Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
f07c2250 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 3 | * All Rights Reserved. |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
93c189c1 | 18 | #include "xfs.h" |
1da177e4 LT |
19 | #include <linux/stddef.h> |
20 | #include <linux/errno.h> | |
5a0e3ad6 | 21 | #include <linux/gfp.h> |
1da177e4 LT |
22 | #include <linux/pagemap.h> |
23 | #include <linux/init.h> | |
24 | #include <linux/vmalloc.h> | |
25 | #include <linux/bio.h> | |
26 | #include <linux/sysctl.h> | |
27 | #include <linux/proc_fs.h> | |
28 | #include <linux/workqueue.h> | |
29 | #include <linux/percpu.h> | |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/hash.h> | |
4df08c52 | 32 | #include <linux/kthread.h> |
b20a3503 | 33 | #include <linux/migrate.h> |
3fcfab16 | 34 | #include <linux/backing-dev.h> |
7dfb7103 | 35 | #include <linux/freezer.h> |
089716aa | 36 | #include <linux/list_sort.h> |
1da177e4 | 37 | |
b7963133 CH |
38 | #include "xfs_sb.h" |
39 | #include "xfs_inum.h" | |
ed3b4d6c | 40 | #include "xfs_log.h" |
b7963133 | 41 | #include "xfs_ag.h" |
b7963133 | 42 | #include "xfs_mount.h" |
0b1b213f | 43 | #include "xfs_trace.h" |
b7963133 | 44 | |
7989cb8e | 45 | static kmem_zone_t *xfs_buf_zone; |
a6867a68 | 46 | STATIC int xfsbufd(void *); |
7f8275d0 | 47 | STATIC int xfsbufd_wakeup(struct shrinker *, int, gfp_t); |
ce8e922c | 48 | STATIC void xfs_buf_delwri_queue(xfs_buf_t *, int); |
8e1f936b RR |
49 | static struct shrinker xfs_buf_shake = { |
50 | .shrink = xfsbufd_wakeup, | |
51 | .seeks = DEFAULT_SEEKS, | |
52 | }; | |
23ea4032 | 53 | |
7989cb8e | 54 | static struct workqueue_struct *xfslogd_workqueue; |
0829c360 | 55 | struct workqueue_struct *xfsdatad_workqueue; |
c626d174 | 56 | struct workqueue_struct *xfsconvertd_workqueue; |
1da177e4 | 57 | |
ce8e922c NS |
58 | #ifdef XFS_BUF_LOCK_TRACKING |
59 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
60 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
61 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 62 | #else |
ce8e922c NS |
63 | # define XB_SET_OWNER(bp) do { } while (0) |
64 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
65 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
66 | #endif |
67 | ||
ce8e922c NS |
68 | #define xb_to_gfp(flags) \ |
69 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \ | |
70 | ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN) | |
1da177e4 | 71 | |
ce8e922c NS |
72 | #define xb_to_km(flags) \ |
73 | (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP) | |
1da177e4 | 74 | |
ce8e922c NS |
75 | #define xfs_buf_allocate(flags) \ |
76 | kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags)) | |
77 | #define xfs_buf_deallocate(bp) \ | |
78 | kmem_zone_free(xfs_buf_zone, (bp)); | |
1da177e4 | 79 | |
73c77e2c JB |
80 | static inline int |
81 | xfs_buf_is_vmapped( | |
82 | struct xfs_buf *bp) | |
83 | { | |
84 | /* | |
85 | * Return true if the buffer is vmapped. | |
86 | * | |
87 | * The XBF_MAPPED flag is set if the buffer should be mapped, but the | |
88 | * code is clever enough to know it doesn't have to map a single page, | |
89 | * so the check has to be both for XBF_MAPPED and bp->b_page_count > 1. | |
90 | */ | |
91 | return (bp->b_flags & XBF_MAPPED) && bp->b_page_count > 1; | |
92 | } | |
93 | ||
94 | static inline int | |
95 | xfs_buf_vmap_len( | |
96 | struct xfs_buf *bp) | |
97 | { | |
98 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
99 | } | |
100 | ||
1da177e4 | 101 | /* |
ce8e922c | 102 | * Page Region interfaces. |
1da177e4 | 103 | * |
ce8e922c NS |
104 | * For pages in filesystems where the blocksize is smaller than the |
105 | * pagesize, we use the page->private field (long) to hold a bitmap | |
106 | * of uptodate regions within the page. | |
1da177e4 | 107 | * |
ce8e922c | 108 | * Each such region is "bytes per page / bits per long" bytes long. |
1da177e4 | 109 | * |
ce8e922c NS |
110 | * NBPPR == number-of-bytes-per-page-region |
111 | * BTOPR == bytes-to-page-region (rounded up) | |
112 | * BTOPRT == bytes-to-page-region-truncated (rounded down) | |
1da177e4 LT |
113 | */ |
114 | #if (BITS_PER_LONG == 32) | |
115 | #define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */ | |
116 | #elif (BITS_PER_LONG == 64) | |
117 | #define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */ | |
118 | #else | |
119 | #error BITS_PER_LONG must be 32 or 64 | |
120 | #endif | |
121 | #define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG) | |
122 | #define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT) | |
123 | #define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT)) | |
124 | ||
125 | STATIC unsigned long | |
126 | page_region_mask( | |
127 | size_t offset, | |
128 | size_t length) | |
129 | { | |
130 | unsigned long mask; | |
131 | int first, final; | |
132 | ||
133 | first = BTOPR(offset); | |
134 | final = BTOPRT(offset + length - 1); | |
135 | first = min(first, final); | |
136 | ||
137 | mask = ~0UL; | |
138 | mask <<= BITS_PER_LONG - (final - first); | |
139 | mask >>= BITS_PER_LONG - (final); | |
140 | ||
141 | ASSERT(offset + length <= PAGE_CACHE_SIZE); | |
142 | ASSERT((final - first) < BITS_PER_LONG && (final - first) >= 0); | |
143 | ||
144 | return mask; | |
145 | } | |
146 | ||
b8f82a4a | 147 | STATIC void |
1da177e4 LT |
148 | set_page_region( |
149 | struct page *page, | |
150 | size_t offset, | |
151 | size_t length) | |
152 | { | |
4c21e2f2 HD |
153 | set_page_private(page, |
154 | page_private(page) | page_region_mask(offset, length)); | |
155 | if (page_private(page) == ~0UL) | |
1da177e4 LT |
156 | SetPageUptodate(page); |
157 | } | |
158 | ||
b8f82a4a | 159 | STATIC int |
1da177e4 LT |
160 | test_page_region( |
161 | struct page *page, | |
162 | size_t offset, | |
163 | size_t length) | |
164 | { | |
165 | unsigned long mask = page_region_mask(offset, length); | |
166 | ||
4c21e2f2 | 167 | return (mask && (page_private(page) & mask) == mask); |
1da177e4 LT |
168 | } |
169 | ||
1da177e4 | 170 | /* |
ce8e922c | 171 | * Internal xfs_buf_t object manipulation |
1da177e4 LT |
172 | */ |
173 | ||
174 | STATIC void | |
ce8e922c NS |
175 | _xfs_buf_initialize( |
176 | xfs_buf_t *bp, | |
1da177e4 | 177 | xfs_buftarg_t *target, |
204ab25f | 178 | xfs_off_t range_base, |
1da177e4 | 179 | size_t range_length, |
ce8e922c | 180 | xfs_buf_flags_t flags) |
1da177e4 LT |
181 | { |
182 | /* | |
ce8e922c | 183 | * We don't want certain flags to appear in b_flags. |
1da177e4 | 184 | */ |
ce8e922c NS |
185 | flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD); |
186 | ||
187 | memset(bp, 0, sizeof(xfs_buf_t)); | |
188 | atomic_set(&bp->b_hold, 1); | |
b4dd330b | 189 | init_completion(&bp->b_iowait); |
ce8e922c | 190 | INIT_LIST_HEAD(&bp->b_list); |
74f75a0c | 191 | RB_CLEAR_NODE(&bp->b_rbnode); |
ce8e922c NS |
192 | init_MUTEX_LOCKED(&bp->b_sema); /* held, no waiters */ |
193 | XB_SET_OWNER(bp); | |
194 | bp->b_target = target; | |
195 | bp->b_file_offset = range_base; | |
1da177e4 LT |
196 | /* |
197 | * Set buffer_length and count_desired to the same value initially. | |
198 | * I/O routines should use count_desired, which will be the same in | |
199 | * most cases but may be reset (e.g. XFS recovery). | |
200 | */ | |
ce8e922c NS |
201 | bp->b_buffer_length = bp->b_count_desired = range_length; |
202 | bp->b_flags = flags; | |
203 | bp->b_bn = XFS_BUF_DADDR_NULL; | |
204 | atomic_set(&bp->b_pin_count, 0); | |
205 | init_waitqueue_head(&bp->b_waiters); | |
206 | ||
207 | XFS_STATS_INC(xb_create); | |
0b1b213f CH |
208 | |
209 | trace_xfs_buf_init(bp, _RET_IP_); | |
1da177e4 LT |
210 | } |
211 | ||
212 | /* | |
ce8e922c NS |
213 | * Allocate a page array capable of holding a specified number |
214 | * of pages, and point the page buf at it. | |
1da177e4 LT |
215 | */ |
216 | STATIC int | |
ce8e922c NS |
217 | _xfs_buf_get_pages( |
218 | xfs_buf_t *bp, | |
1da177e4 | 219 | int page_count, |
ce8e922c | 220 | xfs_buf_flags_t flags) |
1da177e4 LT |
221 | { |
222 | /* Make sure that we have a page list */ | |
ce8e922c NS |
223 | if (bp->b_pages == NULL) { |
224 | bp->b_offset = xfs_buf_poff(bp->b_file_offset); | |
225 | bp->b_page_count = page_count; | |
226 | if (page_count <= XB_PAGES) { | |
227 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 228 | } else { |
ce8e922c NS |
229 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
230 | page_count, xb_to_km(flags)); | |
231 | if (bp->b_pages == NULL) | |
1da177e4 LT |
232 | return -ENOMEM; |
233 | } | |
ce8e922c | 234 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
235 | } |
236 | return 0; | |
237 | } | |
238 | ||
239 | /* | |
ce8e922c | 240 | * Frees b_pages if it was allocated. |
1da177e4 LT |
241 | */ |
242 | STATIC void | |
ce8e922c | 243 | _xfs_buf_free_pages( |
1da177e4 LT |
244 | xfs_buf_t *bp) |
245 | { | |
