
1da177e4 | 224 | xfs_buftarg_t *target, |
204ab25f | 225 | xfs_off_t range_base, |
1da177e4 | 226 | size_t range_length, |
ce8e922c | 227 | xfs_buf_flags_t flags) |
1da177e4 LT |
228 | { |
229 | /* | |
ce8e922c | 230 | * We don't want certain flags to appear in b_flags. |
1da177e4 | 231 | */ |
ce8e922c NS |
232 | flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD); |
233 | ||
234 | memset(bp, 0, sizeof(xfs_buf_t)); | |
235 | atomic_set(&bp->b_hold, 1); | |
b4dd330b | 236 | init_completion(&bp->b_iowait); |
ce8e922c NS |
237 | INIT_LIST_HEAD(&bp->b_list); |
238 | INIT_LIST_HEAD(&bp->b_hash_list); | |
239 | init_MUTEX_LOCKED(&bp->b_sema); /* held, no waiters */ | |
240 | XB_SET_OWNER(bp); | |
241 | bp->b_target = target; | |
242 | bp->b_file_offset = range_base; | |
1da177e4 LT |
243 | /* |
244 | * Set buffer_length and count_desired to the same value initially. | |
245 | * I/O routines should use count_desired, which will be the same in | |
246 | * most cases but may be reset (e.g. XFS recovery). | |
247 | */ | |
ce8e922c NS |
248 | bp->b_buffer_length = bp->b_count_desired = range_length; |
249 | bp->b_flags = flags; | |
250 | bp->b_bn = XFS_BUF_DADDR_NULL; | |
251 | atomic_set(&bp->b_pin_count, 0); | |
252 | init_waitqueue_head(&bp->b_waiters); | |
253 | ||
254 | XFS_STATS_INC(xb_create); | |
0b1b213f CH |
255 | |
256 | trace_xfs_buf_init(bp, _RET_IP_); | |
1da177e4 LT |
257 | } |
258 | ||
259 | /* | |
ce8e922c NS |
260 | * Allocate a page array capable of holding a specified number |
261 | * of pages, and point the page buf at it. | |
1da177e4 LT |
262 | */ |
263 | STATIC int | |
ce8e922c NS |
264 | _xfs_buf_get_pages( |
265 | xfs_buf_t *bp, | |
1da177e4 | 266 | int page_count, |
ce8e922c | 267 | xfs_buf_flags_t flags) |
1da177e4 LT |
268 | { |
269 | /* Make sure that we have a page list */ | |
ce8e922c NS |
270 | if (bp->b_pages == NULL) { |
271 | bp->b_offset = xfs_buf_poff(bp->b_file_offset); | |
272 | bp->b_page_count = page_count; | |
273 | if (page_count <= XB_PAGES) { | |
274 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 275 | } else { |
ce8e922c NS |
276 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
277 | page_count, xb_to_km(flags)); | |
278 | if (bp->b_pages == NULL) | |
1da177e4 LT |
279 | return -ENOMEM; |
280 | } | |
ce8e922c | 281 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
282 | } |
283 | return 0; | |
284 | } | |
285 | ||
286 | /* | |
ce8e922c | 287 | * Frees b_pages if it was allocated. |
1da177e4 LT |
288 | */ |
289 | STATIC void | |
ce8e922c | 290 | _xfs_buf_free_pages( |
1da177e4 LT |
291 | xfs_buf_t *bp) |
292 | { | |
ce8e922c | 293 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 294 | kmem_free(bp->b_pages); |
3fc98b1a | 295 | bp->b_pages = NULL; |
1da177e4 LT |
296 | } |
297 | } | |
298 | ||
299 | /* | |
300 | * Releases the specified buffer. | |
301 | * | |
302 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 303 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
304 | * hashed and refcounted buffers |
305 | */ | |
306 | void | |
ce8e922c | 307 | xfs_buf_free( |
1da177e4 LT |
308 | xfs_buf_t *bp) |
309 | { | |
0b1b213f | 310 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 311 | |
ce8e922c | 312 | ASSERT(list_empty(&bp->b_hash_list)); |
1da177e4 | 313 | |
1fa40b01 | 314 | if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) { |
1da177e4 LT |
315 | uint i; |
316 | ||
ce8e922c | 317 | if ((bp->b_flags & XBF_MAPPED) && (bp->b_page_count > 1)) |
3a011a17 | 318 | free_address(bp->b_addr - bp->b_offset); |
1da177e4 | 319 | |
948ecdb4 NS |
320 | for (i = 0; i < bp->b_page_count; i++) { |
321 | struct page *page = bp->b_pages[i]; | |
322 | ||
1fa40b01 CH |
323 | if (bp->b_flags & _XBF_PAGE_CACHE) |
324 | ASSERT(!PagePrivate(page)); | |
948ecdb4 NS |
325 | page_cache_release(page); |
326 | } | |
1da177e4 | 327 | } |
3fc98b1a | 328 | _xfs_buf_free_pages(bp); |
ce8e922c | 329 | xfs_buf_deallocate(bp); |
1da177e4 LT |
330 | } |
331 | ||
332 | /* | |
333 | * Finds all pages for buffer in question and builds it's page list. | |
334 | */ | |
335 | STATIC int | |
ce8e922c | 336 | _xfs_buf_lookup_pages( |
1da177e4 LT |
337 | xfs_buf_t *bp, |
338 | uint flags) | |
339 | { | |
ce8e922c NS |
340 | struct address_space *mapping = bp->b_target->bt_mapping; |
341 | size_t blocksize = bp->b_target->bt_bsize; | |
342 | size_t size = bp->b_count_desired; | |
1da177e4 | 343 | size_t nbytes, offset; |
ce8e922c | 344 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 LT |
345 | unsigned short page_count, i; |
346 | pgoff_t first; | |
204ab25f | 347 | xfs_off_t end; |
1da177e4 LT |
348 | int error; |
349 | ||
ce8e922c NS |
350 | end = bp->b_file_offset + bp->b_buffer_length; |
351 | page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset); | |
1da177e4 | 352 | |
ce8e922c | 353 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
354 | if (unlikely(error)) |
355 | return error; | |
ce8e922c | 356 | bp->b_flags |= _XBF_PAGE_CACHE; |
1da177e4 | 357 | |
ce8e922c NS |
358 | offset = bp->b_offset; |
359 | first = bp->b_file_offset >> PAGE_CACHE_SHIFT; | |
1da177e4 | 360 | |
ce8e922c | 361 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
362 | struct page *page; |
363 | uint retries = 0; | |
364 | ||
365 | retry: | |
366 | page = find_or_create_page(mapping, first + i, gfp_mask); | |
367 | if (unlikely(page == NULL)) { | |
ce8e922c NS |
368 | if (flags & XBF_READ_AHEAD) { |
369 | bp->b_page_count = i; | |
6ab455ee CH |
370 | for (i = 0; i < bp->b_page_count; i++) |
371 | unlock_page(bp->b_pages[i]); | |
1da177e4 LT |
372 | return -ENOMEM; |
373 | } | |
374 | ||
375 | /* | |
376 | * This could deadlock. | |
377 | * | |
378 | * But until all the XFS lowlevel code is revamped to | |
379 | * handle buffer allocation failures we can't do much. | |
380 | */ | |
381 | if (!(++retries % 100)) | |
382 | printk(KERN_ERR | |
383 | "XFS: possible memory allocation " | |
384 | "deadlock in %s (mode:0x%x)\n", | |
34a622b2 | 385 | __func__, gfp_mask); |
1da177e4 | 386 | |
ce8e922c | 387 | XFS_STATS_INC(xb_page_retries); |
23ea4032 | 388 | xfsbufd_wakeup(0, gfp_mask); |
8aa7e847 | 389 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
390 | goto retry; |
391 | } | |
392 | ||
ce8e922c | 393 | XFS_STATS_INC(xb_page_found); |
1da177e4 LT |
394 | |
395 | nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset); | |
396 | size -= nbytes; | |
397 | ||
948ecdb4 | 398 | ASSERT(!PagePrivate(page)); |
1da177e4 LT |
399 | if (!PageUptodate(page)) { |
400 | page_count--; | |
6ab455ee CH |
401 | if (blocksize >= PAGE_CACHE_SIZE) { |
402 | if (flags & XBF_READ) | |
403 | bp->b_flags |= _XBF_PAGE_LOCKED; | |
404 | } else if (!PagePrivate(page)) { | |
1da177e4 LT |
405 | if (test_page_region(page, offset, nbytes)) |
406 | page_count++; | |
407 | } | |
408 | } | |
409 | ||
ce8e922c | 410 | bp->b_pages[i] = page; |
1da177e4 LT |
411 | offset = 0; |
412 | } | |
413 | ||
6ab455ee CH |
414 | if (!(bp->b_flags & _XBF_PAGE_LOCKED)) { |
415 | for (i = 0; i < bp->b_page_count; i++) | |
416 | unlock_page(bp->b_pages[i]); | |
417 | } | |
418 | ||
ce8e922c NS |
419 | if (page_count == bp->b_page_count) |
420 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 421 | |
1da177e4 LT |
422 | return error; |
423 | } | |
424 | ||
425 | /* | |
