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> |
1da177e4 | 36 | |
b7963133 | 37 | #include "xfs_sb.h" |
7fd36c44 | 38 | #include "xfs_trans_resv.h" |
ed3b4d6c | 39 | #include "xfs_log.h" |
b7963133 | 40 | #include "xfs_ag.h" |
b7963133 | 41 | #include "xfs_mount.h" |
0b1b213f | 42 | #include "xfs_trace.h" |
b7963133 | 43 | |
7989cb8e | 44 | static kmem_zone_t *xfs_buf_zone; |
23ea4032 | 45 | |
7989cb8e | 46 | static struct workqueue_struct *xfslogd_workqueue; |
1da177e4 | 47 | |
ce8e922c NS |
48 | #ifdef XFS_BUF_LOCK_TRACKING |
49 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
50 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
51 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 52 | #else |
ce8e922c NS |
53 | # define XB_SET_OWNER(bp) do { } while (0) |
54 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
55 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
56 | #endif |
57 | ||
ce8e922c | 58 | #define xb_to_gfp(flags) \ |
aa5c158e | 59 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN) |
1da177e4 | 60 | |
1da177e4 | 61 | |
73c77e2c JB |
62 | static inline int |
63 | xfs_buf_is_vmapped( | |
64 | struct xfs_buf *bp) | |
65 | { | |
66 | /* | |
67 | * Return true if the buffer is vmapped. | |
68 | * | |
611c9946 DC |
69 | * b_addr is null if the buffer is not mapped, but the code is clever |
70 | * enough to know it doesn't have to map a single page, so the check has | |
71 | * to be both for b_addr and bp->b_page_count > 1. | |
73c77e2c | 72 | */ |
611c9946 | 73 | return bp->b_addr && bp->b_page_count > 1; |
73c77e2c JB |
74 | } |
75 | ||
76 | static inline int | |
77 | xfs_buf_vmap_len( | |
78 | struct xfs_buf *bp) | |
79 | { | |
80 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
81 | } | |
82 | ||
430cbeb8 DC |
83 | /* |
84 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
85 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
86 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
87 | * to remove the reference that LRU holds on the buffer. | |
88 | * | |
89 | * This prevents build-up of stale buffers on the LRU. | |
90 | */ | |
91 | void | |
92 | xfs_buf_stale( | |
93 | struct xfs_buf *bp) | |
94 | { | |
43ff2122 CH |
95 | ASSERT(xfs_buf_islocked(bp)); |
96 | ||
430cbeb8 | 97 | bp->b_flags |= XBF_STALE; |
43ff2122 CH |
98 | |
99 | /* | |
100 | * Clear the delwri status so that a delwri queue walker will not | |
101 | * flush this buffer to disk now that it is stale. The delwri queue has | |
102 | * a reference to the buffer, so this is safe to do. | |
103 | */ | |
104 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
105 | ||
a4082357 DC |
106 | spin_lock(&bp->b_lock); |
107 | atomic_set(&bp->b_lru_ref, 0); | |
108 | if (!(bp->b_state & XFS_BSTATE_DISPOSE) && | |
e80dfa19 DC |
109 | (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru))) |
110 | atomic_dec(&bp->b_hold); | |
111 | ||
430cbeb8 | 112 | ASSERT(atomic_read(&bp->b_hold) >= 1); |
a4082357 | 113 | spin_unlock(&bp->b_lock); |
430cbeb8 | 114 | } |
1da177e4 | 115 | |
3e85c868 DC |
116 | static int |
117 | xfs_buf_get_maps( | |
118 | struct xfs_buf *bp, | |
119 | int map_count) | |
120 | { | |
121 | ASSERT(bp->b_maps == NULL); | |
122 | bp->b_map_count = map_count; | |
123 | ||
124 | if (map_count == 1) { | |
f4b42421 | 125 | bp->b_maps = &bp->__b_map; |
3e85c868 DC |
126 | return 0; |
127 | } | |
128 | ||
129 | bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map), | |
130 | KM_NOFS); | |
131 | if (!bp->b_maps) | |
132 | return ENOMEM; | |
133 | return 0; | |
134 | } | |
135 | ||
136 | /* | |
137 | * Frees b_pages if it was allocated. | |
138 | */ | |
139 | static void | |
140 | xfs_buf_free_maps( | |
141 | struct xfs_buf *bp) | |
142 | { | |
f4b42421 | 143 | if (bp->b_maps != &bp->__b_map) { |
3e85c868 DC |
144 | kmem_free(bp->b_maps); |
145 | bp->b_maps = NULL; | |
146 | } | |
147 | } | |
148 | ||
4347b9d7 | 149 | struct xfs_buf * |
3e85c868 | 150 | _xfs_buf_alloc( |
4347b9d7 | 151 | struct xfs_buftarg *target, |
3e85c868 DC |
152 | struct xfs_buf_map *map, |
153 | int nmaps, | |
ce8e922c | 154 | xfs_buf_flags_t flags) |
1da177e4 | 155 | { |
4347b9d7 | 156 | struct xfs_buf *bp; |
3e85c868 DC |
157 | int error; |
158 | int i; | |
4347b9d7 | 159 | |
aa5c158e | 160 | bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS); |
4347b9d7 CH |
161 | if (unlikely(!bp)) |
162 | return NULL; | |
163 | ||
1da177e4 | 164 | /* |
12bcb3f7 DC |
165 | * We don't want certain flags to appear in b_flags unless they are |
166 | * specifically set by later operations on the buffer. | |
1da177e4 | 167 | */ |
611c9946 | 168 | flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD); |
ce8e922c | 169 | |
ce8e922c | 170 | atomic_set(&bp->b_hold, 1); |
430cbeb8 | 171 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 172 | init_completion(&bp->b_iowait); |
430cbeb8 | 173 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 174 | INIT_LIST_HEAD(&bp->b_list); |