ce8e922c | 246 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 247 | kmem_free(bp->b_pages); |
3fc98b1a | 248 | bp->b_pages = NULL; |
1da177e4 LT |
249 | } |
250 | } | |
251 | ||
252 | /* | |
253 | * Releases the specified buffer. | |
254 | * | |
255 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 256 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
257 | * hashed and refcounted buffers |
258 | */ | |
259 | void | |
ce8e922c | 260 | xfs_buf_free( |
1da177e4 LT |
261 | xfs_buf_t *bp) |
262 | { | |
0b1b213f | 263 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 264 | |
1fa40b01 | 265 | if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) { |
1da177e4 LT |
266 | uint i; |
267 | ||
73c77e2c | 268 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
269 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
270 | bp->b_page_count); | |
1da177e4 | 271 | |
948ecdb4 NS |
272 | for (i = 0; i < bp->b_page_count; i++) { |
273 | struct page *page = bp->b_pages[i]; | |
274 | ||
1fa40b01 CH |
275 | if (bp->b_flags & _XBF_PAGE_CACHE) |
276 | ASSERT(!PagePrivate(page)); | |
948ecdb4 NS |
277 | page_cache_release(page); |
278 | } | |
1da177e4 | 279 | } |
3fc98b1a | 280 | _xfs_buf_free_pages(bp); |
ce8e922c | 281 | xfs_buf_deallocate(bp); |
1da177e4 LT |
282 | } |
283 | ||
284 | /* | |
285 | * Finds all pages for buffer in question and builds it's page list. | |
286 | */ | |
287 | STATIC int | |
ce8e922c | 288 | _xfs_buf_lookup_pages( |
1da177e4 LT |
289 | xfs_buf_t *bp, |
290 | uint flags) | |
291 | { | |
ce8e922c NS |
292 | struct address_space *mapping = bp->b_target->bt_mapping; |
293 | size_t blocksize = bp->b_target->bt_bsize; | |
294 | size_t size = bp->b_count_desired; | |
1da177e4 | 295 | size_t nbytes, offset; |
ce8e922c | 296 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 LT |
297 | unsigned short page_count, i; |
298 | pgoff_t first; | |
204ab25f | 299 | xfs_off_t end; |
1da177e4 LT |
300 | int error; |
301 | ||
ce8e922c NS |
302 | end = bp->b_file_offset + bp->b_buffer_length; |
303 | page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset); | |
1da177e4 | 304 | |
ce8e922c | 305 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
306 | if (unlikely(error)) |
307 | return error; | |
ce8e922c | 308 | bp->b_flags |= _XBF_PAGE_CACHE; |
1da177e4 | 309 | |
ce8e922c NS |
310 | offset = bp->b_offset; |
311 | first = bp->b_file_offset >> PAGE_CACHE_SHIFT; | |
1da177e4 | 312 | |
ce8e922c | 313 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
314 | struct page *page; |
315 | uint retries = 0; | |
316 | ||
317 | retry: | |
318 | page = find_or_create_page(mapping, first + i, gfp_mask); | |
319 | if (unlikely(page == NULL)) { | |
ce8e922c NS |
320 | if (flags & XBF_READ_AHEAD) { |
321 | bp->b_page_count = i; | |
6ab455ee CH |
322 | for (i = 0; i < bp->b_page_count; i++) |
323 | unlock_page(bp->b_pages[i]); | |
1da177e4 LT |
324 | return -ENOMEM; |
325 | } | |
326 | ||
327 | /* | |
328 | * This could deadlock. | |
329 | * | |
330 | * But until all the XFS lowlevel code is revamped to | |
331 | * handle buffer allocation failures we can't do much. | |
332 | */ | |
333 | if (!(++retries % 100)) | |
334 | printk(KERN_ERR | |
335 | "XFS: possible memory allocation " | |
336 | "deadlock in %s (mode:0x%x)\n", | |
34a622b2 | 337 | __func__, gfp_mask); |
1da177e4 | 338 | |
ce8e922c | 339 | XFS_STATS_INC(xb_page_retries); |
7f8275d0 | 340 | xfsbufd_wakeup(NULL, 0, gfp_mask); |
8aa7e847 | 341 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
342 | goto retry; |
343 | } | |
344 | ||
ce8e922c | 345 | XFS_STATS_INC(xb_page_found); |
1da177e4 LT |
346 | |
347 | nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset); | |
348 | size -= nbytes; | |
349 | ||
948ecdb4 | 350 | ASSERT(!PagePrivate(page)); |
1da177e4 LT |
351 | if (!PageUptodate(page)) { |
352 | page_count--; | |
6ab455ee CH |
353 | if (blocksize >= PAGE_CACHE_SIZE) { |
354 | if (flags & XBF_READ) | |
355 | bp->b_flags |= _XBF_PAGE_LOCKED; | |
356 | } else if (!PagePrivate(page)) { | |
1da177e4 LT |
357 | if (test_page_region(page, offset, nbytes)) |
358 | page_count++; | |
359 | } | |
360 | } | |
361 | ||
ce8e922c | 362 | bp->b_pages[i] = page; |
1da177e4 LT |
363 | offset = 0; |
364 | } | |
365 | ||
6ab455ee CH |
366 | if (!(bp->b_flags & _XBF_PAGE_LOCKED)) { |
367 | for (i = 0; i < bp->b_page_count; i++) | |
368 | unlock_page(bp->b_pages[i]); | |
369 | } | |
370 | ||
ce8e922c NS |
371 | if (page_count == bp->b_page_count) |
372 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 373 | |
1da177e4 LT |
374 | return error; |
375 | } | |
376 | ||
377 | /* | |
378 | * Map buffer into kernel address-space if nessecary. | |
379 | */ | |
380 | STATIC int | |
ce8e922c | 381 | _xfs_buf_map_pages( |
1da177e4 LT |
382 | xfs_buf_t *bp, |
383 | uint flags) | |
384 | { | |
385 | /* A single page buffer is always mappable */ | |
ce8e922c NS |
386 | if (bp->b_page_count == 1) { |
387 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; | |
388 | bp->b_flags |= XBF_MAPPED; | |
389 | } else if (flags & XBF_MAPPED) { | |
8a262e57 AE |
390 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, |
391 | -1, PAGE_KERNEL); | |
ce8e922c | 392 | if (unlikely(bp->b_addr == NULL)) |
1da177e4 | 393 | return -ENOMEM; |
ce8e922c NS |
394 | bp->b_addr += bp->b_offset; |
395 | bp->b_flags |= XBF_MAPPED; | |
1da177e4 LT |
396 | } |
397 | ||
398 | return 0; | |
399 | } | |
400 | ||
401 | /* | |
402 | * Finding and Reading Buffers | |
403 | */ | |
404 | ||
405 | /* | |
ce8e922c | 406 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 LT |
407 | * a given range of an inode. The buffer is returned |
408 | * locked. If other overlapping buffers exist, they are | |
409 | * released before the new buffer is created and locked, | |
410 | * which may imply that this call will block until those buffers | |
411 | * are unlocked. No I/O is implied by this call. | |
412 | */ | |
413 | xfs_buf_t * | |
ce8e922c | 414 | _xfs_buf_find( |
1da177e4 | 415 | xfs_buftarg_t *btp, /* block device target */ |
204ab25f | 416 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 417 | size_t isize, /* length of range */ |
ce8e922c NS |
418 | xfs_buf_flags_t flags, |
419 | xfs_buf_t *new_bp) | |
1da177e4 | 420 | { |
204ab25f | 421 | xfs_off_t range_base; |
1da177e4 | 422 | size_t range_length; |
74f75a0c DC |
423 | struct xfs_perag *pag; |
424 | struct rb_node **rbp; | |
425 | struct rb_node *parent; | |
426 | xfs_buf_t *bp; | |
1da177e4 LT |
427 | |
428 | range_base = (ioff << BBSHIFT); | |
429 | range_length = (isize << BBSHIFT); | |
430 | ||
431 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
ce8e922c | 432 | ASSERT(!(range_length < (1 << btp->bt_sshift))); |
204ab25f | 433 | ASSERT(!(range_base & (xfs_off_t)btp->bt_smask)); |
1da177e4 | 434 | |
74f75a0c DC |
435 | /* get tree root */ |
436 | pag = xfs_perag_get(btp->bt_mount, | |
437 | xfs_daddr_to_agno(btp->bt_mount, ioff)); | |
438 | ||
439 | /* walk tree */ | |
440 | spin_lock(&pag->pag_buf_lock); | |
441 | rbp = &pag->pag_buf_tree.rb_node; | |
442 | parent = NULL; | |
443 | bp = NULL; | |
444 | while (*rbp) { | |
445 | parent = *rbp; | |
446 | bp = rb_entry(parent, struct xfs_buf, b_rbnode); | |
447 | ||
448 | if (range_base < bp->b_file_offset) | |
449 | rbp = &(*rbp)->rb_left; | |
450 | else if (range_base > bp->b_file_offset) | |
451 | rbp = &(*rbp)->rb_right; | |
452 | else { | |
453 | /* | |
454 | * found a block offset match. If the range doesn't | |
455 | * match, the only way this is allowed is if the buffer | |
456 | * in the cache is stale and the transaction that made | |
457 | * it stale has not yet committed. i.e. we are | |
458 | * reallocating a busy extent. Skip this buffer and | |
459 | * continue searching to the right for an exact match. | |
460 | */ | |
461 | if (bp->b_buffer_length != range_length) { | |
462 | ASSERT(bp->b_flags & XBF_STALE); | |
463 | rbp = &(*rbp)->rb_right; | |
464 | continue; | |
465 | } | |
ce8e922c | 466 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
467 | goto found; |
468 | } | |
469 | } | |
470 | ||
471 | /* No match found */ | |
ce8e922c NS |
472 | if (new_bp) { |
473 | _xfs_buf_initialize(new_bp, btp, range_base, | |
1da177e4 | 474 | range_length, flags); |
74f75a0c DC |
475 | rb_link_node(&new_bp->b_rbnode, parent, rbp); |
476 | rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree); | |
477 | /* the buffer keeps the perag reference until it is freed */ | |
478 | new_bp->b_pag = pag; | |
479 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 480 | } else { |
ce8e922c | 481 | XFS_STATS_INC(xb_miss_locked); |
74f75a0c DC |
482 | spin_unlock(&pag->pag_buf_lock); |
483 | xfs_perag_put(pag); | |
1da177e4 | 484 | } |
ce8e922c | 485 | return new_bp; |
1da177e4 LT |
486 | |
487 | found: | |
74f75a0c DC |
488 | spin_unlock(&pag->pag_buf_lock); |
489 | xfs_perag_put(pag); | |
1da177e4 LT |
