]>
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 | |
c141b292 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" | |
42 | ||
7989cb8e | 43 | static kmem_zone_t *xfs_buf_zone; |
a6867a68 | 44 | STATIC int xfsbufd(void *); |
27496a8c | 45 | STATIC int xfsbufd_wakeup(int, gfp_t); |
ce8e922c | 46 | STATIC void xfs_buf_delwri_queue(xfs_buf_t *, int); |
8e1f936b RR |
47 | static struct shrinker xfs_buf_shake = { |
48 | .shrink = xfsbufd_wakeup, | |
49 | .seeks = DEFAULT_SEEKS, | |
50 | }; | |
23ea4032 | 51 | |
7989cb8e | 52 | static struct workqueue_struct *xfslogd_workqueue; |
0829c360 | 53 | struct workqueue_struct *xfsdatad_workqueue; |
1da177e4 | 54 | |
ce8e922c | 55 | #ifdef XFS_BUF_TRACE |
1da177e4 | 56 | void |
ce8e922c NS |
57 | xfs_buf_trace( |
58 | xfs_buf_t *bp, | |
1da177e4 LT |
59 | char *id, |
60 | void *data, | |
61 | void *ra) | |
62 | { | |
ce8e922c NS |
63 | ktrace_enter(xfs_buf_trace_buf, |
64 | bp, id, | |
65 | (void *)(unsigned long)bp->b_flags, | |
66 | (void *)(unsigned long)bp->b_hold.counter, | |
d63f154a | 67 | (void *)(unsigned long)bp->b_sema.count, |
1da177e4 LT |
68 | (void *)current, |
69 | data, ra, | |
ce8e922c NS |
70 | (void *)(unsigned long)((bp->b_file_offset>>32) & 0xffffffff), |
71 | (void *)(unsigned long)(bp->b_file_offset & 0xffffffff), | |
72 | (void *)(unsigned long)bp->b_buffer_length, | |
1da177e4 LT |
73 | NULL, NULL, NULL, NULL, NULL); |
74 | } | |
ce8e922c NS |
75 | ktrace_t *xfs_buf_trace_buf; |
76 | #define XFS_BUF_TRACE_SIZE 4096 | |
77 | #define XB_TRACE(bp, id, data) \ | |
78 | xfs_buf_trace(bp, id, (void *)data, (void *)__builtin_return_address(0)) | |
1da177e4 | 79 | #else |
ce8e922c | 80 | #define XB_TRACE(bp, id, data) do { } while (0) |
1da177e4 LT |
81 | #endif |
82 | ||
ce8e922c NS |
83 | #ifdef XFS_BUF_LOCK_TRACKING |
84 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
85 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
86 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 87 | #else |
ce8e922c NS |
88 | # define XB_SET_OWNER(bp) do { } while (0) |
89 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
90 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
91 | #endif |
92 | ||
ce8e922c NS |
93 | #define xb_to_gfp(flags) \ |
94 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \ | |
95 | ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN) | |
1da177e4 | 96 | |
ce8e922c NS |
97 | #define xb_to_km(flags) \ |
98 | (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP) | |
1da177e4 | 99 | |
ce8e922c NS |
100 | #define xfs_buf_allocate(flags) \ |
101 | kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags)) | |
102 | #define xfs_buf_deallocate(bp) \ | |
103 | kmem_zone_free(xfs_buf_zone, (bp)); | |
1da177e4 LT |
104 | |
105 | /* | |
ce8e922c | 106 | * Page Region interfaces. |
1da177e4 | 107 | * |
ce8e922c NS |
108 | * For pages in filesystems where the blocksize is smaller than the |
109 | * pagesize, we use the page->private field (long) to hold a bitmap | |
110 | * of uptodate regions within the page. | |
1da177e4 | 111 | * |
ce8e922c | 112 | * Each such region is "bytes per page / bits per long" bytes long. |
1da177e4 | 113 | * |
ce8e922c NS |
114 | * NBPPR == number-of-bytes-per-page-region |
115 | * BTOPR == bytes-to-page-region (rounded up) | |
116 | * BTOPRT == bytes-to-page-region-truncated (rounded down) | |
1da177e4 LT |
117 | */ |
118 | #if (BITS_PER_LONG == 32) | |
119 | #define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */ | |
120 | #elif (BITS_PER_LONG == 64) | |
121 | #define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */ | |
122 | #else | |
123 | #error BITS_PER_LONG must be 32 or 64 | |
124 | #endif | |
125 | #define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG) | |
126 | #define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT) | |
127 | #define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT)) | |
128 | ||
129 | STATIC unsigned long | |
130 | page_region_mask( | |
131 | size_t offset, | |
132 | size_t length) | |
133 | { | |
134 | unsigned long mask; | |
135 | int first, final; | |
136 | ||
137 | first = BTOPR(offset); | |
138 | final = BTOPRT(offset + length - 1); | |
139 | first = min(first, final); | |
140 | ||
141 | mask = ~0UL; | |
142 | mask <<= BITS_PER_LONG - (final - first); | |
143 | mask >>= BITS_PER_LONG - (final); | |
144 | ||
145 | ASSERT(offset + length <= PAGE_CACHE_SIZE); | |
146 | ASSERT((final - first) < BITS_PER_LONG && (final - first) >= 0); | |
147 | ||
148 | return mask; | |
149 | } | |
150 | ||
7989cb8e | 151 | STATIC_INLINE void |
1da177e4 LT |
152 | set_page_region( |
153 | struct page *page, | |
154 | size_t offset, | |
155 | size_t length) | |
156 | { | |
4c21e2f2 HD |
157 | set_page_private(page, |
158 | page_private(page) | page_region_mask(offset, length)); | |
159 | if (page_private(page) == ~0UL) | |
1da177e4 LT |
160 | SetPageUptodate(page); |
161 | } | |
162 | ||
7989cb8e | 163 | STATIC_INLINE int |
1da177e4 LT |
164 | test_page_region( |
165 | struct page *page, | |
166 | size_t offset, | |
167 | size_t length) | |
168 | { | |
169 | unsigned long mask = page_region_mask(offset, length); | |
170 | ||
4c21e2f2 | 171 | return (mask && (page_private(page) & mask) == mask); |
1da177e4 LT |
172 | } |
173 | ||
27e88bf6 FB |
174 | /* |
175 | * Mapping of multi-page buffers into contiguous virtual space | |
176 | */ | |
177 | ||
178 | typedef struct a_list { | |
179 | void *vm_addr; | |
180 | struct a_list *next; | |
181 | } a_list_t; | |
182 | ||
183 | static a_list_t *as_free_head; | |
184 | static int as_list_len; | |
185 | static DEFINE_SPINLOCK(as_lock); | |
186 | ||
187 | /* | |
188 | * Try to batch vunmaps because they are costly. | |
189 | */ | |
190 | STATIC void | |
191 | free_address( | |
192 | void *addr) | |
193 | { | |
194 | a_list_t *aentry; | |
195 | ||
196 | #ifdef CONFIG_XEN | |
197 | /* | |
198 | * Xen needs to be able to make sure it can get an exclusive | |
199 | * RO mapping of pages it wants to turn into a pagetable. If | |
200 | * a newly allocated page is also still being vmap()ed by xfs, | |
201 | * it will cause pagetable construction to fail. This is a | |
202 | * quick workaround to always eagerly unmap pages so that Xen | |
203 | * is happy. | |
204 | */ | |
205 | vunmap(addr); | |
206 | return; | |
207 | #endif | |
208 | ||
209 | aentry = kmalloc(sizeof(a_list_t), GFP_NOWAIT); | |
210 | if (likely(aentry)) { | |
211 | spin_lock(&as_lock); | |
212 | aentry->next = as_free_head; | |
213 | aentry->vm_addr = addr; | |
214 | as_free_head = aentry; | |
215 | as_list_len++; | |
216 | spin_unlock(&as_lock); | |
217 | } else { | |
218 | vunmap(addr); | |
219 | } | |
220 | } | |
221 | ||
222 | STATIC void | |
223 | purge_addresses(void) | |
224 | { | |
225 | a_list_t *aentry, *old; | |
226 | ||
227 | if (as_free_head == NULL) | |
228 | return; | |
229 | ||
230 | spin_lock(&as_lock); | |
231 | aentry = as_free_head; | |
232 | as_free_head = NULL; | |
233 | as_list_len = 0; | |
234 | spin_unlock(&as_lock); | |
235 | ||
236 | while ((old = aentry) != NULL) { | |
237 | vunmap(aentry->vm_addr); | |
238 | aentry = aentry->next; | |
239 | kfree(old); | |
240 | } | |
241 | } | |
242 | ||
1da177e4 | 243 | /* |
ce8e922c | 244 | * Internal xfs_buf_t object manipulation |
1da177e4 LT |
245 | */ |
246 | ||
247 | STATIC void | |
ce8e922c NS |
248 | _xfs_buf_initialize( |
249 | xfs_buf_t *bp, | |
1da177e4 | 250 | xfs_buftarg_t *target, |
204ab25f | 251 | xfs_off_t range_base, |
1da177e4 | 252 | size_t range_length, |
ce8e922c | 253 | xfs_buf_flags_t flags) |
1da177e4 LT |
254 | { |
255 | /* | |
ce8e922c | 256 | * We don't want certain flags to appear in b_flags. |
1da177e4 | 257 | */ |
ce8e922c NS |
258 | flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD); |
259 | ||
260 | memset(bp, 0, sizeof(xfs_buf_t)); | |
261 | atomic_set(&bp->b_hold, 1); | |
b4dd330b | 262 | init_completion(&bp->b_iowait); |
ce8e922c NS |
263 | INIT_LIST_HEAD(&bp->b_list); |
264 | INIT_LIST_HEAD(&bp->b_hash_list); | |
265 | init_MUTEX_LOCKED(&bp->b_sema); /* held, no waiters */ | |
266 | XB_SET_OWNER(bp); | |
267 | bp->b_target = target; | |
268 | bp->b_file_offset = range_base; | |