74f75a0c | 175 | RB_CLEAR_NODE(&bp->b_rbnode); |
a731cd11 | 176 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
a4082357 | 177 | spin_lock_init(&bp->b_lock); |
ce8e922c NS |
178 | XB_SET_OWNER(bp); |
179 | bp->b_target = target; | |
3e85c868 | 180 | bp->b_flags = flags; |
de1cbee4 | 181 | |
1da177e4 | 182 | /* |
aa0e8833 DC |
183 | * Set length and io_length to the same value initially. |
184 | * I/O routines should use io_length, which will be the same in | |
1da177e4 LT |
185 | * most cases but may be reset (e.g. XFS recovery). |
186 | */ | |
3e85c868 DC |
187 | error = xfs_buf_get_maps(bp, nmaps); |
188 | if (error) { | |
189 | kmem_zone_free(xfs_buf_zone, bp); | |
190 | return NULL; | |
191 | } | |
192 | ||
193 | bp->b_bn = map[0].bm_bn; | |
194 | bp->b_length = 0; | |
195 | for (i = 0; i < nmaps; i++) { | |
196 | bp->b_maps[i].bm_bn = map[i].bm_bn; | |
197 | bp->b_maps[i].bm_len = map[i].bm_len; | |
198 | bp->b_length += map[i].bm_len; | |
199 | } | |
200 | bp->b_io_length = bp->b_length; | |
201 | ||
ce8e922c NS |
202 | atomic_set(&bp->b_pin_count, 0); |
203 | init_waitqueue_head(&bp->b_waiters); | |
204 | ||
205 | XFS_STATS_INC(xb_create); | |
0b1b213f | 206 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 CH |
207 | |
208 | return bp; | |
1da177e4 LT |
209 | } |
210 | ||
211 | /* | |
ce8e922c NS |
212 | * Allocate a page array capable of holding a specified number |
213 | * of pages, and point the page buf at it. | |
1da177e4 LT |
214 | */ |
215 | STATIC int | |
ce8e922c NS |
216 | _xfs_buf_get_pages( |
217 | xfs_buf_t *bp, | |
1da177e4 | 218 | int page_count, |
ce8e922c | 219 | xfs_buf_flags_t flags) |
1da177e4 LT |
220 | { |
221 | /* Make sure that we have a page list */ | |
ce8e922c | 222 | if (bp->b_pages == NULL) { |
ce8e922c NS |
223 | bp->b_page_count = page_count; |
224 | if (page_count <= XB_PAGES) { | |
225 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 226 | } else { |
ce8e922c | 227 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
aa5c158e | 228 | page_count, KM_NOFS); |
ce8e922c | 229 | if (bp->b_pages == NULL) |
1da177e4 LT |
230 | return -ENOMEM; |
231 | } | |
ce8e922c | 232 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
233 | } |
234 | return 0; | |
235 | } | |
236 | ||
237 | /* | |
ce8e922c | 238 | * Frees b_pages if it was allocated. |
1da177e4 LT |
239 | */ |
240 | STATIC void | |
ce8e922c | 241 | _xfs_buf_free_pages( |
1da177e4 LT |
242 | xfs_buf_t *bp) |
243 | { | |
ce8e922c | 244 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 245 | kmem_free(bp->b_pages); |
3fc98b1a | 246 | bp->b_pages = NULL; |
1da177e4 LT |
247 | } |
248 | } | |
249 | ||
250 | /* | |
251 | * Releases the specified buffer. | |
252 | * | |
253 | * The modification state of any associated pages is left unchanged. | |
b46fe825 | 254 | * The buffer must not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
255 | * hashed and refcounted buffers |
256 | */ | |
257 | void | |
ce8e922c | 258 | xfs_buf_free( |
1da177e4 LT |
259 | xfs_buf_t *bp) |
260 | { | |
0b1b213f | 261 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 262 | |
430cbeb8 DC |
263 | ASSERT(list_empty(&bp->b_lru)); |
264 | ||
0e6e847f | 265 | if (bp->b_flags & _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 | ||
0e6e847f | 275 | __free_page(page); |
948ecdb4 | 276 | } |
0e6e847f DC |
277 | } else if (bp->b_flags & _XBF_KMEM) |
278 | kmem_free(bp->b_addr); | |
3fc98b1a | 279 | _xfs_buf_free_pages(bp); |
3e85c868 | 280 | xfs_buf_free_maps(bp); |
4347b9d7 | 281 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
282 | } |
283 | ||
284 | /* | |
0e6e847f | 285 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
286 | */ |
287 | STATIC int | |
0e6e847f | 288 | xfs_buf_allocate_memory( |
1da177e4 LT |
289 | xfs_buf_t *bp, |
290 | uint flags) | |
291 | { | |
aa0e8833 | 292 | size_t size; |
1da177e4 | 293 | size_t nbytes, offset; |
ce8e922c | 294 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 295 | unsigned short page_count, i; |
795cac72 | 296 | xfs_off_t start, end; |
1da177e4 LT |
297 | int error; |
298 | ||
0e6e847f DC |
299 | /* |
300 | * for buffers that are contained within a single page, just allocate | |
301 | * the memory from the heap - there's no need for the complexity of | |
302 | * page arrays to keep allocation down to order 0. | |
303 | */ | |
795cac72 DC |
304 | size = BBTOB(bp->b_length); |
305 | if (size < PAGE_SIZE) { | |
aa5c158e | 306 | bp->b_addr = kmem_alloc(size, KM_NOFS); |
0e6e847f DC |
307 | if (!bp->b_addr) { |
308 | /* low memory - use alloc_page loop instead */ | |
309 | goto use_alloc_page; | |
310 | } | |
311 | ||
795cac72 | 312 | if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) != |
0e6e847f DC |
313 | ((unsigned long)bp->b_addr & PAGE_MASK)) { |
314 | /* b_addr spans two pages - use alloc_page instead */ | |
315 | kmem_free(bp->b_addr); | |
316 | bp->b_addr = NULL; | |
317 | goto use_alloc_page; | |
318 | } | |