490 | |
491 | /* Attempt to get the semaphore without sleeping, | |
492 | * if this does not work then we need to drop the | |
493 | * spinlock and do a hard attempt on the semaphore. | |
494 | */ | |
ce8e922c NS |
495 | if (down_trylock(&bp->b_sema)) { |
496 | if (!(flags & XBF_TRYLOCK)) { | |
1da177e4 | 497 | /* wait for buffer ownership */ |
ce8e922c NS |
498 | xfs_buf_lock(bp); |
499 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
500 | } else { |
501 | /* We asked for a trylock and failed, no need | |
502 | * to look at file offset and length here, we | |
ce8e922c NS |
503 | * know that this buffer at least overlaps our |
504 | * buffer and is locked, therefore our buffer | |
505 | * either does not exist, or is this buffer. | |
1da177e4 | 506 | */ |
ce8e922c NS |
507 | xfs_buf_rele(bp); |
508 | XFS_STATS_INC(xb_busy_locked); | |
509 | return NULL; | |
1da177e4 LT |
510 | } |
511 | } else { | |
512 | /* trylock worked */ | |
ce8e922c | 513 | XB_SET_OWNER(bp); |
1da177e4 LT |
514 | } |
515 | ||
ce8e922c NS |
516 | if (bp->b_flags & XBF_STALE) { |
517 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
518 | bp->b_flags &= XBF_MAPPED; | |
2f926587 | 519 | } |
0b1b213f CH |
520 | |
521 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ce8e922c NS |
522 | XFS_STATS_INC(xb_get_locked); |
523 | return bp; | |
1da177e4 LT |
524 | } |
525 | ||
526 | /* | |
ce8e922c | 527 | * Assembles a buffer covering the specified range. |
1da177e4 LT |
528 | * Storage in memory for all portions of the buffer will be allocated, |
529 | * although backing storage may not be. | |
530 | */ | |
531 | xfs_buf_t * | |
6ad112bf | 532 | xfs_buf_get( |
1da177e4 | 533 | xfs_buftarg_t *target,/* target for buffer */ |
204ab25f | 534 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 535 | size_t isize, /* length of range */ |
ce8e922c | 536 | xfs_buf_flags_t flags) |
1da177e4 | 537 | { |
ce8e922c | 538 | xfs_buf_t *bp, *new_bp; |
1da177e4 LT |
539 | int error = 0, i; |
540 | ||
ce8e922c NS |
541 | new_bp = xfs_buf_allocate(flags); |
542 | if (unlikely(!new_bp)) | |
1da177e4 LT |
543 | return NULL; |
544 | ||
ce8e922c NS |
545 | bp = _xfs_buf_find(target, ioff, isize, flags, new_bp); |
546 | if (bp == new_bp) { | |
547 | error = _xfs_buf_lookup_pages(bp, flags); | |
1da177e4 LT |
548 | if (error) |
549 | goto no_buffer; | |
550 | } else { | |
ce8e922c NS |
551 | xfs_buf_deallocate(new_bp); |
552 | if (unlikely(bp == NULL)) | |
1da177e4 LT |
553 | return NULL; |
554 | } | |
555 | ||
ce8e922c NS |
556 | for (i = 0; i < bp->b_page_count; i++) |
557 | mark_page_accessed(bp->b_pages[i]); | |
1da177e4 | 558 | |
ce8e922c NS |
559 | if (!(bp->b_flags & XBF_MAPPED)) { |
560 | error = _xfs_buf_map_pages(bp, flags); | |
1da177e4 LT |
561 | if (unlikely(error)) { |
562 | printk(KERN_WARNING "%s: failed to map pages\n", | |
34a622b2 | 563 | __func__); |
1da177e4 LT |
564 | goto no_buffer; |
565 | } | |
566 | } | |
567 | ||
ce8e922c | 568 | XFS_STATS_INC(xb_get); |
1da177e4 LT |
569 | |
570 | /* | |
571 | * Always fill in the block number now, the mapped cases can do | |
572 | * their own overlay of this later. | |
573 | */ | |
ce8e922c NS |
574 | bp->b_bn = ioff; |
575 | bp->b_count_desired = bp->b_buffer_length; | |
1da177e4 | 576 | |
0b1b213f | 577 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 578 | return bp; |
1da177e4 LT |
579 | |
580 | no_buffer: | |
ce8e922c NS |
581 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
582 | xfs_buf_unlock(bp); | |
583 | xfs_buf_rele(bp); | |
1da177e4 LT |
584 | return NULL; |
585 | } | |
586 | ||
5d765b97 CH |
587 | STATIC int |
588 | _xfs_buf_read( | |
589 | xfs_buf_t *bp, | |
590 | xfs_buf_flags_t flags) | |
591 | { | |
592 | int status; | |
593 | ||
5d765b97 CH |
594 | ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE))); |
595 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); | |
596 | ||
597 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \ | |
598 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
599 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \ | |
600 | XBF_READ_AHEAD | _XBF_RUN_QUEUES); | |
601 | ||
602 | status = xfs_buf_iorequest(bp); | |
ec53d1db DC |
603 | if (status || XFS_BUF_ISERROR(bp) || (flags & XBF_ASYNC)) |
604 | return status; | |
605 | return xfs_buf_iowait(bp); | |
5d765b97 CH |
606 | } |
607 | ||
1da177e4 | 608 | xfs_buf_t * |
6ad112bf | 609 | xfs_buf_read( |
1da177e4 | 610 | xfs_buftarg_t *target, |
204ab25f | 611 | xfs_off_t ioff, |
1da177e4 | 612 | size_t isize, |
ce8e922c | 613 | xfs_buf_flags_t flags) |
1da177e4 | 614 | { |
ce8e922c NS |
615 | xfs_buf_t *bp; |
616 | ||
617 | flags |= XBF_READ; | |
618 | ||
6ad112bf | 619 | bp = xfs_buf_get(target, ioff, isize, flags); |
ce8e922c | 620 | if (bp) { |
0b1b213f CH |
621 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
622 | ||
ce8e922c | 623 | if (!XFS_BUF_ISDONE(bp)) { |
ce8e922c | 624 | XFS_STATS_INC(xb_get_read); |
5d765b97 | 625 | _xfs_buf_read(bp, flags); |
ce8e922c | 626 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
627 | /* |
628 | * Read ahead call which is already satisfied, | |
629 | * drop the buffer | |
630 | */ | |
631 | goto no_buffer; | |
632 | } else { | |
1da177e4 | 633 | /* We do not want read in the flags */ |
ce8e922c | 634 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
635 | } |
636 | } | |
637 | ||
ce8e922c | 638 | return bp; |
1da177e4 LT |
639 | |
640 | no_buffer: | |
ce8e922c NS |
641 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
642 | xfs_buf_unlock(bp); | |
643 | xfs_buf_rele(bp); | |
1da177e4 LT |
644 | return NULL; |
645 | } | |
646 | ||
1da177e4 | 647 | /* |
ce8e922c NS |
648 | * If we are not low on memory then do the readahead in a deadlock |
649 | * safe manner. | |
1da177e4 LT |
650 | */ |
651 | void | |
ce8e922c | 652 | xfs_buf_readahead( |
1da177e4 | 653 | xfs_buftarg_t *target, |
204ab25f | 654 | xfs_off_t ioff, |
1da177e4 | 655 | size_t isize, |
ce8e922c | 656 | xfs_buf_flags_t flags) |
1da177e4 LT |
657 | { |
658 | struct backing_dev_info *bdi; | |
659 | ||
ce8e922c | 660 | bdi = target->bt_mapping->backing_dev_info; |
1da177e4 LT |
661 | if (bdi_read_congested(bdi)) |
662 | return; | |
663 | ||
ce8e922c | 664 | flags |= (XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD); |
6ad112bf | 665 | xfs_buf_read(target, ioff, isize, flags); |
1da177e4 LT |
666 | } |
667 | ||
5adc94c2 DC |
668 | /* |
669 | * Read an uncached buffer from disk. Allocates and returns a locked | |
670 | * buffer containing the disk contents or nothing. | |
671 | */ | |
672 | struct xfs_buf * | |
673 | xfs_buf_read_uncached( | |
674 | struct xfs_mount *mp, | |
675 | struct xfs_buftarg *target, | |
676 | xfs_daddr_t daddr, | |
677 | size_t length, | |
678 | int flags) | |
679 | { | |
680 | xfs_buf_t *bp; | |
681 | int error; | |
682 | ||
683 | bp = xfs_buf_get_uncached(target, length, flags); | |
684 | if (!bp) | |
685 | return NULL; | |
686 | ||
687 | /* set up the buffer for a read IO */ | |
688 | xfs_buf_lock(bp); | |
689 | XFS_BUF_SET_ADDR(bp, daddr); | |
690 | XFS_BUF_READ(bp); | |
691 | XFS_BUF_BUSY(bp); | |
692 | ||
693 | xfsbdstrat(mp, bp); | |
694 | error = xfs_iowait(bp); | |
695 | if (error || bp->b_error) { | |
696 | xfs_buf_relse(bp); | |
697 | return NULL; | |
698 | } | |
699 | return bp; | |
700 | } | |
701 | ||
1da177e4 | 702 | xfs_buf_t * |
ce8e922c | 703 | xfs_buf_get_empty( |
1da177e4 LT |
704 | size_t len, |
705 | xfs_buftarg_t *target) | |
706 | { | |
ce8e922c | 707 | xfs_buf_t *bp; |
1da177e4 | 708 | |
ce8e922c NS |
709 | bp = xfs_buf_allocate(0); |
710 | if (bp) | |
711 | _xfs_buf_initialize(bp, target, 0, len, 0); | |
712 | return bp; | |
1da177e4 LT |
713 | } |
714 | ||
715 | static inline struct page * | |
716 | mem_to_page( | |
717 | void *addr) | |
718 | { | |
9e2779fa | 719 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
720 | return virt_to_page(addr); |
721 | } else { | |
722 | return vmalloc_to_page(addr); | |
723 | } | |
724 | } | |
725 | ||
726 | int | |
ce8e922c NS |
727 | xfs_buf_associate_memory( |
728 | xfs_buf_t *bp, | |
1da177e4 LT |
729 | void *mem, |
730 | size_t len) | |
731 | { | |
732 | int rval; | |
733 | int i = 0; | |
d1afb678 LM |
734 | unsigned long pageaddr; |
735 | unsigned long offset; | |
736 | size_t buflen; | |
1da177e4 LT |
737 | int page_count; |
738 | ||
d1afb678 LM |
739 | pageaddr = (unsigned long)mem & PAGE_CACHE_MASK; |
740 | offset = (unsigned long)mem - pageaddr; | |
741 | buflen = PAGE_CACHE_ALIGN(len + offset); | |
742 | page_count = buflen >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
743 | |
744 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
745 | if (bp->b_pages) |
746 | _xfs_buf_free_pages(bp); | |
1da177e4 | 747 | |
ce8e922c NS |
748 | bp->b_pages = NULL; |
749 | bp->b_addr = mem; | |
1da177e4 | 750 | |
36fae17a | 751 | rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK); |