1da177e4 LT |
269 | /* |
270 | * Set buffer_length and count_desired to the same value initially. | |
271 | * I/O routines should use count_desired, which will be the same in | |
272 | * most cases but may be reset (e.g. XFS recovery). | |
273 | */ | |
ce8e922c NS |
274 | bp->b_buffer_length = bp->b_count_desired = range_length; |
275 | bp->b_flags = flags; | |
276 | bp->b_bn = XFS_BUF_DADDR_NULL; | |
277 | atomic_set(&bp->b_pin_count, 0); | |
278 | init_waitqueue_head(&bp->b_waiters); | |
279 | ||
280 | XFS_STATS_INC(xb_create); | |
281 | XB_TRACE(bp, "initialize", target); | |
1da177e4 LT |
282 | } |
283 | ||
284 | /* | |
ce8e922c NS |
285 | * Allocate a page array capable of holding a specified number |
286 | * of pages, and point the page buf at it. | |
1da177e4 LT |
287 | */ |
288 | STATIC int | |
ce8e922c NS |
289 | _xfs_buf_get_pages( |
290 | xfs_buf_t *bp, | |
1da177e4 | 291 | int page_count, |
ce8e922c | 292 | xfs_buf_flags_t flags) |
1da177e4 LT |
293 | { |
294 | /* Make sure that we have a page list */ | |
ce8e922c NS |
295 | if (bp->b_pages == NULL) { |
296 | bp->b_offset = xfs_buf_poff(bp->b_file_offset); | |
297 | bp->b_page_count = page_count; | |
298 | if (page_count <= XB_PAGES) { | |
299 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 300 | } else { |
ce8e922c NS |
301 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
302 | page_count, xb_to_km(flags)); | |
303 | if (bp->b_pages == NULL) | |
1da177e4 LT |
304 | return -ENOMEM; |
305 | } | |
ce8e922c | 306 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
307 | } |
308 | return 0; | |
309 | } | |
310 | ||
311 | /* | |
ce8e922c | 312 | * Frees b_pages if it was allocated. |
1da177e4 LT |
313 | */ |
314 | STATIC void | |
ce8e922c | 315 | _xfs_buf_free_pages( |
1da177e4 LT |
316 | xfs_buf_t *bp) |
317 | { | |
ce8e922c | 318 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 319 | kmem_free(bp->b_pages); |
1da177e4 LT |
320 | } |
321 | } | |
322 | ||
323 | /* | |
324 | * Releases the specified buffer. | |
325 | * | |
326 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 327 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
328 | * hashed and refcounted buffers |
329 | */ | |
330 | void | |
ce8e922c | 331 | xfs_buf_free( |
1da177e4 LT |
332 | xfs_buf_t *bp) |
333 | { | |
ce8e922c | 334 | XB_TRACE(bp, "free", 0); |
1da177e4 | 335 | |
ce8e922c | 336 | ASSERT(list_empty(&bp->b_hash_list)); |
1da177e4 | 337 | |
1fa40b01 | 338 | if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) { |
1da177e4 LT |
339 | uint i; |
340 | ||
ce8e922c | 341 | if ((bp->b_flags & XBF_MAPPED) && (bp->b_page_count > 1)) |
27e88bf6 | 342 | free_address(bp->b_addr - bp->b_offset); |
1da177e4 | 343 | |
948ecdb4 NS |
344 | for (i = 0; i < bp->b_page_count; i++) { |
345 | struct page *page = bp->b_pages[i]; | |
346 | ||
1fa40b01 CH |
347 | if (bp->b_flags & _XBF_PAGE_CACHE) |
348 | ASSERT(!PagePrivate(page)); | |
948ecdb4 NS |
349 | page_cache_release(page); |
350 | } | |
ce8e922c | 351 | _xfs_buf_free_pages(bp); |
1da177e4 LT |
352 | } |
353 | ||
ce8e922c | 354 | xfs_buf_deallocate(bp); |
1da177e4 LT |
355 | } |
356 | ||
357 | /* | |
358 | * Finds all pages for buffer in question and builds it's page list. | |
359 | */ | |
360 | STATIC int | |
ce8e922c | 361 | _xfs_buf_lookup_pages( |
1da177e4 LT |
362 | xfs_buf_t *bp, |
363 | uint flags) | |
364 | { | |
ce8e922c NS |
365 | struct address_space *mapping = bp->b_target->bt_mapping; |
366 | size_t blocksize = bp->b_target->bt_bsize; | |
367 | size_t size = bp->b_count_desired; | |
1da177e4 | 368 | size_t nbytes, offset; |
ce8e922c | 369 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 LT |
370 | unsigned short page_count, i; |
371 | pgoff_t first; | |
204ab25f | 372 | xfs_off_t end; |
1da177e4 LT |
373 | int error; |
374 | ||
ce8e922c NS |
375 | end = bp->b_file_offset + bp->b_buffer_length; |
376 | page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset); | |
1da177e4 | 377 | |
ce8e922c | 378 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
379 | if (unlikely(error)) |
380 | return error; | |
ce8e922c | 381 | bp->b_flags |= _XBF_PAGE_CACHE; |
1da177e4 | 382 | |
ce8e922c NS |
383 | offset = bp->b_offset; |
384 | first = bp->b_file_offset >> PAGE_CACHE_SHIFT; | |
1da177e4 | 385 | |
ce8e922c | 386 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
387 | struct page *page; |
388 | uint retries = 0; | |
389 | ||
390 | retry: | |
391 | page = find_or_create_page(mapping, first + i, gfp_mask); | |
392 | if (unlikely(page == NULL)) { | |
ce8e922c NS |
393 | if (flags & XBF_READ_AHEAD) { |
394 | bp->b_page_count = i; | |
6ab455ee CH |
395 | for (i = 0; i < bp->b_page_count; i++) |
396 | unlock_page(bp->b_pages[i]); | |
1da177e4 LT |
397 | return -ENOMEM; |
398 | } | |
399 | ||
400 | /* | |
401 | * This could deadlock. | |
402 | * | |
403 | * But until all the XFS lowlevel code is revamped to | |
404 | * handle buffer allocation failures we can't do much. | |
405 | */ | |
406 | if (!(++retries % 100)) | |
407 | printk(KERN_ERR | |
408 | "XFS: possible memory allocation " | |
409 | "deadlock in %s (mode:0x%x)\n", | |
34a622b2 | 410 | __func__, gfp_mask); |
1da177e4 | 411 | |
ce8e922c | 412 | XFS_STATS_INC(xb_page_retries); |
23ea4032 | 413 | xfsbufd_wakeup(0, gfp_mask); |
3fcfab16 | 414 | congestion_wait(WRITE, HZ/50); |
1da177e4 LT |
415 | goto retry; |
416 | } | |
417 | ||
ce8e922c | 418 | XFS_STATS_INC(xb_page_found); |
1da177e4 LT |
419 | |
420 | nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset); | |
421 | size -= nbytes; | |
422 | ||
948ecdb4 | 423 | ASSERT(!PagePrivate(page)); |
1da177e4 LT |
424 | if (!PageUptodate(page)) { |
425 | page_count--; | |
6ab455ee CH |
426 | if (blocksize >= PAGE_CACHE_SIZE) { |
427 | if (flags & XBF_READ) | |
428 | bp->b_flags |= _XBF_PAGE_LOCKED; | |
429 | } else if (!PagePrivate(page)) { | |
1da177e4 LT |
430 | if (test_page_region(page, offset, nbytes)) |
431 | page_count++; | |
432 | } | |
433 | } | |
434 | ||
ce8e922c | 435 | bp->b_pages[i] = page; |
1da177e4 LT |
436 | offset = 0; |
437 | } | |
438 | ||
6ab455ee CH |
439 | if (!(bp->b_flags & _XBF_PAGE_LOCKED)) { |
440 | for (i = 0; i < bp->b_page_count; i++) | |
441 | unlock_page(bp->b_pages[i]); | |
442 | } | |
443 | ||
ce8e922c NS |
444 | if (page_count == bp->b_page_count) |
445 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 446 | |
ce8e922c | 447 | XB_TRACE(bp, "lookup_pages", (long)page_count); |
1da177e4 LT |
448 | return error; |
449 | } | |
450 | ||
451 | /* | |
452 | * Map buffer into kernel address-space if nessecary. | |
453 | */ | |
454 | STATIC int | |
ce8e922c | 455 | _xfs_buf_map_pages( |
1da177e4 LT |
456 | xfs_buf_t *bp, |
457 | uint flags) | |
458 | { | |
459 | /* A single page buffer is always mappable */ | |
ce8e922c NS |
460 | if (bp->b_page_count == 1) { |
461 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; | |
462 | bp->b_flags |= XBF_MAPPED; | |
463 | } else if (flags & XBF_MAPPED) { | |
27e88bf6 FB |
464 | if (as_list_len > 64) |
465 | purge_addresses(); | |
7fdf5824 FB |
466 | bp->b_addr = vmap(bp->b_pages, bp->b_page_count, |
467 | VM_MAP, PAGE_KERNEL); | |
ce8e922c | 468 | if (unlikely(bp->b_addr == NULL)) |
1da177e4 | 469 | return -ENOMEM; |
ce8e922c NS |
470 | bp->b_addr += bp->b_offset; |
471 | bp->b_flags |= XBF_MAPPED; | |
1da177e4 LT |
472 | } |
473 | ||
474 | return 0; | |
475 | } | |
476 | ||
477 | /* | |
478 | * Finding and Reading Buffers | |
479 | */ | |
480 | ||
481 | /* | |
ce8e922c | 482 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 LT |
483 | * a given range of an inode. The buffer is returned |
484 | * locked. If other overlapping buffers exist, they are | |
485 | * released before the new buffer is created and locked, | |
486 | * which may imply that this call will block until those buffers | |
487 | * are unlocked. No I/O is implied by this call. | |
488 | */ | |
489 | xfs_buf_t * | |
ce8e922c | 490 | _xfs_buf_find( |
1da177e4 | 491 | xfs_buftarg_t *btp, /* block device target */ |
204ab25f | 492 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 493 | size_t isize, /* length of range */ |
ce8e922c NS |
494 | xfs_buf_flags_t flags, |
