]>
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, | |
61 | (void *)(unsigned long)bp->b_sema.count.counter, | |
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 | ||
168 | /* | |
ce8e922c | 169 | * Mapping of multi-page buffers into contiguous virtual space |
1da177e4 LT |
170 | */ |
171 | ||
172 | typedef struct a_list { | |
173 | void *vm_addr; | |
174 | struct a_list *next; | |
175 | } a_list_t; | |
176 | ||
7989cb8e DC |
177 | static a_list_t *as_free_head; |
178 | static int as_list_len; | |
179 | static DEFINE_SPINLOCK(as_lock); | |
1da177e4 LT |
180 | |
181 | /* | |
ce8e922c | 182 | * Try to batch vunmaps because they are costly. |
1da177e4 LT |
183 | */ |
184 | STATIC void | |
185 | free_address( | |
186 | void *addr) | |
187 | { | |
188 | a_list_t *aentry; | |
189 | ||
7f015072 JF |
190 | #ifdef CONFIG_XEN |
191 | /* | |
192 | * Xen needs to be able to make sure it can get an exclusive | |
193 | * RO mapping of pages it wants to turn into a pagetable. If | |
194 | * a newly allocated page is also still being vmap()ed by xfs, | |
195 | * it will cause pagetable construction to fail. This is a | |
196 | * quick workaround to always eagerly unmap pages so that Xen | |
197 | * is happy. | |
198 | */ | |
199 | vunmap(addr); | |
200 | return; | |
201 | #endif | |
202 | ||
7b04d717 | 203 | aentry = kmalloc(sizeof(a_list_t), GFP_NOWAIT); |
1da177e4 LT |
204 | if (likely(aentry)) { |
205 | spin_lock(&as_lock); | |
206 | aentry->next = as_free_head; | |
207 | aentry->vm_addr = addr; | |
208 | as_free_head = aentry; | |
209 | as_list_len++; | |
210 | spin_unlock(&as_lock); | |
211 | } else { | |
212 | vunmap(addr); | |
213 | } | |
214 | } | |
215 | ||
216 | STATIC void | |
217 | purge_addresses(void) | |
218 | { | |
219 | a_list_t *aentry, *old; | |
220 | ||
221 | if (as_free_head == NULL) | |
222 | return; | |
223 | ||
224 | spin_lock(&as_lock); | |
225 | aentry = as_free_head; | |
226 | as_free_head = NULL; | |
227 | as_list_len = 0; | |
228 | spin_unlock(&as_lock); | |
229 | ||
230 | while ((old = aentry) != NULL) { | |
231 | vunmap(aentry->vm_addr); | |
232 | aentry = aentry->next; | |
233 | kfree(old); | |
234 | } | |
235 | } | |
236 | ||
237 | /* | |
ce8e922c | 238 | * Internal xfs_buf_t object manipulation |
1da177e4 LT |
239 | */ |
240 | ||
241 | STATIC void | |
ce8e922c NS |
242 | _xfs_buf_initialize( |
243 | xfs_buf_t *bp, | |
1da177e4 | 244 | xfs_buftarg_t *target, |
204ab25f | 245 | xfs_off_t range_base, |
1da177e4 | 246 | size_t range_length, |
ce8e922c | 247 | xfs_buf_flags_t flags) |
1da177e4 LT |
248 | { |
249 | /* | |
ce8e922c | 250 | * We don't want certain flags to appear in b_flags. |
1da177e4 | 251 | */ |
ce8e922c NS |
252 | flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD); |
253 | ||
254 | memset(bp, 0, sizeof(xfs_buf_t)); | |
255 | atomic_set(&bp->b_hold, 1); | |
256 | init_MUTEX_LOCKED(&bp->b_iodonesema); | |
257 | INIT_LIST_HEAD(&bp->b_list); | |
258 | INIT_LIST_HEAD(&bp->b_hash_list); | |
259 | init_MUTEX_LOCKED(&bp->b_sema); /* held, no waiters */ | |
260 | XB_SET_OWNER(bp); | |
261 | bp->b_target = target; | |
262 | bp->b_file_offset = range_base; | |
1da177e4 LT |
263 | /* |
264 | * Set buffer_length and count_desired to the same value initially. | |
265 | * I/O routines should use count_desired, which will be the same in | |
266 | * most cases but may be reset (e.g. XFS recovery). | |
267 | */ | |
ce8e922c NS |
268 | bp->b_buffer_length = bp->b_count_desired = range_length; |
269 | bp->b_flags = flags; | |
270 | bp->b_bn = XFS_BUF_DADDR_NULL; | |
271 | atomic_set(&bp->b_pin_count, 0); | |
272 | init_waitqueue_head(&bp->b_waiters); | |
273 | ||
274 | XFS_STATS_INC(xb_create); | |
275 | XB_TRACE(bp, "initialize", target); | |
1da177e4 LT |
276 | } |
277 | ||
278 | /* | |
ce8e922c NS |
279 | * Allocate a page array capable of holding a specified number |
280 | * of pages, and point the page buf at it. | |
1da177e4 LT |
281 | */ |
282 | STATIC int | |
ce8e922c NS |
283 | _xfs_buf_get_pages( |
284 | xfs_buf_t *bp, | |
1da177e4 | 285 | int page_count, |
ce8e922c | 286 | xfs_buf_flags_t flags) |
1da177e4 LT |
287 | { |
288 | /* Make sure that we have a page list */ | |
ce8e922c NS |
289 | if (bp->b_pages == NULL) { |
290 | bp->b_offset = xfs_buf_poff(bp->b_file_offset); | |
291 | bp->b_page_count = page_count; | |
292 | if (page_count <= XB_PAGES) { | |
293 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 294 | } else { |
ce8e922c NS |
295 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
296 | page_count, xb_to_km(flags)); | |
297 | if (bp->b_pages == NULL) | |
1da177e4 LT |
298 | return -ENOMEM; |
299 | } | |
ce8e922c | 300 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
301 | } |
302 | return 0; | |
303 | } | |
304 | ||
305 | /* | |
ce8e922c | 306 | * Frees b_pages if it was allocated. |
1da177e4 LT |
307 | */ |
308 | STATIC void | |
ce8e922c | 309 | _xfs_buf_free_pages( |
1da177e4 LT |
310 | xfs_buf_t *bp) |
311 | { | |
ce8e922c NS |
312 | if (bp->b_pages != bp->b_page_array) { |
313 | kmem_free(bp->b_pages, | |
314 | bp->b_page_count * sizeof(struct page *)); | |
1da177e4 LT |
315 | } |
316 | } | |
317 | ||
318 | /* | |
319 | * Releases the specified buffer. | |
320 | * | |
321 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 322 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
323 | * hashed and refcounted buffers |
324 | */ | |
325 | void | |
ce8e922c | 326 | xfs_buf_free( |
1da177e4 LT |
327 | xfs_buf_t *bp) |
328 | { | |
ce8e922c | 329 | XB_TRACE(bp, "free", 0); |
1da177e4 | 330 | |
ce8e922c | 331 | ASSERT(list_empty(&bp->b_hash_list)); |
1da177e4 | 332 | |
1fa40b01 | 333 | if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) { |
1da177e4 LT |
334 | uint i; |
335 | ||
ce8e922c NS |
336 | if ((bp->b_flags & XBF_MAPPED) && (bp->b_page_count > 1)) |
337 | free_address(bp->b_addr - bp->b_offset); | |
1da177e4 | 338 | |
948ecdb4 NS |
339 | for (i = 0; i < bp->b_page_count; i++) { |
340 | struct page *page = bp->b_pages[i]; | |
341 | ||
1fa40b01 CH |
342 | if (bp->b_flags & _XBF_PAGE_CACHE) |
343 | ASSERT(!PagePrivate(page)); | |
948ecdb4 NS |
344 | page_cache_release(page); |
345 | } | |
ce8e922c | 346 | _xfs_buf_free_pages(bp); |
1da177e4 LT |
347 | } |
348 | ||
ce8e922c | 349 | xfs_buf_deallocate(bp); |
1da177e4 LT |
350 | } |
351 | ||
352 | /* | |
353 | * Finds all pages for buffer in question and builds it's page list. | |
354 | */ | |
355 | STATIC int | |
ce8e922c | 356 | _xfs_buf_lookup_pages( |
1da177e4 LT |
357 | xfs_buf_t *bp, |
358 | uint flags) | |
359 | { | |
ce8e922c NS |
360 | struct address_space *mapping = bp->b_target->bt_mapping; |
361 | size_t blocksize = bp->b_target->bt_bsize; | |
362 | size_t size = bp->b_count_desired; | |
1da177e4 | 363 | size_t nbytes, offset; |
ce8e922c | 364 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 LT |
365 | unsigned short page_count, i; |
366 | pgoff_t first; | |
204ab25f | 367 | xfs_off_t end; |
1da177e4 LT |
368 | int error; |
369 | ||
ce8e922c NS |
370 | end = bp->b_file_offset + bp->b_buffer_length; |
371 | page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset); | |
1da177e4 | 372 | |
ce8e922c | 373 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
374 | if (unlikely(error)) |
375 | return error; | |
ce8e922c | 376 | bp->b_flags |= _XBF_PAGE_CACHE; |
1da177e4 | 377 | |
ce8e922c NS |
378 | offset = bp->b_offset; |
379 | first = bp->b_file_offset >> PAGE_CACHE_SHIFT; | |
1da177e4 | 380 | |
ce8e922c | 381 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
382 | struct page *page; |
383 | uint retries = 0; | |
384 | ||
385 | retry: | |
386 | page = find_or_create_page(mapping, first + i, gfp_mask); | |