1da177e4 LT |
752 | if (rval) |
753 | return rval; | |
754 | ||
ce8e922c | 755 | bp->b_offset = offset; |
d1afb678 LM |
756 | |
757 | for (i = 0; i < bp->b_page_count; i++) { | |
758 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
759 | pageaddr += PAGE_CACHE_SIZE; | |
1da177e4 | 760 | } |
1da177e4 | 761 | |
d1afb678 LM |
762 | bp->b_count_desired = len; |
763 | bp->b_buffer_length = buflen; | |
ce8e922c | 764 | bp->b_flags |= XBF_MAPPED; |
6ab455ee | 765 | bp->b_flags &= ~_XBF_PAGE_LOCKED; |
1da177e4 LT |
766 | |
767 | return 0; | |
768 | } | |
769 | ||
770 | xfs_buf_t * | |
686865f7 DC |
771 | xfs_buf_get_uncached( |
772 | struct xfs_buftarg *target, | |
1da177e4 | 773 | size_t len, |
686865f7 | 774 | int flags) |
1da177e4 | 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 | 789 | for (i = 0; i < page_count; i++) { |
686865f7 | 790 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
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", | |
34a622b2 | 799 | __func__); |
1da177e4 | 800 | goto fail_free_mem; |
1fa40b01 | 801 | } |
1da177e4 | 802 | |
ce8e922c | 803 | xfs_buf_unlock(bp); |
1da177e4 | 804 | |
686865f7 | 805 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
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 | { |
0b1b213f | 827 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 828 | atomic_inc(&bp->b_hold); |
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 | { |
74f75a0c | 839 | struct xfs_perag *pag = bp->b_pag; |
1da177e4 | 840 | |
0b1b213f | 841 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 842 | |
74f75a0c | 843 | if (!pag) { |
fad3aa1e | 844 | ASSERT(!bp->b_relse); |
74f75a0c | 845 | ASSERT(RB_EMPTY_NODE(&bp->b_rbnode)); |
fad3aa1e NS |
846 | if (atomic_dec_and_test(&bp->b_hold)) |
847 | xfs_buf_free(bp); | |
848 | return; | |
849 | } | |
850 | ||
74f75a0c | 851 | ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode)); |
3790689f | 852 | ASSERT(atomic_read(&bp->b_hold) > 0); |
74f75a0c | 853 | if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) { |
ce8e922c NS |
854 | if (bp->b_relse) { |
855 | atomic_inc(&bp->b_hold); | |
74f75a0c DC |
856 | spin_unlock(&pag->pag_buf_lock); |
857 | bp->b_relse(bp); | |
1da177e4 | 858 | } else { |
ce8e922c | 859 | ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q))); |
74f75a0c DC |
860 | rb_erase(&bp->b_rbnode, &pag->pag_buf_tree); |
861 | spin_unlock(&pag->pag_buf_lock); | |
862 | xfs_perag_put(pag); | |
ce8e922c | 863 | xfs_buf_free(bp); |
1da177e4 LT |
864 | } |
865 | } | |
866 | } | |
867 | ||
868 | ||
869 | /* | |
870 | * Mutual exclusion on buffers. Locking model: | |
871 | * | |
872 | * Buffers associated with inodes for which buffer locking | |
873 | * is not enabled are not protected by semaphores, and are | |
874 | * assumed to be exclusively owned by the caller. There is a | |
875 | * spinlock in the buffer, used by the caller when concurrent | |
876 | * access is possible. | |
877 | */ | |
878 | ||
879 | /* | |
ce8e922c NS |
880 | * Locks a buffer object, if it is not already locked. |
881 | * Note that this in no way locks the underlying pages, so it is only | |
882 | * useful for synchronizing concurrent use of buffer objects, not for | |
883 | * synchronizing independent access to the underlying pages. | |
1da177e4 LT |
884 | */ |
885 | int | |
ce8e922c NS |
886 | xfs_buf_cond_lock( |
887 | xfs_buf_t *bp) | |
1da177e4 LT |
888 | { |
889 | int locked; | |
890 | ||
ce8e922c | 891 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 892 | if (locked) |
ce8e922c | 893 | XB_SET_OWNER(bp); |
0b1b213f CH |
894 | |
895 | trace_xfs_buf_cond_lock(bp, _RET_IP_); | |
ce8e922c | 896 | return locked ? 0 : -EBUSY; |
1da177e4 LT |
897 | } |
898 | ||
1da177e4 | 899 | int |
ce8e922c NS |
900 | xfs_buf_lock_value( |
901 | xfs_buf_t *bp) | |
1da177e4 | 902 | { |
adaa693b | 903 | return bp->b_sema.count; |
1da177e4 | 904 | } |
1da177e4 LT |
905 | |
906 | /* | |
ce8e922c NS |
907 | * Locks a buffer object. |
908 | * Note that this in no way locks the underlying pages, so it is only | |
909 | * useful for synchronizing concurrent use of buffer objects, not for | |
910 | * synchronizing independent access to the underlying pages. | |
ed3b4d6c DC |
911 | * |
912 | * If we come across a stale, pinned, locked buffer, we know that we | |
913 | * are being asked to lock a buffer that has been reallocated. Because | |
914 | * it is pinned, we know that the log has not been pushed to disk and | |
915 | * hence it will still be locked. Rather than sleeping until someone | |
916 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 917 | */ |
ce8e922c NS |
918 | void |
919 | xfs_buf_lock( | |
920 | xfs_buf_t *bp) | |
1da177e4 | 921 | { |
0b1b213f CH |
922 | trace_xfs_buf_lock(bp, _RET_IP_); |
923 | ||
ed3b4d6c | 924 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 925 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
926 | if (atomic_read(&bp->b_io_remaining)) |
927 | blk_run_address_space(bp->b_target->bt_mapping); | |
928 | down(&bp->b_sema); | |
929 | XB_SET_OWNER(bp); | |
0b1b213f CH |
930 | |
931 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
932 | } |
933 | ||
934 | /* | |
ce8e922c | 935 | * Releases the lock on the buffer object. |
2f926587 | 936 | * If the buffer is marked delwri but is not queued, do so before we |
ce8e922c | 937 | * unlock the buffer as we need to set flags correctly. We also need to |
2f926587 DC |
938 | * take a reference for the delwri queue because the unlocker is going to |
939 | * drop their's and they don't know we just queued it. | |
1da177e4 LT |
940 | */ |
941 | void | |
ce8e922c NS |
942 | xfs_buf_unlock( |
943 | xfs_buf_t *bp) | |
1da177e4 | 944 | { |
ce8e922c NS |
945 | if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) { |
946 | atomic_inc(&bp->b_hold); | |
947 | bp->b_flags |= XBF_ASYNC; | |
948 | xfs_buf_delwri_queue(bp, 0); | |
2f926587 DC |
949 | } |
950 | ||
ce8e922c NS |
951 | XB_CLEAR_OWNER(bp); |
952 | up(&bp->b_sema); | |
0b1b213f CH |
953 | |
954 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
955 | } |
956 | ||
ce8e922c NS |
957 | STATIC void |
958 | xfs_buf_wait_unpin( | |
959 | xfs_buf_t *bp) | |
1da177e4 LT |
960 | { |
961 | DECLARE_WAITQUEUE (wait, current); | |
962 | ||
ce8e922c | 963 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
964 | return; |
965 | ||
ce8e922c | 966 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
967 | for (;;) { |
968 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 969 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 970 | break; |
ce8e922c NS |
971 | if (atomic_read(&bp->b_io_remaining)) |
972 | blk_run_address_space(bp->b_target->bt_mapping); | |
1da177e4 LT |
973 | schedule(); |
974 | } | |
ce8e922c | 975 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
976 | set_current_state(TASK_RUNNING); |
977 | } | |
978 | ||
979 | /* | |
980 | * Buffer Utility Routines | |
981 | */ | |
982 | ||
1da177e4 | 983 | STATIC void |
ce8e922c | 984 | xfs_buf_iodone_work( |
c4028958 | 985 | struct work_struct *work) |
1da177e4 | 986 | { |
c4028958 DH |
987 | xfs_buf_t *bp = |
988 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 989 | |
0bfefc46 DC |
990 | /* |
991 | * We can get an EOPNOTSUPP to ordered writes. Here we clear the | |
992 | * ordered flag and reissue them. Because we can't tell the higher | |
993 | * layers directly that they should not issue ordered I/O anymore, they | |
73f6aa4d | 994 | * need to check if the _XFS_BARRIER_FAILED flag was set during I/O completion. |
0bfefc46 DC |
995 | */ |
996 | if ((bp->b_error == EOPNOTSUPP) && | |
997 | (bp->b_flags & (XBF_ORDERED|XBF_ASYNC)) == (XBF_ORDERED|XBF_ASYNC)) { | |
0b1b213f | 998 | trace_xfs_buf_ordered_retry(bp, _RET_IP_); |
0bfefc46 | 999 | bp->b_flags &= ~XBF_ORDERED; |
73f6aa4d | 1000 | bp->b_flags |= _XFS_BARRIER_FAILED; |
0bfefc46 DC |
1001 | xfs_buf_iorequest(bp); |
1002 | } else if (bp->b_iodone) | |
ce8e922c NS |
1003 | (*(bp->b_iodone))(bp); |
1004 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
1005 | xfs_buf_relse(bp); |
1006 | } | |
1007 | ||
1008 | void | |
ce8e922c NS |
1009 | xfs_buf_ioend( |
1010 | xfs_buf_t *bp, | |
1da177e4 LT |
1011 | int schedule) |
1012 | { | |
0b1b213f CH |
1013 | trace_xfs_buf_iodone(bp, _RET_IP_); |
1014 | ||
77be55a5 | 1015 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
ce8e922c NS |
1016 | if (bp->b_error == 0) |
1017 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1018 | |
ce8e922c | 1019 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 1020 | if (schedule) { |
c4028958 | 1021 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 1022 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 1023 | } else { |
c4028958 | 1024 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
1025 | } |
1026 | } else { | |
b4dd330b | 1027 | complete(&bp->b_iowait); |