495 | xfs_buf_t *new_bp) | |
1da177e4 | 496 | { |
204ab25f | 497 | xfs_off_t range_base; |
1da177e4 LT |
498 | size_t range_length; |
499 | xfs_bufhash_t *hash; | |
ce8e922c | 500 | xfs_buf_t *bp, *n; |
1da177e4 LT |
501 | |
502 | range_base = (ioff << BBSHIFT); | |
503 | range_length = (isize << BBSHIFT); | |
504 | ||
505 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
ce8e922c | 506 | ASSERT(!(range_length < (1 << btp->bt_sshift))); |
204ab25f | 507 | ASSERT(!(range_base & (xfs_off_t)btp->bt_smask)); |
1da177e4 LT |
508 | |
509 | hash = &btp->bt_hash[hash_long((unsigned long)ioff, btp->bt_hashshift)]; | |
510 | ||
511 | spin_lock(&hash->bh_lock); | |
512 | ||
ce8e922c NS |
513 | list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) { |
514 | ASSERT(btp == bp->b_target); | |
515 | if (bp->b_file_offset == range_base && | |
516 | bp->b_buffer_length == range_length) { | |
1da177e4 | 517 | /* |
ce8e922c | 518 | * If we look at something, bring it to the |
1da177e4 LT |
519 | * front of the list for next time. |
520 | */ | |
ce8e922c NS |
521 | atomic_inc(&bp->b_hold); |
522 | list_move(&bp->b_hash_list, &hash->bh_list); | |
1da177e4 LT |
523 | goto found; |
524 | } | |
525 | } | |
526 | ||
527 | /* No match found */ | |
ce8e922c NS |
528 | if (new_bp) { |
529 | _xfs_buf_initialize(new_bp, btp, range_base, | |
1da177e4 | 530 | range_length, flags); |
ce8e922c NS |
531 | new_bp->b_hash = hash; |
532 | list_add(&new_bp->b_hash_list, &hash->bh_list); | |
1da177e4 | 533 | } else { |
ce8e922c | 534 | XFS_STATS_INC(xb_miss_locked); |
1da177e4 LT |
535 | } |
536 | ||
537 | spin_unlock(&hash->bh_lock); | |
ce8e922c | 538 | return new_bp; |
1da177e4 LT |
539 | |
540 | found: | |
541 | spin_unlock(&hash->bh_lock); | |
542 | ||
543 | /* Attempt to get the semaphore without sleeping, | |
544 | * if this does not work then we need to drop the | |
545 | * spinlock and do a hard attempt on the semaphore. | |
546 | */ | |
ce8e922c NS |
547 | if (down_trylock(&bp->b_sema)) { |
548 | if (!(flags & XBF_TRYLOCK)) { | |
1da177e4 | 549 | /* wait for buffer ownership */ |
ce8e922c NS |
550 | XB_TRACE(bp, "get_lock", 0); |
551 | xfs_buf_lock(bp); | |
552 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
553 | } else { |
554 | /* We asked for a trylock and failed, no need | |
555 | * to look at file offset and length here, we | |
ce8e922c NS |
556 | * know that this buffer at least overlaps our |
557 | * buffer and is locked, therefore our buffer | |
558 | * either does not exist, or is this buffer. | |
1da177e4 | 559 | */ |
ce8e922c NS |
560 | xfs_buf_rele(bp); |
561 | XFS_STATS_INC(xb_busy_locked); | |
562 | return NULL; | |
1da177e4 LT |
563 | } |
564 | } else { | |
565 | /* trylock worked */ | |
ce8e922c | 566 | XB_SET_OWNER(bp); |
1da177e4 LT |
567 | } |
568 | ||
ce8e922c NS |
569 | if (bp->b_flags & XBF_STALE) { |
570 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
571 | bp->b_flags &= XBF_MAPPED; | |
2f926587 | 572 | } |
ce8e922c NS |
573 | XB_TRACE(bp, "got_lock", 0); |
574 | XFS_STATS_INC(xb_get_locked); | |
575 | return bp; | |
1da177e4 LT |
576 | } |
577 | ||
578 | /* | |
ce8e922c | 579 | * Assembles a buffer covering the specified range. |
1da177e4 LT |
580 | * Storage in memory for all portions of the buffer will be allocated, |
581 | * although backing storage may not be. | |
582 | */ | |
583 | xfs_buf_t * | |
ce8e922c | 584 | xfs_buf_get_flags( |
1da177e4 | 585 | xfs_buftarg_t *target,/* target for buffer */ |
204ab25f | 586 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 587 | size_t isize, /* length of range */ |
ce8e922c | 588 | xfs_buf_flags_t flags) |
1da177e4 | 589 | { |
ce8e922c | 590 | xfs_buf_t *bp, *new_bp; |
1da177e4 LT |
591 | int error = 0, i; |
592 | ||
ce8e922c NS |
593 | new_bp = xfs_buf_allocate(flags); |
594 | if (unlikely(!new_bp)) | |
1da177e4 LT |
595 | return NULL; |
596 | ||
ce8e922c NS |
597 | bp = _xfs_buf_find(target, ioff, isize, flags, new_bp); |
598 | if (bp == new_bp) { | |
599 | error = _xfs_buf_lookup_pages(bp, flags); | |
1da177e4 LT |
600 | if (error) |
601 | goto no_buffer; | |
602 | } else { | |
ce8e922c NS |
603 | xfs_buf_deallocate(new_bp); |
604 | if (unlikely(bp == NULL)) | |
1da177e4 LT |
605 | return NULL; |
606 | } | |
607 | ||
ce8e922c NS |
608 | for (i = 0; i < bp->b_page_count; i++) |
609 | mark_page_accessed(bp->b_pages[i]); | |
1da177e4 | 610 | |
ce8e922c NS |
611 | if (!(bp->b_flags & XBF_MAPPED)) { |
612 | error = _xfs_buf_map_pages(bp, flags); | |
1da177e4 LT |
613 | if (unlikely(error)) { |
614 | printk(KERN_WARNING "%s: failed to map pages\n", | |
34a622b2 | 615 | __func__); |
1da177e4 LT |
616 | goto no_buffer; |
617 | } | |
618 | } | |
619 | ||
ce8e922c | 620 | XFS_STATS_INC(xb_get); |
1da177e4 LT |
621 | |
622 | /* | |
623 | * Always fill in the block number now, the mapped cases can do | |
624 | * their own overlay of this later. | |
625 | */ | |
ce8e922c NS |
626 | bp->b_bn = ioff; |
627 | bp->b_count_desired = bp->b_buffer_length; | |
1da177e4 | 628 | |
ce8e922c NS |
629 | XB_TRACE(bp, "get", (unsigned long)flags); |
630 | return bp; | |
1da177e4 LT |
631 | |
632 | no_buffer: | |
ce8e922c NS |
633 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
634 | xfs_buf_unlock(bp); | |
635 | xfs_buf_rele(bp); | |
1da177e4 LT |
636 | return NULL; |
637 | } | |
638 | ||
5d765b97 CH |
639 | STATIC int |
640 | _xfs_buf_read( | |
641 | xfs_buf_t *bp, | |
642 | xfs_buf_flags_t flags) | |
643 | { | |
644 | int status; | |
645 | ||
646 | XB_TRACE(bp, "_xfs_buf_read", (unsigned long)flags); | |
647 | ||
648 | ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE))); | |
649 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); | |
650 | ||
651 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \ | |
652 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
653 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \ | |
654 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
655 | ||
656 | status = xfs_buf_iorequest(bp); | |
657 | if (!status && !(flags & XBF_ASYNC)) | |
658 | status = xfs_buf_iowait(bp); | |
659 | return status; | |
660 | } | |
661 | ||
1da177e4 LT |
662 | xfs_buf_t * |
663 | xfs_buf_read_flags( | |
664 | xfs_buftarg_t *target, | |
204ab25f | 665 | xfs_off_t ioff, |
1da177e4 | 666 | size_t isize, |
ce8e922c | 667 | xfs_buf_flags_t flags) |
1da177e4 | 668 | { |
ce8e922c NS |
669 | xfs_buf_t *bp; |
670 | ||
671 | flags |= XBF_READ; | |
672 | ||
673 | bp = xfs_buf_get_flags(target, ioff, isize, flags); | |
674 | if (bp) { | |
675 | if (!XFS_BUF_ISDONE(bp)) { | |
676 | XB_TRACE(bp, "read", (unsigned long)flags); | |
677 | XFS_STATS_INC(xb_get_read); | |
5d765b97 | 678 | _xfs_buf_read(bp, flags); |
ce8e922c NS |
679 | } else if (flags & XBF_ASYNC) { |
680 | XB_TRACE(bp, "read_async", (unsigned long)flags); | |
1da177e4 LT |
681 | /* |
682 | * Read ahead call which is already satisfied, | |
683 | * drop the buffer | |
684 | */ | |
685 | goto no_buffer; | |
686 | } else { | |
ce8e922c | 687 | XB_TRACE(bp, "read_done", (unsigned long)flags); |
1da177e4 | 688 | /* We do not want read in the flags */ |
ce8e922c | 689 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
690 | } |
691 | } | |
692 | ||
ce8e922c | 693 | return bp; |
1da177e4 LT |
694 | |
695 | no_buffer: | |
ce8e922c NS |
696 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
697 | xfs_buf_unlock(bp); | |
698 | xfs_buf_rele(bp); | |
1da177e4 LT |
699 | return NULL; |
700 | } | |
701 | ||
1da177e4 | 702 | /* |
ce8e922c NS |
703 | * If we are not low on memory then do the readahead in a deadlock |
704 | * safe manner. | |
1da177e4 LT |
705 | */ |
706 | void | |
ce8e922c | 707 | xfs_buf_readahead( |
1da177e4 | 708 | xfs_buftarg_t *target, |
204ab25f | 709 | xfs_off_t ioff, |
1da177e4 | 710 | size_t isize, |
ce8e922c | 711 | xfs_buf_flags_t flags) |
1da177e4 LT |
712 | { |
713 | struct backing_dev_info *bdi; | |
714 | ||
ce8e922c | 715 | bdi = target->bt_mapping->backing_dev_info; |
1da177e4 LT |
716 | if (bdi_read_congested(bdi)) |
717 | return; | |
718 | ||
ce8e922c | 719 | flags |= (XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD); |
1da177e4 LT |
720 | xfs_buf_read_flags(target, ioff, isize, flags); |