387 | if (unlikely(page == NULL)) { | |
ce8e922c NS |
388 | if (flags & XBF_READ_AHEAD) { |
389 | bp->b_page_count = i; | |
390 | for (i = 0; i < bp->b_page_count; i++) | |
391 | unlock_page(bp->b_pages[i]); | |
1da177e4 LT |
392 | return -ENOMEM; |
393 | } | |
394 | ||
395 | /* | |
396 | * This could deadlock. | |
397 | * | |
398 | * But until all the XFS lowlevel code is revamped to | |
399 | * handle buffer allocation failures we can't do much. | |
400 | */ | |
401 | if (!(++retries % 100)) | |
402 | printk(KERN_ERR | |
403 | "XFS: possible memory allocation " | |
404 | "deadlock in %s (mode:0x%x)\n", | |
405 | __FUNCTION__, gfp_mask); | |
406 | ||
ce8e922c | 407 | XFS_STATS_INC(xb_page_retries); |
23ea4032 | 408 | xfsbufd_wakeup(0, gfp_mask); |
3fcfab16 | 409 | congestion_wait(WRITE, HZ/50); |
1da177e4 LT |
410 | goto retry; |
411 | } | |
412 | ||
ce8e922c | 413 | XFS_STATS_INC(xb_page_found); |
1da177e4 LT |
414 | |
415 | nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset); | |
416 | size -= nbytes; | |
417 | ||
948ecdb4 | 418 | ASSERT(!PagePrivate(page)); |
1da177e4 LT |
419 | if (!PageUptodate(page)) { |
420 | page_count--; | |
421 | if (blocksize >= PAGE_CACHE_SIZE) { | |
ce8e922c NS |
422 | if (flags & XBF_READ) |
423 | bp->b_locked = 1; | |
1da177e4 LT |
424 | } else if (!PagePrivate(page)) { |
425 | if (test_page_region(page, offset, nbytes)) | |
426 | page_count++; | |
427 | } | |
428 | } | |
429 | ||
ce8e922c | 430 | bp->b_pages[i] = page; |
1da177e4 LT |
431 | offset = 0; |
432 | } | |
433 | ||
ce8e922c NS |
434 | if (!bp->b_locked) { |
435 | for (i = 0; i < bp->b_page_count; i++) | |
436 | unlock_page(bp->b_pages[i]); | |
1da177e4 LT |
437 | } |
438 | ||
ce8e922c NS |
439 | if (page_count == bp->b_page_count) |
440 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 441 | |
ce8e922c | 442 | XB_TRACE(bp, "lookup_pages", (long)page_count); |
1da177e4 LT |
443 | return error; |
444 | } | |
445 | ||
446 | /* | |
447 | * Map buffer into kernel address-space if nessecary. | |
448 | */ | |
449 | STATIC int | |
ce8e922c | 450 | _xfs_buf_map_pages( |
1da177e4 LT |
451 | xfs_buf_t *bp, |
452 | uint flags) | |
453 | { | |
454 | /* A single page buffer is always mappable */ | |
ce8e922c NS |
455 | if (bp->b_page_count == 1) { |
456 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; | |
457 | bp->b_flags |= XBF_MAPPED; | |
458 | } else if (flags & XBF_MAPPED) { | |
1da177e4 LT |
459 | if (as_list_len > 64) |
460 | purge_addresses(); | |
ce8e922c NS |
461 | bp->b_addr = vmap(bp->b_pages, bp->b_page_count, |
462 | VM_MAP, PAGE_KERNEL); | |
463 | if (unlikely(bp->b_addr == NULL)) | |
1da177e4 | 464 | return -ENOMEM; |
ce8e922c NS |
465 | bp->b_addr += bp->b_offset; |
466 | bp->b_flags |= XBF_MAPPED; | |
1da177e4 LT |
467 | } |
468 | ||
469 | return 0; | |
470 | } | |
471 | ||
472 | /* | |
473 | * Finding and Reading Buffers | |
474 | */ | |
475 | ||
476 | /* | |
ce8e922c | 477 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 LT |
478 | * a given range of an inode. The buffer is returned |
479 | * locked. If other overlapping buffers exist, they are | |
480 | * released before the new buffer is created and locked, | |
481 | * which may imply that this call will block until those buffers | |
482 | * are unlocked. No I/O is implied by this call. | |
483 | */ | |
484 | xfs_buf_t * | |
ce8e922c | 485 | _xfs_buf_find( |
1da177e4 | 486 | xfs_buftarg_t *btp, /* block device target */ |
204ab25f | 487 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 488 | size_t isize, /* length of range */ |
ce8e922c NS |
489 | xfs_buf_flags_t flags, |
490 | xfs_buf_t *new_bp) | |
1da177e4 | 491 | { |
204ab25f | 492 | xfs_off_t range_base; |
1da177e4 LT |
493 | size_t range_length; |
494 | xfs_bufhash_t *hash; | |
ce8e922c | 495 | xfs_buf_t *bp, *n; |
1da177e4 LT |
496 | |
497 | range_base = (ioff << BBSHIFT); | |
498 | range_length = (isize << BBSHIFT); | |
499 | ||
500 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
ce8e922c | 501 | ASSERT(!(range_length < (1 << btp->bt_sshift))); |
204ab25f | 502 | ASSERT(!(range_base & (xfs_off_t)btp->bt_smask)); |
1da177e4 LT |
503 | |
504 | hash = &btp->bt_hash[hash_long((unsigned long)ioff, btp->bt_hashshift)]; | |
505 | ||
506 | spin_lock(&hash->bh_lock); | |
507 | ||
ce8e922c NS |
508 | list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) { |
509 | ASSERT(btp == bp->b_target); | |
510 | if (bp->b_file_offset == range_base && | |
511 | bp->b_buffer_length == range_length) { | |
1da177e4 | 512 | /* |
ce8e922c | 513 | * If we look at something, bring it to the |
1da177e4 LT |
514 | * front of the list for next time. |
515 | */ | |
ce8e922c NS |
516 | atomic_inc(&bp->b_hold); |
517 | list_move(&bp->b_hash_list, &hash->bh_list); | |
1da177e4 LT |
518 | goto found; |
519 | } | |
520 | } | |
521 | ||
522 | /* No match found */ | |
ce8e922c NS |
523 | if (new_bp) { |
524 | _xfs_buf_initialize(new_bp, btp, range_base, | |
1da177e4 | 525 | range_length, flags); |
ce8e922c NS |
526 | new_bp->b_hash = hash; |
527 | list_add(&new_bp->b_hash_list, &hash->bh_list); | |
1da177e4 | 528 | } else { |
ce8e922c | 529 | XFS_STATS_INC(xb_miss_locked); |
1da177e4 LT |
530 | } |
531 | ||
532 | spin_unlock(&hash->bh_lock); | |
ce8e922c | 533 | return new_bp; |
1da177e4 LT |
534 | |
535 | found: | |
536 | spin_unlock(&hash->bh_lock); | |
537 | ||
538 | /* Attempt to get the semaphore without sleeping, | |
539 | * if this does not work then we need to drop the | |
540 | * spinlock and do a hard attempt on the semaphore. | |
541 | */ | |
ce8e922c NS |
542 | if (down_trylock(&bp->b_sema)) { |
543 | if (!(flags & XBF_TRYLOCK)) { | |
1da177e4 | 544 | /* wait for buffer ownership */ |
ce8e922c NS |
545 | XB_TRACE(bp, "get_lock", 0); |
546 | xfs_buf_lock(bp); | |
547 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
548 | } else { |
549 | /* We asked for a trylock and failed, no need | |
550 | * to look at file offset and length here, we | |
ce8e922c NS |
551 | * know that this buffer at least overlaps our |
552 | * buffer and is locked, therefore our buffer | |
553 | * either does not exist, or is this buffer. | |
1da177e4 | 554 | */ |
ce8e922c NS |
555 | xfs_buf_rele(bp); |
556 | XFS_STATS_INC(xb_busy_locked); | |
557 | return NULL; | |
1da177e4 LT |
558 | } |
559 | } else { | |
560 | /* trylock worked */ | |
ce8e922c | 561 | XB_SET_OWNER(bp); |
1da177e4 LT |
562 | } |
563 | ||
ce8e922c NS |
564 | if (bp->b_flags & XBF_STALE) { |
565 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
566 | bp->b_flags &= XBF_MAPPED; | |
2f926587 | 567 | } |
ce8e922c NS |
568 | XB_TRACE(bp, "got_lock", 0); |
569 | XFS_STATS_INC(xb_get_locked); | |
570 | return bp; | |
1da177e4 LT |
571 | } |
572 | ||
573 | /* | |
ce8e922c | 574 | * Assembles a buffer covering the specified range. |
1da177e4 LT |
575 | * Storage in memory for all portions of the buffer will be allocated, |
576 | * although backing storage may not be. | |
577 | */ | |
578 | xfs_buf_t * | |
ce8e922c | 579 | xfs_buf_get_flags( |
1da177e4 | 580 | xfs_buftarg_t *target,/* target for buffer */ |
204ab25f | 581 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 582 | size_t isize, /* length of range */ |
ce8e922c | 583 | xfs_buf_flags_t flags) |
1da177e4 | 584 | { |
ce8e922c | 585 | xfs_buf_t *bp, *new_bp; |
1da177e4 LT |
586 | int error = 0, i; |
587 | ||
ce8e922c NS |
588 | new_bp = xfs_buf_allocate(flags); |
589 | if (unlikely(!new_bp)) | |
1da177e4 LT |
590 | return NULL; |
591 | ||
ce8e922c NS |
592 | bp = _xfs_buf_find(target, ioff, isize, flags, new_bp); |
593 | if (bp == new_bp) { | |