426 | * Map buffer into kernel address-space if nessecary. | |
427 | */ | |
428 | STATIC int | |
ce8e922c | 429 | _xfs_buf_map_pages( |
1da177e4 LT |
430 | xfs_buf_t *bp, |
431 | uint flags) | |
432 | { | |
433 | /* A single page buffer is always mappable */ | |
ce8e922c NS |
434 | if (bp->b_page_count == 1) { |
435 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; | |
436 | bp->b_flags |= XBF_MAPPED; | |
437 | } else if (flags & XBF_MAPPED) { | |
3a011a17 FB |
438 | if (as_list_len > 64) |
439 | purge_addresses(); | |
cf7dab80 FB |
440 | bp->b_addr = vmap(bp->b_pages, bp->b_page_count, |
441 | VM_MAP, PAGE_KERNEL); | |
ce8e922c | 442 | if (unlikely(bp->b_addr == NULL)) |
1da177e4 | 443 | return -ENOMEM; |
ce8e922c NS |
444 | bp->b_addr += bp->b_offset; |
445 | bp->b_flags |= XBF_MAPPED; | |
1da177e4 LT |
446 | } |
447 | ||
448 | return 0; | |
449 | } | |
450 | ||
451 | /* | |
452 | * Finding and Reading Buffers | |
453 | */ | |
454 | ||
455 | /* | |
ce8e922c | 456 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 LT |
457 | * a given range of an inode. The buffer is returned |
458 | * locked. If other overlapping buffers exist, they are | |
459 | * released before the new buffer is created and locked, | |
460 | * which may imply that this call will block until those buffers | |
461 | * are unlocked. No I/O is implied by this call. | |
462 | */ | |
463 | xfs_buf_t * | |
ce8e922c | 464 | _xfs_buf_find( |
1da177e4 | 465 | xfs_buftarg_t *btp, /* block device target */ |
204ab25f | 466 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 467 | size_t isize, /* length of range */ |
ce8e922c NS |
468 | xfs_buf_flags_t flags, |
469 | xfs_buf_t *new_bp) | |
1da177e4 | 470 | { |
204ab25f | 471 | xfs_off_t range_base; |
1da177e4 LT |
472 | size_t range_length; |
473 | xfs_bufhash_t *hash; | |
ce8e922c | 474 | xfs_buf_t *bp, *n; |
1da177e4 LT |
475 | |
476 | range_base = (ioff << BBSHIFT); | |
477 | range_length = (isize << BBSHIFT); | |
478 | ||
479 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
ce8e922c | 480 | ASSERT(!(range_length < (1 << btp->bt_sshift))); |
204ab25f | 481 | ASSERT(!(range_base & (xfs_off_t)btp->bt_smask)); |
1da177e4 LT |
482 | |
483 | hash = &btp->bt_hash[hash_long((unsigned long)ioff, btp->bt_hashshift)]; | |
484 | ||
485 | spin_lock(&hash->bh_lock); | |
486 | ||
ce8e922c NS |
487 | list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) { |
488 | ASSERT(btp == bp->b_target); | |
489 | if (bp->b_file_offset == range_base && | |
490 | bp->b_buffer_length == range_length) { | |
1da177e4 | 491 | /* |
ce8e922c | 492 | * If we look at something, bring it to the |
1da177e4 LT |
493 | * front of the list for next time. |
494 | */ | |
ce8e922c NS |
495 | atomic_inc(&bp->b_hold); |
496 | list_move(&bp->b_hash_list, &hash->bh_list); | |
1da177e4 LT |
497 | goto found; |
498 | } | |
499 | } | |
500 | ||
501 | /* No match found */ | |
ce8e922c NS |
502 | if (new_bp) { |
503 | _xfs_buf_initialize(new_bp, btp, range_base, | |
1da177e4 | 504 | range_length, flags); |
ce8e922c NS |
505 | new_bp->b_hash = hash; |
506 | list_add(&new_bp->b_hash_list, &hash->bh_list); | |
1da177e4 | 507 | } else { |
ce8e922c | 508 | XFS_STATS_INC(xb_miss_locked); |
1da177e4 LT |
509 | } |
510 | ||
511 | spin_unlock(&hash->bh_lock); | |
ce8e922c | 512 | return new_bp; |
1da177e4 LT |
513 | |
514 | found: | |
515 | spin_unlock(&hash->bh_lock); | |
516 | ||
517 | /* Attempt to get the semaphore without sleeping, | |
518 | * if this does not work then we need to drop the | |
519 | * spinlock and do a hard attempt on the semaphore. | |
520 | */ | |
ce8e922c NS |
521 | if (down_trylock(&bp->b_sema)) { |
522 | if (!(flags & XBF_TRYLOCK)) { | |
1da177e4 | 523 | /* wait for buffer ownership */ |
ce8e922c NS |
524 | xfs_buf_lock(bp); |
525 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
526 | } else { |
527 | /* We asked for a trylock and failed, no need | |
528 | * to look at file offset and length here, we | |
ce8e922c NS |
529 | * know that this buffer at least overlaps our |
530 | * buffer and is locked, therefore our buffer | |
531 | * either does not exist, or is this buffer. | |
1da177e4 | 532 | */ |
ce8e922c NS |
533 | xfs_buf_rele(bp); |
534 | XFS_STATS_INC(xb_busy_locked); | |
535 | return NULL; | |
1da177e4 LT |
536 | } |
537 | } else { | |
538 | /* trylock worked */ | |
ce8e922c | 539 | XB_SET_OWNER(bp); |
1da177e4 LT |
540 | } |
541 | ||
ce8e922c NS |
542 | if (bp->b_flags & XBF_STALE) { |
543 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
544 | bp->b_flags &= XBF_MAPPED; | |
2f926587 | 545 | } |
0b1b213f CH |
546 | |
547 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ce8e922c NS |
548 | XFS_STATS_INC(xb_get_locked); |
549 | return bp; | |
1da177e4 LT |
550 | } |
551 | ||
552 | /* | |
ce8e922c | 553 | * Assembles a buffer covering the specified range. |
1da177e4 LT |
554 | * Storage in memory for all portions of the buffer will be allocated, |
555 | * although backing storage may not be. | |
556 | */ | |
557 | xfs_buf_t * | |
6ad112bf | 558 | xfs_buf_get( |
1da177e4 | 559 | xfs_buftarg_t *target,/* target for buffer */ |
204ab25f | 560 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 561 | size_t isize, /* length of range */ |
ce8e922c | 562 | xfs_buf_flags_t flags) |
1da177e4 | 563 | { |
ce8e922c | 564 | xfs_buf_t *bp, *new_bp; |
1da177e4 LT |
565 | int error = 0, i; |
566 | ||
ce8e922c NS |
567 | new_bp = xfs_buf_allocate(flags); |
568 | if (unlikely(!new_bp)) | |
1da177e4 LT |
569 | return NULL; |
570 | ||
ce8e922c NS |
571 | bp = _xfs_buf_find(target, ioff, isize, flags, new_bp); |
572 | if (bp == new_bp) { | |
573 | error = _xfs_buf_lookup_pages(bp, flags); | |
1da177e4 LT |
574 | if (error) |
575 | goto no_buffer; | |
576 | } else { | |
ce8e922c NS |
577 | xfs_buf_deallocate(new_bp); |
578 | if (unlikely(bp == NULL)) | |
1da177e4 LT |
579 | return NULL; |
580 | } | |
581 | ||
ce8e922c NS |
582 | for (i = 0; i < bp->b_page_count; i++) |
583 | mark_page_accessed(bp->b_pages[i]); | |
1da177e4 | 584 | |
ce8e922c NS |
585 | if (!(bp->b_flags & XBF_MAPPED)) { |
586 | error = _xfs_buf_map_pages(bp, flags); | |
1da177e4 LT |
587 | if (unlikely(error)) { |
588 | printk(KERN_WARNING "%s: failed to map pages\n", | |
34a622b2 | 589 | __func__); |
1da177e4 LT |
590 | goto no_buffer; |
591 | } | |
592 | } | |
593 | ||
ce8e922c | 594 | XFS_STATS_INC(xb_get); |
1da177e4 LT |
595 | |
596 | /* | |
597 | * Always fill in the block number now, the mapped cases can do | |
598 | * their own overlay of this later. | |
599 | */ | |
ce8e922c NS |
600 | bp->b_bn = ioff; |
601 | bp->b_count_desired = bp->b_buffer_length; | |
1da177e4 | 602 | |
0b1b213f | 603 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 604 | return bp; |
1da177e4 LT |
605 | |
606 | no_buffer: | |
ce8e922c NS |
607 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
608 | xfs_buf_unlock(bp); | |
609 | xfs_buf_rele(bp); | |
1da177e4 LT |
610 | return NULL; |
611 | } | |
612 | ||
5d765b97 CH |
613 | STATIC int |
614 | _xfs_buf_read( | |
615 | xfs_buf_t *bp, | |
616 | xfs_buf_flags_t flags) | |
617 | { | |
618 | int status; | |
619 | ||
5d765b97 CH |
620 | ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE))); |
621 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); | |