721 | } | |
722 | ||
723 | xfs_buf_t * | |
ce8e922c | 724 | xfs_buf_get_empty( |
1da177e4 LT |
725 | size_t len, |
726 | xfs_buftarg_t *target) | |
727 | { | |
ce8e922c | 728 | xfs_buf_t *bp; |
1da177e4 | 729 | |
ce8e922c NS |
730 | bp = xfs_buf_allocate(0); |
731 | if (bp) | |
732 | _xfs_buf_initialize(bp, target, 0, len, 0); | |
733 | return bp; | |
1da177e4 LT |
734 | } |
735 | ||
736 | static inline struct page * | |
737 | mem_to_page( | |
738 | void *addr) | |
739 | { | |
9e2779fa | 740 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
741 | return virt_to_page(addr); |
742 | } else { | |
743 | return vmalloc_to_page(addr); | |
744 | } | |
745 | } | |
746 | ||
747 | int | |
ce8e922c NS |
748 | xfs_buf_associate_memory( |
749 | xfs_buf_t *bp, | |
1da177e4 LT |
750 | void *mem, |
751 | size_t len) | |
752 | { | |
753 | int rval; | |
754 | int i = 0; | |
d1afb678 LM |
755 | unsigned long pageaddr; |
756 | unsigned long offset; | |
757 | size_t buflen; | |
1da177e4 LT |
758 | int page_count; |
759 | ||
d1afb678 LM |
760 | pageaddr = (unsigned long)mem & PAGE_CACHE_MASK; |
761 | offset = (unsigned long)mem - pageaddr; | |
762 | buflen = PAGE_CACHE_ALIGN(len + offset); | |
763 | page_count = buflen >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
764 | |
765 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
766 | if (bp->b_pages) |
767 | _xfs_buf_free_pages(bp); | |
1da177e4 | 768 | |
ce8e922c NS |
769 | bp->b_pages = NULL; |
770 | bp->b_addr = mem; | |
1da177e4 | 771 | |
ce8e922c | 772 | rval = _xfs_buf_get_pages(bp, page_count, 0); |
1da177e4 LT |
773 | if (rval) |
774 | return rval; | |
775 | ||
ce8e922c | 776 | bp->b_offset = offset; |
d1afb678 LM |
777 | |
778 | for (i = 0; i < bp->b_page_count; i++) { | |
779 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
780 | pageaddr += PAGE_CACHE_SIZE; | |
1da177e4 | 781 | } |
1da177e4 | 782 | |
d1afb678 LM |
783 | bp->b_count_desired = len; |
784 | bp->b_buffer_length = buflen; | |
ce8e922c | 785 | bp->b_flags |= XBF_MAPPED; |
6ab455ee | 786 | bp->b_flags &= ~_XBF_PAGE_LOCKED; |
1da177e4 LT |
787 | |
788 | return 0; | |
789 | } | |
790 | ||
791 | xfs_buf_t * | |
ce8e922c | 792 | xfs_buf_get_noaddr( |
1da177e4 LT |
793 | size_t len, |
794 | xfs_buftarg_t *target) | |
795 | { | |
1fa40b01 CH |
796 | unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT; |
797 | int error, i; | |
1da177e4 | 798 | xfs_buf_t *bp; |
1da177e4 | 799 | |
ce8e922c | 800 | bp = xfs_buf_allocate(0); |
1da177e4 LT |
801 | if (unlikely(bp == NULL)) |
802 | goto fail; | |
ce8e922c | 803 | _xfs_buf_initialize(bp, target, 0, len, 0); |
1da177e4 | 804 | |
1fa40b01 CH |
805 | error = _xfs_buf_get_pages(bp, page_count, 0); |
806 | if (error) | |
1da177e4 LT |
807 | goto fail_free_buf; |
808 | ||
1fa40b01 CH |
809 | for (i = 0; i < page_count; i++) { |
810 | bp->b_pages[i] = alloc_page(GFP_KERNEL); | |
811 | if (!bp->b_pages[i]) | |
812 | goto fail_free_mem; | |
1da177e4 | 813 | } |
1fa40b01 | 814 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 815 | |
1fa40b01 CH |
816 | error = _xfs_buf_map_pages(bp, XBF_MAPPED); |
817 | if (unlikely(error)) { | |
818 | printk(KERN_WARNING "%s: failed to map pages\n", | |
34a622b2 | 819 | __func__); |
1da177e4 | 820 | goto fail_free_mem; |
1fa40b01 | 821 | } |
1da177e4 | 822 | |
ce8e922c | 823 | xfs_buf_unlock(bp); |
1da177e4 | 824 | |
1fa40b01 | 825 | XB_TRACE(bp, "no_daddr", len); |
1da177e4 | 826 | return bp; |
1fa40b01 | 827 | |
1da177e4 | 828 | fail_free_mem: |
1fa40b01 CH |
829 | while (--i >= 0) |
830 | __free_page(bp->b_pages[i]); | |
ca165b88 | 831 | _xfs_buf_free_pages(bp); |
1da177e4 | 832 | fail_free_buf: |
ca165b88 | 833 | xfs_buf_deallocate(bp); |
1da177e4 LT |
834 | fail: |
835 | return NULL; | |
836 | } | |
837 | ||
838 | /* | |
1da177e4 LT |
839 | * Increment reference count on buffer, to hold the buffer concurrently |
840 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
841 | * Must hold the buffer already to call this function. |
842 | */ | |
843 | void | |
ce8e922c NS |
844 | xfs_buf_hold( |
845 | xfs_buf_t *bp) | |
1da177e4 | 846 | { |
ce8e922c NS |
847 | atomic_inc(&bp->b_hold); |
848 | XB_TRACE(bp, "hold", 0); | |
1da177e4 LT |
849 | } |
850 | ||
851 | /* | |
ce8e922c NS |
852 | * Releases a hold on the specified buffer. If the |
853 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
854 | */ |
855 | void | |
ce8e922c NS |
856 | xfs_buf_rele( |
857 | xfs_buf_t *bp) | |
1da177e4 | 858 | { |
ce8e922c | 859 | xfs_bufhash_t *hash = bp->b_hash; |
1da177e4 | 860 | |
ce8e922c | 861 | XB_TRACE(bp, "rele", bp->b_relse); |
1da177e4 | 862 | |
fad3aa1e NS |
863 | if (unlikely(!hash)) { |
864 | ASSERT(!bp->b_relse); | |
865 | if (atomic_dec_and_test(&bp->b_hold)) | |
866 | xfs_buf_free(bp); | |
867 | return; | |
868 | } | |
869 | ||
3790689f | 870 | ASSERT(atomic_read(&bp->b_hold) > 0); |
ce8e922c NS |
871 | if (atomic_dec_and_lock(&bp->b_hold, &hash->bh_lock)) { |
872 | if (bp->b_relse) { | |
873 | atomic_inc(&bp->b_hold); | |
1da177e4 | 874 | spin_unlock(&hash->bh_lock); |
ce8e922c NS |
875 | (*(bp->b_relse)) (bp); |
876 | } else if (bp->b_flags & XBF_FS_MANAGED) { | |
1da177e4 | 877 | spin_unlock(&hash->bh_lock); |
1da177e4 | 878 | } else { |
ce8e922c NS |
879 | ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q))); |
880 | list_del_init(&bp->b_hash_list); | |
1da177e4 | 881 | spin_unlock(&hash->bh_lock); |
ce8e922c | 882 | xfs_buf_free(bp); |
1da177e4 LT |
883 | } |
884 | } | |
885 | } | |
886 | ||
887 | ||
888 | /* | |
889 | * Mutual exclusion on buffers. Locking model: | |
890 | * | |
891 | * Buffers associated with inodes for which buffer locking | |
892 | * is not enabled are not protected by semaphores, and are | |
893 | * assumed to be exclusively owned by the caller. There is a | |
894 | * spinlock in the buffer, used by the caller when concurrent | |
895 | * access is possible. | |
896 | */ | |
897 | ||
898 | /* | |
ce8e922c NS |
899 | * Locks a buffer object, if it is not already locked. |
900 | * Note that this in no way locks the underlying pages, so it is only | |
901 | * useful for synchronizing concurrent use of buffer objects, not for | |
902 | * synchronizing independent access to the underlying pages. | |
1da177e4 LT |
903 | */ |
904 | int | |
ce8e922c NS |
905 | xfs_buf_cond_lock( |
906 | xfs_buf_t *bp) | |
1da177e4 LT |
907 | { |
908 | int locked; | |
909 | ||
ce8e922c | 910 | locked = down_trylock(&bp->b_sema) == 0; |
1da177e4 | 911 | if (locked) { |
ce8e922c | 912 | XB_SET_OWNER(bp); |
1da177e4 | 913 | } |
ce8e922c NS |
914 | XB_TRACE(bp, "cond_lock", (long)locked); |
915 | return locked ? 0 : -EBUSY; | |
1da177e4 LT |
916 | } |
917 | ||
918 | #if defined(DEBUG) || defined(XFS_BLI_TRACE) | |
1da177e4 | 919 | int |
ce8e922c NS |
920 | xfs_buf_lock_value( |
921 | xfs_buf_t *bp) | |
1da177e4 | 922 | { |
adaa693b | 923 | return bp->b_sema.count; |
1da177e4 LT |
924 | } |
925 | #endif | |
926 | ||
927 | /* | |
ce8e922c NS |
928 | * Locks a buffer object. |
929 | * Note that this in no way locks the underlying pages, so it is only | |
930 | * useful for synchronizing concurrent use of buffer objects, not for | |
931 | * synchronizing independent access to the underlying pages. | |
1da177e4 | 932 | */ |
ce8e922c NS |
933 | void |
934 | xfs_buf_lock( | |
935 | xfs_buf_t *bp) | |
1da177e4 | 936 | { |
ce8e922c NS |
937 | XB_TRACE(bp, "lock", 0); |
938 | if (atomic_read(&bp->b_io_remaining)) | |
939 | blk_run_address_space(bp->b_target->bt_mapping); | |
940 | down(&bp->b_sema); | |
941 | XB_SET_OWNER(bp); | |
942 | XB_TRACE(bp, "locked", 0); | |
1da177e4 LT |
943 | } |
944 | ||
945 | /* | |
ce8e922c | 946 | * Releases the lock on the buffer object. |
2f926587 | 947 | * If the buffer is marked delwri but is not queued, do so before we |
ce8e922c | 948 | * unlock the buffer as we need to set flags correctly. We also need to |
2f926587 DC |
949 | * take a reference for the delwri queue because the unlocker is going to |
950 | * drop their's and they don't know we just queued it. | |
1da177e4 LT |
951 | */ |
952 | void | |
ce8e922c NS |
953 | xfs_buf_unlock( |
954 | xfs_buf_t *bp) | |
1da177e4 | 955 | { |
ce8e922c NS |