594 | error = _xfs_buf_lookup_pages(bp, flags); | |
1da177e4 LT |
595 | if (error) |
596 | goto no_buffer; | |
597 | } else { | |
ce8e922c NS |
598 | xfs_buf_deallocate(new_bp); |
599 | if (unlikely(bp == NULL)) | |
1da177e4 LT |
600 | return NULL; |
601 | } | |
602 | ||
ce8e922c NS |
603 | for (i = 0; i < bp->b_page_count; i++) |
604 | mark_page_accessed(bp->b_pages[i]); | |
1da177e4 | 605 | |
ce8e922c NS |
606 | if (!(bp->b_flags & XBF_MAPPED)) { |
607 | error = _xfs_buf_map_pages(bp, flags); | |
1da177e4 LT |
608 | if (unlikely(error)) { |
609 | printk(KERN_WARNING "%s: failed to map pages\n", | |
610 | __FUNCTION__); | |
611 | goto no_buffer; | |
612 | } | |
613 | } | |
614 | ||
ce8e922c | 615 | XFS_STATS_INC(xb_get); |
1da177e4 LT |
616 | |
617 | /* | |
618 | * Always fill in the block number now, the mapped cases can do | |
619 | * their own overlay of this later. | |
620 | */ | |
ce8e922c NS |
621 | bp->b_bn = ioff; |
622 | bp->b_count_desired = bp->b_buffer_length; | |
1da177e4 | 623 | |
ce8e922c NS |
624 | XB_TRACE(bp, "get", (unsigned long)flags); |
625 | return bp; | |
1da177e4 LT |
626 | |
627 | no_buffer: | |
ce8e922c NS |
628 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
629 | xfs_buf_unlock(bp); | |
630 | xfs_buf_rele(bp); | |
1da177e4 LT |
631 | return NULL; |
632 | } | |
633 | ||
634 | xfs_buf_t * | |
635 | xfs_buf_read_flags( | |
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 | ||
645 | bp = xfs_buf_get_flags(target, ioff, isize, flags); | |
646 | if (bp) { | |
647 | if (!XFS_BUF_ISDONE(bp)) { | |
648 | XB_TRACE(bp, "read", (unsigned long)flags); | |
649 | XFS_STATS_INC(xb_get_read); | |
650 | xfs_buf_iostart(bp, flags); | |
651 | } else if (flags & XBF_ASYNC) { | |
652 | XB_TRACE(bp, "read_async", (unsigned long)flags); | |
1da177e4 LT |
653 | /* |
654 | * Read ahead call which is already satisfied, | |
655 | * drop the buffer | |
656 | */ | |
657 | goto no_buffer; | |
658 | } else { | |
ce8e922c | 659 | XB_TRACE(bp, "read_done", (unsigned long)flags); |
1da177e4 | 660 | /* We do not want read in the flags */ |
ce8e922c | 661 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
662 | } |
663 | } | |
664 | ||
ce8e922c | 665 | return bp; |
1da177e4 LT |
666 | |
667 | no_buffer: | |
ce8e922c NS |
668 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
669 | xfs_buf_unlock(bp); | |
670 | xfs_buf_rele(bp); | |
1da177e4 LT |
671 | return NULL; |
672 | } | |
673 | ||
1da177e4 | 674 | /* |
ce8e922c NS |
675 | * If we are not low on memory then do the readahead in a deadlock |
676 | * safe manner. | |
1da177e4 LT |
677 | */ |
678 | void | |
ce8e922c | 679 | xfs_buf_readahead( |
1da177e4 | 680 | xfs_buftarg_t *target, |
204ab25f | 681 | xfs_off_t ioff, |
1da177e4 | 682 | size_t isize, |
ce8e922c | 683 | xfs_buf_flags_t flags) |
1da177e4 LT |
684 | { |
685 | struct backing_dev_info *bdi; | |
686 | ||
ce8e922c | 687 | bdi = target->bt_mapping->backing_dev_info; |
1da177e4 LT |
688 | if (bdi_read_congested(bdi)) |
689 | return; | |
690 | ||
ce8e922c | 691 | flags |= (XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD); |
1da177e4 LT |
692 | xfs_buf_read_flags(target, ioff, isize, flags); |
693 | } | |
694 | ||
695 | xfs_buf_t * | |
ce8e922c | 696 | xfs_buf_get_empty( |
1da177e4 LT |
697 | size_t len, |
698 | xfs_buftarg_t *target) | |
699 | { | |
ce8e922c | 700 | xfs_buf_t *bp; |
1da177e4 | 701 | |
ce8e922c NS |
702 | bp = xfs_buf_allocate(0); |
703 | if (bp) | |
704 | _xfs_buf_initialize(bp, target, 0, len, 0); | |
705 | return bp; | |
1da177e4 LT |
706 | } |
707 | ||
708 | static inline struct page * | |
709 | mem_to_page( | |
710 | void *addr) | |
711 | { | |
712 | if (((unsigned long)addr < VMALLOC_START) || | |
713 | ((unsigned long)addr >= VMALLOC_END)) { | |
714 | return virt_to_page(addr); | |
715 | } else { | |
716 | return vmalloc_to_page(addr); | |
717 | } | |
718 | } | |
719 | ||
720 | int | |
ce8e922c NS |
721 | xfs_buf_associate_memory( |
722 | xfs_buf_t *bp, | |
1da177e4 LT |
723 | void *mem, |
724 | size_t len) | |
725 | { | |
726 | int rval; | |
727 | int i = 0; | |
728 | size_t ptr; | |
729 | size_t end, end_cur; | |
730 | off_t offset; | |
731 | int page_count; | |
732 | ||
733 | page_count = PAGE_CACHE_ALIGN(len) >> PAGE_CACHE_SHIFT; | |
734 | offset = (off_t) mem - ((off_t)mem & PAGE_CACHE_MASK); | |
735 | if (offset && (len > PAGE_CACHE_SIZE)) | |
736 | page_count++; | |
737 | ||
738 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
739 | if (bp->b_pages) |
740 | _xfs_buf_free_pages(bp); | |
1da177e4 | 741 | |
ce8e922c NS |
742 | bp->b_pages = NULL; |
743 | bp->b_addr = mem; | |
1da177e4 | 744 | |
ce8e922c | 745 | rval = _xfs_buf_get_pages(bp, page_count, 0); |
1da177e4 LT |
746 | if (rval) |
747 | return rval; | |
748 | ||
ce8e922c | 749 | bp->b_offset = offset; |
1da177e4 LT |
750 | ptr = (size_t) mem & PAGE_CACHE_MASK; |
751 | end = PAGE_CACHE_ALIGN((size_t) mem + len); | |
752 | end_cur = end; | |
753 | /* set up first page */ | |
ce8e922c | 754 | bp->b_pages[0] = mem_to_page(mem); |
1da177e4 LT |
755 | |
756 | ptr += PAGE_CACHE_SIZE; | |
ce8e922c | 757 | bp->b_page_count = ++i; |
1da177e4 | 758 | while (ptr < end) { |
ce8e922c NS |
759 | bp->b_pages[i] = mem_to_page((void *)ptr); |
760 | bp->b_page_count = ++i; | |
1da177e4 LT |
761 | ptr += PAGE_CACHE_SIZE; |
762 | } | |
ce8e922c | 763 | bp->b_locked = 0; |
1da177e4 | 764 | |
ce8e922c NS |
765 | bp->b_count_desired = bp->b_buffer_length = len; |
766 | bp->b_flags |= XBF_MAPPED; | |
1da177e4 LT |
767 | |
768 | return 0; | |
769 | } | |
770 | ||
771 | xfs_buf_t * | |
ce8e922c | 772 | xfs_buf_get_noaddr( |
1da177e4 LT |
773 | size_t len, |
774 | xfs_buftarg_t *target) | |
775 | { | |
1fa40b01 CH |
776 | unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT; |
777 | int error, i; | |
1da177e4 | 778 | xfs_buf_t *bp; |
1da177e4 | 779 | |
ce8e922c | 780 | bp = xfs_buf_allocate(0); |
1da177e4 LT |
781 | if (unlikely(bp == NULL)) |
782 | goto fail; | |
ce8e922c | 783 | _xfs_buf_initialize(bp, target, 0, len, 0); |
1da177e4 | 784 | |
1fa40b01 CH |
785 | error = _xfs_buf_get_pages(bp, page_count, 0); |
786 | if (error) | |
1da177e4 LT |
787 | goto fail_free_buf; |
788 | ||
1fa40b01 CH |
789 | for (i = 0; i < page_count; i++) { |
790 | bp->b_pages[i] = alloc_page(GFP_KERNEL); | |
791 | if (!bp->b_pages[i]) | |
792 | goto fail_free_mem; | |
1da177e4 | 793 | } |
1fa40b01 | 794 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 795 | |
1fa40b01 CH |
796 | error = _xfs_buf_map_pages(bp, XBF_MAPPED); |
797 | if (unlikely(error)) { | |
798 | printk(KERN_WARNING "%s: failed to map pages\n", | |
799 | __FUNCTION__); | |
1da177e4 | 800 | goto fail_free_mem; |
1fa40b01 | 801 | } |
1da177e4 | 802 | |
ce8e922c | 803 | xfs_buf_unlock(bp); |
1da177e4 | 804 | |
1fa40b01 | 805 | XB_TRACE(bp, "no_daddr", len); |
1da177e4 | 806 | return bp; |
1fa40b01 | 807 | |
1da177e4 | 808 | fail_free_mem: |
1fa40b01 CH |
809 | while (--i >= 0) |
810 | __free_page(bp->b_pages[i]); | |
ca165b88 | 811 | _xfs_buf_free_pages(bp); |
1da177e4 | 812 | fail_free_buf: |
ca165b88 | 813 | xfs_buf_deallocate(bp); |
1da177e4 LT |
814 | fail: |
815 | return NULL; | |
816 | } | |
817 | ||
818 | /* | |
1da177e4 LT |
819 | * Increment reference count on buffer, to hold the buffer concurrently |
820 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
821 | * Must hold the buffer already to call this function. |
822 | */ | |
823 | void | |
ce8e922c NS |
824 | xfs_buf_hold( |
825 | xfs_buf_t *bp) | |
1da177e4 | 826 | { |
ce8e922c NS |
827 | atomic_inc(&bp->b_hold); |
828 | XB_TRACE(bp, "hold", 0); | |
1da177e4 LT |
829 | } |
830 | ||
831 | /* | |
ce8e922c NS |