622 | ||
623 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \ | |
624 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
625 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \ | |
626 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
627 | ||
628 | status = xfs_buf_iorequest(bp); | |
629 | if (!status && !(flags & XBF_ASYNC)) | |
630 | status = xfs_buf_iowait(bp); | |
631 | return status; | |
632 | } | |
633 | ||
1da177e4 | 634 | xfs_buf_t * |
6ad112bf | 635 | xfs_buf_read( |
1da177e4 | 636 | xfs_buftarg_t *target, |
204ab25f | 637 | xfs_off_t ioff, |
1da177e4 | 638 | size_t isize, |
ce8e922c | 639 | xfs_buf_flags_t flags) |
1da177e4 | 640 | { |
ce8e922c NS |
641 | xfs_buf_t *bp; |
642 | ||
643 | flags |= XBF_READ; | |
644 | ||
6ad112bf | 645 | bp = xfs_buf_get(target, ioff, isize, flags); |
ce8e922c | 646 | if (bp) { |
0b1b213f CH |
647 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
648 | ||
ce8e922c | 649 | if (!XFS_BUF_ISDONE(bp)) { |
ce8e922c | 650 | XFS_STATS_INC(xb_get_read); |
5d765b97 | 651 | _xfs_buf_read(bp, flags); |
ce8e922c | 652 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
653 | /* |
654 | * Read ahead call which is already satisfied, | |
655 | * drop the buffer | |
656 | */ | |
657 | goto no_buffer; | |
658 | } else { | |
1da177e4 | 659 | /* We do not want read in the flags */ |
ce8e922c | 660 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
661 | } |
662 | } | |
663 | ||
ce8e922c | 664 | return bp; |
1da177e4 LT |
665 | |
666 | no_buffer: | |
ce8e922c NS |
667 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
668 | xfs_buf_unlock(bp); | |
669 | xfs_buf_rele(bp); | |
1da177e4 LT |
670 | return NULL; |
671 | } | |
672 | ||
1da177e4 | 673 | /* |
ce8e922c NS |
674 | * If we are not low on memory then do the readahead in a deadlock |
675 | * safe manner. | |
1da177e4 LT |
676 | */ |
677 | void | |
ce8e922c | 678 | xfs_buf_readahead( |
1da177e4 | 679 | xfs_buftarg_t *target, |
204ab25f | 680 | xfs_off_t ioff, |
1da177e4 | 681 | size_t isize, |
ce8e922c | 682 | xfs_buf_flags_t flags) |
1da177e4 LT |
683 | { |
684 | struct backing_dev_info *bdi; | |
685 | ||
ce8e922c | 686 | bdi = target->bt_mapping->backing_dev_info; |
1da177e4 LT |
687 | if (bdi_read_congested(bdi)) |
688 | return; | |
689 | ||
ce8e922c | 690 | flags |= (XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD); |
6ad112bf | 691 | xfs_buf_read(target, ioff, isize, flags); |
1da177e4 LT |
692 | } |
693 | ||
694 | xfs_buf_t * | |
ce8e922c | 695 | xfs_buf_get_empty( |
1da177e4 LT |
696 | size_t len, |
697 | xfs_buftarg_t *target) | |
698 | { | |
ce8e922c | 699 | xfs_buf_t *bp; |
1da177e4 | 700 | |
ce8e922c NS |
701 | bp = xfs_buf_allocate(0); |
702 | if (bp) | |
703 | _xfs_buf_initialize(bp, target, 0, len, 0); | |
704 | return bp; | |
1da177e4 LT |
705 | } |
706 | ||
707 | static inline struct page * | |
708 | mem_to_page( | |
709 | void *addr) | |
710 | { | |
9e2779fa | 711 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
712 | return virt_to_page(addr); |
713 | } else { | |
714 | return vmalloc_to_page(addr); | |
715 | } | |
716 | } | |
717 | ||
718 | int | |
ce8e922c NS |
719 | xfs_buf_associate_memory( |
720 | xfs_buf_t *bp, | |
1da177e4 LT |
721 | void *mem, |
722 | size_t len) | |
723 | { | |
724 | int rval; | |
725 | int i = 0; | |
d1afb678 LM |
726 | unsigned long pageaddr; |
727 | unsigned long offset; | |
728 | size_t buflen; | |
1da177e4 LT |
729 | int page_count; |
730 | ||
d1afb678 LM |
731 | pageaddr = (unsigned long)mem & PAGE_CACHE_MASK; |
732 | offset = (unsigned long)mem - pageaddr; | |
733 | buflen = PAGE_CACHE_ALIGN(len + offset); | |
734 | page_count = buflen >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
735 | |
736 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
737 | if (bp->b_pages) |
738 | _xfs_buf_free_pages(bp); | |
1da177e4 | 739 | |
ce8e922c NS |
740 | bp->b_pages = NULL; |
741 | bp->b_addr = mem; | |
1da177e4 | 742 | |
36fae17a | 743 | rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK); |
1da177e4 LT |
744 | if (rval) |
745 | return rval; | |
746 | ||
ce8e922c | 747 | bp->b_offset = offset; |
d1afb678 LM |
748 | |
749 | for (i = 0; i < bp->b_page_count; i++) { | |
750 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
751 | pageaddr += PAGE_CACHE_SIZE; | |
1da177e4 | 752 | } |
1da177e4 | 753 | |
d1afb678 LM |
754 | bp->b_count_desired = len; |
755 | bp->b_buffer_length = buflen; | |
ce8e922c | 756 | bp->b_flags |= XBF_MAPPED; |
6ab455ee | 757 | bp->b_flags &= ~_XBF_PAGE_LOCKED; |
1da177e4 LT |
758 | |
759 | return 0; | |
760 | } | |
761 | ||
762 | xfs_buf_t * | |
ce8e922c | 763 | xfs_buf_get_noaddr( |
1da177e4 LT |
764 | size_t len, |
765 | xfs_buftarg_t *target) | |
766 | { | |
1fa40b01 CH |
767 | unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT; |
768 | int error, i; | |
1da177e4 | 769 | xfs_buf_t *bp; |
1da177e4 | 770 | |
ce8e922c | 771 | bp = xfs_buf_allocate(0); |
1da177e4 LT |
772 | if (unlikely(bp == NULL)) |
773 | goto fail; | |
ce8e922c | 774 | _xfs_buf_initialize(bp, target, 0, len, 0); |
1da177e4 | 775 | |
1fa40b01 CH |
776 | error = _xfs_buf_get_pages(bp, page_count, 0); |
777 | if (error) | |
1da177e4 LT |
778 | goto fail_free_buf; |
779 | ||
1fa40b01 CH |
780 | for (i = 0; i < page_count; i++) { |
781 | bp->b_pages[i] = alloc_page(GFP_KERNEL); | |
782 | if (!bp->b_pages[i]) | |
783 | goto fail_free_mem; | |
1da177e4 | 784 | } |
1fa40b01 | 785 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 786 | |
1fa40b01 CH |
787 | error = _xfs_buf_map_pages(bp, XBF_MAPPED); |
788 | if (unlikely(error)) { | |
789 | printk(KERN_WARNING "%s: failed to map pages\n", | |
34a622b2 | 790 | __func__); |
1da177e4 | 791 | goto fail_free_mem; |
1fa40b01 | 792 | } |
1da177e4 | 793 | |
ce8e922c | 794 | xfs_buf_unlock(bp); |
1da177e4 | 795 | |
0b1b213f | 796 | trace_xfs_buf_get_noaddr(bp, _RET_IP_); |
1da177e4 | 797 | return bp; |
1fa40b01 | 798 | |
1da177e4 | 799 | fail_free_mem: |
1fa40b01 CH |
800 | while (--i >= 0) |
801 | __free_page(bp->b_pages[i]); | |
ca165b88 | 802 | _xfs_buf_free_pages(bp); |
1da177e4 | 803 | fail_free_buf: |
ca165b88 | 804 | xfs_buf_deallocate(bp); |
1da177e4 LT |
805 | fail: |
806 | return NULL; | |
807 | } | |
808 | ||
809 | /* | |
1da177e4 LT |
810 | * Increment reference count on buffer, to hold the buffer concurrently |
811 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
812 | * Must hold the buffer already to call this function. |
813 | */ | |
814 | void | |
ce8e922c NS |
815 | xfs_buf_hold( |
816 | xfs_buf_t *bp) | |
1da177e4 | 817 | { |
0b1b213f | 818 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 819 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
820 | } |
821 | ||
822 | /* | |
ce8e922c NS |
823 | * Releases a hold on the specified buffer. If the |
824 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
825 | */ |
826 | void | |
ce8e922c NS |
827 | xfs_buf_rele( |
828 | xfs_buf_t *bp) | |
1da177e4 | 829 | { |
ce8e922c | 830 | xfs_bufhash_t *hash = bp->b_hash; |
1da177e4 | 831 | |
0b1b213f | 832 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 833 | |
fad3aa1e NS |
834 | if (unlikely(!hash)) { |
835 | ASSERT(!bp->b_relse); | |
836 | if (atomic_dec_and_test(&bp->b_hold)) | |
837 | xfs_buf_free(bp); | |