1da177e4 LT |
1028 | } |
1029 | } | |
1030 | ||
1da177e4 | 1031 | void |
ce8e922c NS |
1032 | xfs_buf_ioerror( |
1033 | xfs_buf_t *bp, | |
1034 | int error) | |
1da177e4 LT |
1035 | { |
1036 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c | 1037 | bp->b_error = (unsigned short)error; |
0b1b213f | 1038 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1039 | } |
1040 | ||
1da177e4 | 1041 | int |
64e0bc7d CH |
1042 | xfs_bwrite( |
1043 | struct xfs_mount *mp, | |
5d765b97 | 1044 | struct xfs_buf *bp) |
1da177e4 | 1045 | { |
8c38366f | 1046 | int error; |
1da177e4 | 1047 | |
64e0bc7d | 1048 | bp->b_flags |= XBF_WRITE; |
8c38366f | 1049 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ); |
1da177e4 | 1050 | |
5d765b97 | 1051 | xfs_buf_delwri_dequeue(bp); |
939d723b | 1052 | xfs_bdstrat_cb(bp); |
1da177e4 | 1053 | |
8c38366f CH |
1054 | error = xfs_buf_iowait(bp); |
1055 | if (error) | |
1056 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1057 | xfs_buf_relse(bp); | |
64e0bc7d | 1058 | return error; |
5d765b97 | 1059 | } |
1da177e4 | 1060 | |
5d765b97 CH |
1061 | void |
1062 | xfs_bdwrite( | |
1063 | void *mp, | |
1064 | struct xfs_buf *bp) | |
1065 | { | |
0b1b213f | 1066 | trace_xfs_buf_bdwrite(bp, _RET_IP_); |
1da177e4 | 1067 | |
5d765b97 CH |
1068 | bp->b_flags &= ~XBF_READ; |
1069 | bp->b_flags |= (XBF_DELWRI | XBF_ASYNC); | |
1070 | ||
1071 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1072 | } |
1073 | ||
4e23471a CH |
1074 | /* |
1075 | * Called when we want to stop a buffer from getting written or read. | |
1076 | * We attach the EIO error, muck with its flags, and call biodone | |
1077 | * so that the proper iodone callbacks get called. | |
1078 | */ | |
1079 | STATIC int | |
1080 | xfs_bioerror( | |
1081 | xfs_buf_t *bp) | |
1082 | { | |
1083 | #ifdef XFSERRORDEBUG | |
1084 | ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone); | |
1085 | #endif | |
1086 | ||
1087 | /* | |
1088 | * No need to wait until the buffer is unpinned, we aren't flushing it. | |
1089 | */ | |
1090 | XFS_BUF_ERROR(bp, EIO); | |
1091 | ||
1092 | /* | |
1093 | * We're calling biodone, so delete XBF_DONE flag. | |
1094 | */ | |
1095 | XFS_BUF_UNREAD(bp); | |
1096 | XFS_BUF_UNDELAYWRITE(bp); | |
1097 | XFS_BUF_UNDONE(bp); | |
1098 | XFS_BUF_STALE(bp); | |
1099 | ||
4e23471a CH |
1100 | xfs_biodone(bp); |
1101 | ||
1102 | return EIO; | |
1103 | } | |
1104 | ||
1105 | /* | |
1106 | * Same as xfs_bioerror, except that we are releasing the buffer | |
1107 | * here ourselves, and avoiding the biodone call. | |
1108 | * This is meant for userdata errors; metadata bufs come with | |
1109 | * iodone functions attached, so that we can track down errors. | |
1110 | */ | |
1111 | STATIC int | |
1112 | xfs_bioerror_relse( | |
1113 | struct xfs_buf *bp) | |
1114 | { | |
1115 | int64_t fl = XFS_BUF_BFLAGS(bp); | |
1116 | /* | |
1117 | * No need to wait until the buffer is unpinned. | |
1118 | * We aren't flushing it. | |
1119 | * | |
1120 | * chunkhold expects B_DONE to be set, whether | |
1121 | * we actually finish the I/O or not. We don't want to | |
1122 | * change that interface. | |
1123 | */ | |
1124 | XFS_BUF_UNREAD(bp); | |
1125 | XFS_BUF_UNDELAYWRITE(bp); | |
1126 | XFS_BUF_DONE(bp); | |
1127 | XFS_BUF_STALE(bp); | |
1128 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
0cadda1c | 1129 | if (!(fl & XBF_ASYNC)) { |
4e23471a CH |
1130 | /* |
1131 | * Mark b_error and B_ERROR _both_. | |
1132 | * Lot's of chunkcache code assumes that. | |
1133 | * There's no reason to mark error for | |
1134 | * ASYNC buffers. | |
1135 | */ | |
1136 | XFS_BUF_ERROR(bp, EIO); | |
1137 | XFS_BUF_FINISH_IOWAIT(bp); | |
1138 | } else { | |
1139 | xfs_buf_relse(bp); | |
1140 | } | |
1141 | ||
1142 | return EIO; | |
1143 | } | |
1144 | ||
1145 | ||
1146 | /* | |
1147 | * All xfs metadata buffers except log state machine buffers | |
1148 | * get this attached as their b_bdstrat callback function. | |
1149 | * This is so that we can catch a buffer | |
1150 | * after prematurely unpinning it to forcibly shutdown the filesystem. | |
1151 | */ | |
1152 | int | |
1153 | xfs_bdstrat_cb( | |
1154 | struct xfs_buf *bp) | |
1155 | { | |
ebad861b | 1156 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
4e23471a CH |
1157 | trace_xfs_bdstrat_shut(bp, _RET_IP_); |
1158 | /* | |
1159 | * Metadata write that didn't get logged but | |
1160 | * written delayed anyway. These aren't associated | |
1161 | * with a transaction, and can be ignored. | |
1162 | */ | |
1163 | if (!bp->b_iodone && !XFS_BUF_ISREAD(bp)) | |
1164 | return xfs_bioerror_relse(bp); | |
1165 | else | |
1166 | return xfs_bioerror(bp); | |
1167 | } | |
1168 | ||
1169 | xfs_buf_iorequest(bp); | |
1170 | return 0; | |
1171 | } | |
1172 | ||
1173 | /* | |
1174 | * Wrapper around bdstrat so that we can stop data from going to disk in case | |
1175 | * we are shutting down the filesystem. Typically user data goes thru this | |
1176 | * path; one of the exceptions is the superblock. | |
1177 | */ | |
1178 | void | |
1179 | xfsbdstrat( | |
1180 | struct xfs_mount *mp, | |
1181 | struct xfs_buf *bp) | |
1182 | { | |
1183 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1184 | trace_xfs_bdstrat_shut(bp, _RET_IP_); | |
1185 | xfs_bioerror_relse(bp); | |
1186 | return; | |
1187 | } | |
1188 | ||
1189 | xfs_buf_iorequest(bp); | |
1190 | } | |
1191 | ||
b8f82a4a | 1192 | STATIC void |
ce8e922c NS |
1193 | _xfs_buf_ioend( |
1194 | xfs_buf_t *bp, | |
1da177e4 LT |
1195 | int schedule) |
1196 | { | |
6ab455ee CH |
1197 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) { |
1198 | bp->b_flags &= ~_XBF_PAGE_LOCKED; | |
ce8e922c | 1199 | xfs_buf_ioend(bp, schedule); |
6ab455ee | 1200 | } |
1da177e4 LT |
1201 | } |
1202 | ||
782e3b3b | 1203 | STATIC void |
ce8e922c | 1204 | xfs_buf_bio_end_io( |
1da177e4 | 1205 | struct bio *bio, |
1da177e4 LT |
1206 | int error) |
1207 | { | |
ce8e922c NS |
1208 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1209 | unsigned int blocksize = bp->b_target->bt_bsize; | |
eedb5530 | 1210 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1da177e4 | 1211 | |
cfbe5267 | 1212 | xfs_buf_ioerror(bp, -error); |
1da177e4 | 1213 | |
73c77e2c JB |
1214 | if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
1215 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); | |
1216 | ||
eedb5530 | 1217 | do { |
1da177e4 LT |
1218 | struct page *page = bvec->bv_page; |
1219 | ||
948ecdb4 | 1220 | ASSERT(!PagePrivate(page)); |
ce8e922c NS |
1221 | if (unlikely(bp->b_error)) { |
1222 | if (bp->b_flags & XBF_READ) | |
eedb5530 | 1223 | ClearPageUptodate(page); |
ce8e922c | 1224 | } else if (blocksize >= PAGE_CACHE_SIZE) { |
1da177e4 LT |
1225 | SetPageUptodate(page); |
1226 | } else if (!PagePrivate(page) && | |
ce8e922c | 1227 | (bp->b_flags & _XBF_PAGE_CACHE)) { |
1da177e4 LT |
1228 | set_page_region(page, bvec->bv_offset, bvec->bv_len); |
1229 | } | |
1230 | ||
eedb5530 NS |
1231 | if (--bvec >= bio->bi_io_vec) |
1232 | prefetchw(&bvec->bv_page->flags); | |
6ab455ee CH |
1233 | |
1234 | if (bp->b_flags & _XBF_PAGE_LOCKED) | |
1235 | unlock_page(page); | |
eedb5530 | 1236 | } while (bvec >= bio->bi_io_vec); |
1da177e4 | 1237 | |
ce8e922c | 1238 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1239 | bio_put(bio); |
1da177e4 LT |
1240 | } |
1241 | ||
1242 | STATIC void | |
ce8e922c NS |
1243 | _xfs_buf_ioapply( |
1244 | xfs_buf_t *bp) | |
1da177e4 | 1245 | { |
a9759f2d | 1246 | int rw, map_i, total_nr_pages, nr_pages; |
1da177e4 | 1247 | struct bio *bio; |
ce8e922c NS |
1248 | int offset = bp->b_offset; |
1249 | int size = bp->b_count_desired; | |
1250 | sector_t sector = bp->b_bn; | |
1251 | unsigned int blocksize = bp->b_target->bt_bsize; | |
1da177e4 | 1252 | |
ce8e922c | 1253 | total_nr_pages = bp->b_page_count; |
1da177e4 LT |
1254 | map_i = 0; |
1255 | ||
ce8e922c NS |
1256 | if (bp->b_flags & XBF_ORDERED) { |
1257 | ASSERT(!(bp->b_flags & XBF_READ)); | |
f538d4da | 1258 | rw = WRITE_BARRIER; |
2ee1abad | 1259 | } else if (bp->b_flags & XBF_LOG_BUFFER) { |
51bdd706 NS |
1260 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); |
1261 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1262 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC; | |
2ee1abad DC |
1263 | } else if (bp->b_flags & _XBF_RUN_QUEUES) { |
1264 | ASSERT(!(bp->b_flags & XBF_READ_AHEAD)); | |
1265 | bp->b_flags &= ~_XBF_RUN_QUEUES; | |
1266 | rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META; | |
51bdd706 NS |
1267 | } else { |
1268 | rw = (bp->b_flags & XBF_WRITE) ? WRITE : | |
1269 | (bp->b_flags & XBF_READ_AHEAD) ? READA : READ; | |
f538d4da CH |
1270 | } |
1271 | ||
ce8e922c | 1272 | /* Special code path for reading a sub page size buffer in -- |
1da177e4 LT |
1273 | * we populate up the whole page, and hence the other metadata |
1274 | * in the same page. This optimization is only valid when the | |