832 | * Releases a hold on the specified buffer. If the |
833 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
834 | */ |
835 | void | |
ce8e922c NS |
836 | xfs_buf_rele( |
837 | xfs_buf_t *bp) | |
1da177e4 | 838 | { |
ce8e922c | 839 | xfs_bufhash_t *hash = bp->b_hash; |
1da177e4 | 840 | |
ce8e922c | 841 | XB_TRACE(bp, "rele", bp->b_relse); |
1da177e4 | 842 | |
fad3aa1e NS |
843 | if (unlikely(!hash)) { |
844 | ASSERT(!bp->b_relse); | |
845 | if (atomic_dec_and_test(&bp->b_hold)) | |
846 | xfs_buf_free(bp); | |
847 | return; | |
848 | } | |
849 | ||
ce8e922c NS |
850 | if (atomic_dec_and_lock(&bp->b_hold, &hash->bh_lock)) { |
851 | if (bp->b_relse) { | |
852 | atomic_inc(&bp->b_hold); | |
1da177e4 | 853 | spin_unlock(&hash->bh_lock); |
ce8e922c NS |
854 | (*(bp->b_relse)) (bp); |
855 | } else if (bp->b_flags & XBF_FS_MANAGED) { | |
1da177e4 | 856 | spin_unlock(&hash->bh_lock); |
1da177e4 | 857 | } else { |
ce8e922c NS |
858 | ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q))); |
859 | list_del_init(&bp->b_hash_list); | |
1da177e4 | 860 | spin_unlock(&hash->bh_lock); |
ce8e922c | 861 | xfs_buf_free(bp); |
1da177e4 | 862 | } |
2f926587 DC |
863 | } else { |
864 | /* | |
865 | * Catch reference count leaks | |
866 | */ | |
ce8e922c | 867 | ASSERT(atomic_read(&bp->b_hold) >= 0); |
1da177e4 LT |
868 | } |
869 | } | |
870 | ||
871 | ||
872 | /* | |
873 | * Mutual exclusion on buffers. Locking model: | |
874 | * | |
875 | * Buffers associated with inodes for which buffer locking | |
876 | * is not enabled are not protected by semaphores, and are | |
877 | * assumed to be exclusively owned by the caller. There is a | |
878 | * spinlock in the buffer, used by the caller when concurrent | |
879 | * access is possible. | |
880 | */ | |
881 | ||
882 | /* | |
ce8e922c NS |
883 | * Locks a buffer object, if it is not already locked. |
884 | * Note that this in no way locks the underlying pages, so it is only | |
885 | * useful for synchronizing concurrent use of buffer objects, not for | |
886 | * synchronizing independent access to the underlying pages. | |
1da177e4 LT |
887 | */ |
888 | int | |
ce8e922c NS |
889 | xfs_buf_cond_lock( |
890 | xfs_buf_t *bp) | |
1da177e4 LT |
891 | { |
892 | int locked; | |
893 | ||
ce8e922c | 894 | locked = down_trylock(&bp->b_sema) == 0; |
1da177e4 | 895 | if (locked) { |
ce8e922c | 896 | XB_SET_OWNER(bp); |
1da177e4 | 897 | } |
ce8e922c NS |
898 | XB_TRACE(bp, "cond_lock", (long)locked); |
899 | return locked ? 0 : -EBUSY; | |
1da177e4 LT |
900 | } |
901 | ||
902 | #if defined(DEBUG) || defined(XFS_BLI_TRACE) | |
1da177e4 | 903 | int |
ce8e922c NS |
904 | xfs_buf_lock_value( |
905 | xfs_buf_t *bp) | |
1da177e4 | 906 | { |
ce8e922c | 907 | return atomic_read(&bp->b_sema.count); |
1da177e4 LT |
908 | } |
909 | #endif | |
910 | ||
911 | /* | |
ce8e922c NS |
912 | * Locks a buffer object. |
913 | * Note that this in no way locks the underlying pages, so it is only | |
914 | * useful for synchronizing concurrent use of buffer objects, not for | |
915 | * synchronizing independent access to the underlying pages. | |
1da177e4 | 916 | */ |
ce8e922c NS |
917 | void |
918 | xfs_buf_lock( | |
919 | xfs_buf_t *bp) | |
1da177e4 | 920 | { |
ce8e922c NS |
921 | XB_TRACE(bp, "lock", 0); |
922 | if (atomic_read(&bp->b_io_remaining)) | |
923 | blk_run_address_space(bp->b_target->bt_mapping); | |
924 | down(&bp->b_sema); | |
925 | XB_SET_OWNER(bp); | |
926 | XB_TRACE(bp, "locked", 0); | |
1da177e4 LT |
927 | } |
928 | ||
929 | /* | |
ce8e922c | 930 | * Releases the lock on the buffer object. |
2f926587 | 931 | * If the buffer is marked delwri but is not queued, do so before we |
ce8e922c | 932 | * unlock the buffer as we need to set flags correctly. We also need to |
2f926587 DC |
933 | * take a reference for the delwri queue because the unlocker is going to |
934 | * drop their's and they don't know we just queued it. | |
1da177e4 LT |
935 | */ |
936 | void | |
ce8e922c NS |
937 | xfs_buf_unlock( |
938 | xfs_buf_t *bp) | |
1da177e4 | 939 | { |
ce8e922c NS |
940 | if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) { |
941 | atomic_inc(&bp->b_hold); | |
942 | bp->b_flags |= XBF_ASYNC; | |
943 | xfs_buf_delwri_queue(bp, 0); | |
2f926587 DC |
944 | } |
945 | ||
ce8e922c NS |
946 | XB_CLEAR_OWNER(bp); |
947 | up(&bp->b_sema); | |
948 | XB_TRACE(bp, "unlock", 0); | |
1da177e4 LT |
949 | } |
950 | ||
951 | ||
952 | /* | |
953 | * Pinning Buffer Storage in Memory | |
ce8e922c | 954 | * Ensure that no attempt to force a buffer to disk will succeed. |
1da177e4 LT |
955 | */ |
956 | void | |
ce8e922c NS |
957 | xfs_buf_pin( |
958 | xfs_buf_t *bp) | |
1da177e4 | 959 | { |
ce8e922c NS |
960 | atomic_inc(&bp->b_pin_count); |
961 | XB_TRACE(bp, "pin", (long)bp->b_pin_count.counter); | |
1da177e4 LT |
962 | } |
963 | ||
1da177e4 | 964 | void |
ce8e922c NS |
965 | xfs_buf_unpin( |
966 | xfs_buf_t *bp) | |
1da177e4 | 967 | { |
ce8e922c NS |
968 | if (atomic_dec_and_test(&bp->b_pin_count)) |
969 | wake_up_all(&bp->b_waiters); | |
970 | XB_TRACE(bp, "unpin", (long)bp->b_pin_count.counter); | |
1da177e4 LT |
971 | } |
972 | ||
973 | int | |
ce8e922c NS |
974 | xfs_buf_ispin( |
975 | xfs_buf_t *bp) | |
1da177e4 | 976 | { |
ce8e922c | 977 | return atomic_read(&bp->b_pin_count); |
1da177e4 LT |
978 | } |
979 | ||
ce8e922c NS |
980 | STATIC void |
981 | xfs_buf_wait_unpin( | |
982 | xfs_buf_t *bp) | |
1da177e4 LT |
983 | { |
984 | DECLARE_WAITQUEUE (wait, current); | |
985 | ||
ce8e922c | 986 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
987 | return; |
988 | ||
ce8e922c | 989 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
990 | for (;;) { |
991 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 992 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 993 | break; |
ce8e922c NS |
994 | if (atomic_read(&bp->b_io_remaining)) |
995 | blk_run_address_space(bp->b_target->bt_mapping); | |
1da177e4 LT |
996 | schedule(); |
997 | } | |
ce8e922c | 998 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
999 | set_current_state(TASK_RUNNING); |
1000 | } | |
1001 | ||
1002 | /* | |
1003 | * Buffer Utility Routines | |
1004 | */ | |
1005 | ||
1da177e4 | 1006 | STATIC void |
ce8e922c | 1007 | xfs_buf_iodone_work( |
c4028958 | 1008 | struct work_struct *work) |
1da177e4 | 1009 | { |
c4028958 DH |
1010 | xfs_buf_t *bp = |
1011 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 1012 | |
0bfefc46 DC |
1013 | /* |
1014 | * We can get an EOPNOTSUPP to ordered writes. Here we clear the | |
1015 | * ordered flag and reissue them. Because we can't tell the higher | |
1016 | * layers directly that they should not issue ordered I/O anymore, they | |
1017 | * need to check if the ordered flag was cleared during I/O completion. | |
1018 | */ | |
1019 | if ((bp->b_error == EOPNOTSUPP) && | |
1020 | (bp->b_flags & (XBF_ORDERED|XBF_ASYNC)) == (XBF_ORDERED|XBF_ASYNC)) { | |
1021 | XB_TRACE(bp, "ordered_retry", bp->b_iodone); | |
1022 | bp->b_flags &= ~XBF_ORDERED; | |
1023 | xfs_buf_iorequest(bp); | |
1024 | } else if (bp->b_iodone) | |
ce8e922c NS |
1025 | (*(bp->b_iodone))(bp); |
1026 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
1027 | xfs_buf_relse(bp); |
1028 | } | |
1029 | ||
1030 | void | |
ce8e922c NS |
1031 | xfs_buf_ioend( |
1032 | xfs_buf_t *bp, | |
1da177e4 LT |
1033 | int schedule) |
1034 | { | |
ce8e922c NS |
1035 | bp->b_flags &= ~(XBF_READ | XBF_WRITE); |
1036 | if (bp->b_error == 0) | |
1037 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1038 | |
ce8e922c | 1039 | XB_TRACE(bp, "iodone", bp->b_iodone); |
1da177e4 | 1040 | |
ce8e922c | 1041 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 1042 | if (schedule) { |