319 | bp->b_offset = offset_in_page(bp->b_addr); | |
320 | bp->b_pages = bp->b_page_array; | |
321 | bp->b_pages[0] = virt_to_page(bp->b_addr); | |
322 | bp->b_page_count = 1; | |
611c9946 | 323 | bp->b_flags |= _XBF_KMEM; |
0e6e847f DC |
324 | return 0; |
325 | } | |
326 | ||
327 | use_alloc_page: | |
f4b42421 MT |
328 | start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT; |
329 | end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1) | |
cbb7baab | 330 | >> PAGE_SHIFT; |
795cac72 | 331 | page_count = end - start; |
ce8e922c | 332 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
333 | if (unlikely(error)) |
334 | return error; | |
1da177e4 | 335 | |
ce8e922c | 336 | offset = bp->b_offset; |
0e6e847f | 337 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 338 | |
ce8e922c | 339 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
340 | struct page *page; |
341 | uint retries = 0; | |
0e6e847f DC |
342 | retry: |
343 | page = alloc_page(gfp_mask); | |
1da177e4 | 344 | if (unlikely(page == NULL)) { |
ce8e922c NS |
345 | if (flags & XBF_READ_AHEAD) { |
346 | bp->b_page_count = i; | |
0e6e847f DC |
347 | error = ENOMEM; |
348 | goto out_free_pages; | |
1da177e4 LT |
349 | } |
350 | ||
351 | /* | |
352 | * This could deadlock. | |
353 | * | |
354 | * But until all the XFS lowlevel code is revamped to | |
355 | * handle buffer allocation failures we can't do much. | |
356 | */ | |
357 | if (!(++retries % 100)) | |
4f10700a DC |
358 | xfs_err(NULL, |
359 | "possible memory allocation deadlock in %s (mode:0x%x)", | |
34a622b2 | 360 | __func__, gfp_mask); |
1da177e4 | 361 | |
ce8e922c | 362 | XFS_STATS_INC(xb_page_retries); |
8aa7e847 | 363 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
364 | goto retry; |
365 | } | |
366 | ||
ce8e922c | 367 | XFS_STATS_INC(xb_page_found); |
1da177e4 | 368 | |
0e6e847f | 369 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 370 | size -= nbytes; |
ce8e922c | 371 | bp->b_pages[i] = page; |
1da177e4 LT |
372 | offset = 0; |
373 | } | |
0e6e847f | 374 | return 0; |
1da177e4 | 375 | |
0e6e847f DC |
376 | out_free_pages: |
377 | for (i = 0; i < bp->b_page_count; i++) | |
378 | __free_page(bp->b_pages[i]); | |
1da177e4 LT |
379 | return error; |
380 | } | |
381 | ||
382 | /* | |
25985edc | 383 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
384 | */ |
385 | STATIC int | |
ce8e922c | 386 | _xfs_buf_map_pages( |
1da177e4 LT |
387 | xfs_buf_t *bp, |
388 | uint flags) | |
389 | { | |
0e6e847f | 390 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 391 | if (bp->b_page_count == 1) { |
0e6e847f | 392 | /* A single page buffer is always mappable */ |
ce8e922c | 393 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
611c9946 DC |
394 | } else if (flags & XBF_UNMAPPED) { |
395 | bp->b_addr = NULL; | |
396 | } else { | |
a19fb380 DC |
397 | int retried = 0; |
398 | ||
399 | do { | |
400 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
401 | -1, PAGE_KERNEL); | |
402 | if (bp->b_addr) | |
403 | break; | |
404 | vm_unmap_aliases(); | |
405 | } while (retried++ <= 1); | |
406 | ||
407 | if (!bp->b_addr) | |
1da177e4 | 408 | return -ENOMEM; |
ce8e922c | 409 | bp->b_addr += bp->b_offset; |
1da177e4 LT |
410 | } |
411 | ||
412 | return 0; | |
413 | } | |
414 | ||
415 | /* | |
416 | * Finding and Reading Buffers | |
417 | */ | |
418 | ||
419 | /* | |
ce8e922c | 420 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 | 421 | * a given range of an inode. The buffer is returned |
eabbaf11 | 422 | * locked. No I/O is implied by this call. |
1da177e4 LT |
423 | */ |
424 | xfs_buf_t * | |
ce8e922c | 425 | _xfs_buf_find( |
e70b73f8 | 426 | struct xfs_buftarg *btp, |
3e85c868 DC |
427 | struct xfs_buf_map *map, |
428 | int nmaps, | |
ce8e922c NS |
429 | xfs_buf_flags_t flags, |
430 | xfs_buf_t *new_bp) | |
1da177e4 | 431 | { |
e70b73f8 | 432 | size_t numbytes; |
74f75a0c DC |
433 | struct xfs_perag *pag; |
434 | struct rb_node **rbp; | |
435 | struct rb_node *parent; | |
436 | xfs_buf_t *bp; | |
3e85c868 | 437 | xfs_daddr_t blkno = map[0].bm_bn; |
10616b80 | 438 | xfs_daddr_t eofs; |
3e85c868 DC |
439 | int numblks = 0; |
440 | int i; | |
1da177e4 | 441 | |
3e85c868 DC |
442 | for (i = 0; i < nmaps; i++) |
443 | numblks += map[i].bm_len; | |
e70b73f8 | 444 | numbytes = BBTOB(numblks); |
1da177e4 LT |
445 | |
446 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
e70b73f8 | 447 | ASSERT(!(numbytes < (1 << btp->bt_sshift))); |
de1cbee4 | 448 | ASSERT(!(BBTOB(blkno) & (xfs_off_t)btp->bt_smask)); |
1da177e4 | 449 | |
10616b80 DC |
450 | /* |
451 | * Corrupted block numbers can get through to here, unfortunately, so we | |
452 | * have to check that the buffer falls within the filesystem bounds. | |
453 | */ | |
454 | eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks); | |
455 | if (blkno >= eofs) { | |
456 | /* | |
457 | * XXX (dgc): we should really be returning EFSCORRUPTED here, | |
458 | * but none of the higher level infrastructure supports | |