956 | if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) { |
957 | atomic_inc(&bp->b_hold); | |
958 | bp->b_flags |= XBF_ASYNC; | |
959 | xfs_buf_delwri_queue(bp, 0); | |
2f926587 DC |
960 | } |
961 | ||
ce8e922c NS |
962 | XB_CLEAR_OWNER(bp); |
963 | up(&bp->b_sema); | |
964 | XB_TRACE(bp, "unlock", 0); | |
1da177e4 LT |
965 | } |
966 | ||
967 | ||
968 | /* | |
969 | * Pinning Buffer Storage in Memory | |
ce8e922c | 970 | * Ensure that no attempt to force a buffer to disk will succeed. |
1da177e4 LT |
971 | */ |
972 | void | |
ce8e922c NS |
973 | xfs_buf_pin( |
974 | xfs_buf_t *bp) | |
1da177e4 | 975 | { |
ce8e922c NS |
976 | atomic_inc(&bp->b_pin_count); |
977 | XB_TRACE(bp, "pin", (long)bp->b_pin_count.counter); | |
1da177e4 LT |
978 | } |
979 | ||
1da177e4 | 980 | void |
ce8e922c NS |
981 | xfs_buf_unpin( |
982 | xfs_buf_t *bp) | |
1da177e4 | 983 | { |
ce8e922c NS |
984 | if (atomic_dec_and_test(&bp->b_pin_count)) |
985 | wake_up_all(&bp->b_waiters); | |
986 | XB_TRACE(bp, "unpin", (long)bp->b_pin_count.counter); | |
1da177e4 LT |
987 | } |
988 | ||
989 | int | |
ce8e922c NS |
990 | xfs_buf_ispin( |
991 | xfs_buf_t *bp) | |
1da177e4 | 992 | { |
ce8e922c | 993 | return atomic_read(&bp->b_pin_count); |
1da177e4 LT |
994 | } |
995 | ||
ce8e922c NS |
996 | STATIC void |
997 | xfs_buf_wait_unpin( | |
998 | xfs_buf_t *bp) | |
1da177e4 LT |
999 | { |
1000 | DECLARE_WAITQUEUE (wait, current); | |
1001 | ||
ce8e922c | 1002 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
1003 | return; |
1004 | ||
ce8e922c | 1005 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1006 | for (;;) { |
1007 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 1008 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 1009 | break; |
ce8e922c NS |
1010 | if (atomic_read(&bp->b_io_remaining)) |
1011 | blk_run_address_space(bp->b_target->bt_mapping); | |
1da177e4 LT |
1012 | schedule(); |
1013 | } | |
ce8e922c | 1014 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
1015 | set_current_state(TASK_RUNNING); |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * Buffer Utility Routines | |
1020 | */ | |
1021 | ||
1da177e4 | 1022 | STATIC void |
ce8e922c | 1023 | xfs_buf_iodone_work( |
c4028958 | 1024 | struct work_struct *work) |
1da177e4 | 1025 | { |
c4028958 DH |
1026 | xfs_buf_t *bp = |
1027 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 1028 | |
0bfefc46 DC |
1029 | /* |
1030 | * We can get an EOPNOTSUPP to ordered writes. Here we clear the | |
1031 | * ordered flag and reissue them. Because we can't tell the higher | |
1032 | * layers directly that they should not issue ordered I/O anymore, they | |
73f6aa4d | 1033 | * need to check if the _XFS_BARRIER_FAILED flag was set during I/O completion. |
0bfefc46 DC |
1034 | */ |
1035 | if ((bp->b_error == EOPNOTSUPP) && | |
1036 | (bp->b_flags & (XBF_ORDERED|XBF_ASYNC)) == (XBF_ORDERED|XBF_ASYNC)) { | |
1037 | XB_TRACE(bp, "ordered_retry", bp->b_iodone); | |
1038 | bp->b_flags &= ~XBF_ORDERED; | |
73f6aa4d | 1039 | bp->b_flags |= _XFS_BARRIER_FAILED; |
0bfefc46 DC |
1040 | xfs_buf_iorequest(bp); |
1041 | } else if (bp->b_iodone) | |
ce8e922c NS |
1042 | (*(bp->b_iodone))(bp); |
1043 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
1044 | xfs_buf_relse(bp); |
1045 | } | |
1046 | ||
1047 | void | |
ce8e922c NS |
1048 | xfs_buf_ioend( |
1049 | xfs_buf_t *bp, | |
1da177e4 LT |
1050 | int schedule) |
1051 | { | |
77be55a5 | 1052 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
ce8e922c NS |
1053 | if (bp->b_error == 0) |
1054 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1055 | |
ce8e922c | 1056 | XB_TRACE(bp, "iodone", bp->b_iodone); |
1da177e4 | 1057 | |
ce8e922c | 1058 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 1059 | if (schedule) { |
c4028958 | 1060 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 1061 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 1062 | } else { |
c4028958 | 1063 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
1064 | } |
1065 | } else { | |
b4dd330b | 1066 | complete(&bp->b_iowait); |
1da177e4 LT |
1067 | } |
1068 | } | |
1069 | ||
1da177e4 | 1070 | void |
ce8e922c NS |
1071 | xfs_buf_ioerror( |
1072 | xfs_buf_t *bp, | |
1073 | int error) | |
1da177e4 LT |
1074 | { |
1075 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c NS |
1076 | bp->b_error = (unsigned short)error; |
1077 | XB_TRACE(bp, "ioerror", (unsigned long)error); | |
1da177e4 LT |
1078 | } |
1079 | ||
1da177e4 | 1080 | int |
5d765b97 CH |
1081 | xfs_bawrite( |
1082 | void *mp, | |
1083 | struct xfs_buf *bp) | |
1da177e4 | 1084 | { |
5d765b97 | 1085 | XB_TRACE(bp, "bawrite", 0); |
1da177e4 | 1086 | |
5d765b97 | 1087 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); |
1da177e4 | 1088 | |
5d765b97 | 1089 | xfs_buf_delwri_dequeue(bp); |
1da177e4 | 1090 | |
5d765b97 CH |
1091 | bp->b_flags &= ~(XBF_READ | XBF_DELWRI | XBF_READ_AHEAD); |
1092 | bp->b_flags |= (XBF_WRITE | XBF_ASYNC | _XBF_RUN_QUEUES); | |
1da177e4 | 1093 | |
15ac08a8 | 1094 | bp->b_mount = mp; |
5d765b97 CH |
1095 | bp->b_strat = xfs_bdstrat_cb; |
1096 | return xfs_bdstrat_cb(bp); | |
1097 | } | |
1da177e4 | 1098 | |
5d765b97 CH |
1099 | void |
1100 | xfs_bdwrite( | |
1101 | void *mp, | |
1102 | struct xfs_buf *bp) | |
1103 | { | |
1104 | XB_TRACE(bp, "bdwrite", 0); | |
1da177e4 | 1105 | |
5d765b97 | 1106 | bp->b_strat = xfs_bdstrat_cb; |
15ac08a8 | 1107 | bp->b_mount = mp; |
1da177e4 | 1108 | |
5d765b97 CH |
1109 | bp->b_flags &= ~XBF_READ; |
1110 | bp->b_flags |= (XBF_DELWRI | XBF_ASYNC); | |
1111 | ||
1112 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1113 | } |
1114 | ||
7989cb8e | 1115 | STATIC_INLINE void |
ce8e922c NS |
1116 | _xfs_buf_ioend( |
1117 | xfs_buf_t *bp, | |
1da177e4 LT |
1118 | int schedule) |
1119 | { | |
6ab455ee CH |
1120 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
1121 | bp->b_flags &= ~_XBF_PAGE_LOCKED; | |
ce8e922c | 1122 | xfs_buf_ioend(bp, schedule); |
6ab455ee | 1123 | } |
1da177e4 LT |
1124 | } |
1125 | ||
782e3b3b | 1126 | STATIC void |
ce8e922c | 1127 | xfs_buf_bio_end_io( |
1da177e4 | 1128 | struct bio *bio, |
1da177e4 LT |
1129 | int error) |
1130 | { | |
ce8e922c NS |
1131 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1132 | unsigned int blocksize = bp->b_target->bt_bsize; | |
eedb5530 | 1133 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1da177e4 | 1134 | |
cfbe5267 | 1135 | xfs_buf_ioerror(bp, -error); |
1da177e4 | 1136 | |
eedb5530 | 1137 | do { |
1da177e4 LT |
1138 | struct page *page = bvec->bv_page; |
1139 | ||
948ecdb4 | 1140 | ASSERT(!PagePrivate(page)); |
ce8e922c NS |
1141 | if (unlikely(bp->b_error)) { |
1142 | if (bp->b_flags & XBF_READ) | |
eedb5530 | 1143 | ClearPageUptodate(page); |
ce8e922c | 1144 | } else if (blocksize >= PAGE_CACHE_SIZE) { |
1da177e4 LT |
1145 | SetPageUptodate(page); |
1146 | } else if (!PagePrivate(page) && | |
ce8e922c | 1147 | (bp->b_flags & _XBF_PAGE_CACHE)) { |
1da177e4 LT |
1148 | set_page_region(page, bvec->bv_offset, bvec->bv_len); |
1149 | } | |
1150 | ||
eedb5530 NS |
1151 | if (--bvec >= bio->bi_io_vec) |
1152 | prefetchw(&bvec->bv_page->flags); | |
6ab455ee CH |
1153 | |
1154 | if (bp->b_flags & _XBF_PAGE_LOCKED) | |
1155 | unlock_page(page); | |
eedb5530 | 1156 | } while (bvec >= bio->bi_io_vec); |
1da177e4 | 1157 | |
ce8e922c | 1158 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1159 | bio_put(bio); |
1da177e4 LT |
1160 | } |
1161 | ||
1162 | STATIC void | |
ce8e922c NS |
1163 | _xfs_buf_ioapply( |
1164 | xfs_buf_t *bp) | |
1da177e4 | 1165 | { |
a9759f2d | 1166 | int rw, map_i, total_nr_pages, nr_pages; |
1da177e4 | 1167 | struct bio *bio; |
ce8e922c NS |
1168 | int offset = bp->b_offset; |
1169 | int size = bp->b_count_desired; | |
1170 | sector_t sector = bp->b_bn; | |
1171 | unsigned int blocksize = bp->b_target->bt_bsize; | |
1da177e4 | 1172 | |
ce8e922c | 1173 | total_nr_pages = bp->b_page_count; |
1da177e4 LT |
1174 | map_i = 0; |
1175 | ||
ce8e922c NS |
1176 | if (bp->b_flags & XBF_ORDERED) { |
1177 | ASSERT(!(bp->b_flags & XBF_READ)); | |
f538d4da | 1178 | rw = WRITE_BARRIER; |
51bdd706 NS |
1179 | } else if (bp->b_flags & _XBF_RUN_QUEUES) { |