ce8e922c | 1275 | * filesystem block size is not smaller than the page size. |
1da177e4 | 1276 | */ |
ce8e922c | 1277 | if ((bp->b_buffer_length < PAGE_CACHE_SIZE) && |
6ab455ee CH |
1278 | ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) == |
1279 | (XBF_READ|_XBF_PAGE_LOCKED)) && | |
ce8e922c | 1280 | (blocksize >= PAGE_CACHE_SIZE)) { |
1da177e4 LT |
1281 | bio = bio_alloc(GFP_NOIO, 1); |
1282 | ||
ce8e922c | 1283 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1284 | bio->bi_sector = sector - (offset >> BBSHIFT); |
ce8e922c NS |
1285 | bio->bi_end_io = xfs_buf_bio_end_io; |
1286 | bio->bi_private = bp; | |
1da177e4 | 1287 | |
ce8e922c | 1288 | bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0); |
1da177e4 LT |
1289 | size = 0; |
1290 | ||
ce8e922c | 1291 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1292 | |
1293 | goto submit_io; | |
1294 | } | |
1295 | ||
1da177e4 | 1296 | next_chunk: |
ce8e922c | 1297 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1298 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1299 | if (nr_pages > total_nr_pages) | |
1300 | nr_pages = total_nr_pages; | |
1301 | ||
1302 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1303 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1304 | bio->bi_sector = sector; |
ce8e922c NS |
1305 | bio->bi_end_io = xfs_buf_bio_end_io; |
1306 | bio->bi_private = bp; | |
1da177e4 LT |
1307 | |
1308 | for (; size && nr_pages; nr_pages--, map_i++) { | |
ce8e922c | 1309 | int rbytes, nbytes = PAGE_CACHE_SIZE - offset; |
1da177e4 LT |
1310 | |
1311 | if (nbytes > size) | |
1312 | nbytes = size; | |
1313 | ||
ce8e922c NS |
1314 | rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset); |
1315 | if (rbytes < nbytes) | |
1da177e4 LT |
1316 | break; |
1317 | ||
1318 | offset = 0; | |
1319 | sector += nbytes >> BBSHIFT; | |
1320 | size -= nbytes; | |
1321 | total_nr_pages--; | |
1322 | } | |
1323 | ||
1324 | submit_io: | |
1325 | if (likely(bio->bi_size)) { | |
73c77e2c JB |
1326 | if (xfs_buf_is_vmapped(bp)) { |
1327 | flush_kernel_vmap_range(bp->b_addr, | |
1328 | xfs_buf_vmap_len(bp)); | |
1329 | } | |
1da177e4 LT |
1330 | submit_bio(rw, bio); |
1331 | if (size) | |
1332 | goto next_chunk; | |
1333 | } else { | |
ec53d1db DC |
1334 | /* |
1335 | * if we get here, no pages were added to the bio. However, | |
1336 | * we can't just error out here - if the pages are locked then | |
1337 | * we have to unlock them otherwise we can hang on a later | |
1338 | * access to the page. | |
1339 | */ | |
ce8e922c | 1340 | xfs_buf_ioerror(bp, EIO); |
ec53d1db DC |
1341 | if (bp->b_flags & _XBF_PAGE_LOCKED) { |
1342 | int i; | |
1343 | for (i = 0; i < bp->b_page_count; i++) | |
1344 | unlock_page(bp->b_pages[i]); | |
1345 | } | |
1346 | bio_put(bio); | |
1da177e4 LT |
1347 | } |
1348 | } | |
1349 | ||
1da177e4 | 1350 | int |
ce8e922c NS |
1351 | xfs_buf_iorequest( |
1352 | xfs_buf_t *bp) | |
1da177e4 | 1353 | { |
0b1b213f | 1354 | trace_xfs_buf_iorequest(bp, _RET_IP_); |
1da177e4 | 1355 | |
ce8e922c NS |
1356 | if (bp->b_flags & XBF_DELWRI) { |
1357 | xfs_buf_delwri_queue(bp, 1); | |
1da177e4 LT |
1358 | return 0; |
1359 | } | |
1360 | ||
ce8e922c NS |
1361 | if (bp->b_flags & XBF_WRITE) { |
1362 | xfs_buf_wait_unpin(bp); | |
1da177e4 LT |
1363 | } |
1364 | ||
ce8e922c | 1365 | xfs_buf_hold(bp); |
1da177e4 LT |
1366 | |
1367 | /* Set the count to 1 initially, this will stop an I/O | |
1368 | * completion callout which happens before we have started | |
ce8e922c | 1369 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1370 | */ |
ce8e922c NS |
1371 | atomic_set(&bp->b_io_remaining, 1); |
1372 | _xfs_buf_ioapply(bp); | |
1373 | _xfs_buf_ioend(bp, 0); | |
1da177e4 | 1374 | |
ce8e922c | 1375 | xfs_buf_rele(bp); |
1da177e4 LT |
1376 | return 0; |
1377 | } | |
1378 | ||
1379 | /* | |
ce8e922c NS |
1380 | * Waits for I/O to complete on the buffer supplied. |
1381 | * It returns immediately if no I/O is pending. | |
1382 | * It returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1383 | */ |
1384 | int | |
ce8e922c NS |
1385 | xfs_buf_iowait( |
1386 | xfs_buf_t *bp) | |
1da177e4 | 1387 | { |
0b1b213f CH |
1388 | trace_xfs_buf_iowait(bp, _RET_IP_); |
1389 | ||
ce8e922c NS |
1390 | if (atomic_read(&bp->b_io_remaining)) |
1391 | blk_run_address_space(bp->b_target->bt_mapping); | |
b4dd330b | 1392 | wait_for_completion(&bp->b_iowait); |
0b1b213f CH |
1393 | |
1394 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
ce8e922c | 1395 | return bp->b_error; |
1da177e4 LT |
1396 | } |
1397 | ||
ce8e922c NS |
1398 | xfs_caddr_t |
1399 | xfs_buf_offset( | |
1400 | xfs_buf_t *bp, | |
1da177e4 LT |
1401 | size_t offset) |
1402 | { | |
1403 | struct page *page; | |
1404 | ||
ce8e922c NS |
1405 | if (bp->b_flags & XBF_MAPPED) |
1406 | return XFS_BUF_PTR(bp) + offset; | |
1da177e4 | 1407 | |
ce8e922c NS |
1408 | offset += bp->b_offset; |
1409 | page = bp->b_pages[offset >> PAGE_CACHE_SHIFT]; | |
1410 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1)); | |
1da177e4 LT |
1411 | } |
1412 | ||
1413 | /* | |
1da177e4 LT |
1414 | * Move data into or out of a buffer. |
1415 | */ | |
1416 | void | |
ce8e922c NS |
1417 | xfs_buf_iomove( |
1418 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1419 | size_t boff, /* starting buffer offset */ |
1420 | size_t bsize, /* length to copy */ | |
b9c48649 | 1421 | void *data, /* data address */ |
ce8e922c | 1422 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 LT |
1423 | { |
1424 | size_t bend, cpoff, csize; | |
1425 | struct page *page; | |
1426 | ||
1427 | bend = boff + bsize; | |
1428 | while (boff < bend) { | |
ce8e922c NS |
1429 | page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)]; |
1430 | cpoff = xfs_buf_poff(boff + bp->b_offset); | |
1da177e4 | 1431 | csize = min_t(size_t, |
ce8e922c | 1432 | PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff); |
1da177e4 LT |
1433 | |
1434 | ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE)); | |
1435 | ||
1436 | switch (mode) { | |
ce8e922c | 1437 | case XBRW_ZERO: |
1da177e4 LT |
1438 | memset(page_address(page) + cpoff, 0, csize); |
1439 | break; | |
ce8e922c | 1440 | case XBRW_READ: |
1da177e4 LT |
1441 | memcpy(data, page_address(page) + cpoff, csize); |
1442 | break; | |
ce8e922c | 1443 | case XBRW_WRITE: |
1da177e4 LT |
1444 | memcpy(page_address(page) + cpoff, data, csize); |
1445 | } | |
1446 | ||
1447 | boff += csize; | |
1448 | data += csize; | |
1449 | } | |
1450 | } | |
1451 | ||
1452 | /* | |
ce8e922c | 1453 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1454 | */ |
1455 | ||
1456 | /* | |
ce8e922c NS |
1457 | * Wait for any bufs with callbacks that have been submitted but |
1458 | * have not yet returned... walk the hash list for the target. | |
1da177e4 LT |
1459 | */ |
1460 | void | |
1461 | xfs_wait_buftarg( | |
74f75a0c | 1462 | struct xfs_buftarg *btp) |
1da177e4 | 1463 | { |
74f75a0c DC |
1464 | struct xfs_perag *pag; |
1465 | uint i; | |
1466 | ||
1467 | for (i = 0; i < btp->bt_mount->m_sb.sb_agcount; i++) { | |
1468 | pag = xfs_perag_get(btp->bt_mount, i); | |
1469 | spin_lock(&pag->pag_buf_lock); | |
1470 | while (rb_first(&pag->pag_buf_tree)) { | |
1471 | spin_unlock(&pag->pag_buf_lock); | |
26af6552 | 1472 | delay(100); |
74f75a0c | 1473 | spin_lock(&pag->pag_buf_lock); |
1da177e4 | 1474 | } |
74f75a0c DC |
1475 | spin_unlock(&pag->pag_buf_lock); |
1476 | xfs_perag_put(pag); | |
1da177e4 LT |
1477 | } |
1478 | } | |
1479 | ||
a6867a68 | 1480 | /* |
ce8e922c | 1481 | * buftarg list for delwrite queue processing |
a6867a68 | 1482 | */ |
e6a0e9cd | 1483 | static LIST_HEAD(xfs_buftarg_list); |
7989cb8e | 1484 | static DEFINE_SPINLOCK(xfs_buftarg_lock); |
a6867a68 DC |
1485 | |
1486 | STATIC void | |
1487 | xfs_register_buftarg( | |
1488 | xfs_buftarg_t *btp) | |
1489 | { | |
1490 | spin_lock(&xfs_buftarg_lock); | |
1491 | list_add(&btp->bt_list, &xfs_buftarg_list); | |
1492 | spin_unlock(&xfs_buftarg_lock); | |
1493 | } | |
1494 | ||
1495 | STATIC void | |
1496 | xfs_unregister_buftarg( | |
1497 | xfs_buftarg_t *btp) | |
1498 | { | |
1499 | spin_lock(&xfs_buftarg_lock); | |
1500 | list_del(&btp->bt_list); | |
1501 | spin_unlock(&xfs_buftarg_lock); | |
1502 | } | |
1503 | ||
1da177e4 LT |
1504 | void |
1505 | xfs_free_buftarg( | |
b7963133 CH |
1506 | struct xfs_mount *mp, |
1507 | struct xfs_buftarg *btp) | |
1da177e4 LT |
1508 | { |
1509 | xfs_flush_buftarg(btp, 1); | |
b7963133 CH |
1510 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1511 | xfs_blkdev_issue_flush(btp); | |
ce8e922c | 1512 | iput(btp->bt_mapping->host); |
a6867a68 | 1513 | |
ce8e922c NS |
1514 | /* Unregister the buftarg first so that we don't get a |
1515 | * wakeup finding a non-existent task | |
1516 | */ | |
a6867a68 DC |
1517 | xfs_unregister_buftarg(btp); |
1518 | kthread_stop(btp->bt_task); | |
1519 | ||