c4028958 | 1043 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 1044 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 1045 | } else { |
c4028958 | 1046 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
1047 | } |
1048 | } else { | |
ce8e922c | 1049 | up(&bp->b_iodonesema); |
1da177e4 LT |
1050 | } |
1051 | } | |
1052 | ||
1da177e4 | 1053 | void |
ce8e922c NS |
1054 | xfs_buf_ioerror( |
1055 | xfs_buf_t *bp, | |
1056 | int error) | |
1da177e4 LT |
1057 | { |
1058 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c NS |
1059 | bp->b_error = (unsigned short)error; |
1060 | XB_TRACE(bp, "ioerror", (unsigned long)error); | |
1da177e4 LT |
1061 | } |
1062 | ||
1063 | /* | |
ce8e922c NS |
1064 | * Initiate I/O on a buffer, based on the flags supplied. |
1065 | * The b_iodone routine in the buffer supplied will only be called | |
1da177e4 | 1066 | * when all of the subsidiary I/O requests, if any, have been completed. |
1da177e4 LT |
1067 | */ |
1068 | int | |
ce8e922c NS |
1069 | xfs_buf_iostart( |
1070 | xfs_buf_t *bp, | |
1071 | xfs_buf_flags_t flags) | |
1da177e4 LT |
1072 | { |
1073 | int status = 0; | |
1074 | ||
ce8e922c | 1075 | XB_TRACE(bp, "iostart", (unsigned long)flags); |
1da177e4 | 1076 | |
ce8e922c NS |
1077 | if (flags & XBF_DELWRI) { |
1078 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_ASYNC); | |
1079 | bp->b_flags |= flags & (XBF_DELWRI | XBF_ASYNC); | |
1080 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1081 | return status; |
1082 | } | |
1083 | ||
ce8e922c NS |
1084 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \ |
1085 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
1086 | bp->b_flags |= flags & (XBF_READ | XBF_WRITE | XBF_ASYNC | \ | |
1087 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
1da177e4 | 1088 | |
ce8e922c | 1089 | BUG_ON(bp->b_bn == XFS_BUF_DADDR_NULL); |
1da177e4 LT |
1090 | |
1091 | /* For writes allow an alternate strategy routine to precede | |
1092 | * the actual I/O request (which may not be issued at all in | |
1093 | * a shutdown situation, for example). | |
1094 | */ | |
ce8e922c NS |
1095 | status = (flags & XBF_WRITE) ? |
1096 | xfs_buf_iostrategy(bp) : xfs_buf_iorequest(bp); | |
1da177e4 LT |
1097 | |
1098 | /* Wait for I/O if we are not an async request. | |
1099 | * Note: async I/O request completion will release the buffer, | |
1100 | * and that can already be done by this point. So using the | |
1101 | * buffer pointer from here on, after async I/O, is invalid. | |
1102 | */ | |
ce8e922c NS |
1103 | if (!status && !(flags & XBF_ASYNC)) |
1104 | status = xfs_buf_iowait(bp); | |
1da177e4 LT |
1105 | |
1106 | return status; | |
1107 | } | |
1108 | ||
7989cb8e | 1109 | STATIC_INLINE int |
ce8e922c NS |
1110 | _xfs_buf_iolocked( |
1111 | xfs_buf_t *bp) | |
1da177e4 | 1112 | { |
ce8e922c NS |
1113 | ASSERT(bp->b_flags & (XBF_READ | XBF_WRITE)); |
1114 | if (bp->b_flags & XBF_READ) | |
1115 | return bp->b_locked; | |
1da177e4 LT |
1116 | return 0; |
1117 | } | |
1118 | ||
7989cb8e | 1119 | STATIC_INLINE void |
ce8e922c NS |
1120 | _xfs_buf_ioend( |
1121 | xfs_buf_t *bp, | |
1da177e4 LT |
1122 | int schedule) |
1123 | { | |
ce8e922c NS |
1124 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
1125 | bp->b_locked = 0; | |
1126 | xfs_buf_ioend(bp, schedule); | |
1da177e4 LT |
1127 | } |
1128 | } | |
1129 | ||
782e3b3b | 1130 | STATIC void |
ce8e922c | 1131 | xfs_buf_bio_end_io( |
1da177e4 | 1132 | struct bio *bio, |
1da177e4 LT |
1133 | int error) |
1134 | { | |
ce8e922c NS |
1135 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1136 | unsigned int blocksize = bp->b_target->bt_bsize; | |
eedb5530 | 1137 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1da177e4 | 1138 | |
1da177e4 | 1139 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
ce8e922c | 1140 | bp->b_error = EIO; |
1da177e4 | 1141 | |
eedb5530 | 1142 | do { |
1da177e4 LT |
1143 | struct page *page = bvec->bv_page; |
1144 | ||
948ecdb4 | 1145 | ASSERT(!PagePrivate(page)); |
ce8e922c NS |
1146 | if (unlikely(bp->b_error)) { |
1147 | if (bp->b_flags & XBF_READ) | |
eedb5530 | 1148 | ClearPageUptodate(page); |
ce8e922c | 1149 | } else if (blocksize >= PAGE_CACHE_SIZE) { |
1da177e4 LT |
1150 | SetPageUptodate(page); |
1151 | } else if (!PagePrivate(page) && | |
ce8e922c | 1152 | (bp->b_flags & _XBF_PAGE_CACHE)) { |
1da177e4 LT |
1153 | set_page_region(page, bvec->bv_offset, bvec->bv_len); |
1154 | } | |
1155 | ||
eedb5530 NS |
1156 | if (--bvec >= bio->bi_io_vec) |
1157 | prefetchw(&bvec->bv_page->flags); | |
1158 | ||
ce8e922c | 1159 | if (_xfs_buf_iolocked(bp)) { |
1da177e4 LT |
1160 | unlock_page(page); |
1161 | } | |
eedb5530 | 1162 | } while (bvec >= bio->bi_io_vec); |
1da177e4 | 1163 | |
ce8e922c | 1164 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1165 | bio_put(bio); |
1da177e4 LT |
1166 | } |
1167 | ||
1168 | STATIC void | |
ce8e922c NS |
1169 | _xfs_buf_ioapply( |
1170 | xfs_buf_t *bp) | |
1da177e4 LT |
1171 | { |
1172 | int i, rw, map_i, total_nr_pages, nr_pages; | |
1173 | struct bio *bio; | |
ce8e922c NS |
1174 | int offset = bp->b_offset; |
1175 | int size = bp->b_count_desired; | |
1176 | sector_t sector = bp->b_bn; | |
1177 | unsigned int blocksize = bp->b_target->bt_bsize; | |
1178 | int locking = _xfs_buf_iolocked(bp); | |
1da177e4 | 1179 | |
ce8e922c | 1180 | total_nr_pages = bp->b_page_count; |
1da177e4 LT |
1181 | map_i = 0; |
1182 | ||
ce8e922c NS |
1183 | if (bp->b_flags & XBF_ORDERED) { |
1184 | ASSERT(!(bp->b_flags & XBF_READ)); | |
f538d4da | 1185 | rw = WRITE_BARRIER; |
51bdd706 NS |
1186 | } else if (bp->b_flags & _XBF_RUN_QUEUES) { |
1187 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); | |
1188 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1189 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC; | |
1190 | } else { | |
1191 | rw = (bp->b_flags & XBF_WRITE) ? WRITE : | |
1192 | (bp->b_flags & XBF_READ_AHEAD) ? READA : READ; | |
f538d4da CH |
1193 | } |
1194 | ||
ce8e922c | 1195 | /* Special code path for reading a sub page size buffer in -- |
1da177e4 LT |
1196 | * we populate up the whole page, and hence the other metadata |
1197 | * in the same page. This optimization is only valid when the | |
ce8e922c | 1198 | * filesystem block size is not smaller than the page size. |
1da177e4 | 1199 | */ |
ce8e922c NS |
1200 | if ((bp->b_buffer_length < PAGE_CACHE_SIZE) && |
1201 | (bp->b_flags & XBF_READ) && locking && | |
1202 | (blocksize >= PAGE_CACHE_SIZE)) { | |
1da177e4 LT |
1203 | bio = bio_alloc(GFP_NOIO, 1); |
1204 | ||
ce8e922c | 1205 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1206 | bio->bi_sector = sector - (offset >> BBSHIFT); |
ce8e922c NS |
1207 | bio->bi_end_io = xfs_buf_bio_end_io; |
1208 | bio->bi_private = bp; | |
1da177e4 | 1209 | |
ce8e922c | 1210 | bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0); |
1da177e4 LT |
1211 | size = 0; |
1212 | ||
ce8e922c | 1213 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1214 | |
1215 | goto submit_io; | |
1216 | } | |
1217 | ||
1218 | /* Lock down the pages which we need to for the request */ | |
ce8e922c | 1219 | if (locking && (bp->b_flags & XBF_WRITE) && (bp->b_locked == 0)) { |
1da177e4 LT |
1220 | for (i = 0; size; i++) { |
1221 | int nbytes = PAGE_CACHE_SIZE - offset; | |
ce8e922c | 1222 | struct page *page = bp->b_pages[i]; |
1da177e4 LT |
1223 | |
1224 | if (nbytes > size) | |
1225 | nbytes = size; | |
1226 | ||
1227 | lock_page(page); | |
1228 | ||
1229 | size -= nbytes; | |
1230 | offset = 0; | |
1231 | } | |
ce8e922c NS |
1232 | offset = bp->b_offset; |
1233 | size = bp->b_count_desired; | |
1da177e4 LT |
1234 | } |
1235 | ||
1236 | next_chunk: | |