838 | return; | |
839 | } | |
840 | ||
3790689f | 841 | ASSERT(atomic_read(&bp->b_hold) > 0); |
ce8e922c NS |
842 | if (atomic_dec_and_lock(&bp->b_hold, &hash->bh_lock)) { |
843 | if (bp->b_relse) { | |
844 | atomic_inc(&bp->b_hold); | |
1da177e4 | 845 | spin_unlock(&hash->bh_lock); |
ce8e922c NS |
846 | (*(bp->b_relse)) (bp); |
847 | } else if (bp->b_flags & XBF_FS_MANAGED) { | |
1da177e4 | 848 | spin_unlock(&hash->bh_lock); |
1da177e4 | 849 | } else { |
ce8e922c NS |
850 | ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q))); |
851 | list_del_init(&bp->b_hash_list); | |
1da177e4 | 852 | spin_unlock(&hash->bh_lock); |
ce8e922c | 853 | xfs_buf_free(bp); |
1da177e4 LT |
854 | } |
855 | } | |
856 | } | |
857 | ||
858 | ||
859 | /* | |
860 | * Mutual exclusion on buffers. Locking model: | |
861 | * | |
862 | * Buffers associated with inodes for which buffer locking | |
863 | * is not enabled are not protected by semaphores, and are | |
864 | * assumed to be exclusively owned by the caller. There is a | |
865 | * spinlock in the buffer, used by the caller when concurrent | |
866 | * access is possible. | |
867 | */ | |
868 | ||
869 | /* | |
ce8e922c NS |
870 | * Locks a buffer object, if it is not already locked. |
871 | * Note that this in no way locks the underlying pages, so it is only | |
872 | * useful for synchronizing concurrent use of buffer objects, not for | |
873 | * synchronizing independent access to the underlying pages. | |
1da177e4 LT |
874 | */ |
875 | int | |
ce8e922c NS |
876 | xfs_buf_cond_lock( |
877 | xfs_buf_t *bp) | |
1da177e4 LT |
878 | { |
879 | int locked; | |
880 | ||
ce8e922c | 881 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 882 | if (locked) |
ce8e922c | 883 | XB_SET_OWNER(bp); |
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. | |
1da177e4 | 901 | */ |
ce8e922c NS |
902 | void |
903 | xfs_buf_lock( | |
904 | xfs_buf_t *bp) | |
1da177e4 | 905 | { |
0b1b213f CH |
906 | trace_xfs_buf_lock(bp, _RET_IP_); |
907 | ||
ce8e922c NS |
908 | if (atomic_read(&bp->b_io_remaining)) |
909 | blk_run_address_space(bp->b_target->bt_mapping); | |
910 | down(&bp->b_sema); | |
911 | XB_SET_OWNER(bp); | |
0b1b213f CH |
912 | |
913 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
914 | } |
915 | ||
916 | /* | |
ce8e922c | 917 | * Releases the lock on the buffer object. |
2f926587 | 918 | * If the buffer is marked delwri but is not queued, do so before we |
ce8e922c | 919 | * unlock the buffer as we need to set flags correctly. We also need to |
2f926587 DC |
920 | * take a reference for the delwri queue because the unlocker is going to |
921 | * drop their's and they don't know we just queued it. | |
1da177e4 LT |
922 | */ |
923 | void | |
ce8e922c NS |
924 | xfs_buf_unlock( |
925 | xfs_buf_t *bp) | |
1da177e4 | 926 | { |
ce8e922c NS |
927 | if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) { |
928 | atomic_inc(&bp->b_hold); | |
929 | bp->b_flags |= XBF_ASYNC; | |
930 | xfs_buf_delwri_queue(bp, 0); | |
2f926587 DC |
931 | } |
932 | ||
ce8e922c NS |
933 | XB_CLEAR_OWNER(bp); |
934 | up(&bp->b_sema); | |
0b1b213f CH |
935 | |
936 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
937 | } |
938 | ||
939 | ||
940 | /* | |
941 | * Pinning Buffer Storage in Memory | |
ce8e922c | 942 | * Ensure that no attempt to force a buffer to disk will succeed. |
1da177e4 LT |
943 | */ |
944 | void | |
ce8e922c NS |
945 | xfs_buf_pin( |
946 | xfs_buf_t *bp) | |
1da177e4 | 947 | { |
0b1b213f | 948 | trace_xfs_buf_pin(bp, _RET_IP_); |
ce8e922c | 949 | atomic_inc(&bp->b_pin_count); |
1da177e4 LT |
950 | } |
951 | ||
1da177e4 | 952 | void |
ce8e922c NS |
953 | xfs_buf_unpin( |
954 | xfs_buf_t *bp) | |
1da177e4 | 955 | { |
0b1b213f CH |
956 | trace_xfs_buf_unpin(bp, _RET_IP_); |
957 | ||
ce8e922c NS |
958 | if (atomic_dec_and_test(&bp->b_pin_count)) |
959 | wake_up_all(&bp->b_waiters); | |
1da177e4 LT |
960 | } |
961 | ||
962 | int | |
ce8e922c NS |
963 | xfs_buf_ispin( |
964 | xfs_buf_t *bp) | |
1da177e4 | 965 | { |
ce8e922c | 966 | return atomic_read(&bp->b_pin_count); |
1da177e4 LT |
967 | } |
968 | ||
ce8e922c NS |
969 | STATIC void |
970 | xfs_buf_wait_unpin( | |
971 | xfs_buf_t *bp) | |
1da177e4 LT |
972 | { |
973 | DECLARE_WAITQUEUE (wait, current); | |
974 | ||
ce8e922c | 975 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
976 | return; |
977 | ||
ce8e922c | 978 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
979 | for (;;) { |
980 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 981 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 982 | break; |
ce8e922c NS |
983 | if (atomic_read(&bp->b_io_remaining)) |
984 | blk_run_address_space(bp->b_target->bt_mapping); | |
1da177e4 LT |
985 | schedule(); |
986 | } | |
ce8e922c | 987 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
988 | set_current_state(TASK_RUNNING); |
989 | } | |
990 | ||
991 | /* | |
992 | * Buffer Utility Routines | |
993 | */ | |
994 | ||
1da177e4 | 995 | STATIC void |
ce8e922c | 996 | xfs_buf_iodone_work( |
c4028958 | 997 | struct work_struct *work) |
1da177e4 | 998 | { |
c4028958 DH |
999 | xfs_buf_t *bp = |
1000 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 1001 | |
0bfefc46 DC |
1002 | /* |
1003 | * We can get an EOPNOTSUPP to ordered writes. Here we clear the | |
1004 | * ordered flag and reissue them. Because we can't tell the higher | |
1005 | * layers directly that they should not issue ordered I/O anymore, they | |
73f6aa4d | 1006 | * need to check if the _XFS_BARRIER_FAILED flag was set during I/O completion. |
0bfefc46 DC |
1007 | */ |
1008 | if ((bp->b_error == EOPNOTSUPP) && | |
1009 | (bp->b_flags & (XBF_ORDERED|XBF_ASYNC)) == (XBF_ORDERED|XBF_ASYNC)) { | |
0b1b213f | 1010 | trace_xfs_buf_ordered_retry(bp, _RET_IP_); |
0bfefc46 | 1011 | bp->b_flags &= ~XBF_ORDERED; |
73f6aa4d | 1012 | bp->b_flags |= _XFS_BARRIER_FAILED; |
0bfefc46 DC |
1013 | xfs_buf_iorequest(bp); |
1014 | } else if (bp->b_iodone) | |
ce8e922c NS |
1015 | (*(bp->b_iodone))(bp); |
1016 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
1017 | xfs_buf_relse(bp); |
1018 | } | |
1019 | ||
1020 | void | |
ce8e922c NS |
1021 | xfs_buf_ioend( |
1022 | xfs_buf_t *bp, | |
1da177e4 LT |
1023 | int schedule) |
1024 | { | |
0b1b213f CH |
1025 | trace_xfs_buf_iodone(bp, _RET_IP_); |
1026 | ||
77be55a5 | 1027 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
ce8e922c NS |
1028 | if (bp->b_error == 0) |
1029 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1030 | |
ce8e922c | 1031 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 1032 | if (schedule) { |
c4028958 | 1033 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 1034 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 1035 | } else { |
c4028958 | 1036 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
1037 | } |
1038 | } else { | |
b4dd330b | 1039 | complete(&bp->b_iowait); |
1da177e4 LT |
1040 | } |
1041 | } | |
1042 | ||
1da177e4 | 1043 | void |
ce8e922c NS |
1044 | xfs_buf_ioerror( |
1045 | xfs_buf_t *bp, | |
1046 | int error) | |
1da177e4 LT |
1047 | { |
1048 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c | 1049 | bp->b_error = (unsigned short)error; |
0b1b213f | 1050 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1051 | } |
1052 | ||
1da177e4 | 1053 | int |
5d765b97 CH |