f0e2d93c | 1520 | kmem_free(btp); |
1da177e4 LT |
1521 | } |
1522 | ||
1da177e4 LT |
1523 | STATIC int |
1524 | xfs_setsize_buftarg_flags( | |
1525 | xfs_buftarg_t *btp, | |
1526 | unsigned int blocksize, | |
1527 | unsigned int sectorsize, | |
1528 | int verbose) | |
1529 | { | |
ce8e922c NS |
1530 | btp->bt_bsize = blocksize; |
1531 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1532 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1533 | |
ce8e922c | 1534 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
1da177e4 LT |
1535 | printk(KERN_WARNING |
1536 | "XFS: Cannot set_blocksize to %u on device %s\n", | |
1537 | sectorsize, XFS_BUFTARG_NAME(btp)); | |
1538 | return EINVAL; | |
1539 | } | |
1540 | ||
1541 | if (verbose && | |
1542 | (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) { | |
1543 | printk(KERN_WARNING | |
1544 | "XFS: %u byte sectors in use on device %s. " | |
1545 | "This is suboptimal; %u or greater is ideal.\n", | |
1546 | sectorsize, XFS_BUFTARG_NAME(btp), | |
1547 | (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG); | |
1548 | } | |
1549 | ||
1550 | return 0; | |
1551 | } | |
1552 | ||
1553 | /* | |
ce8e922c NS |
1554 | * When allocating the initial buffer target we have not yet |
1555 | * read in the superblock, so don't know what sized sectors | |
1556 | * are being used is at this early stage. Play safe. | |
1557 | */ | |
1da177e4 LT |
1558 | STATIC int |
1559 | xfs_setsize_buftarg_early( | |
1560 | xfs_buftarg_t *btp, | |
1561 | struct block_device *bdev) | |
1562 | { | |
1563 | return xfs_setsize_buftarg_flags(btp, | |
e1defc4f | 1564 | PAGE_CACHE_SIZE, bdev_logical_block_size(bdev), 0); |
1da177e4 LT |
1565 | } |
1566 | ||
1567 | int | |
1568 | xfs_setsize_buftarg( | |
1569 | xfs_buftarg_t *btp, | |
1570 | unsigned int blocksize, | |
1571 | unsigned int sectorsize) | |
1572 | { | |
1573 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1574 | } | |
1575 | ||
1576 | STATIC int | |
1577 | xfs_mapping_buftarg( | |
1578 | xfs_buftarg_t *btp, | |
1579 | struct block_device *bdev) | |
1580 | { | |
1581 | struct backing_dev_info *bdi; | |
1582 | struct inode *inode; | |
1583 | struct address_space *mapping; | |
f5e54d6e | 1584 | static const struct address_space_operations mapping_aops = { |
1da177e4 | 1585 | .sync_page = block_sync_page, |
e965f963 | 1586 | .migratepage = fail_migrate_page, |
1da177e4 LT |
1587 | }; |
1588 | ||
1589 | inode = new_inode(bdev->bd_inode->i_sb); | |
1590 | if (!inode) { | |
1591 | printk(KERN_WARNING | |
1592 | "XFS: Cannot allocate mapping inode for device %s\n", | |
1593 | XFS_BUFTARG_NAME(btp)); | |
1594 | return ENOMEM; | |
1595 | } | |
1596 | inode->i_mode = S_IFBLK; | |
1597 | inode->i_bdev = bdev; | |
1598 | inode->i_rdev = bdev->bd_dev; | |
1599 | bdi = blk_get_backing_dev_info(bdev); | |
1600 | if (!bdi) | |
1601 | bdi = &default_backing_dev_info; | |
1602 | mapping = &inode->i_data; | |
1603 | mapping->a_ops = &mapping_aops; | |
1604 | mapping->backing_dev_info = bdi; | |
1605 | mapping_set_gfp_mask(mapping, GFP_NOFS); | |
ce8e922c | 1606 | btp->bt_mapping = mapping; |
1da177e4 LT |
1607 | return 0; |
1608 | } | |
1609 | ||
a6867a68 DC |
1610 | STATIC int |
1611 | xfs_alloc_delwrite_queue( | |
e2a07812 JE |
1612 | xfs_buftarg_t *btp, |
1613 | const char *fsname) | |
a6867a68 DC |
1614 | { |
1615 | int error = 0; | |
1616 | ||
1617 | INIT_LIST_HEAD(&btp->bt_list); | |
1618 | INIT_LIST_HEAD(&btp->bt_delwrite_queue); | |
007c61c6 | 1619 | spin_lock_init(&btp->bt_delwrite_lock); |
a6867a68 | 1620 | btp->bt_flags = 0; |
e2a07812 | 1621 | btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd/%s", fsname); |
a6867a68 DC |
1622 | if (IS_ERR(btp->bt_task)) { |
1623 | error = PTR_ERR(btp->bt_task); | |
1624 | goto out_error; | |
1625 | } | |
1626 | xfs_register_buftarg(btp); | |
1627 | out_error: | |
1628 | return error; | |
1629 | } | |
1630 | ||
1da177e4 LT |
1631 | xfs_buftarg_t * |
1632 | xfs_alloc_buftarg( | |
ebad861b | 1633 | struct xfs_mount *mp, |
1da177e4 | 1634 | struct block_device *bdev, |
e2a07812 JE |
1635 | int external, |
1636 | const char *fsname) | |
1da177e4 LT |
1637 | { |
1638 | xfs_buftarg_t *btp; | |
1639 | ||
1640 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1641 | ||
ebad861b | 1642 | btp->bt_mount = mp; |
ce8e922c NS |
1643 | btp->bt_dev = bdev->bd_dev; |
1644 | btp->bt_bdev = bdev; | |
1da177e4 LT |
1645 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1646 | goto error; | |
1647 | if (xfs_mapping_buftarg(btp, bdev)) | |
1648 | goto error; | |
e2a07812 | 1649 | if (xfs_alloc_delwrite_queue(btp, fsname)) |
a6867a68 | 1650 | goto error; |
1da177e4 LT |
1651 | return btp; |
1652 | ||
1653 | error: | |
f0e2d93c | 1654 | kmem_free(btp); |
1da177e4 LT |
1655 | return NULL; |
1656 | } | |
1657 | ||
1658 | ||
1659 | /* | |
ce8e922c | 1660 | * Delayed write buffer handling |
1da177e4 | 1661 | */ |
1da177e4 | 1662 | STATIC void |
ce8e922c NS |
1663 | xfs_buf_delwri_queue( |
1664 | xfs_buf_t *bp, | |
1da177e4 LT |
1665 | int unlock) |
1666 | { | |
ce8e922c NS |
1667 | struct list_head *dwq = &bp->b_target->bt_delwrite_queue; |
1668 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; | |
a6867a68 | 1669 | |
0b1b213f CH |
1670 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
1671 | ||
ce8e922c | 1672 | ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC)); |
1da177e4 | 1673 | |
a6867a68 | 1674 | spin_lock(dwlk); |
1da177e4 | 1675 | /* If already in the queue, dequeue and place at tail */ |
ce8e922c NS |
1676 | if (!list_empty(&bp->b_list)) { |
1677 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1678 | if (unlock) | |
1679 | atomic_dec(&bp->b_hold); | |
1680 | list_del(&bp->b_list); | |
1da177e4 LT |
1681 | } |
1682 | ||
c9c12971 DC |
1683 | if (list_empty(dwq)) { |
1684 | /* start xfsbufd as it is about to have something to do */ | |
1685 | wake_up_process(bp->b_target->bt_task); | |
1686 | } | |
1687 | ||
ce8e922c NS |
1688 | bp->b_flags |= _XBF_DELWRI_Q; |
1689 | list_add_tail(&bp->b_list, dwq); | |
1690 | bp->b_queuetime = jiffies; | |
a6867a68 | 1691 | spin_unlock(dwlk); |
1da177e4 LT |
1692 | |
1693 | if (unlock) | |
ce8e922c | 1694 | xfs_buf_unlock(bp); |
1da177e4 LT |
1695 | } |
1696 | ||
1697 | void | |
ce8e922c NS |
1698 | xfs_buf_delwri_dequeue( |
1699 | xfs_buf_t *bp) | |
1da177e4 | 1700 | { |
ce8e922c | 1701 | spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock; |
1da177e4 LT |
1702 | int dequeued = 0; |
1703 | ||
a6867a68 | 1704 | spin_lock(dwlk); |
ce8e922c NS |
1705 | if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) { |
1706 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1707 | list_del_init(&bp->b_list); | |
1da177e4 LT |
1708 | dequeued = 1; |
1709 | } | |
ce8e922c | 1710 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q); |
a6867a68 | 1711 | spin_unlock(dwlk); |
1da177e4 LT |
1712 | |
1713 | if (dequeued) | |
ce8e922c | 1714 | xfs_buf_rele(bp); |
1da177e4 | 1715 | |
0b1b213f | 1716 | trace_xfs_buf_delwri_dequeue(bp, _RET_IP_); |
1da177e4 LT |
1717 | } |
1718 | ||
d808f617 DC |
1719 | /* |
1720 | * If a delwri buffer needs to be pushed before it has aged out, then promote | |
1721 | * it to the head of the delwri queue so that it will be flushed on the next | |
1722 | * xfsbufd run. We do this by resetting the queuetime of the buffer to be older | |
1723 | * than the age currently needed to flush the buffer. Hence the next time the | |
1724 | * xfsbufd sees it is guaranteed to be considered old enough to flush. | |
1725 | */ | |
1726 | void | |
1727 | xfs_buf_delwri_promote( | |
1728 | struct xfs_buf *bp) | |
1729 | { | |
1730 | struct xfs_buftarg *btp = bp->b_target; | |
1731 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1; | |
1732 | ||
1733 | ASSERT(bp->b_flags & XBF_DELWRI); | |
1734 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1735 | ||
1736 | /* | |
1737 | * Check the buffer age before locking the delayed write queue as we | |
1738 | * don't need to promote buffers that are already past the flush age. | |
1739 | */ | |
1740 | if (bp->b_queuetime < jiffies - age) | |
1741 | return; | |
1742 | bp->b_queuetime = jiffies - age; | |
1743 | spin_lock(&btp->bt_delwrite_lock); | |
1744 | list_move(&bp->b_list, &btp->bt_delwrite_queue); | |
1745 | spin_unlock(&btp->bt_delwrite_lock); | |
1746 | } | |
1747 | ||
1da177e4 | 1748 | STATIC void |
ce8e922c | 1749 | xfs_buf_runall_queues( |
1da177e4 LT |
1750 | struct workqueue_struct *queue) |
1751 | { | |
1752 | flush_workqueue(queue); | |
1753 | } | |
1754 | ||
1da177e4 | 1755 | STATIC int |
23ea4032 | 1756 | xfsbufd_wakeup( |
7f8275d0 | 1757 | struct shrinker *shrink, |
15c84a47 NS |
1758 | int priority, |
1759 | gfp_t mask) | |
1da177e4 | 1760 | { |
da7f93e9 | 1761 | xfs_buftarg_t *btp; |
a6867a68 DC |
1762 | |
1763 | spin_lock(&xfs_buftarg_lock); | |