459 | * returning a specific error on buffer lookup failures. | |
460 | */ | |
461 | xfs_alert(btp->bt_mount, | |
462 | "%s: Block out of range: block 0x%llx, EOFS 0x%llx ", | |
463 | __func__, blkno, eofs); | |
7bc0dc27 | 464 | WARN_ON(1); |
10616b80 DC |
465 | return NULL; |
466 | } | |
467 | ||
74f75a0c DC |
468 | /* get tree root */ |
469 | pag = xfs_perag_get(btp->bt_mount, | |
e70b73f8 | 470 | xfs_daddr_to_agno(btp->bt_mount, blkno)); |
74f75a0c DC |
471 | |
472 | /* walk tree */ | |
473 | spin_lock(&pag->pag_buf_lock); | |
474 | rbp = &pag->pag_buf_tree.rb_node; | |
475 | parent = NULL; | |
476 | bp = NULL; | |
477 | while (*rbp) { | |
478 | parent = *rbp; | |
479 | bp = rb_entry(parent, struct xfs_buf, b_rbnode); | |
480 | ||
de1cbee4 | 481 | if (blkno < bp->b_bn) |
74f75a0c | 482 | rbp = &(*rbp)->rb_left; |
de1cbee4 | 483 | else if (blkno > bp->b_bn) |
74f75a0c DC |
484 | rbp = &(*rbp)->rb_right; |
485 | else { | |
486 | /* | |
de1cbee4 | 487 | * found a block number match. If the range doesn't |
74f75a0c DC |
488 | * match, the only way this is allowed is if the buffer |
489 | * in the cache is stale and the transaction that made | |
490 | * it stale has not yet committed. i.e. we are | |
491 | * reallocating a busy extent. Skip this buffer and | |
492 | * continue searching to the right for an exact match. | |
493 | */ | |
4e94b71b | 494 | if (bp->b_length != numblks) { |
74f75a0c DC |
495 | ASSERT(bp->b_flags & XBF_STALE); |
496 | rbp = &(*rbp)->rb_right; | |
497 | continue; | |
498 | } | |
ce8e922c | 499 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
500 | goto found; |
501 | } | |
502 | } | |
503 | ||
504 | /* No match found */ | |
ce8e922c | 505 | if (new_bp) { |
74f75a0c DC |
506 | rb_link_node(&new_bp->b_rbnode, parent, rbp); |
507 | rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree); | |
508 | /* the buffer keeps the perag reference until it is freed */ | |
509 | new_bp->b_pag = pag; | |
510 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 511 | } else { |
ce8e922c | 512 | XFS_STATS_INC(xb_miss_locked); |
74f75a0c DC |
513 | spin_unlock(&pag->pag_buf_lock); |
514 | xfs_perag_put(pag); | |
1da177e4 | 515 | } |
ce8e922c | 516 | return new_bp; |
1da177e4 LT |
517 | |
518 | found: | |
74f75a0c DC |
519 | spin_unlock(&pag->pag_buf_lock); |
520 | xfs_perag_put(pag); | |
1da177e4 | 521 | |
0c842ad4 CH |
522 | if (!xfs_buf_trylock(bp)) { |
523 | if (flags & XBF_TRYLOCK) { | |
ce8e922c NS |
524 | xfs_buf_rele(bp); |
525 | XFS_STATS_INC(xb_busy_locked); | |
526 | return NULL; | |
1da177e4 | 527 | } |
0c842ad4 CH |
528 | xfs_buf_lock(bp); |
529 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
530 | } |
531 | ||
0e6e847f DC |
532 | /* |
533 | * if the buffer is stale, clear all the external state associated with | |
534 | * it. We need to keep flags such as how we allocated the buffer memory | |
535 | * intact here. | |
536 | */ | |
ce8e922c NS |
537 | if (bp->b_flags & XBF_STALE) { |
538 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
cfb02852 | 539 | ASSERT(bp->b_iodone == NULL); |
611c9946 | 540 | bp->b_flags &= _XBF_KMEM | _XBF_PAGES; |
1813dd64 | 541 | bp->b_ops = NULL; |
2f926587 | 542 | } |
0b1b213f CH |
543 | |
544 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ce8e922c NS |
545 | XFS_STATS_INC(xb_get_locked); |
546 | return bp; | |
1da177e4 LT |
547 | } |
548 | ||
549 | /* | |
3815832a DC |
550 | * Assembles a buffer covering the specified range. The code is optimised for |
551 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
552 | * more hits than misses. | |
1da177e4 | 553 | */ |
3815832a | 554 | struct xfs_buf * |
6dde2707 DC |
555 | xfs_buf_get_map( |
556 | struct xfs_buftarg *target, | |
557 | struct xfs_buf_map *map, | |
558 | int nmaps, | |
ce8e922c | 559 | xfs_buf_flags_t flags) |
1da177e4 | 560 | { |
3815832a DC |
561 | struct xfs_buf *bp; |
562 | struct xfs_buf *new_bp; | |
0e6e847f | 563 | int error = 0; |
1da177e4 | 564 | |
6dde2707 | 565 | bp = _xfs_buf_find(target, map, nmaps, flags, NULL); |
3815832a DC |
566 | if (likely(bp)) |
567 | goto found; | |
568 | ||
6dde2707 | 569 | new_bp = _xfs_buf_alloc(target, map, nmaps, flags); |
ce8e922c | 570 | if (unlikely(!new_bp)) |
1da177e4 LT |
571 | return NULL; |
572 | ||
fe2429b0 DC |
573 | error = xfs_buf_allocate_memory(new_bp, flags); |
574 | if (error) { | |
3e85c868 | 575 | xfs_buf_free(new_bp); |
fe2429b0 DC |
576 | return NULL; |
577 | } | |
578 | ||
6dde2707 | 579 | bp = _xfs_buf_find(target, map, nmaps, flags, new_bp); |
3815832a | 580 | if (!bp) { |
fe2429b0 | 581 | xfs_buf_free(new_bp); |
3815832a DC |
582 | return NULL; |
583 | } | |
584 | ||
fe2429b0 DC |
585 | if (bp != new_bp) |
586 | xfs_buf_free(new_bp); | |
1da177e4 | 587 | |
3815832a | 588 | found: |
611c9946 | 589 | if (!bp->b_addr) { |
ce8e922c | 590 | error = _xfs_buf_map_pages(bp, flags); |
1da177e4 | 591 | if (unlikely(error)) { |
4f10700a DC |