1054 | xfs_bawrite( |
1055 | void *mp, | |
1056 | struct xfs_buf *bp) | |
1da177e4 | 1057 | { |
0b1b213f | 1058 | trace_xfs_buf_bawrite(bp, _RET_IP_); |
1da177e4 | 1059 | |
5d765b97 | 1060 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); |
1da177e4 | 1061 | |
5d765b97 | 1062 | xfs_buf_delwri_dequeue(bp); |
1da177e4 | 1063 | |
5d765b97 CH |
1064 | bp->b_flags &= ~(XBF_READ | XBF_DELWRI | XBF_READ_AHEAD); |
1065 | bp->b_flags |= (XBF_WRITE | XBF_ASYNC | _XBF_RUN_QUEUES); | |
1da177e4 | 1066 | |
15ac08a8 | 1067 | bp->b_mount = mp; |
5d765b97 CH |
1068 | bp->b_strat = xfs_bdstrat_cb; |
1069 | return xfs_bdstrat_cb(bp); | |
1070 | } | |
1da177e4 | 1071 | |
5d765b97 CH |
1072 | void |
1073 | xfs_bdwrite( | |
1074 | void *mp, | |
1075 | struct xfs_buf *bp) | |
1076 | { | |
0b1b213f | 1077 | trace_xfs_buf_bdwrite(bp, _RET_IP_); |
1da177e4 | 1078 | |
5d765b97 | 1079 | bp->b_strat = xfs_bdstrat_cb; |
15ac08a8 | 1080 | bp->b_mount = mp; |
1da177e4 | 1081 | |
5d765b97 CH |
1082 | bp->b_flags &= ~XBF_READ; |
1083 | bp->b_flags |= (XBF_DELWRI | XBF_ASYNC); | |
1084 | ||
1085 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1086 | } |
1087 | ||
b8f82a4a | 1088 | STATIC void |
ce8e922c NS |
1089 | _xfs_buf_ioend( |
1090 | xfs_buf_t *bp, | |
1da177e4 LT |
1091 | int schedule) |
1092 | { | |
6ab455ee CH |
1093 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
1094 | bp->b_flags &= ~_XBF_PAGE_LOCKED; | |
ce8e922c | 1095 | xfs_buf_ioend(bp, schedule); |
6ab455ee | 1096 | } |
1da177e4 LT |
1097 | } |
1098 | ||
782e3b3b | 1099 | STATIC void |
ce8e922c | 1100 | xfs_buf_bio_end_io( |
1da177e4 | 1101 | struct bio *bio, |
1da177e4 LT |
1102 | int error) |
1103 | { | |
ce8e922c NS |
1104 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1105 | unsigned int blocksize = bp->b_target->bt_bsize; | |
eedb5530 | 1106 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1da177e4 | 1107 | |
cfbe5267 | 1108 | xfs_buf_ioerror(bp, -error); |
1da177e4 | 1109 | |
eedb5530 | 1110 | do { |
1da177e4 LT |
1111 | struct page *page = bvec->bv_page; |
1112 | ||
948ecdb4 | 1113 | ASSERT(!PagePrivate(page)); |
ce8e922c NS |
1114 | if (unlikely(bp->b_error)) { |
1115 | if (bp->b_flags & XBF_READ) | |
eedb5530 | 1116 | ClearPageUptodate(page); |
ce8e922c | 1117 | } else if (blocksize >= PAGE_CACHE_SIZE) { |
1da177e4 LT |
1118 | SetPageUptodate(page); |
1119 | } else if (!PagePrivate(page) && | |
ce8e922c | 1120 | (bp->b_flags & _XBF_PAGE_CACHE)) { |
1da177e4 LT |
1121 | set_page_region(page, bvec->bv_offset, bvec->bv_len); |
1122 | } | |
1123 | ||
eedb5530 NS |
1124 | if (--bvec >= bio->bi_io_vec) |
1125 | prefetchw(&bvec->bv_page->flags); | |
6ab455ee CH |
1126 | |
1127 | if (bp->b_flags & _XBF_PAGE_LOCKED) | |
1128 | unlock_page(page); | |
eedb5530 | 1129 | } while (bvec >= bio->bi_io_vec); |
1da177e4 | 1130 | |
ce8e922c | 1131 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1132 | bio_put(bio); |
1da177e4 LT |
1133 | } |
1134 | ||
1135 | STATIC void | |
ce8e922c NS |
1136 | _xfs_buf_ioapply( |
1137 | xfs_buf_t *bp) | |
1da177e4 | 1138 | { |
a9759f2d | 1139 | int rw, map_i, total_nr_pages, nr_pages; |
1da177e4 | 1140 | struct bio *bio; |
ce8e922c NS |
1141 | int offset = bp->b_offset; |
1142 | int size = bp->b_count_desired; | |
1143 | sector_t sector = bp->b_bn; | |
1144 | unsigned int blocksize = bp->b_target->bt_bsize; | |
1da177e4 | 1145 | |
ce8e922c | 1146 | total_nr_pages = bp->b_page_count; |
1da177e4 LT |
1147 | map_i = 0; |
1148 | ||
ce8e922c NS |
1149 | if (bp->b_flags & XBF_ORDERED) { |
1150 | ASSERT(!(bp->b_flags & XBF_READ)); | |
f538d4da | 1151 | rw = WRITE_BARRIER; |
2ee1abad | 1152 | } else if (bp->b_flags & XBF_LOG_BUFFER) { |
51bdd706 NS |
1153 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); |
1154 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1155 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC; | |
2ee1abad DC |
1156 | } else if (bp->b_flags & _XBF_RUN_QUEUES) { |
1157 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); | |
1158 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1159 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META; | |
51bdd706 NS |
1160 | } else { |
1161 | rw = (bp->b_flags & XBF_WRITE) ? WRITE : | |
1162 | (bp->b_flags & XBF_READ_AHEAD) ? READA : READ; | |
f538d4da CH |
1163 | } |
1164 | ||
ce8e922c | 1165 | /* Special code path for reading a sub page size buffer in -- |
1da177e4 LT |
1166 | * we populate up the whole page, and hence the other metadata |
1167 | * in the same page. This optimization is only valid when the | |
ce8e922c | 1168 | * filesystem block size is not smaller than the page size. |
1da177e4 | 1169 | */ |
ce8e922c | 1170 | if ((bp->b_buffer_length < PAGE_CACHE_SIZE) && |
6ab455ee CH |
1171 | ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) == |
1172 | (XBF_READ|_XBF_PAGE_LOCKED)) && | |
ce8e922c | 1173 | (blocksize >= PAGE_CACHE_SIZE)) { |
1da177e4 LT |
1174 | bio = bio_alloc(GFP_NOIO, 1); |
1175 | ||
ce8e922c | 1176 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1177 | bio->bi_sector = sector - (offset >> BBSHIFT); |
ce8e922c NS |
1178 | bio->bi_end_io = xfs_buf_bio_end_io; |
1179 | bio->bi_private = bp; | |
1da177e4 | 1180 | |
ce8e922c | 1181 | bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0); |
1da177e4 LT |
1182 | size = 0; |
1183 | ||
ce8e922c | 1184 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1185 | |
1186 | goto submit_io; | |
1187 | } | |
1188 | ||
1da177e4 | 1189 | next_chunk: |
ce8e922c | 1190 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1191 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1192 | if (nr_pages > total_nr_pages) | |
1193 | nr_pages = total_nr_pages; | |
1194 | ||
1195 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1196 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1197 | bio->bi_sector = sector; |
ce8e922c NS |
1198 | bio->bi_end_io = xfs_buf_bio_end_io; |
1199 | bio->bi_private = bp; | |
1da177e4 LT |
1200 | |
1201 | for (; size && nr_pages; nr_pages--, map_i++) { | |
ce8e922c | 1202 | int rbytes, nbytes = PAGE_CACHE_SIZE - offset; |
1da177e4 LT |
1203 | |
1204 | if (nbytes > size) | |
1205 | nbytes = size; | |
1206 | ||
ce8e922c NS |
1207 | rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset); |
1208 | if (rbytes < nbytes) | |
1da177e4 LT |
1209 | break; |
1210 | ||
1211 | offset = 0; | |
1212 | sector += nbytes >> BBSHIFT; | |
1213 | size -= nbytes; | |
1214 | total_nr_pages--; | |
1215 | } | |
1216 | ||
1217 | submit_io: | |
1218 | if (likely(bio->bi_size)) { | |
1219 | submit_bio(rw, bio); | |
1220 | if (size) | |
1221 | goto next_chunk; | |
1222 | } else { | |
1223 | bio_put(bio); | |
ce8e922c | 1224 | xfs_buf_ioerror(bp, EIO); |
1da177e4 LT |
1225 | } |
1226 | } | |
1227 | ||
1da177e4 | 1228 | int |
ce8e922c NS |
1229 | xfs_buf_iorequest( |
1230 | xfs_buf_t *bp) | |
1da177e4 | 1231 | { |
0b1b213f | 1232 | trace_xfs_buf_iorequest(bp, _RET_IP_); |
1da177e4 | 1233 | |
ce8e922c NS |
1234 | if (bp->b_flags & XBF_DELWRI) { |
1235 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1236 | return 0; |
1237 | } | |
1238 | ||
ce8e922c NS |
1239 | if (bp->b_flags & XBF_WRITE) { |
1240 | xfs_buf_wait_unpin(bp); | |
1da177e4 LT |
1241 | } |
1242 | ||
ce8e922c | 1243 | xfs_buf_hold(bp); |
1da177e4 LT |