da7f93e9 | 1764 | list_for_each_entry(btp, &xfs_buftarg_list, bt_list) { |
ce8e922c | 1765 | if (test_bit(XBT_FORCE_SLEEP, &btp->bt_flags)) |
a6867a68 | 1766 | continue; |
c9c12971 DC |
1767 | if (list_empty(&btp->bt_delwrite_queue)) |
1768 | continue; | |
ce8e922c | 1769 | set_bit(XBT_FORCE_FLUSH, &btp->bt_flags); |
a6867a68 DC |
1770 | wake_up_process(btp->bt_task); |
1771 | } | |
1772 | spin_unlock(&xfs_buftarg_lock); | |
1da177e4 LT |
1773 | return 0; |
1774 | } | |
1775 | ||
585e6d88 DC |
1776 | /* |
1777 | * Move as many buffers as specified to the supplied list | |
1778 | * idicating if we skipped any buffers to prevent deadlocks. | |
1779 | */ | |
1780 | STATIC int | |
1781 | xfs_buf_delwri_split( | |
1782 | xfs_buftarg_t *target, | |
1783 | struct list_head *list, | |
5e6a07df | 1784 | unsigned long age) |
585e6d88 DC |
1785 | { |
1786 | xfs_buf_t *bp, *n; | |
1787 | struct list_head *dwq = &target->bt_delwrite_queue; | |
1788 | spinlock_t *dwlk = &target->bt_delwrite_lock; | |
1789 | int skipped = 0; | |
5e6a07df | 1790 | int force; |
585e6d88 | 1791 | |
5e6a07df | 1792 | force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
585e6d88 DC |
1793 | INIT_LIST_HEAD(list); |
1794 | spin_lock(dwlk); | |
1795 | list_for_each_entry_safe(bp, n, dwq, b_list) { | |
0b1b213f | 1796 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
585e6d88 DC |
1797 | ASSERT(bp->b_flags & XBF_DELWRI); |
1798 | ||
4d16e924 | 1799 | if (!XFS_BUF_ISPINNED(bp) && !xfs_buf_cond_lock(bp)) { |
5e6a07df | 1800 | if (!force && |
585e6d88 DC |
1801 | time_before(jiffies, bp->b_queuetime + age)) { |
1802 | xfs_buf_unlock(bp); | |
1803 | break; | |
1804 | } | |
1805 | ||
1806 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q| | |
1807 | _XBF_RUN_QUEUES); | |
1808 | bp->b_flags |= XBF_WRITE; | |
1809 | list_move_tail(&bp->b_list, list); | |
1810 | } else | |
1811 | skipped++; | |
1812 | } | |
1813 | spin_unlock(dwlk); | |
1814 | ||
1815 | return skipped; | |
1816 | ||
1817 | } | |
1818 | ||
089716aa DC |
1819 | /* |
1820 | * Compare function is more complex than it needs to be because | |
1821 | * the return value is only 32 bits and we are doing comparisons | |
1822 | * on 64 bit values | |
1823 | */ | |
1824 | static int | |
1825 | xfs_buf_cmp( | |
1826 | void *priv, | |
1827 | struct list_head *a, | |
1828 | struct list_head *b) | |
1829 | { | |
1830 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1831 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1832 | xfs_daddr_t diff; | |
1833 | ||
1834 | diff = ap->b_bn - bp->b_bn; | |
1835 | if (diff < 0) | |
1836 | return -1; | |
1837 | if (diff > 0) | |
1838 | return 1; | |
1839 | return 0; | |
1840 | } | |
1841 | ||
1842 | void | |
1843 | xfs_buf_delwri_sort( | |
1844 | xfs_buftarg_t *target, | |
1845 | struct list_head *list) | |
1846 | { | |
1847 | list_sort(NULL, list, xfs_buf_cmp); | |
1848 | } | |
1849 | ||
1da177e4 | 1850 | STATIC int |
23ea4032 | 1851 | xfsbufd( |
585e6d88 | 1852 | void *data) |
1da177e4 | 1853 | { |
089716aa | 1854 | xfs_buftarg_t *target = (xfs_buftarg_t *)data; |
1da177e4 | 1855 | |
1da177e4 LT |
1856 | current->flags |= PF_MEMALLOC; |
1857 | ||
978c7b2f RW |
1858 | set_freezable(); |
1859 | ||
1da177e4 | 1860 | do { |
c9c12971 DC |
1861 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10); |
1862 | long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10); | |
089716aa DC |
1863 | int count = 0; |
1864 | struct list_head tmp; | |
c9c12971 | 1865 | |
3e1d1d28 | 1866 | if (unlikely(freezing(current))) { |
ce8e922c | 1867 | set_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
3e1d1d28 | 1868 | refrigerator(); |
abd0cf7a | 1869 | } else { |
ce8e922c | 1870 | clear_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
abd0cf7a | 1871 | } |
1da177e4 | 1872 | |
c9c12971 DC |
1873 | /* sleep for a long time if there is nothing to do. */ |
1874 | if (list_empty(&target->bt_delwrite_queue)) | |
1875 | tout = MAX_SCHEDULE_TIMEOUT; | |
1876 | schedule_timeout_interruptible(tout); | |
1da177e4 | 1877 | |
c9c12971 | 1878 | xfs_buf_delwri_split(target, &tmp, age); |
089716aa | 1879 | list_sort(NULL, &tmp, xfs_buf_cmp); |
1da177e4 | 1880 | while (!list_empty(&tmp)) { |
089716aa DC |
1881 | struct xfs_buf *bp; |
1882 | bp = list_first_entry(&tmp, struct xfs_buf, b_list); | |
ce8e922c | 1883 | list_del_init(&bp->b_list); |
939d723b | 1884 | xfs_bdstrat_cb(bp); |
585e6d88 | 1885 | count++; |
1da177e4 | 1886 | } |
f07c2250 NS |
1887 | if (count) |
1888 | blk_run_address_space(target->bt_mapping); | |
1da177e4 | 1889 | |
4df08c52 | 1890 | } while (!kthread_should_stop()); |
1da177e4 | 1891 | |
4df08c52 | 1892 | return 0; |
1da177e4 LT |
1893 | } |
1894 | ||
1895 | /* | |
ce8e922c NS |
1896 | * Go through all incore buffers, and release buffers if they belong to |
1897 | * the given device. This is used in filesystem error handling to | |
1898 | * preserve the consistency of its metadata. | |
1da177e4 LT |
1899 | */ |
1900 | int | |
1901 | xfs_flush_buftarg( | |
585e6d88 DC |
1902 | xfs_buftarg_t *target, |
1903 | int wait) | |
1da177e4 | 1904 | { |
089716aa | 1905 | xfs_buf_t *bp; |
585e6d88 | 1906 | int pincount = 0; |
089716aa DC |
1907 | LIST_HEAD(tmp_list); |
1908 | LIST_HEAD(wait_list); | |
1da177e4 | 1909 | |
c626d174 | 1910 | xfs_buf_runall_queues(xfsconvertd_workqueue); |
ce8e922c NS |
1911 | xfs_buf_runall_queues(xfsdatad_workqueue); |
1912 | xfs_buf_runall_queues(xfslogd_workqueue); | |
1da177e4 | 1913 | |
5e6a07df | 1914 | set_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
089716aa | 1915 | pincount = xfs_buf_delwri_split(target, &tmp_list, 0); |
1da177e4 LT |
1916 | |
1917 | /* | |
089716aa DC |
1918 | * Dropped the delayed write list lock, now walk the temporary list. |
1919 | * All I/O is issued async and then if we need to wait for completion | |
1920 | * we do that after issuing all the IO. | |
1da177e4 | 1921 | */ |
089716aa DC |
1922 | list_sort(NULL, &tmp_list, xfs_buf_cmp); |
1923 | while (!list_empty(&tmp_list)) { | |
1924 | bp = list_first_entry(&tmp_list, struct xfs_buf, b_list); | |
585e6d88 | 1925 | ASSERT(target == bp->b_target); |
089716aa DC |
1926 | list_del_init(&bp->b_list); |
1927 | if (wait) { | |
ce8e922c | 1928 | bp->b_flags &= ~XBF_ASYNC; |
089716aa DC |
1929 | list_add(&bp->b_list, &wait_list); |
1930 | } | |
939d723b | 1931 | xfs_bdstrat_cb(bp); |
1da177e4 LT |
1932 | } |
1933 | ||
089716aa DC |
1934 | if (wait) { |
1935 | /* Expedite and wait for IO to complete. */ | |
f07c2250 | 1936 | blk_run_address_space(target->bt_mapping); |
089716aa DC |
1937 | while (!list_empty(&wait_list)) { |
1938 | bp = list_first_entry(&wait_list, struct xfs_buf, b_list); | |
f07c2250 | 1939 | |
089716aa DC |
1940 | list_del_init(&bp->b_list); |
1941 | xfs_iowait(bp); | |
1942 | xfs_buf_relse(bp); | |
1943 | } | |
1da177e4 LT |
1944 | } |
1945 | ||
1da177e4 LT |
1946 | return pincount; |
1947 | } | |
1948 | ||
04d8b284 | 1949 | int __init |
ce8e922c | 1950 | xfs_buf_init(void) |
1da177e4 | 1951 | { |
8758280f NS |
1952 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1953 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1954 | if (!xfs_buf_zone) |
0b1b213f | 1955 | goto out; |
04d8b284 | 1956 | |
51749e47 DC |
1957 | xfslogd_workqueue = alloc_workqueue("xfslogd", |
1958 | WQ_RESCUER | WQ_HIGHPRI, 1); | |
23ea4032 | 1959 | if (!xfslogd_workqueue) |
04d8b284 | 1960 | goto out_free_buf_zone; |
1da177e4 | 1961 | |
b4337692 | 1962 | xfsdatad_workqueue = create_workqueue("xfsdatad"); |
23ea4032 CH |
1963 | if (!xfsdatad_workqueue) |
1964 | goto out_destroy_xfslogd_workqueue; | |
1da177e4 | 1965 | |
c626d174 DC |
1966 | xfsconvertd_workqueue = create_workqueue("xfsconvertd"); |
1967 | if (!xfsconvertd_workqueue) | |
1968 | goto out_destroy_xfsdatad_workqueue; | |
1969 | ||
8e1f936b | 1970 | register_shrinker(&xfs_buf_shake); |
23ea4032 | 1971 | return 0; |
1da177e4 | 1972 | |
c626d174 DC |
1973 | out_destroy_xfsdatad_workqueue: |
1974 | destroy_workqueue(xfsdatad_workqueue); | |
23ea4032 CH |
1975 | out_destroy_xfslogd_workqueue: |
1976 | destroy_workqueue(xfslogd_workqueue); | |
23ea4032 | 1977 | out_free_buf_zone: |
ce8e922c | 1978 | kmem_zone_destroy(xfs_buf_zone); |
0b1b213f | 1979 | out: |
8758280f | 1980 | return -ENOMEM; |
1da177e4 LT |
1981 | } |
1982 | ||
1da177e4 | 1983 | void |
ce8e922c | 1984 | xfs_buf_terminate(void) |
1da177e4 | 1985 | { |
8e1f936b | 1986 | unregister_shrinker(&xfs_buf_shake); |
c626d174 | 1987 | destroy_workqueue(xfsconvertd_workqueue); |
04d8b284 CH |
1988 | destroy_workqueue(xfsdatad_workqueue); |
1989 | destroy_workqueue(xfslogd_workqueue); | |
ce8e922c | 1990 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 1991 | } |
e6a0e9cd TS |
1992 | |
1993 | #ifdef CONFIG_KDB_MODULES | |
1994 | struct list_head * | |
1995 | xfs_get_buftarg_list(void) | |
1996 | { | |
1997 | return &xfs_buftarg_list; | |
1998 | } | |
1999 | #endif |