1180 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); | |
1181 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1182 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC; | |
1183 | } else { | |
1184 | rw = (bp->b_flags & XBF_WRITE) ? WRITE : | |
1185 | (bp->b_flags & XBF_READ_AHEAD) ? READA : READ; | |
f538d4da CH |
1186 | } |
1187 | ||
ce8e922c | 1188 | /* Special code path for reading a sub page size buffer in -- |
1da177e4 LT |
1189 | * we populate up the whole page, and hence the other metadata |
1190 | * in the same page. This optimization is only valid when the | |
ce8e922c | 1191 | * filesystem block size is not smaller than the page size. |
1da177e4 | 1192 | */ |
ce8e922c | 1193 | if ((bp->b_buffer_length < PAGE_CACHE_SIZE) && |
6ab455ee CH |
1194 | ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) == |
1195 | (XBF_READ|_XBF_PAGE_LOCKED)) && | |
ce8e922c | 1196 | (blocksize >= PAGE_CACHE_SIZE)) { |
1da177e4 LT |
1197 | bio = bio_alloc(GFP_NOIO, 1); |
1198 | ||
ce8e922c | 1199 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1200 | bio->bi_sector = sector - (offset >> BBSHIFT); |
ce8e922c NS |
1201 | bio->bi_end_io = xfs_buf_bio_end_io; |
1202 | bio->bi_private = bp; | |
1da177e4 | 1203 | |
ce8e922c | 1204 | bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0); |
1da177e4 LT |
1205 | size = 0; |
1206 | ||
ce8e922c | 1207 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1208 | |
1209 | goto submit_io; | |
1210 | } | |
1211 | ||
1da177e4 | 1212 | next_chunk: |
ce8e922c | 1213 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1214 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1215 | if (nr_pages > total_nr_pages) | |
1216 | nr_pages = total_nr_pages; | |
1217 | ||
1218 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1219 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1220 | bio->bi_sector = sector; |
ce8e922c NS |
1221 | bio->bi_end_io = xfs_buf_bio_end_io; |
1222 | bio->bi_private = bp; | |
1da177e4 LT |
1223 | |
1224 | for (; size && nr_pages; nr_pages--, map_i++) { | |
ce8e922c | 1225 | int rbytes, nbytes = PAGE_CACHE_SIZE - offset; |
1da177e4 LT |
1226 | |
1227 | if (nbytes > size) | |
1228 | nbytes = size; | |
1229 | ||
ce8e922c NS |
1230 | rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset); |
1231 | if (rbytes < nbytes) | |
1da177e4 LT |
1232 | break; |
1233 | ||
1234 | offset = 0; | |
1235 | sector += nbytes >> BBSHIFT; | |
1236 | size -= nbytes; | |
1237 | total_nr_pages--; | |
1238 | } | |
1239 | ||
1240 | submit_io: | |
1241 | if (likely(bio->bi_size)) { | |
1242 | submit_bio(rw, bio); | |
1243 | if (size) | |
1244 | goto next_chunk; | |
1245 | } else { | |
1246 | bio_put(bio); | |
ce8e922c | 1247 | xfs_buf_ioerror(bp, EIO); |
1da177e4 LT |
1248 | } |
1249 | } | |
1250 | ||
1da177e4 | 1251 | int |
ce8e922c NS |
1252 | xfs_buf_iorequest( |
1253 | xfs_buf_t *bp) | |
1da177e4 | 1254 | { |
ce8e922c | 1255 | XB_TRACE(bp, "iorequest", 0); |
1da177e4 | 1256 | |
ce8e922c NS |
1257 | if (bp->b_flags & XBF_DELWRI) { |
1258 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1259 | return 0; |
1260 | } | |
1261 | ||
ce8e922c NS |
1262 | if (bp->b_flags & XBF_WRITE) { |
1263 | xfs_buf_wait_unpin(bp); | |
1da177e4 LT |
1264 | } |
1265 | ||
ce8e922c | 1266 | xfs_buf_hold(bp); |
1da177e4 LT |
1267 | |
1268 | /* Set the count to 1 initially, this will stop an I/O | |
1269 | * completion callout which happens before we have started | |
ce8e922c | 1270 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1271 | */ |
ce8e922c NS |
1272 | atomic_set(&bp->b_io_remaining, 1); |
1273 | _xfs_buf_ioapply(bp); | |
1274 | _xfs_buf_ioend(bp, 0); | |
1da177e4 | 1275 | |
ce8e922c | 1276 | xfs_buf_rele(bp); |
1da177e4 LT |
1277 | return 0; |
1278 | } | |
1279 | ||
1280 | /* | |
ce8e922c NS |
1281 | * Waits for I/O to complete on the buffer supplied. |
1282 | * It returns immediately if no I/O is pending. | |
1283 | * It returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1284 | */ |
1285 | int | |
ce8e922c NS |
1286 | xfs_buf_iowait( |
1287 | xfs_buf_t *bp) | |
1da177e4 | 1288 | { |
ce8e922c NS |
1289 | XB_TRACE(bp, "iowait", 0); |
1290 | if (atomic_read(&bp->b_io_remaining)) | |
1291 | blk_run_address_space(bp->b_target->bt_mapping); | |
b4dd330b | 1292 | wait_for_completion(&bp->b_iowait); |
ce8e922c NS |
1293 | XB_TRACE(bp, "iowaited", (long)bp->b_error); |
1294 | return bp->b_error; | |
1da177e4 LT |
1295 | } |
1296 | ||
ce8e922c NS |
1297 | xfs_caddr_t |
1298 | xfs_buf_offset( | |
1299 | xfs_buf_t *bp, | |
1da177e4 LT |
1300 | size_t offset) |
1301 | { | |
1302 | struct page *page; | |
1303 | ||
ce8e922c NS |
1304 | if (bp->b_flags & XBF_MAPPED) |
1305 | return XFS_BUF_PTR(bp) + offset; | |
1da177e4 | 1306 | |
ce8e922c NS |
1307 | offset += bp->b_offset; |
1308 | page = bp->b_pages[offset >> PAGE_CACHE_SHIFT]; | |
1309 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1)); | |
1da177e4 LT |
1310 | } |
1311 | ||
1312 | /* | |
1da177e4 LT |
1313 | * Move data into or out of a buffer. |
1314 | */ | |
1315 | void | |
ce8e922c NS |
1316 | xfs_buf_iomove( |
1317 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1318 | size_t boff, /* starting buffer offset */ |
1319 | size_t bsize, /* length to copy */ | |
1320 | caddr_t data, /* data address */ | |
ce8e922c | 1321 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 LT |
1322 | { |
1323 | size_t bend, cpoff, csize; | |
1324 | struct page *page; | |
1325 | ||
1326 | bend = boff + bsize; | |
1327 | while (boff < bend) { | |
ce8e922c NS |
1328 | page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)]; |
1329 | cpoff = xfs_buf_poff(boff + bp->b_offset); | |
1da177e4 | 1330 | csize = min_t(size_t, |
ce8e922c | 1331 | PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff); |
1da177e4 LT |
1332 | |
1333 | ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE)); | |
1334 | ||
1335 | switch (mode) { | |
ce8e922c | 1336 | case XBRW_ZERO: |
1da177e4 LT |
1337 | memset(page_address(page) + cpoff, 0, csize); |
1338 | break; | |
ce8e922c | 1339 | case XBRW_READ: |
1da177e4 LT |
1340 | memcpy(data, page_address(page) + cpoff, csize); |
1341 | break; | |
ce8e922c | 1342 | case XBRW_WRITE: |
1da177e4 LT |
1343 | memcpy(page_address(page) + cpoff, data, csize); |
1344 | } | |
1345 | ||
1346 | boff += csize; | |
1347 | data += csize; | |
1348 | } | |
1349 | } | |
1350 | ||
1351 | /* | |
ce8e922c | 1352 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1353 | */ |
1354 | ||
1355 | /* | |
ce8e922c NS |
1356 | * Wait for any bufs with callbacks that have been submitted but |
1357 | * have not yet returned... walk the hash list for the target. | |
1da177e4 LT |
1358 | */ |
1359 | void | |
1360 | xfs_wait_buftarg( | |
1361 | xfs_buftarg_t *btp) | |
1362 | { | |
1363 | xfs_buf_t *bp, *n; | |
1364 | xfs_bufhash_t *hash; | |
1365 | uint i; | |
1366 | ||
1367 | for (i = 0; i < (1 << btp->bt_hashshift); i++) { | |
1368 | hash = &btp->bt_hash[i]; | |
1369 | again: | |
1370 | spin_lock(&hash->bh_lock); | |
ce8e922c NS |
1371 | list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) { |
1372 | ASSERT(btp == bp->b_target); | |
1373 | if (!(bp->b_flags & XBF_FS_MANAGED)) { | |
1da177e4 | 1374 | spin_unlock(&hash->bh_lock); |
2f926587 DC |
1375 | /* |
1376 | * Catch superblock reference count leaks | |
1377 | * immediately | |
1378 | */ | |
ce8e922c | 1379 | BUG_ON(bp->b_bn == 0); |
1da177e4 LT |
1380 | delay(100); |
1381 | goto again; | |
1382 | } | |
1383 | } | |
1384 | spin_unlock(&hash->bh_lock); | |
1385 | } | |
1386 | } | |
1387 | ||
1388 | /* | |
ce8e922c NS |
1389 | * Allocate buffer hash table for a given target. |
1390 | * For devices containing metadata (i.e. not the log/realtime devices) | |
1391 | * we need to allocate a much larger hash table. | |
1da177e4 LT |
1392 | */ |
1393 | STATIC void | |
1394 | xfs_alloc_bufhash( | |
1395 | xfs_buftarg_t *btp, | |
1396 | int external) | |
1397 | { | |
1398 | unsigned int i; | |
1399 | ||
1400 | btp->bt_hashshift = external ? 3 : 8; /* 8 or 256 buckets */ | |
1401 | btp->bt_hashmask = (1 << btp->bt_hashshift) - 1; | |
1402 | btp->bt_hash = kmem_zalloc((1 << btp->bt_hashshift) * | |