1244 | |
1245 | /* Set the count to 1 initially, this will stop an I/O | |
1246 | * completion callout which happens before we have started | |
ce8e922c | 1247 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1248 | */ |
ce8e922c NS |
1249 | atomic_set(&bp->b_io_remaining, 1); |
1250 | _xfs_buf_ioapply(bp); | |
1251 | _xfs_buf_ioend(bp, 0); | |
1da177e4 | 1252 | |
ce8e922c | 1253 | xfs_buf_rele(bp); |
1da177e4 LT |
1254 | return 0; |
1255 | } | |
1256 | ||
1257 | /* | |
ce8e922c NS |
1258 | * Waits for I/O to complete on the buffer supplied. |
1259 | * It returns immediately if no I/O is pending. | |
1260 | * It returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1261 | */ |
1262 | int | |
ce8e922c NS |
1263 | xfs_buf_iowait( |
1264 | xfs_buf_t *bp) | |
1da177e4 | 1265 | { |
0b1b213f CH |
1266 | trace_xfs_buf_iowait(bp, _RET_IP_); |
1267 | ||
ce8e922c NS |
1268 | if (atomic_read(&bp->b_io_remaining)) |
1269 | blk_run_address_space(bp->b_target->bt_mapping); | |
b4dd330b | 1270 | wait_for_completion(&bp->b_iowait); |
0b1b213f CH |
1271 | |
1272 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
ce8e922c | 1273 | return bp->b_error; |
1da177e4 LT |
1274 | } |
1275 | ||
ce8e922c NS |
1276 | xfs_caddr_t |
1277 | xfs_buf_offset( | |
1278 | xfs_buf_t *bp, | |
1da177e4 LT |
1279 | size_t offset) |
1280 | { | |
1281 | struct page *page; | |
1282 | ||
ce8e922c NS |
1283 | if (bp->b_flags & XBF_MAPPED) |
1284 | return XFS_BUF_PTR(bp) + offset; | |
1da177e4 | 1285 | |
ce8e922c NS |
1286 | offset += bp->b_offset; |
1287 | page = bp->b_pages[offset >> PAGE_CACHE_SHIFT]; | |
1288 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1)); | |
1da177e4 LT |
1289 | } |
1290 | ||
1291 | /* | |
1da177e4 LT |
1292 | * Move data into or out of a buffer. |
1293 | */ | |
1294 | void | |
ce8e922c NS |
1295 | xfs_buf_iomove( |
1296 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1297 | size_t boff, /* starting buffer offset */ |
1298 | size_t bsize, /* length to copy */ | |
1299 | caddr_t data, /* data address */ | |
ce8e922c | 1300 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 LT |
1301 | { |
1302 | size_t bend, cpoff, csize; | |
1303 | struct page *page; | |
1304 | ||
1305 | bend = boff + bsize; | |
1306 | while (boff < bend) { | |
ce8e922c NS |
1307 | page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)]; |
1308 | cpoff = xfs_buf_poff(boff + bp->b_offset); | |
1da177e4 | 1309 | csize = min_t(size_t, |
ce8e922c | 1310 | PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff); |
1da177e4 LT |
1311 | |
1312 | ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE)); | |
1313 | ||
1314 | switch (mode) { | |
ce8e922c | 1315 | case XBRW_ZERO: |
1da177e4 LT |
1316 | memset(page_address(page) + cpoff, 0, csize); |
1317 | break; | |
ce8e922c | 1318 | case XBRW_READ: |
1da177e4 LT |
1319 | memcpy(data, page_address(page) + cpoff, csize); |
1320 | break; | |
ce8e922c | 1321 | case XBRW_WRITE: |
1da177e4 LT |
1322 | memcpy(page_address(page) + cpoff, data, csize); |
1323 | } | |
1324 | ||
1325 | boff += csize; | |
1326 | data += csize; | |
1327 | } | |
1328 | } | |
1329 | ||
1330 | /* | |
ce8e922c | 1331 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1332 | */ |
1333 | ||
1334 | /* | |
ce8e922c NS |
1335 | * Wait for any bufs with callbacks that have been submitted but |
1336 | * have not yet returned... walk the hash list for the target. | |
1da177e4 LT |
1337 | */ |
1338 | void | |
1339 | xfs_wait_buftarg( | |
1340 | xfs_buftarg_t *btp) | |
1341 | { | |
1342 | xfs_buf_t *bp, *n; | |
1343 | xfs_bufhash_t *hash; | |
1344 | uint i; | |
1345 | ||
1346 | for (i = 0; i < (1 << btp->bt_hashshift); i++) { | |
1347 | hash = &btp->bt_hash[i]; | |
1348 | again: | |
1349 | spin_lock(&hash->bh_lock); | |
ce8e922c NS |
1350 | list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) { |
1351 | ASSERT(btp == bp->b_target); | |
1352 | if (!(bp->b_flags & XBF_FS_MANAGED)) { | |
1da177e4 | 1353 | spin_unlock(&hash->bh_lock); |
2f926587 DC |
1354 | /* |
1355 | * Catch superblock reference count leaks | |
1356 | * immediately | |
1357 | */ | |
ce8e922c | 1358 | BUG_ON(bp->b_bn == 0); |
1da177e4 LT |
1359 | delay(100); |
1360 | goto again; | |
1361 | } | |
1362 | } | |
1363 | spin_unlock(&hash->bh_lock); | |
1364 | } | |
1365 | } | |
1366 | ||
1367 | /* | |
ce8e922c NS |
1368 | * Allocate buffer hash table for a given target. |
1369 | * For devices containing metadata (i.e. not the log/realtime devices) | |
1370 | * we need to allocate a much larger hash table. | |
1da177e4 LT |
1371 | */ |
1372 | STATIC void | |
1373 | xfs_alloc_bufhash( | |
1374 | xfs_buftarg_t *btp, | |
1375 | int external) | |
1376 | { | |
1377 | unsigned int i; | |
1378 | ||
1379 | btp->bt_hashshift = external ? 3 : 8; /* 8 or 256 buckets */ | |
1380 | btp->bt_hashmask = (1 << btp->bt_hashshift) - 1; | |
1381 | btp->bt_hash = kmem_zalloc((1 << btp->bt_hashshift) * | |
93c189c1 | 1382 | sizeof(xfs_bufhash_t), KM_SLEEP | KM_LARGE); |
1da177e4 LT |
1383 | for (i = 0; i < (1 << btp->bt_hashshift); i++) { |
1384 | spin_lock_init(&btp->bt_hash[i].bh_lock); | |
1385 | INIT_LIST_HEAD(&btp->bt_hash[i].bh_list); | |
1386 | } | |
1387 | } | |
1388 | ||
1389 | STATIC void | |
1390 | xfs_free_bufhash( | |
1391 | xfs_buftarg_t *btp) | |
1392 | { | |
f0e2d93c | 1393 | kmem_free(btp->bt_hash); |
1da177e4 LT |
1394 | btp->bt_hash = NULL; |
1395 | } | |
1396 | ||
a6867a68 | 1397 | /* |
ce8e922c | 1398 | * buftarg list for delwrite queue processing |
a6867a68 | 1399 | */ |
e6a0e9cd | 1400 | static LIST_HEAD(xfs_buftarg_list); |
7989cb8e | 1401 | static DEFINE_SPINLOCK(xfs_buftarg_lock); |
a6867a68 DC |
1402 | |
1403 | STATIC void | |
1404 | xfs_register_buftarg( | |
1405 | xfs_buftarg_t *btp) | |
1406 | { | |
1407 | spin_lock(&xfs_buftarg_lock); | |
1408 | list_add(&btp->bt_list, &xfs_buftarg_list); | |
1409 | spin_unlock(&xfs_buftarg_lock); | |
1410 | } | |
1411 | ||
1412 | STATIC void | |
1413 | xfs_unregister_buftarg( | |
1414 | xfs_buftarg_t *btp) | |
1415 | { | |
1416 | spin_lock(&xfs_buftarg_lock); | |
1417 | list_del(&btp->bt_list); | |
1418 | spin_unlock(&xfs_buftarg_lock); | |
1419 | } | |
1420 | ||
1da177e4 LT |
1421 | void |
1422 | xfs_free_buftarg( | |
b7963133 CH |
1423 | struct xfs_mount *mp, |
1424 | struct xfs_buftarg *btp) | |
1da177e4 LT |
1425 | { |
1426 | xfs_flush_buftarg(btp, 1); | |
b7963133 CH |
1427 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1428 | xfs_blkdev_issue_flush(btp); | |
1da177e4 | 1429 | xfs_free_bufhash(btp); |
ce8e922c | 1430 | iput(btp->bt_mapping->host); |
a6867a68 | 1431 | |
ce8e922c NS |
1432 | /* Unregister the buftarg first so that we don't get a |
1433 | * wakeup finding a non-existent task | |
1434 | */ | |
a6867a68 DC |
1435 | xfs_unregister_buftarg(btp); |
1436 | kthread_stop(btp->bt_task); | |
1437 | ||
f0e2d93c | 1438 | kmem_free(btp); |
1da177e4 LT |
1439 | } |
1440 | ||
1da177e4 LT |
1441 | STATIC int |
1442 | xfs_setsize_buftarg_flags( | |
1443 | xfs_buftarg_t *btp, | |
1444 | unsigned int blocksize, | |
1445 | unsigned int sectorsize, | |
1446 | int verbose) | |
1447 | { | |
ce8e922c NS |
1448 | btp->bt_bsize = blocksize; |
1449 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1450 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1451 | |