592 | xfs_warn(target->bt_mount, |
593 | "%s: failed to map pages\n", __func__); | |
a8acad70 DC |
594 | xfs_buf_relse(bp); |
595 | return NULL; | |
1da177e4 LT |
596 | } |
597 | } | |
598 | ||
ce8e922c | 599 | XFS_STATS_INC(xb_get); |
0b1b213f | 600 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 601 | return bp; |
1da177e4 LT |
602 | } |
603 | ||
5d765b97 CH |
604 | STATIC int |
605 | _xfs_buf_read( | |
606 | xfs_buf_t *bp, | |
607 | xfs_buf_flags_t flags) | |
608 | { | |
43ff2122 | 609 | ASSERT(!(flags & XBF_WRITE)); |
f4b42421 | 610 | ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL); |
5d765b97 | 611 | |
43ff2122 | 612 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD); |
1d5ae5df | 613 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 614 | |
0e95f19a DC |
615 | xfs_buf_iorequest(bp); |
616 | if (flags & XBF_ASYNC) | |
617 | return 0; | |
ec53d1db | 618 | return xfs_buf_iowait(bp); |
5d765b97 CH |
619 | } |
620 | ||
1da177e4 | 621 | xfs_buf_t * |
6dde2707 DC |
622 | xfs_buf_read_map( |
623 | struct xfs_buftarg *target, | |
624 | struct xfs_buf_map *map, | |
625 | int nmaps, | |
c3f8fc73 | 626 | xfs_buf_flags_t flags, |
1813dd64 | 627 | const struct xfs_buf_ops *ops) |
1da177e4 | 628 | { |
6dde2707 | 629 | struct xfs_buf *bp; |
ce8e922c NS |
630 | |
631 | flags |= XBF_READ; | |
632 | ||
6dde2707 | 633 | bp = xfs_buf_get_map(target, map, nmaps, flags); |
ce8e922c | 634 | if (bp) { |
0b1b213f CH |
635 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
636 | ||
ce8e922c | 637 | if (!XFS_BUF_ISDONE(bp)) { |
ce8e922c | 638 | XFS_STATS_INC(xb_get_read); |
1813dd64 | 639 | bp->b_ops = ops; |
5d765b97 | 640 | _xfs_buf_read(bp, flags); |
ce8e922c | 641 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
642 | /* |
643 | * Read ahead call which is already satisfied, | |
644 | * drop the buffer | |
645 | */ | |
a8acad70 DC |
646 | xfs_buf_relse(bp); |
647 | return NULL; | |
1da177e4 | 648 | } else { |
1da177e4 | 649 | /* We do not want read in the flags */ |
ce8e922c | 650 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
651 | } |
652 | } | |
653 | ||
ce8e922c | 654 | return bp; |
1da177e4 LT |
655 | } |
656 | ||
1da177e4 | 657 | /* |
ce8e922c NS |
658 | * If we are not low on memory then do the readahead in a deadlock |
659 | * safe manner. | |
1da177e4 LT |
660 | */ |
661 | void | |
6dde2707 DC |
662 | xfs_buf_readahead_map( |
663 | struct xfs_buftarg *target, | |
664 | struct xfs_buf_map *map, | |
c3f8fc73 | 665 | int nmaps, |
1813dd64 | 666 | const struct xfs_buf_ops *ops) |
1da177e4 | 667 | { |
0e6e847f | 668 | if (bdi_read_congested(target->bt_bdi)) |
1da177e4 LT |
669 | return; |
670 | ||
6dde2707 | 671 | xfs_buf_read_map(target, map, nmaps, |
1813dd64 | 672 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD, ops); |
1da177e4 LT |
673 | } |
674 | ||
5adc94c2 DC |
675 | /* |
676 | * Read an uncached buffer from disk. Allocates and returns a locked | |
677 | * buffer containing the disk contents or nothing. | |
678 | */ | |
679 | struct xfs_buf * | |
680 | xfs_buf_read_uncached( | |
5adc94c2 DC |
681 | struct xfs_buftarg *target, |
682 | xfs_daddr_t daddr, | |
e70b73f8 | 683 | size_t numblks, |
c3f8fc73 | 684 | int flags, |
1813dd64 | 685 | const struct xfs_buf_ops *ops) |
5adc94c2 | 686 | { |
eab4e633 | 687 | struct xfs_buf *bp; |
5adc94c2 | 688 | |
e70b73f8 | 689 | bp = xfs_buf_get_uncached(target, numblks, flags); |
5adc94c2 DC |
690 | if (!bp) |
691 | return NULL; | |
692 | ||
693 | /* set up the buffer for a read IO */ | |
3e85c868 DC |
694 | ASSERT(bp->b_map_count == 1); |
695 | bp->b_bn = daddr; | |
696 | bp->b_maps[0].bm_bn = daddr; | |
cbb7baab | 697 | bp->b_flags |= XBF_READ; |
1813dd64 | 698 | bp->b_ops = ops; |
5adc94c2 | 699 | |
e70b73f8 | 700 | xfsbdstrat(target->bt_mount, bp); |
eab4e633 | 701 | xfs_buf_iowait(bp); |
5adc94c2 | 702 | return bp; |
1da177e4 LT |
703 | } |
704 | ||
44396476 DC |
705 | /* |
706 | * Return a buffer allocated as an empty buffer and associated to external | |
707 | * memory via xfs_buf_associate_memory() back to it's empty state. | |
708 | */ | |
709 | void | |
710 | xfs_buf_set_empty( | |
711 | struct xfs_buf *bp, | |
e70b73f8 | 712 | size_t numblks) |
44396476 DC |
713 | { |
714 | if (bp->b_pages) | |
715 | _xfs_buf_free_pages(bp); | |
716 | ||
717 | bp->b_pages = NULL; | |
718 | bp->b_page_count = 0; | |
719 | bp->b_addr = NULL; | |
4e94b71b | 720 | bp->b_length = numblks; |
aa0e8833 | 721 | bp->b_io_length = numblks; |
3e85c868 DC |
722 | |
723 | ASSERT(bp->b_map_count == 1); | |
44396476 | 724 | bp->b_bn = XFS_BUF_DADDR_NULL; |
3e85c868 DC |
725 | bp->b_maps[0].bm_bn = XFS_BUF_DADDR_NULL; |
726 | bp->b_maps[0].bm_len = bp->b_length; | |
44396476 DC |
727 | } |
728 | ||
1da177e4 LT |
729 | static inline struct page * |
730 | mem_to_page( | |
731 | void *addr) | |
732 | { | |
9e2779fa | 733 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
734 | return virt_to_page(addr); |
735 | } else { | |