93c189c1 | 1403 | sizeof(xfs_bufhash_t), KM_SLEEP | KM_LARGE); |
1da177e4 LT |
1404 | for (i = 0; i < (1 << btp->bt_hashshift); i++) { |
1405 | spin_lock_init(&btp->bt_hash[i].bh_lock); | |
1406 | INIT_LIST_HEAD(&btp->bt_hash[i].bh_list); | |
1407 | } | |
1408 | } | |
1409 | ||
1410 | STATIC void | |
1411 | xfs_free_bufhash( | |
1412 | xfs_buftarg_t *btp) | |
1413 | { | |
f0e2d93c | 1414 | kmem_free(btp->bt_hash); |
1da177e4 LT |
1415 | btp->bt_hash = NULL; |
1416 | } | |
1417 | ||
a6867a68 | 1418 | /* |
ce8e922c | 1419 | * buftarg list for delwrite queue processing |
a6867a68 | 1420 | */ |
e6a0e9cd | 1421 | static LIST_HEAD(xfs_buftarg_list); |
7989cb8e | 1422 | static DEFINE_SPINLOCK(xfs_buftarg_lock); |
a6867a68 DC |
1423 | |
1424 | STATIC void | |
1425 | xfs_register_buftarg( | |
1426 | xfs_buftarg_t *btp) | |
1427 | { | |
1428 | spin_lock(&xfs_buftarg_lock); | |
1429 | list_add(&btp->bt_list, &xfs_buftarg_list); | |
1430 | spin_unlock(&xfs_buftarg_lock); | |
1431 | } | |
1432 | ||
1433 | STATIC void | |
1434 | xfs_unregister_buftarg( | |
1435 | xfs_buftarg_t *btp) | |
1436 | { | |
1437 | spin_lock(&xfs_buftarg_lock); | |
1438 | list_del(&btp->bt_list); | |
1439 | spin_unlock(&xfs_buftarg_lock); | |
1440 | } | |
1441 | ||
1da177e4 LT |
1442 | void |
1443 | xfs_free_buftarg( | |
c141b292 CH |
1444 | struct xfs_mount *mp, |
1445 | struct xfs_buftarg *btp) | |
1da177e4 LT |
1446 | { |
1447 | xfs_flush_buftarg(btp, 1); | |
c141b292 CH |
1448 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1449 | xfs_blkdev_issue_flush(btp); | |
1da177e4 | 1450 | xfs_free_bufhash(btp); |
ce8e922c | 1451 | iput(btp->bt_mapping->host); |
a6867a68 | 1452 | |
ce8e922c NS |
1453 | /* Unregister the buftarg first so that we don't get a |
1454 | * wakeup finding a non-existent task | |
1455 | */ | |
a6867a68 DC |
1456 | xfs_unregister_buftarg(btp); |
1457 | kthread_stop(btp->bt_task); | |
1458 | ||
f0e2d93c | 1459 | kmem_free(btp); |
1da177e4 LT |
1460 | } |
1461 | ||
1da177e4 LT |
1462 | STATIC int |
1463 | xfs_setsize_buftarg_flags( | |
1464 | xfs_buftarg_t *btp, | |
1465 | unsigned int blocksize, | |
1466 | unsigned int sectorsize, | |
1467 | int verbose) | |
1468 | { | |
ce8e922c NS |
1469 | btp->bt_bsize = blocksize; |
1470 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1471 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1472 | |
ce8e922c | 1473 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
1da177e4 LT |
1474 | printk(KERN_WARNING |
1475 | "XFS: Cannot set_blocksize to %u on device %s\n", | |
1476 | sectorsize, XFS_BUFTARG_NAME(btp)); | |
1477 | return EINVAL; | |
1478 | } | |
1479 | ||
1480 | if (verbose && | |
1481 | (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) { | |
1482 | printk(KERN_WARNING | |
1483 | "XFS: %u byte sectors in use on device %s. " | |
1484 | "This is suboptimal; %u or greater is ideal.\n", | |
1485 | sectorsize, XFS_BUFTARG_NAME(btp), | |
1486 | (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG); | |
1487 | } | |
1488 | ||
1489 | return 0; | |
1490 | } | |
1491 | ||
1492 | /* | |
ce8e922c NS |
1493 | * When allocating the initial buffer target we have not yet |
1494 | * read in the superblock, so don't know what sized sectors | |
1495 | * are being used is at this early stage. Play safe. | |
1496 | */ | |
1da177e4 LT |
1497 | STATIC int |
1498 | xfs_setsize_buftarg_early( | |
1499 | xfs_buftarg_t *btp, | |
1500 | struct block_device *bdev) | |
1501 | { | |
1502 | return xfs_setsize_buftarg_flags(btp, | |
1503 | PAGE_CACHE_SIZE, bdev_hardsect_size(bdev), 0); | |
1504 | } | |
1505 | ||
1506 | int | |
1507 | xfs_setsize_buftarg( | |
1508 | xfs_buftarg_t *btp, | |
1509 | unsigned int blocksize, | |
1510 | unsigned int sectorsize) | |
1511 | { | |
1512 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1513 | } | |
1514 | ||
1515 | STATIC int | |
1516 | xfs_mapping_buftarg( | |
1517 | xfs_buftarg_t *btp, | |
1518 | struct block_device *bdev) | |
1519 | { | |
1520 | struct backing_dev_info *bdi; | |
1521 | struct inode *inode; | |
1522 | struct address_space *mapping; | |
f5e54d6e | 1523 | static const struct address_space_operations mapping_aops = { |
1da177e4 | 1524 | .sync_page = block_sync_page, |
e965f963 | 1525 | .migratepage = fail_migrate_page, |
1da177e4 LT |
1526 | }; |
1527 | ||
1528 | inode = new_inode(bdev->bd_inode->i_sb); | |
1529 | if (!inode) { | |
1530 | printk(KERN_WARNING | |
1531 | "XFS: Cannot allocate mapping inode for device %s\n", | |
1532 | XFS_BUFTARG_NAME(btp)); | |
1533 | return ENOMEM; | |
1534 | } | |
1535 | inode->i_mode = S_IFBLK; | |
1536 | inode->i_bdev = bdev; | |
1537 | inode->i_rdev = bdev->bd_dev; | |
1538 | bdi = blk_get_backing_dev_info(bdev); | |
1539 | if (!bdi) | |
1540 | bdi = &default_backing_dev_info; | |
1541 | mapping = &inode->i_data; | |
1542 | mapping->a_ops = &mapping_aops; | |
1543 | mapping->backing_dev_info = bdi; | |
1544 | mapping_set_gfp_mask(mapping, GFP_NOFS); | |
ce8e922c | 1545 | btp->bt_mapping = mapping; |
1da177e4 LT |
1546 | return 0; |
1547 | } | |
1548 | ||
a6867a68 DC |
1549 | STATIC int |
1550 | xfs_alloc_delwrite_queue( | |
1551 | xfs_buftarg_t *btp) | |
1552 | { | |
1553 | int error = 0; | |
1554 | ||
1555 | INIT_LIST_HEAD(&btp->bt_list); | |
1556 | INIT_LIST_HEAD(&btp->bt_delwrite_queue); | |
007c61c6 | 1557 | spin_lock_init(&btp->bt_delwrite_lock); |
a6867a68 DC |
1558 | btp->bt_flags = 0; |
1559 | btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd"); | |
1560 | if (IS_ERR(btp->bt_task)) { | |
1561 | error = PTR_ERR(btp->bt_task); | |
1562 | goto out_error; | |
1563 | } | |
1564 | xfs_register_buftarg(btp); | |
1565 | out_error: | |
1566 | return error; | |
1567 | } | |
1568 | ||
1da177e4 LT |
1569 | xfs_buftarg_t * |
1570 | xfs_alloc_buftarg( | |
1571 | struct block_device *bdev, | |
1572 | int external) | |
1573 | { | |
1574 | xfs_buftarg_t *btp; | |
1575 | ||
1576 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1577 | ||
ce8e922c NS |
1578 | btp->bt_dev = bdev->bd_dev; |
1579 | btp->bt_bdev = bdev; | |
1da177e4 LT |
1580 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1581 | goto error; | |
1582 | if (xfs_mapping_buftarg(btp, bdev)) | |
1583 | goto error; | |
a6867a68 DC |
1584 | if (xfs_alloc_delwrite_queue(btp)) |
1585 | goto error; | |
1da177e4 LT |
1586 | xfs_alloc_bufhash(btp, external); |
1587 | return btp; | |
1588 | ||
1589 | error: | |
f0e2d93c | 1590 | kmem_free(btp); |
1da177e4 LT |
1591 | return NULL; |
1592 | } | |
1593 | ||
1594 | ||
1595 | /* | |
ce8e922c | 1596 | * Delayed write buffer handling |
1da177e4 | 1597 | */ |
1da177e4 | 1598 | STATIC void |
ce8e922c NS |
1599 | xfs_buf_delwri_queue( |
1600 | xfs_buf_t *bp, | |
1da177e4 LT |
1601 | int unlock) |
1602 | { | |
ce8e922c NS |
1603 | struct list_head *dwq = &bp->b_target->bt_delwrite_queue; |
1604 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; | |
a6867a68 | 1605 | |
ce8e922c NS |
1606 | XB_TRACE(bp, "delwri_q", (long)unlock); |
1607 | ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC)); | |
1da177e4 | 1608 | |
a6867a68 | 1609 | spin_lock(dwlk); |
1da177e4 | 1610 | /* If already in the queue, dequeue and place at tail */ |
ce8e922c NS |
1611 | if (!list_empty(&bp->b_list)) { |
1612 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1613 | if (unlock) | |
1614 | atomic_dec(&bp->b_hold); | |
1615 | list_del(&bp->b_list); | |
1da177e4 LT |
1616 | } |
1617 | ||
ce8e922c NS |
1618 | bp->b_flags |= _XBF_DELWRI_Q; |
1619 | list_add_tail(&bp->b_list, dwq); | |
1620 | bp->b_queuetime = jiffies; | |
a6867a68 | 1621 | spin_unlock(dwlk); |
1da177e4 LT |
1622 | |
1623 | if (unlock) | |
ce8e922c | 1624 | xfs_buf_unlock(bp); |
1da177e4 LT |
1625 | } |
1626 | ||
1627 | void | |
ce8e922c NS |
1628 | xfs_buf_delwri_dequeue( |
1629 | xfs_buf_t *bp) | |
1da177e4 | 1630 | { |
ce8e922c | 1631 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; |
1da177e4 LT |
1632 | int dequeued = 0; |
1633 | ||
a6867a68 | 1634 | spin_lock(dwlk); |
ce8e922c NS |
1635 | if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) { |
1636 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1637 | list_del_init(&bp->b_list); | |
1da177e4 LT |
1638 | dequeued = 1; |
1639 | } | |