ce8e922c | 1237 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1238 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1239 | if (nr_pages > total_nr_pages) | |
1240 | nr_pages = total_nr_pages; | |
1241 | ||
1242 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1243 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1244 | bio->bi_sector = sector; |
ce8e922c NS |
1245 | bio->bi_end_io = xfs_buf_bio_end_io; |
1246 | bio->bi_private = bp; | |
1da177e4 LT |
1247 | |
1248 | for (; size && nr_pages; nr_pages--, map_i++) { | |
ce8e922c | 1249 | int rbytes, nbytes = PAGE_CACHE_SIZE - offset; |
1da177e4 LT |
1250 | |
1251 | if (nbytes > size) | |
1252 | nbytes = size; | |
1253 | ||
ce8e922c NS |
1254 | rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset); |
1255 | if (rbytes < nbytes) | |
1da177e4 LT |
1256 | break; |
1257 | ||
1258 | offset = 0; | |
1259 | sector += nbytes >> BBSHIFT; | |
1260 | size -= nbytes; | |
1261 | total_nr_pages--; | |
1262 | } | |
1263 | ||
1264 | submit_io: | |
1265 | if (likely(bio->bi_size)) { | |
1266 | submit_bio(rw, bio); | |
1267 | if (size) | |
1268 | goto next_chunk; | |
1269 | } else { | |
1270 | bio_put(bio); | |
ce8e922c | 1271 | xfs_buf_ioerror(bp, EIO); |
1da177e4 LT |
1272 | } |
1273 | } | |
1274 | ||
1da177e4 | 1275 | int |
ce8e922c NS |
1276 | xfs_buf_iorequest( |
1277 | xfs_buf_t *bp) | |
1da177e4 | 1278 | { |
ce8e922c | 1279 | XB_TRACE(bp, "iorequest", 0); |
1da177e4 | 1280 | |
ce8e922c NS |
1281 | if (bp->b_flags & XBF_DELWRI) { |
1282 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1283 | return 0; |
1284 | } | |
1285 | ||
ce8e922c NS |
1286 | if (bp->b_flags & XBF_WRITE) { |
1287 | xfs_buf_wait_unpin(bp); | |
1da177e4 LT |
1288 | } |
1289 | ||
ce8e922c | 1290 | xfs_buf_hold(bp); |
1da177e4 LT |
1291 | |
1292 | /* Set the count to 1 initially, this will stop an I/O | |
1293 | * completion callout which happens before we have started | |
ce8e922c | 1294 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1295 | */ |
ce8e922c NS |
1296 | atomic_set(&bp->b_io_remaining, 1); |
1297 | _xfs_buf_ioapply(bp); | |
1298 | _xfs_buf_ioend(bp, 0); | |
1da177e4 | 1299 | |
ce8e922c | 1300 | xfs_buf_rele(bp); |
1da177e4 LT |
1301 | return 0; |
1302 | } | |
1303 | ||
1304 | /* | |
ce8e922c NS |
1305 | * Waits for I/O to complete on the buffer supplied. |
1306 | * It returns immediately if no I/O is pending. | |
1307 | * It returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1308 | */ |
1309 | int | |
ce8e922c NS |
1310 | xfs_buf_iowait( |
1311 | xfs_buf_t *bp) | |
1da177e4 | 1312 | { |
ce8e922c NS |
1313 | XB_TRACE(bp, "iowait", 0); |
1314 | if (atomic_read(&bp->b_io_remaining)) | |
1315 | blk_run_address_space(bp->b_target->bt_mapping); | |
1316 | down(&bp->b_iodonesema); | |
1317 | XB_TRACE(bp, "iowaited", (long)bp->b_error); | |
1318 | return bp->b_error; | |
1da177e4 LT |
1319 | } |
1320 | ||
ce8e922c NS |
1321 | xfs_caddr_t |
1322 | xfs_buf_offset( | |
1323 | xfs_buf_t *bp, | |
1da177e4 LT |
1324 | size_t offset) |
1325 | { | |
1326 | struct page *page; | |
1327 | ||
ce8e922c NS |
1328 | if (bp->b_flags & XBF_MAPPED) |
1329 | return XFS_BUF_PTR(bp) + offset; | |
1da177e4 | 1330 | |
ce8e922c NS |
1331 | offset += bp->b_offset; |
1332 | page = bp->b_pages[offset >> PAGE_CACHE_SHIFT]; | |
1333 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1)); | |
1da177e4 LT |
1334 | } |
1335 | ||
1336 | /* | |
1da177e4 LT |
1337 | * Move data into or out of a buffer. |
1338 | */ | |
1339 | void | |
ce8e922c NS |
1340 | xfs_buf_iomove( |
1341 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1342 | size_t boff, /* starting buffer offset */ |
1343 | size_t bsize, /* length to copy */ | |
1344 | caddr_t data, /* data address */ | |
ce8e922c | 1345 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 LT |
1346 | { |
1347 | size_t bend, cpoff, csize; | |
1348 | struct page *page; | |
1349 | ||
1350 | bend = boff + bsize; | |
1351 | while (boff < bend) { | |
ce8e922c NS |
1352 | page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)]; |
1353 | cpoff = xfs_buf_poff(boff + bp->b_offset); | |
1da177e4 | 1354 | csize = min_t(size_t, |
ce8e922c | 1355 | PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff); |
1da177e4 LT |
1356 | |
1357 | ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE)); | |
1358 | ||
1359 | switch (mode) { | |
ce8e922c | 1360 | case XBRW_ZERO: |
1da177e4 LT |
1361 | memset(page_address(page) + cpoff, 0, csize); |
1362 | break; | |
ce8e922c | 1363 | case XBRW_READ: |
1da177e4 LT |
1364 | memcpy(data, page_address(page) + cpoff, csize); |
1365 | break; | |
ce8e922c | 1366 | case XBRW_WRITE: |
1da177e4 LT |
1367 | memcpy(page_address(page) + cpoff, data, csize); |
1368 | } | |
1369 | ||
1370 | boff += csize; | |
1371 | data += csize; | |
1372 | } | |
1373 | } | |
1374 | ||
1375 | /* | |
ce8e922c | 1376 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1377 | */ |
1378 | ||
1379 | /* | |
ce8e922c NS |
1380 | * Wait for any bufs with callbacks that have been submitted but |
1381 | * have not yet returned... walk the hash list for the target. | |
1da177e4 LT |
1382 | */ |
1383 | void | |
1384 | xfs_wait_buftarg( | |
1385 | xfs_buftarg_t *btp) | |
1386 | { | |
1387 | xfs_buf_t *bp, *n; | |
1388 | xfs_bufhash_t *hash; | |
1389 | uint i; | |
1390 | ||
1391 | for (i = 0; i < (1 << btp->bt_hashshift); i++) { | |
1392 | hash = &btp->bt_hash[i]; | |
1393 | again: | |
1394 | spin_lock(&hash->bh_lock); | |
ce8e922c NS |
1395 | list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) { |
1396 | ASSERT(btp == bp->b_target); | |
1397 | if (!(bp->b_flags & XBF_FS_MANAGED)) { | |
1da177e4 | 1398 | spin_unlock(&hash->bh_lock); |
2f926587 DC |
1399 | /* |
1400 | * Catch superblock reference count leaks | |
1401 | * immediately | |
1402 | */ | |
ce8e922c | 1403 | BUG_ON(bp->b_bn == 0); |
1da177e4 LT |
1404 | delay(100); |
1405 | goto again; | |
1406 | } | |
1407 | } | |
1408 | spin_unlock(&hash->bh_lock); | |
1409 | } | |
1410 | } | |
1411 | ||
1412 | /* | |
ce8e922c NS |
1413 | * Allocate buffer hash table for a given target. |
1414 | * For devices containing metadata (i.e. not the log/realtime devices) | |
1415 | * we need to allocate a much larger hash table. | |
1da177e4 LT |
1416 | */ |
1417 | STATIC void | |
1418 | xfs_alloc_bufhash( | |
1419 | xfs_buftarg_t *btp, | |
1420 | int external) | |
1421 | { | |
1422 | unsigned int i; | |
1423 | ||
1424 | btp->bt_hashshift = external ? 3 : 8; /* 8 or 256 buckets */ | |
1425 | btp->bt_hashmask = (1 << btp->bt_hashshift) - 1; | |
1426 | btp->bt_hash = kmem_zalloc((1 << btp->bt_hashshift) * | |
93c189c1 | 1427 | sizeof(xfs_bufhash_t), KM_SLEEP | KM_LARGE); |
1da177e4 LT |
1428 | for (i = 0; i < (1 << btp->bt_hashshift); i++) { |
1429 | spin_lock_init(&btp->bt_hash[i].bh_lock); | |
1430 | INIT_LIST_HEAD(&btp->bt_hash[i].bh_list); | |
1431 | } | |
1432 | } | |
1433 | ||
1434 | STATIC void | |
1435 | xfs_free_bufhash( | |
1436 | xfs_buftarg_t *btp) | |
1437 | { | |
ce8e922c | 1438 | kmem_free(btp->bt_hash, (1<<btp->bt_hashshift) * sizeof(xfs_bufhash_t)); |
1da177e4 LT |
1439 | btp->bt_hash = NULL; |
1440 | } | |
1441 | ||
a6867a68 | 1442 | /* |
ce8e922c | 1443 | * buftarg list for delwrite queue processing |
a6867a68 | 1444 | */ |
e6a0e9cd | 1445 | static LIST_HEAD(xfs_buftarg_list); |
7989cb8e | 1446 | static DEFINE_SPINLOCK(xfs_buftarg_lock); |
a6867a68 DC |
1447 | |
1448 | STATIC void | |
1449 | xfs_register_buftarg( | |
1450 | xfs_buftarg_t *btp) | |
1451 | { | |
1452 | spin_lock(&xfs_buftarg_lock); | |
1453 | list_add(&btp->bt_list, &xfs_buftarg_list); | |
1454 | spin_unlock(&xfs_buftarg_lock); | |