736 | return vmalloc_to_page(addr); | |
737 | } | |
738 | } | |
739 | ||
740 | int | |
ce8e922c NS |
741 | xfs_buf_associate_memory( |
742 | xfs_buf_t *bp, | |
1da177e4 LT |
743 | void *mem, |
744 | size_t len) | |
745 | { | |
746 | int rval; | |
747 | int i = 0; | |
d1afb678 LM |
748 | unsigned long pageaddr; |
749 | unsigned long offset; | |
750 | size_t buflen; | |
1da177e4 LT |
751 | int page_count; |
752 | ||
0e6e847f | 753 | pageaddr = (unsigned long)mem & PAGE_MASK; |
d1afb678 | 754 | offset = (unsigned long)mem - pageaddr; |
0e6e847f DC |
755 | buflen = PAGE_ALIGN(len + offset); |
756 | page_count = buflen >> PAGE_SHIFT; | |
1da177e4 LT |
757 | |
758 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
759 | if (bp->b_pages) |
760 | _xfs_buf_free_pages(bp); | |
1da177e4 | 761 | |
ce8e922c NS |
762 | bp->b_pages = NULL; |
763 | bp->b_addr = mem; | |
1da177e4 | 764 | |
aa5c158e | 765 | rval = _xfs_buf_get_pages(bp, page_count, 0); |
1da177e4 LT |
766 | if (rval) |
767 | return rval; | |
768 | ||
ce8e922c | 769 | bp->b_offset = offset; |
d1afb678 LM |
770 | |
771 | for (i = 0; i < bp->b_page_count; i++) { | |
772 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
0e6e847f | 773 | pageaddr += PAGE_SIZE; |
1da177e4 | 774 | } |
1da177e4 | 775 | |
aa0e8833 | 776 | bp->b_io_length = BTOBB(len); |
4e94b71b | 777 | bp->b_length = BTOBB(buflen); |
1da177e4 LT |
778 | |
779 | return 0; | |
780 | } | |
781 | ||
782 | xfs_buf_t * | |
686865f7 DC |
783 | xfs_buf_get_uncached( |
784 | struct xfs_buftarg *target, | |
e70b73f8 | 785 | size_t numblks, |
686865f7 | 786 | int flags) |
1da177e4 | 787 | { |
e70b73f8 | 788 | unsigned long page_count; |
1fa40b01 | 789 | int error, i; |
3e85c868 DC |
790 | struct xfs_buf *bp; |
791 | DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks); | |
1da177e4 | 792 | |
3e85c868 | 793 | bp = _xfs_buf_alloc(target, &map, 1, 0); |
1da177e4 LT |
794 | if (unlikely(bp == NULL)) |
795 | goto fail; | |
1da177e4 | 796 | |
e70b73f8 | 797 | page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT; |
1fa40b01 CH |
798 | error = _xfs_buf_get_pages(bp, page_count, 0); |
799 | if (error) | |
1da177e4 LT |
800 | goto fail_free_buf; |
801 | ||
1fa40b01 | 802 | for (i = 0; i < page_count; i++) { |
686865f7 | 803 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
804 | if (!bp->b_pages[i]) |
805 | goto fail_free_mem; | |
1da177e4 | 806 | } |
1fa40b01 | 807 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 808 | |
611c9946 | 809 | error = _xfs_buf_map_pages(bp, 0); |
1fa40b01 | 810 | if (unlikely(error)) { |
4f10700a DC |
811 | xfs_warn(target->bt_mount, |
812 | "%s: failed to map pages\n", __func__); | |
1da177e4 | 813 | goto fail_free_mem; |
1fa40b01 | 814 | } |
1da177e4 | 815 | |
686865f7 | 816 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 817 | return bp; |
1fa40b01 | 818 | |
1da177e4 | 819 | fail_free_mem: |
1fa40b01 CH |
820 | while (--i >= 0) |
821 | __free_page(bp->b_pages[i]); | |
ca165b88 | 822 | _xfs_buf_free_pages(bp); |
1da177e4 | 823 | fail_free_buf: |
3e85c868 | 824 | xfs_buf_free_maps(bp); |
4347b9d7 | 825 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
826 | fail: |
827 | return NULL; | |
828 | } | |
829 | ||
830 | /* | |
1da177e4 LT |
831 | * Increment reference count on buffer, to hold the buffer concurrently |
832 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
833 | * Must hold the buffer already to call this function. |
834 | */ | |
835 | void | |
ce8e922c NS |
836 | xfs_buf_hold( |
837 | xfs_buf_t *bp) | |
1da177e4 | 838 | { |
0b1b213f | 839 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 840 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
841 | } |
842 | ||
843 | /* | |
ce8e922c NS |
844 | * Releases a hold on the specified buffer. If the |
845 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
846 | */ |
847 | void | |
ce8e922c NS |
848 | xfs_buf_rele( |
849 | xfs_buf_t *bp) | |
1da177e4 | 850 | { |
74f75a0c | 851 | struct xfs_perag *pag = bp->b_pag; |
1da177e4 | 852 | |
0b1b213f | 853 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 854 | |
74f75a0c | 855 | if (!pag) { |
430cbeb8 | 856 | ASSERT(list_empty(&bp->b_lru)); |
74f75a0c | 857 | ASSERT(RB_EMPTY_NODE(&bp->b_rbnode)); |
fad3aa1e NS |
858 | if (atomic_dec_and_test(&bp->b_hold)) |
859 | xfs_buf_free(bp); | |
860 | return; | |
861 | } | |
862 | ||
74f75a0c | 863 | ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode)); |
430cbeb8 | 864 | |
3790689f | 865 | ASSERT(atomic_read(&bp->b_hold) > 0); |
74f75a0c | 866 | if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) { |
a4082357 DC |
867 | spin_lock(&bp->b_lock); |
868 | if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) { | |
869 | /* | |
870 | * If the buffer is added to the LRU take a new | |
871 | * reference to the buffer for the LRU and clear the | |
872 | * (now stale) dispose list state flag | |
873 | */ | |
874 | if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) { | |