ce8e922c | 1640 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q); |
a6867a68 | 1641 | spin_unlock(dwlk); |
1da177e4 LT |
1642 | |
1643 | if (dequeued) | |
ce8e922c | 1644 | xfs_buf_rele(bp); |
1da177e4 | 1645 | |
ce8e922c | 1646 | XB_TRACE(bp, "delwri_dq", (long)dequeued); |
1da177e4 LT |
1647 | } |
1648 | ||
1649 | STATIC void | |
ce8e922c | 1650 | xfs_buf_runall_queues( |
1da177e4 LT |
1651 | struct workqueue_struct *queue) |
1652 | { | |
1653 | flush_workqueue(queue); | |
1654 | } | |
1655 | ||
1da177e4 | 1656 | STATIC int |
23ea4032 | 1657 | xfsbufd_wakeup( |
15c84a47 NS |
1658 | int priority, |
1659 | gfp_t mask) | |
1da177e4 | 1660 | { |
da7f93e9 | 1661 | xfs_buftarg_t *btp; |
a6867a68 DC |
1662 | |
1663 | spin_lock(&xfs_buftarg_lock); | |
da7f93e9 | 1664 | list_for_each_entry(btp, &xfs_buftarg_list, bt_list) { |
ce8e922c | 1665 | if (test_bit(XBT_FORCE_SLEEP, &btp->bt_flags)) |
a6867a68 | 1666 | continue; |
ce8e922c | 1667 | set_bit(XBT_FORCE_FLUSH, &btp->bt_flags); |
a6867a68 DC |
1668 | wake_up_process(btp->bt_task); |
1669 | } | |
1670 | spin_unlock(&xfs_buftarg_lock); | |
1da177e4 LT |
1671 | return 0; |
1672 | } | |
1673 | ||
585e6d88 DC |
1674 | /* |
1675 | * Move as many buffers as specified to the supplied list | |
1676 | * idicating if we skipped any buffers to prevent deadlocks. | |
1677 | */ | |
1678 | STATIC int | |
1679 | xfs_buf_delwri_split( | |
1680 | xfs_buftarg_t *target, | |
1681 | struct list_head *list, | |
5e6a07df | 1682 | unsigned long age) |
585e6d88 DC |
1683 | { |
1684 | xfs_buf_t *bp, *n; | |
1685 | struct list_head *dwq = &target->bt_delwrite_queue; | |
1686 | spinlock_t *dwlk = &target->bt_delwrite_lock; | |
1687 | int skipped = 0; | |
5e6a07df | 1688 | int force; |
585e6d88 | 1689 | |
5e6a07df | 1690 | force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
585e6d88 DC |
1691 | INIT_LIST_HEAD(list); |
1692 | spin_lock(dwlk); | |
1693 | list_for_each_entry_safe(bp, n, dwq, b_list) { | |
1694 | XB_TRACE(bp, "walkq1", (long)xfs_buf_ispin(bp)); | |
1695 | ASSERT(bp->b_flags & XBF_DELWRI); | |
1696 | ||
1697 | if (!xfs_buf_ispin(bp) && !xfs_buf_cond_lock(bp)) { | |
5e6a07df | 1698 | if (!force && |
585e6d88 DC |
1699 | time_before(jiffies, bp->b_queuetime + age)) { |
1700 | xfs_buf_unlock(bp); | |
1701 | break; | |
1702 | } | |
1703 | ||
1704 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q| | |
1705 | _XBF_RUN_QUEUES); | |
1706 | bp->b_flags |= XBF_WRITE; | |
1707 | list_move_tail(&bp->b_list, list); | |
1708 | } else | |
1709 | skipped++; | |
1710 | } | |
1711 | spin_unlock(dwlk); | |
1712 | ||
1713 | return skipped; | |
1714 | ||
1715 | } | |
1716 | ||
1da177e4 | 1717 | STATIC int |
23ea4032 | 1718 | xfsbufd( |
585e6d88 | 1719 | void *data) |
1da177e4 | 1720 | { |
585e6d88 DC |
1721 | struct list_head tmp; |
1722 | xfs_buftarg_t *target = (xfs_buftarg_t *)data; | |
1723 | int count; | |
1724 | xfs_buf_t *bp; | |
1da177e4 | 1725 | |
1da177e4 LT |
1726 | current->flags |= PF_MEMALLOC; |
1727 | ||
978c7b2f RW |
1728 | set_freezable(); |
1729 | ||
1da177e4 | 1730 | do { |
3e1d1d28 | 1731 | if (unlikely(freezing(current))) { |
ce8e922c | 1732 | set_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
3e1d1d28 | 1733 | refrigerator(); |
abd0cf7a | 1734 | } else { |
ce8e922c | 1735 | clear_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
abd0cf7a | 1736 | } |
1da177e4 | 1737 | |
15c84a47 NS |
1738 | schedule_timeout_interruptible( |
1739 | xfs_buf_timer_centisecs * msecs_to_jiffies(10)); | |
1da177e4 | 1740 | |
585e6d88 | 1741 | xfs_buf_delwri_split(target, &tmp, |
5e6a07df | 1742 | xfs_buf_age_centisecs * msecs_to_jiffies(10)); |
1da177e4 | 1743 | |
585e6d88 | 1744 | count = 0; |
1da177e4 | 1745 | while (!list_empty(&tmp)) { |
ce8e922c NS |
1746 | bp = list_entry(tmp.next, xfs_buf_t, b_list); |
1747 | ASSERT(target == bp->b_target); | |
1da177e4 | 1748 | |
ce8e922c NS |
1749 | list_del_init(&bp->b_list); |
1750 | xfs_buf_iostrategy(bp); | |
585e6d88 | 1751 | count++; |
1da177e4 LT |
1752 | } |
1753 | ||
27e88bf6 FB |
1754 | if (as_list_len > 0) |
1755 | purge_addresses(); | |
f07c2250 NS |
1756 | if (count) |
1757 | blk_run_address_space(target->bt_mapping); | |
1da177e4 | 1758 | |
4df08c52 | 1759 | } while (!kthread_should_stop()); |
1da177e4 | 1760 | |
4df08c52 | 1761 | return 0; |
1da177e4 LT |
1762 | } |
1763 | ||
1764 | /* | |
ce8e922c NS |
1765 | * Go through all incore buffers, and release buffers if they belong to |
1766 | * the given device. This is used in filesystem error handling to | |
1767 | * preserve the consistency of its metadata. | |
1da177e4 LT |
1768 | */ |
1769 | int | |
1770 | xfs_flush_buftarg( | |
585e6d88 DC |
1771 | xfs_buftarg_t *target, |
1772 | int wait) | |
1da177e4 | 1773 | { |
585e6d88 DC |
1774 | struct list_head tmp; |
1775 | xfs_buf_t *bp, *n; | |
1776 | int pincount = 0; | |
1da177e4 | 1777 | |
ce8e922c NS |
1778 | xfs_buf_runall_queues(xfsdatad_workqueue); |
1779 | xfs_buf_runall_queues(xfslogd_workqueue); | |
1da177e4 | 1780 | |
5e6a07df DC |
1781 | set_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
1782 | pincount = xfs_buf_delwri_split(target, &tmp, 0); | |
1da177e4 LT |
1783 | |
1784 | /* | |
1785 | * Dropped the delayed write list lock, now walk the temporary list | |
1786 | */ | |
ce8e922c | 1787 | list_for_each_entry_safe(bp, n, &tmp, b_list) { |
585e6d88 | 1788 | ASSERT(target == bp->b_target); |
1da177e4 | 1789 | if (wait) |
ce8e922c | 1790 | bp->b_flags &= ~XBF_ASYNC; |
1da177e4 | 1791 | else |
ce8e922c | 1792 | list_del_init(&bp->b_list); |
1da177e4 | 1793 | |
ce8e922c | 1794 | xfs_buf_iostrategy(bp); |
1da177e4 LT |
1795 | } |
1796 | ||
f07c2250 NS |
1797 | if (wait) |
1798 | blk_run_address_space(target->bt_mapping); | |
1799 | ||
1da177e4 LT |
1800 | /* |
1801 | * Remaining list items must be flushed before returning | |
1802 | */ | |
1803 | while (!list_empty(&tmp)) { | |
ce8e922c | 1804 | bp = list_entry(tmp.next, xfs_buf_t, b_list); |
1da177e4 | 1805 | |
ce8e922c NS |
1806 | list_del_init(&bp->b_list); |
1807 | xfs_iowait(bp); | |
1808 | xfs_buf_relse(bp); | |
1da177e4 LT |
1809 | } |
1810 | ||
1da177e4 LT |
1811 | return pincount; |
1812 | } | |
1813 | ||
04d8b284 | 1814 | int __init |
ce8e922c | 1815 | xfs_buf_init(void) |
1da177e4 | 1816 | { |
ce8e922c | 1817 | #ifdef XFS_BUF_TRACE |
5695ef46 | 1818 | xfs_buf_trace_buf = ktrace_alloc(XFS_BUF_TRACE_SIZE, KM_NOFS); |
04d8b284 CH |
1819 | #endif |
1820 | ||
8758280f NS |
1821 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1822 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1823 | if (!xfs_buf_zone) |
04d8b284 CH |
1824 | goto out_free_trace_buf; |
1825 | ||
b4337692 | 1826 | xfslogd_workqueue = create_workqueue("xfslogd"); |
23ea4032 | 1827 | if (!xfslogd_workqueue) |
04d8b284 | 1828 | goto out_free_buf_zone; |
1da177e4 | 1829 | |
b4337692 | 1830 | xfsdatad_workqueue = create_workqueue("xfsdatad"); |
23ea4032 CH |
1831 | if (!xfsdatad_workqueue) |
1832 | goto out_destroy_xfslogd_workqueue; | |
1da177e4 | 1833 | |
8e1f936b | 1834 | register_shrinker(&xfs_buf_shake); |
23ea4032 | 1835 | return 0; |
1da177e4 | 1836 | |
23ea4032 CH |
1837 | out_destroy_xfslogd_workqueue: |
1838 | destroy_workqueue(xfslogd_workqueue); | |
23ea4032 | 1839 | out_free_buf_zone: |
ce8e922c | 1840 | kmem_zone_destroy(xfs_buf_zone); |
04d8b284 | 1841 | out_free_trace_buf: |
ce8e922c NS |
1842 | #ifdef XFS_BUF_TRACE |
1843 | ktrace_free(xfs_buf_trace_buf); | |
23ea4032 | 1844 | #endif |
8758280f | 1845 | return -ENOMEM; |
1da177e4 LT |
1846 | } |
1847 | ||
1da177e4 | 1848 | void |
ce8e922c | 1849 | xfs_buf_terminate(void) |
1da177e4 | 1850 | { |
8e1f936b | 1851 | unregister_shrinker(&xfs_buf_shake); |
04d8b284 CH |
1852 | destroy_workqueue(xfsdatad_workqueue); |
1853 | destroy_workqueue(xfslogd_workqueue); | |
ce8e922c NS |
1854 | kmem_zone_destroy(xfs_buf_zone); |
1855 | #ifdef XFS_BUF_TRACE | |
1856 | ktrace_free(xfs_buf_trace_buf); | |
1da177e4 | 1857 | #endif |
1da177e4 | 1858 | } |
e6a0e9cd TS |
1859 | |
1860 | #ifdef CONFIG_KDB_MODULES | |
1861 | struct list_head * | |
1862 | xfs_get_buftarg_list(void) | |
1863 | { | |
1864 | return &xfs_buftarg_list; | |
1865 | } | |
1866 | #endif |