ce8e922c | 1452 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
1da177e4 LT |
1453 | printk(KERN_WARNING |
1454 | "XFS: Cannot set_blocksize to %u on device %s\n", | |
1455 | sectorsize, XFS_BUFTARG_NAME(btp)); | |
1456 | return EINVAL; | |
1457 | } | |
1458 | ||
1459 | if (verbose && | |
1460 | (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) { | |
1461 | printk(KERN_WARNING | |
1462 | "XFS: %u byte sectors in use on device %s. " | |
1463 | "This is suboptimal; %u or greater is ideal.\n", | |
1464 | sectorsize, XFS_BUFTARG_NAME(btp), | |
1465 | (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG); | |
1466 | } | |
1467 | ||
1468 | return 0; | |
1469 | } | |
1470 | ||
1471 | /* | |
ce8e922c NS |
1472 | * When allocating the initial buffer target we have not yet |
1473 | * read in the superblock, so don't know what sized sectors | |
1474 | * are being used is at this early stage. Play safe. | |
1475 | */ | |
1da177e4 LT |
1476 | STATIC int |
1477 | xfs_setsize_buftarg_early( | |
1478 | xfs_buftarg_t *btp, | |
1479 | struct block_device *bdev) | |
1480 | { | |
1481 | return xfs_setsize_buftarg_flags(btp, | |
e1defc4f | 1482 | PAGE_CACHE_SIZE, bdev_logical_block_size(bdev), 0); |
1da177e4 LT |
1483 | } |
1484 | ||
1485 | int | |
1486 | xfs_setsize_buftarg( | |
1487 | xfs_buftarg_t *btp, | |
1488 | unsigned int blocksize, | |
1489 | unsigned int sectorsize) | |
1490 | { | |
1491 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1492 | } | |
1493 | ||
1494 | STATIC int | |
1495 | xfs_mapping_buftarg( | |
1496 | xfs_buftarg_t *btp, | |
1497 | struct block_device *bdev) | |
1498 | { | |
1499 | struct backing_dev_info *bdi; | |
1500 | struct inode *inode; | |
1501 | struct address_space *mapping; | |
f5e54d6e | 1502 | static const struct address_space_operations mapping_aops = { |
1da177e4 | 1503 | .sync_page = block_sync_page, |
e965f963 | 1504 | .migratepage = fail_migrate_page, |
1da177e4 LT |
1505 | }; |
1506 | ||
1507 | inode = new_inode(bdev->bd_inode->i_sb); | |
1508 | if (!inode) { | |
1509 | printk(KERN_WARNING | |
1510 | "XFS: Cannot allocate mapping inode for device %s\n", | |
1511 | XFS_BUFTARG_NAME(btp)); | |
1512 | return ENOMEM; | |
1513 | } | |
1514 | inode->i_mode = S_IFBLK; | |
1515 | inode->i_bdev = bdev; | |
1516 | inode->i_rdev = bdev->bd_dev; | |
1517 | bdi = blk_get_backing_dev_info(bdev); | |
1518 | if (!bdi) | |
1519 | bdi = &default_backing_dev_info; | |
1520 | mapping = &inode->i_data; | |
1521 | mapping->a_ops = &mapping_aops; | |
1522 | mapping->backing_dev_info = bdi; | |
1523 | mapping_set_gfp_mask(mapping, GFP_NOFS); | |
ce8e922c | 1524 | btp->bt_mapping = mapping; |
1da177e4 LT |
1525 | return 0; |
1526 | } | |
1527 | ||
a6867a68 DC |
1528 | STATIC int |
1529 | xfs_alloc_delwrite_queue( | |
1530 | xfs_buftarg_t *btp) | |
1531 | { | |
1532 | int error = 0; | |
1533 | ||
1534 | INIT_LIST_HEAD(&btp->bt_list); | |
1535 | INIT_LIST_HEAD(&btp->bt_delwrite_queue); | |
007c61c6 | 1536 | spin_lock_init(&btp->bt_delwrite_lock); |
a6867a68 DC |
1537 | btp->bt_flags = 0; |
1538 | btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd"); | |
1539 | if (IS_ERR(btp->bt_task)) { | |
1540 | error = PTR_ERR(btp->bt_task); | |
1541 | goto out_error; | |
1542 | } | |
1543 | xfs_register_buftarg(btp); | |
1544 | out_error: | |
1545 | return error; | |
1546 | } | |
1547 | ||
1da177e4 LT |
1548 | xfs_buftarg_t * |
1549 | xfs_alloc_buftarg( | |
1550 | struct block_device *bdev, | |
1551 | int external) | |
1552 | { | |
1553 | xfs_buftarg_t *btp; | |
1554 | ||
1555 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1556 | ||
ce8e922c NS |
1557 | btp->bt_dev = bdev->bd_dev; |
1558 | btp->bt_bdev = bdev; | |
1da177e4 LT |
1559 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1560 | goto error; | |
1561 | if (xfs_mapping_buftarg(btp, bdev)) | |
1562 | goto error; | |
a6867a68 DC |
1563 | if (xfs_alloc_delwrite_queue(btp)) |
1564 | goto error; | |
1da177e4 LT |
1565 | xfs_alloc_bufhash(btp, external); |
1566 | return btp; | |
1567 | ||
1568 | error: | |
f0e2d93c | 1569 | kmem_free(btp); |
1da177e4 LT |
1570 | return NULL; |
1571 | } | |
1572 | ||
1573 | ||
1574 | /* | |
ce8e922c | 1575 | * Delayed write buffer handling |
1da177e4 | 1576 | */ |
1da177e4 | 1577 | STATIC void |
ce8e922c NS |
1578 | xfs_buf_delwri_queue( |
1579 | xfs_buf_t *bp, | |
1da177e4 LT |
1580 | int unlock) |
1581 | { | |
ce8e922c NS |
1582 | struct list_head *dwq = &bp->b_target->bt_delwrite_queue; |
1583 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; | |
a6867a68 | 1584 | |
0b1b213f CH |
1585 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
1586 | ||
ce8e922c | 1587 | ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC)); |
1da177e4 | 1588 | |
a6867a68 | 1589 | spin_lock(dwlk); |
1da177e4 | 1590 | /* If already in the queue, dequeue and place at tail */ |
ce8e922c NS |
1591 | if (!list_empty(&bp->b_list)) { |
1592 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1593 | if (unlock) | |
1594 | atomic_dec(&bp->b_hold); | |
1595 | list_del(&bp->b_list); | |
1da177e4 LT |
1596 | } |
1597 | ||
c9c12971 DC |
1598 | if (list_empty(dwq)) { |
1599 | /* start xfsbufd as it is about to have something to do */ | |
1600 | wake_up_process(bp->b_target->bt_task); | |
1601 | } | |
1602 | ||
ce8e922c NS |
1603 | bp->b_flags |= _XBF_DELWRI_Q; |
1604 | list_add_tail(&bp->b_list, dwq); | |
1605 | bp->b_queuetime = jiffies; | |
a6867a68 | 1606 | spin_unlock(dwlk); |
1da177e4 LT |
1607 | |
1608 | if (unlock) | |
ce8e922c | 1609 | xfs_buf_unlock(bp); |
1da177e4 LT |
1610 | } |
1611 | ||
1612 | void | |
ce8e922c NS |
1613 | xfs_buf_delwri_dequeue( |
1614 | xfs_buf_t *bp) | |
1da177e4 | 1615 | { |
ce8e922c | 1616 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; |
1da177e4 LT |
1617 | int dequeued = 0; |
1618 | ||
a6867a68 | 1619 | spin_lock(dwlk); |
ce8e922c NS |
1620 | if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) { |
1621 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1622 | list_del_init(&bp->b_list); | |
1da177e4 LT |
1623 | dequeued = 1; |
1624 | } | |
ce8e922c | 1625 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q); |
a6867a68 | 1626 | spin_unlock(dwlk); |
1da177e4 LT |
1627 | |
1628 | if (dequeued) | |
ce8e922c | 1629 | xfs_buf_rele(bp); |
1da177e4 | 1630 | |
0b1b213f | 1631 | trace_xfs_buf_delwri_dequeue(bp, _RET_IP_); |
1da177e4 LT |
1632 | } |
1633 | ||
1634 | STATIC void | |
ce8e922c | 1635 | xfs_buf_runall_queues( |
1da177e4 LT |
1636 | struct workqueue_struct *queue) |
1637 | { | |
1638 | flush_workqueue(queue); | |
1639 | } | |
1640 | ||
1da177e4 | 1641 | STATIC int |
23ea4032 | 1642 | xfsbufd_wakeup( |
15c84a47 NS |
1643 | int priority, |
1644 | gfp_t mask) | |
1da177e4 | 1645 | { |
da7f93e9 | 1646 | xfs_buftarg_t *btp; |
a6867a68 DC |
1647 | |
1648 | spin_lock(&xfs_buftarg_lock); | |
da7f93e9 | 1649 | list_for_each_entry(btp, &xfs_buftarg_list, bt_list) { |
ce8e922c | 1650 | if (test_bit(XBT_FORCE_SLEEP, &btp->bt_flags)) |
a6867a68 | 1651 | continue; |
c9c12971 DC |
1652 | if (list_empty(&btp->bt_delwrite_queue)) |
1653 | continue; | |
ce8e922c | 1654 | set_bit(XBT_FORCE_FLUSH, &btp->bt_flags); |
a6867a68 DC |
1655 | wake_up_process(btp->bt_task); |
1656 | } | |
1657 | spin_unlock(&xfs_buftarg_lock); | |
1da177e4 LT |