1455 | } | |
1456 | ||
1457 | STATIC void | |
1458 | xfs_unregister_buftarg( | |
1459 | xfs_buftarg_t *btp) | |
1460 | { | |
1461 | spin_lock(&xfs_buftarg_lock); | |
1462 | list_del(&btp->bt_list); | |
1463 | spin_unlock(&xfs_buftarg_lock); | |
1464 | } | |
1465 | ||
1da177e4 LT |
1466 | void |
1467 | xfs_free_buftarg( | |
1468 | xfs_buftarg_t *btp, | |
1469 | int external) | |
1470 | { | |
1471 | xfs_flush_buftarg(btp, 1); | |
f4a9f28a | 1472 | xfs_blkdev_issue_flush(btp); |
1da177e4 | 1473 | if (external) |
ce8e922c | 1474 | xfs_blkdev_put(btp->bt_bdev); |
1da177e4 | 1475 | xfs_free_bufhash(btp); |
ce8e922c | 1476 | iput(btp->bt_mapping->host); |
a6867a68 | 1477 | |
ce8e922c NS |
1478 | /* Unregister the buftarg first so that we don't get a |
1479 | * wakeup finding a non-existent task | |
1480 | */ | |
a6867a68 DC |
1481 | xfs_unregister_buftarg(btp); |
1482 | kthread_stop(btp->bt_task); | |
1483 | ||
1da177e4 LT |
1484 | kmem_free(btp, sizeof(*btp)); |
1485 | } | |
1486 | ||
1da177e4 LT |
1487 | STATIC int |
1488 | xfs_setsize_buftarg_flags( | |
1489 | xfs_buftarg_t *btp, | |
1490 | unsigned int blocksize, | |
1491 | unsigned int sectorsize, | |
1492 | int verbose) | |
1493 | { | |
ce8e922c NS |
1494 | btp->bt_bsize = blocksize; |
1495 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1496 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1497 | |
ce8e922c | 1498 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
1da177e4 LT |
1499 | printk(KERN_WARNING |
1500 | "XFS: Cannot set_blocksize to %u on device %s\n", | |
1501 | sectorsize, XFS_BUFTARG_NAME(btp)); | |
1502 | return EINVAL; | |
1503 | } | |
1504 | ||
1505 | if (verbose && | |
1506 | (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) { | |
1507 | printk(KERN_WARNING | |
1508 | "XFS: %u byte sectors in use on device %s. " | |
1509 | "This is suboptimal; %u or greater is ideal.\n", | |
1510 | sectorsize, XFS_BUFTARG_NAME(btp), | |
1511 | (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG); | |
1512 | } | |
1513 | ||
1514 | return 0; | |
1515 | } | |
1516 | ||
1517 | /* | |
ce8e922c NS |
1518 | * When allocating the initial buffer target we have not yet |
1519 | * read in the superblock, so don't know what sized sectors | |
1520 | * are being used is at this early stage. Play safe. | |
1521 | */ | |
1da177e4 LT |
1522 | STATIC int |
1523 | xfs_setsize_buftarg_early( | |
1524 | xfs_buftarg_t *btp, | |
1525 | struct block_device *bdev) | |
1526 | { | |
1527 | return xfs_setsize_buftarg_flags(btp, | |
1528 | PAGE_CACHE_SIZE, bdev_hardsect_size(bdev), 0); | |
1529 | } | |
1530 | ||
1531 | int | |
1532 | xfs_setsize_buftarg( | |
1533 | xfs_buftarg_t *btp, | |
1534 | unsigned int blocksize, | |
1535 | unsigned int sectorsize) | |
1536 | { | |
1537 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1538 | } | |
1539 | ||
1540 | STATIC int | |
1541 | xfs_mapping_buftarg( | |
1542 | xfs_buftarg_t *btp, | |
1543 | struct block_device *bdev) | |
1544 | { | |
1545 | struct backing_dev_info *bdi; | |
1546 | struct inode *inode; | |
1547 | struct address_space *mapping; | |
f5e54d6e | 1548 | static const struct address_space_operations mapping_aops = { |
1da177e4 | 1549 | .sync_page = block_sync_page, |
e965f963 | 1550 | .migratepage = fail_migrate_page, |
1da177e4 LT |
1551 | }; |
1552 | ||
1553 | inode = new_inode(bdev->bd_inode->i_sb); | |
1554 | if (!inode) { | |
1555 | printk(KERN_WARNING | |
1556 | "XFS: Cannot allocate mapping inode for device %s\n", | |
1557 | XFS_BUFTARG_NAME(btp)); | |
1558 | return ENOMEM; | |
1559 | } | |
1560 | inode->i_mode = S_IFBLK; | |
1561 | inode->i_bdev = bdev; | |
1562 | inode->i_rdev = bdev->bd_dev; | |
1563 | bdi = blk_get_backing_dev_info(bdev); | |
1564 | if (!bdi) | |
1565 | bdi = &default_backing_dev_info; | |
1566 | mapping = &inode->i_data; | |
1567 | mapping->a_ops = &mapping_aops; | |
1568 | mapping->backing_dev_info = bdi; | |
1569 | mapping_set_gfp_mask(mapping, GFP_NOFS); | |
ce8e922c | 1570 | btp->bt_mapping = mapping; |
1da177e4 LT |
1571 | return 0; |
1572 | } | |
1573 | ||
a6867a68 DC |
1574 | STATIC int |
1575 | xfs_alloc_delwrite_queue( | |
1576 | xfs_buftarg_t *btp) | |
1577 | { | |
1578 | int error = 0; | |
1579 | ||
1580 | INIT_LIST_HEAD(&btp->bt_list); | |
1581 | INIT_LIST_HEAD(&btp->bt_delwrite_queue); | |
1582 | spinlock_init(&btp->bt_delwrite_lock, "delwri_lock"); | |
1583 | btp->bt_flags = 0; | |
1584 | btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd"); | |
1585 | if (IS_ERR(btp->bt_task)) { | |
1586 | error = PTR_ERR(btp->bt_task); | |
1587 | goto out_error; | |
1588 | } | |
1589 | xfs_register_buftarg(btp); | |
1590 | out_error: | |
1591 | return error; | |
1592 | } | |
1593 | ||
1da177e4 LT |
1594 | xfs_buftarg_t * |
1595 | xfs_alloc_buftarg( | |
1596 | struct block_device *bdev, | |
1597 | int external) | |
1598 | { | |
1599 | xfs_buftarg_t *btp; | |
1600 | ||
1601 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1602 | ||
ce8e922c NS |
1603 | btp->bt_dev = bdev->bd_dev; |
1604 | btp->bt_bdev = bdev; | |
1da177e4 LT |
1605 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1606 | goto error; | |
1607 | if (xfs_mapping_buftarg(btp, bdev)) | |
1608 | goto error; | |
a6867a68 DC |
1609 | if (xfs_alloc_delwrite_queue(btp)) |
1610 | goto error; | |
1da177e4 LT |
1611 | xfs_alloc_bufhash(btp, external); |
1612 | return btp; | |
1613 | ||
1614 | error: | |
1615 | kmem_free(btp, sizeof(*btp)); | |
1616 | return NULL; | |
1617 | } | |
1618 | ||
1619 | ||
1620 | /* | |
ce8e922c | 1621 | * Delayed write buffer handling |
1da177e4 | 1622 | */ |
1da177e4 | 1623 | STATIC void |
ce8e922c NS |
1624 | xfs_buf_delwri_queue( |
1625 | xfs_buf_t *bp, | |
1da177e4 LT |
1626 | int unlock) |
1627 | { | |
ce8e922c NS |
1628 | struct list_head *dwq = &bp->b_target->bt_delwrite_queue; |
1629 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; | |
a6867a68 | 1630 | |
ce8e922c NS |
1631 | XB_TRACE(bp, "delwri_q", (long)unlock); |
1632 | ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC)); | |
1da177e4 | 1633 | |
a6867a68 | 1634 | spin_lock(dwlk); |
1da177e4 | 1635 | /* If already in the queue, dequeue and place at tail */ |
ce8e922c NS |
1636 | if (!list_empty(&bp->b_list)) { |
1637 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1638 | if (unlock) | |
1639 | atomic_dec(&bp->b_hold); | |
1640 | list_del(&bp->b_list); | |
1da177e4 LT |
1641 | } |
1642 | ||
ce8e922c NS |
1643 | bp->b_flags |= _XBF_DELWRI_Q; |
1644 | list_add_tail(&bp->b_list, dwq); | |
1645 | bp->b_queuetime = jiffies; | |
a6867a68 | 1646 | spin_unlock(dwlk); |
1da177e4 LT |
1647 | |
1648 | if (unlock) | |
ce8e922c | 1649 | xfs_buf_unlock(bp); |
1da177e4 LT |
1650 | } |
1651 | ||
1652 | void | |
ce8e922c NS |
1653 | xfs_buf_delwri_dequeue( |
1654 | xfs_buf_t *bp) | |
1da177e4 | 1655 | { |
ce8e922c | 1656 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; |
1da177e4 LT |
1657 | int dequeued = 0; |
1658 | ||
a6867a68 | 1659 | spin_lock(dwlk); |
ce8e922c NS |
1660 | if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) { |
1661 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1662 | list_del_init(&bp->b_list); | |
1da177e4 LT |
1663 | dequeued = 1; |
1664 | } | |
ce8e922c | 1665 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q); |
a6867a68 | 1666 | spin_unlock(dwlk); |
1da177e4 LT |
1667 | |
1668 | if (dequeued) | |
ce8e922c | 1669 | xfs_buf_rele(bp); |
1da177e4 | 1670 | |
ce8e922c | 1671 | XB_TRACE(bp, "delwri_dq", (long)dequeued); |
1da177e4 LT |
1672 | } |
1673 | ||
1674 | STATIC void | |
ce8e922c | 1675 | xfs_buf_runall_queues( |
1da177e4 LT |
1676 | struct workqueue_struct *queue) |
1677 | { | |