875 | bp->b_state &= ~XFS_BSTATE_DISPOSE; | |
876 | atomic_inc(&bp->b_hold); | |
877 | } | |
878 | spin_unlock(&bp->b_lock); | |
430cbeb8 | 879 | spin_unlock(&pag->pag_buf_lock); |
1da177e4 | 880 | } else { |
a4082357 DC |
881 | /* |
882 | * most of the time buffers will already be removed from | |
883 | * the LRU, so optimise that case by checking for the | |
884 | * XFS_BSTATE_DISPOSE flag indicating the last list the | |
885 | * buffer was on was the disposal list | |
886 | */ | |
887 | if (!(bp->b_state & XFS_BSTATE_DISPOSE)) { | |
888 | list_lru_del(&bp->b_target->bt_lru, &bp->b_lru); | |
889 | } else { | |
890 | ASSERT(list_empty(&bp->b_lru)); | |
891 | } | |
892 | spin_unlock(&bp->b_lock); | |
893 | ||
43ff2122 | 894 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
74f75a0c DC |
895 | rb_erase(&bp->b_rbnode, &pag->pag_buf_tree); |
896 | spin_unlock(&pag->pag_buf_lock); | |
897 | xfs_perag_put(pag); | |
ce8e922c | 898 | xfs_buf_free(bp); |
1da177e4 LT |
899 | } |
900 | } | |
901 | } | |
902 | ||
903 | ||
904 | /* | |
0e6e847f | 905 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
906 | * |
907 | * If we come across a stale, pinned, locked buffer, we know that we are | |
908 | * being asked to lock a buffer that has been reallocated. Because it is | |
909 | * pinned, we know that the log has not been pushed to disk and hence it | |
910 | * will still be locked. Rather than continuing to have trylock attempts | |
911 | * fail until someone else pushes the log, push it ourselves before | |
912 | * returning. This means that the xfsaild will not get stuck trying | |
913 | * to push on stale inode buffers. | |
1da177e4 LT |
914 | */ |
915 | int | |
0c842ad4 CH |
916 | xfs_buf_trylock( |
917 | struct xfs_buf *bp) | |
1da177e4 LT |
918 | { |
919 | int locked; | |
920 | ||
ce8e922c | 921 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 922 | if (locked) |
ce8e922c | 923 | XB_SET_OWNER(bp); |
0b1b213f | 924 | |
0c842ad4 CH |
925 | trace_xfs_buf_trylock(bp, _RET_IP_); |
926 | return locked; | |
1da177e4 | 927 | } |
1da177e4 LT |
928 | |
929 | /* | |
0e6e847f | 930 | * Lock a buffer object. |
ed3b4d6c DC |
931 | * |
932 | * If we come across a stale, pinned, locked buffer, we know that we | |
933 | * are being asked to lock a buffer that has been reallocated. Because | |
934 | * it is pinned, we know that the log has not been pushed to disk and | |
935 | * hence it will still be locked. Rather than sleeping until someone | |
936 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 937 | */ |
ce8e922c NS |
938 | void |
939 | xfs_buf_lock( | |
0c842ad4 | 940 | struct xfs_buf *bp) |
1da177e4 | 941 | { |
0b1b213f CH |
942 | trace_xfs_buf_lock(bp, _RET_IP_); |
943 | ||
ed3b4d6c | 944 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 945 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
946 | down(&bp->b_sema); |
947 | XB_SET_OWNER(bp); | |
0b1b213f CH |
948 | |
949 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
950 | } |
951 | ||
1da177e4 | 952 | void |
ce8e922c | 953 | xfs_buf_unlock( |
0c842ad4 | 954 | struct xfs_buf *bp) |
1da177e4 | 955 | { |
ce8e922c NS |
956 | XB_CLEAR_OWNER(bp); |
957 | up(&bp->b_sema); | |
0b1b213f CH |
958 | |
959 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
960 | } |
961 | ||
ce8e922c NS |
962 | STATIC void |
963 | xfs_buf_wait_unpin( | |
964 | xfs_buf_t *bp) | |
1da177e4 LT |
965 | { |
966 | DECLARE_WAITQUEUE (wait, current); | |
967 | ||
ce8e922c | 968 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
969 | return; |
970 | ||
ce8e922c | 971 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
972 | for (;;) { |
973 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 974 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 975 | break; |
7eaceacc | 976 | io_schedule(); |
1da177e4 | 977 | } |
ce8e922c | 978 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
979 | set_current_state(TASK_RUNNING); |
980 | } | |
981 | ||
982 | /* | |
983 | * Buffer Utility Routines | |
984 | */ | |
985 | ||
1da177e4 | 986 | STATIC void |
ce8e922c | 987 | xfs_buf_iodone_work( |
c4028958 | 988 | struct work_struct *work) |
1da177e4 | 989 | { |
1813dd64 | 990 | struct xfs_buf *bp = |
c4028958 | 991 | container_of(work, xfs_buf_t, b_iodone_work); |
1813dd64 DC |
992 | bool read = !!(bp->b_flags & XBF_READ); |
993 | ||
994 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); | |
d5929de8 DC |
995 | |
996 | /* only validate buffers that were read without errors */ | |
997 | if (read && bp->b_ops && !bp->b_error && (bp->b_flags & XBF_DONE)) | |
1813dd64 | 998 | bp->b_ops->verify_read(bp); |
1da177e4 | 999 | |
80f6c29d | 1000 | if (bp->b_iodone) |
ce8e922c NS |
1001 | (*(bp->b_iodone))(bp); |
1002 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 | 1003 | xfs_buf_relse(bp); |
1813dd64 DC |
1004 | else { |
1005 | ASSERT(read && bp->b_ops); | |
1006 | complete(&bp->b_iowait); | |
1007 | } | |
1da177e4 LT |