1658 | return 0; |
1659 | } | |
1660 | ||
585e6d88 DC |
1661 | /* |
1662 | * Move as many buffers as specified to the supplied list | |
1663 | * idicating if we skipped any buffers to prevent deadlocks. | |
1664 | */ | |
1665 | STATIC int | |
1666 | xfs_buf_delwri_split( | |
1667 | xfs_buftarg_t *target, | |
1668 | struct list_head *list, | |
5e6a07df | 1669 | unsigned long age) |
585e6d88 DC |
1670 | { |
1671 | xfs_buf_t *bp, *n; | |
1672 | struct list_head *dwq = &target->bt_delwrite_queue; | |
1673 | spinlock_t *dwlk = &target->bt_delwrite_lock; | |
1674 | int skipped = 0; | |
5e6a07df | 1675 | int force; |
585e6d88 | 1676 | |
5e6a07df | 1677 | force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
585e6d88 DC |
1678 | INIT_LIST_HEAD(list); |
1679 | spin_lock(dwlk); | |
1680 | list_for_each_entry_safe(bp, n, dwq, b_list) { | |
0b1b213f | 1681 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
585e6d88 DC |
1682 | ASSERT(bp->b_flags & XBF_DELWRI); |
1683 | ||
1684 | if (!xfs_buf_ispin(bp) && !xfs_buf_cond_lock(bp)) { | |
5e6a07df | 1685 | if (!force && |
585e6d88 DC |
1686 | time_before(jiffies, bp->b_queuetime + age)) { |
1687 | xfs_buf_unlock(bp); | |
1688 | break; | |
1689 | } | |
1690 | ||
1691 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q| | |
1692 | _XBF_RUN_QUEUES); | |
1693 | bp->b_flags |= XBF_WRITE; | |
1694 | list_move_tail(&bp->b_list, list); | |
1695 | } else | |
1696 | skipped++; | |
1697 | } | |
1698 | spin_unlock(dwlk); | |
1699 | ||
1700 | return skipped; | |
1701 | ||
1702 | } | |
1703 | ||
1da177e4 | 1704 | STATIC int |
23ea4032 | 1705 | xfsbufd( |
585e6d88 | 1706 | void *data) |
1da177e4 | 1707 | { |
585e6d88 DC |
1708 | struct list_head tmp; |
1709 | xfs_buftarg_t *target = (xfs_buftarg_t *)data; | |
1710 | int count; | |
1711 | xfs_buf_t *bp; | |
1da177e4 | 1712 | |
1da177e4 LT |
1713 | current->flags |= PF_MEMALLOC; |
1714 | ||
978c7b2f RW |
1715 | set_freezable(); |
1716 | ||
1da177e4 | 1717 | do { |
c9c12971 DC |
1718 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10); |
1719 | long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10); | |
1720 | ||
3e1d1d28 | 1721 | if (unlikely(freezing(current))) { |
ce8e922c | 1722 | set_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
3e1d1d28 | 1723 | refrigerator(); |
abd0cf7a | 1724 | } else { |
ce8e922c | 1725 | clear_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
abd0cf7a | 1726 | } |
1da177e4 | 1727 | |
c9c12971 DC |
1728 | /* sleep for a long time if there is nothing to do. */ |
1729 | if (list_empty(&target->bt_delwrite_queue)) | |
1730 | tout = MAX_SCHEDULE_TIMEOUT; | |
1731 | schedule_timeout_interruptible(tout); | |
1da177e4 | 1732 | |
c9c12971 | 1733 | xfs_buf_delwri_split(target, &tmp, age); |
585e6d88 | 1734 | count = 0; |
1da177e4 | 1735 | while (!list_empty(&tmp)) { |
ce8e922c NS |
1736 | bp = list_entry(tmp.next, xfs_buf_t, b_list); |
1737 | ASSERT(target == bp->b_target); | |
1da177e4 | 1738 | |
ce8e922c NS |
1739 | list_del_init(&bp->b_list); |
1740 | xfs_buf_iostrategy(bp); | |
585e6d88 | 1741 | count++; |
1da177e4 LT |
1742 | } |
1743 | ||
3a011a17 FB |
1744 | if (as_list_len > 0) |
1745 | purge_addresses(); | |
f07c2250 NS |
1746 | if (count) |
1747 | blk_run_address_space(target->bt_mapping); | |
1da177e4 | 1748 | |
4df08c52 | 1749 | } while (!kthread_should_stop()); |
1da177e4 | 1750 | |
4df08c52 | 1751 | return 0; |
1da177e4 LT |
1752 | } |
1753 | ||
1754 | /* | |
ce8e922c NS |
1755 | * Go through all incore buffers, and release buffers if they belong to |
1756 | * the given device. This is used in filesystem error handling to | |
1757 | * preserve the consistency of its metadata. | |
1da177e4 LT |
1758 | */ |
1759 | int | |
1760 | xfs_flush_buftarg( | |
585e6d88 DC |
1761 | xfs_buftarg_t *target, |
1762 | int wait) | |
1da177e4 | 1763 | { |
585e6d88 DC |
1764 | struct list_head tmp; |
1765 | xfs_buf_t *bp, *n; | |
1766 | int pincount = 0; | |
1da177e4 | 1767 | |
c626d174 | 1768 | xfs_buf_runall_queues(xfsconvertd_workqueue); |
ce8e922c NS |
1769 | xfs_buf_runall_queues(xfsdatad_workqueue); |
1770 | xfs_buf_runall_queues(xfslogd_workqueue); | |
1da177e4 | 1771 | |
5e6a07df DC |
1772 | set_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
1773 | pincount = xfs_buf_delwri_split(target, &tmp, 0); | |
1da177e4 LT |
1774 | |
1775 | /* | |
1776 | * Dropped the delayed write list lock, now walk the temporary list | |
1777 | */ | |
ce8e922c | 1778 | list_for_each_entry_safe(bp, n, &tmp, b_list) { |
585e6d88 | 1779 | ASSERT(target == bp->b_target); |
1da177e4 | 1780 | if (wait) |
ce8e922c | 1781 | bp->b_flags &= ~XBF_ASYNC; |
1da177e4 | 1782 | else |
ce8e922c | 1783 | list_del_init(&bp->b_list); |
1da177e4 | 1784 | |
ce8e922c | 1785 | xfs_buf_iostrategy(bp); |
1da177e4 LT |
1786 | } |
1787 | ||
f07c2250 NS |
1788 | if (wait) |
1789 | blk_run_address_space(target->bt_mapping); | |
1790 | ||
1da177e4 LT |
1791 | /* |
1792 | * Remaining list items must be flushed before returning | |
1793 | */ | |
1794 | while (!list_empty(&tmp)) { | |
ce8e922c | 1795 | bp = list_entry(tmp.next, xfs_buf_t, b_list); |
1da177e4 | 1796 | |
ce8e922c NS |
1797 | list_del_init(&bp->b_list); |
1798 | xfs_iowait(bp); | |
1799 | xfs_buf_relse(bp); | |
1da177e4 LT |
1800 | } |
1801 | ||
1da177e4 LT |
1802 | return pincount; |
1803 | } | |
1804 | ||
04d8b284 | 1805 | int __init |
ce8e922c | 1806 | xfs_buf_init(void) |
1da177e4 | 1807 | { |
8758280f NS |
1808 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1809 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1810 | if (!xfs_buf_zone) |
0b1b213f | 1811 | goto out; |
04d8b284 | 1812 | |
b4337692 | 1813 | xfslogd_workqueue = create_workqueue("xfslogd"); |
23ea4032 | 1814 | if (!xfslogd_workqueue) |
04d8b284 | 1815 | goto out_free_buf_zone; |
1da177e4 | 1816 | |
b4337692 | 1817 | xfsdatad_workqueue = create_workqueue("xfsdatad"); |
23ea4032 CH |
1818 | if (!xfsdatad_workqueue) |
1819 | goto out_destroy_xfslogd_workqueue; | |
1da177e4 | 1820 | |
c626d174 DC |
1821 | xfsconvertd_workqueue = create_workqueue("xfsconvertd"); |
1822 | if (!xfsconvertd_workqueue) | |
1823 | goto out_destroy_xfsdatad_workqueue; | |
1824 | ||
8e1f936b | 1825 | register_shrinker(&xfs_buf_shake); |
23ea4032 | 1826 | return 0; |
1da177e4 | 1827 | |
c626d174 DC |
1828 | out_destroy_xfsdatad_workqueue: |
1829 | destroy_workqueue(xfsdatad_workqueue); | |
23ea4032 CH |
1830 | out_destroy_xfslogd_workqueue: |
1831 | destroy_workqueue(xfslogd_workqueue); | |
23ea4032 | 1832 | out_free_buf_zone: |
ce8e922c | 1833 | kmem_zone_destroy(xfs_buf_zone); |
0b1b213f | 1834 | out: |
8758280f | 1835 | return -ENOMEM; |
1da177e4 LT |
1836 | } |
1837 | ||
1da177e4 | 1838 | void |
ce8e922c | 1839 | xfs_buf_terminate(void) |
1da177e4 | 1840 | { |
8e1f936b | 1841 | unregister_shrinker(&xfs_buf_shake); |
c626d174 | 1842 | destroy_workqueue(xfsconvertd_workqueue); |
04d8b284 CH |
1843 | destroy_workqueue(xfsdatad_workqueue); |
1844 | destroy_workqueue(xfslogd_workqueue); | |
ce8e922c | 1845 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 1846 | } |
e6a0e9cd TS |
1847 | |
1848 | #ifdef CONFIG_KDB_MODULES | |
1849 | struct list_head * | |
1850 | xfs_get_buftarg_list(void) | |
1851 | { | |
1852 | return &xfs_buftarg_list; | |
1853 | } | |
1854 | #endif |