1678 | flush_workqueue(queue); | |
1679 | } | |
1680 | ||
1da177e4 | 1681 | STATIC int |
23ea4032 | 1682 | xfsbufd_wakeup( |
15c84a47 NS |
1683 | int priority, |
1684 | gfp_t mask) | |
1da177e4 | 1685 | { |
da7f93e9 | 1686 | xfs_buftarg_t *btp; |
a6867a68 DC |
1687 | |
1688 | spin_lock(&xfs_buftarg_lock); | |
da7f93e9 | 1689 | list_for_each_entry(btp, &xfs_buftarg_list, bt_list) { |
ce8e922c | 1690 | if (test_bit(XBT_FORCE_SLEEP, &btp->bt_flags)) |
a6867a68 | 1691 | continue; |
ce8e922c | 1692 | set_bit(XBT_FORCE_FLUSH, &btp->bt_flags); |
a6867a68 DC |
1693 | wake_up_process(btp->bt_task); |
1694 | } | |
1695 | spin_unlock(&xfs_buftarg_lock); | |
1da177e4 LT |
1696 | return 0; |
1697 | } | |
1698 | ||
585e6d88 DC |
1699 | /* |
1700 | * Move as many buffers as specified to the supplied list | |
1701 | * idicating if we skipped any buffers to prevent deadlocks. | |
1702 | */ | |
1703 | STATIC int | |
1704 | xfs_buf_delwri_split( | |
1705 | xfs_buftarg_t *target, | |
1706 | struct list_head *list, | |
5e6a07df | 1707 | unsigned long age) |
585e6d88 DC |
1708 | { |
1709 | xfs_buf_t *bp, *n; | |
1710 | struct list_head *dwq = &target->bt_delwrite_queue; | |
1711 | spinlock_t *dwlk = &target->bt_delwrite_lock; | |
1712 | int skipped = 0; | |
5e6a07df | 1713 | int force; |
585e6d88 | 1714 | |
5e6a07df | 1715 | force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
585e6d88 DC |
1716 | INIT_LIST_HEAD(list); |
1717 | spin_lock(dwlk); | |
1718 | list_for_each_entry_safe(bp, n, dwq, b_list) { | |
1719 | XB_TRACE(bp, "walkq1", (long)xfs_buf_ispin(bp)); | |
1720 | ASSERT(bp->b_flags & XBF_DELWRI); | |
1721 | ||
1722 | if (!xfs_buf_ispin(bp) && !xfs_buf_cond_lock(bp)) { | |
5e6a07df | 1723 | if (!force && |
585e6d88 DC |
1724 | time_before(jiffies, bp->b_queuetime + age)) { |
1725 | xfs_buf_unlock(bp); | |
1726 | break; | |
1727 | } | |
1728 | ||
1729 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q| | |
1730 | _XBF_RUN_QUEUES); | |
1731 | bp->b_flags |= XBF_WRITE; | |
1732 | list_move_tail(&bp->b_list, list); | |
1733 | } else | |
1734 | skipped++; | |
1735 | } | |
1736 | spin_unlock(dwlk); | |
1737 | ||
1738 | return skipped; | |
1739 | ||
1740 | } | |
1741 | ||
1da177e4 | 1742 | STATIC int |
23ea4032 | 1743 | xfsbufd( |
585e6d88 | 1744 | void *data) |
1da177e4 | 1745 | { |
585e6d88 DC |
1746 | struct list_head tmp; |
1747 | xfs_buftarg_t *target = (xfs_buftarg_t *)data; | |
1748 | int count; | |
1749 | xfs_buf_t *bp; | |
1da177e4 | 1750 | |
1da177e4 LT |
1751 | current->flags |= PF_MEMALLOC; |
1752 | ||
1da177e4 | 1753 | do { |
3e1d1d28 | 1754 | if (unlikely(freezing(current))) { |
ce8e922c | 1755 | set_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
3e1d1d28 | 1756 | refrigerator(); |
abd0cf7a | 1757 | } else { |
ce8e922c | 1758 | clear_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
abd0cf7a | 1759 | } |
1da177e4 | 1760 | |
15c84a47 NS |
1761 | schedule_timeout_interruptible( |
1762 | xfs_buf_timer_centisecs * msecs_to_jiffies(10)); | |
1da177e4 | 1763 | |
585e6d88 | 1764 | xfs_buf_delwri_split(target, &tmp, |
5e6a07df | 1765 | xfs_buf_age_centisecs * msecs_to_jiffies(10)); |
1da177e4 | 1766 | |
585e6d88 | 1767 | count = 0; |
1da177e4 | 1768 | while (!list_empty(&tmp)) { |
ce8e922c NS |
1769 | bp = list_entry(tmp.next, xfs_buf_t, b_list); |
1770 | ASSERT(target == bp->b_target); | |
1da177e4 | 1771 | |
ce8e922c NS |
1772 | list_del_init(&bp->b_list); |
1773 | xfs_buf_iostrategy(bp); | |
585e6d88 | 1774 | count++; |
1da177e4 LT |
1775 | } |
1776 | ||
1777 | if (as_list_len > 0) | |
1778 | purge_addresses(); | |
f07c2250 NS |
1779 | if (count) |
1780 | blk_run_address_space(target->bt_mapping); | |
1da177e4 | 1781 | |
4df08c52 | 1782 | } while (!kthread_should_stop()); |
1da177e4 | 1783 | |
4df08c52 | 1784 | return 0; |
1da177e4 LT |
1785 | } |
1786 | ||
1787 | /* | |
ce8e922c NS |
1788 | * Go through all incore buffers, and release buffers if they belong to |
1789 | * the given device. This is used in filesystem error handling to | |
1790 | * preserve the consistency of its metadata. | |
1da177e4 LT |
1791 | */ |
1792 | int | |
1793 | xfs_flush_buftarg( | |
585e6d88 DC |
1794 | xfs_buftarg_t *target, |
1795 | int wait) | |
1da177e4 | 1796 | { |
585e6d88 DC |
1797 | struct list_head tmp; |
1798 | xfs_buf_t *bp, *n; | |
1799 | int pincount = 0; | |
1da177e4 | 1800 | |
ce8e922c NS |
1801 | xfs_buf_runall_queues(xfsdatad_workqueue); |
1802 | xfs_buf_runall_queues(xfslogd_workqueue); | |
1da177e4 | 1803 | |
5e6a07df DC |
1804 | set_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
1805 | pincount = xfs_buf_delwri_split(target, &tmp, 0); | |
1da177e4 LT |
1806 | |
1807 | /* | |
1808 | * Dropped the delayed write list lock, now walk the temporary list | |
1809 | */ | |
ce8e922c | 1810 | list_for_each_entry_safe(bp, n, &tmp, b_list) { |
585e6d88 | 1811 | ASSERT(target == bp->b_target); |
1da177e4 | 1812 | if (wait) |
ce8e922c | 1813 | bp->b_flags &= ~XBF_ASYNC; |
1da177e4 | 1814 | else |
ce8e922c | 1815 | list_del_init(&bp->b_list); |
1da177e4 | 1816 | |
ce8e922c | 1817 | xfs_buf_iostrategy(bp); |
1da177e4 LT |
1818 | } |
1819 | ||
f07c2250 NS |
1820 | if (wait) |
1821 | blk_run_address_space(target->bt_mapping); | |
1822 | ||
1da177e4 LT |
1823 | /* |
1824 | * Remaining list items must be flushed before returning | |
1825 | */ | |
1826 | while (!list_empty(&tmp)) { | |
ce8e922c | 1827 | bp = list_entry(tmp.next, xfs_buf_t, b_list); |
1da177e4 | 1828 | |
ce8e922c NS |
1829 | list_del_init(&bp->b_list); |
1830 | xfs_iowait(bp); | |
1831 | xfs_buf_relse(bp); | |
1da177e4 LT |
1832 | } |
1833 | ||
1da177e4 LT |
1834 | return pincount; |
1835 | } | |
1836 | ||
04d8b284 | 1837 | int __init |
ce8e922c | 1838 | xfs_buf_init(void) |
1da177e4 | 1839 | { |
ce8e922c NS |
1840 | #ifdef XFS_BUF_TRACE |
1841 | xfs_buf_trace_buf = ktrace_alloc(XFS_BUF_TRACE_SIZE, KM_SLEEP); | |
04d8b284 CH |
1842 | #endif |
1843 | ||
8758280f NS |
1844 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1845 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1846 | if (!xfs_buf_zone) |
04d8b284 CH |
1847 | goto out_free_trace_buf; |
1848 | ||
b4337692 | 1849 | xfslogd_workqueue = create_workqueue("xfslogd"); |
23ea4032 | 1850 | if (!xfslogd_workqueue) |
04d8b284 | 1851 | goto out_free_buf_zone; |
1da177e4 | 1852 | |
b4337692 | 1853 | xfsdatad_workqueue = create_workqueue("xfsdatad"); |
23ea4032 CH |
1854 | if (!xfsdatad_workqueue) |
1855 | goto out_destroy_xfslogd_workqueue; | |
1da177e4 | 1856 | |
8e1f936b | 1857 | register_shrinker(&xfs_buf_shake); |
23ea4032 | 1858 | return 0; |
1da177e4 | 1859 | |
23ea4032 CH |
1860 | out_destroy_xfslogd_workqueue: |
1861 | destroy_workqueue(xfslogd_workqueue); | |
23ea4032 | 1862 | out_free_buf_zone: |
ce8e922c | 1863 | kmem_zone_destroy(xfs_buf_zone); |
04d8b284 | 1864 | out_free_trace_buf: |
ce8e922c NS |
1865 | #ifdef XFS_BUF_TRACE |
1866 | ktrace_free(xfs_buf_trace_buf); | |
23ea4032 | 1867 | #endif |
8758280f | 1868 | return -ENOMEM; |
1da177e4 LT |
1869 | } |
1870 | ||
1da177e4 | 1871 | void |
ce8e922c | 1872 | xfs_buf_terminate(void) |
1da177e4 | 1873 | { |
8e1f936b | 1874 | unregister_shrinker(&xfs_buf_shake); |
04d8b284 CH |
1875 | destroy_workqueue(xfsdatad_workqueue); |
1876 | destroy_workqueue(xfslogd_workqueue); | |
ce8e922c NS |
1877 | kmem_zone_destroy(xfs_buf_zone); |
1878 | #ifdef XFS_BUF_TRACE | |
1879 | ktrace_free(xfs_buf_trace_buf); | |
1da177e4 | 1880 | #endif |
1da177e4 | 1881 | } |
e6a0e9cd TS |
1882 | |
1883 | #ifdef CONFIG_KDB_MODULES | |
1884 | struct list_head * | |
1885 | xfs_get_buftarg_list(void) | |
1886 | { | |
1887 | return &xfs_buftarg_list; | |
1888 | } | |
1889 | #endif |