1008 | } |
1009 | ||
1010 | void | |
ce8e922c | 1011 | xfs_buf_ioend( |
1813dd64 DC |
1012 | struct xfs_buf *bp, |
1013 | int schedule) | |
1da177e4 | 1014 | { |
1813dd64 DC |
1015 | bool read = !!(bp->b_flags & XBF_READ); |
1016 | ||
0b1b213f CH |
1017 | trace_xfs_buf_iodone(bp, _RET_IP_); |
1018 | ||
ce8e922c NS |
1019 | if (bp->b_error == 0) |
1020 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 1021 | |
1813dd64 | 1022 | if (bp->b_iodone || (read && bp->b_ops) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 1023 | if (schedule) { |
c4028958 | 1024 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 1025 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 1026 | } else { |
c4028958 | 1027 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
1028 | } |
1029 | } else { | |
1813dd64 | 1030 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
b4dd330b | 1031 | complete(&bp->b_iowait); |
1da177e4 LT |
1032 | } |
1033 | } | |
1034 | ||
1da177e4 | 1035 | void |
ce8e922c NS |
1036 | xfs_buf_ioerror( |
1037 | xfs_buf_t *bp, | |
1038 | int error) | |
1da177e4 LT |
1039 | { |
1040 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c | 1041 | bp->b_error = (unsigned short)error; |
0b1b213f | 1042 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1043 | } |
1044 | ||
901796af CH |
1045 | void |
1046 | xfs_buf_ioerror_alert( | |
1047 | struct xfs_buf *bp, | |
1048 | const char *func) | |
1049 | { | |
1050 | xfs_alert(bp->b_target->bt_mount, | |
aa0e8833 DC |
1051 | "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d", |
1052 | (__uint64_t)XFS_BUF_ADDR(bp), func, bp->b_error, bp->b_length); | |
901796af CH |
1053 | } |
1054 | ||
4e23471a CH |
1055 | /* |
1056 | * Called when we want to stop a buffer from getting written or read. | |
1a1a3e97 | 1057 | * We attach the EIO error, muck with its flags, and call xfs_buf_ioend |
4e23471a CH |
1058 | * so that the proper iodone callbacks get called. |
1059 | */ | |
1060 | STATIC int | |
1061 | xfs_bioerror( | |
1062 | xfs_buf_t *bp) | |
1063 | { | |
1064 | #ifdef XFSERRORDEBUG | |
1065 | ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone); | |
1066 | #endif | |
1067 | ||
1068 | /* | |
1069 | * No need to wait until the buffer is unpinned, we aren't flushing it. | |
1070 | */ | |
5a52c2a5 | 1071 | xfs_buf_ioerror(bp, EIO); |
4e23471a CH |
1072 | |
1073 | /* | |
1a1a3e97 | 1074 | * We're calling xfs_buf_ioend, so delete XBF_DONE flag. |
4e23471a CH |
1075 | */ |
1076 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1077 | XFS_BUF_UNDONE(bp); |
c867cb61 | 1078 | xfs_buf_stale(bp); |
4e23471a | 1079 | |
1a1a3e97 | 1080 | xfs_buf_ioend(bp, 0); |
4e23471a CH |
1081 | |
1082 | return EIO; | |
1083 | } | |
1084 | ||
1085 | /* | |
1086 | * Same as xfs_bioerror, except that we are releasing the buffer | |
1a1a3e97 | 1087 | * here ourselves, and avoiding the xfs_buf_ioend call. |
4e23471a CH |
1088 | * This is meant for userdata errors; metadata bufs come with |
1089 | * iodone functions attached, so that we can track down errors. | |
1090 | */ | |
1091 | STATIC int | |
1092 | xfs_bioerror_relse( | |
1093 | struct xfs_buf *bp) | |
1094 | { | |
ed43233b | 1095 | int64_t fl = bp->b_flags; |
4e23471a CH |
1096 | /* |
1097 | * No need to wait until the buffer is unpinned. | |
1098 | * We aren't flushing it. | |
1099 | * | |
1100 | * chunkhold expects B_DONE to be set, whether | |
1101 | * we actually finish the I/O or not. We don't want to | |
1102 | * change that interface. | |
1103 | */ | |
1104 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1105 | XFS_BUF_DONE(bp); |
c867cb61 | 1106 | xfs_buf_stale(bp); |
cb669ca5 | 1107 | bp->b_iodone = NULL; |
0cadda1c | 1108 | if (!(fl & XBF_ASYNC)) { |
4e23471a CH |
1109 | /* |
1110 | * Mark b_error and B_ERROR _both_. | |
1111 | * Lot's of chunkcache code assumes that. | |
1112 | * There's no reason to mark error for | |
1113 | * ASYNC buffers. | |
1114 | */ | |
5a52c2a5 | 1115 | xfs_buf_ioerror(bp, EIO); |
5fde0326 | 1116 | complete(&bp->b_iowait); |
4e23471a CH |
1117 | } else { |
1118 | xfs_buf_relse(bp); | |
1119 | } | |
1120 | ||
1121 | return EIO; | |
1122 | } | |
1123 | ||
a2dcf5df | 1124 | STATIC int |
4e23471a CH |
1125 | xfs_bdstrat_cb( |
1126 | struct xfs_buf *bp) | |
1127 | { | |
ebad861b | 1128 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
4e23471a CH |
1129 | trace_xfs_bdstrat_shut(bp, _RET_IP_); |
1130 | /* | |
1131 | * Metadata write that didn't get logged but | |
1132 | * written delayed anyway. These aren't associated | |
1133 | * with a transaction, and can be ignored. | |
1134 | */ | |
1135 | if (!bp->b_iodone && !XFS_BUF_ISREAD(bp)) | |
1136 | return xfs_bioerror_relse(bp); | |
1137 | else | |
1138 | return xfs_bioerror(bp); | |
1139 | } | |
1140 | ||
1141 | xfs_buf_iorequest(bp); | |
1142 | return 0; | |
1143 | } | |
1144 | ||
a2dcf5df CH |
1145 | int |
1146 | xfs_bwrite( | |
1147 | struct xfs_buf *bp) | |
1148 | { | |
1149 | int error; | |
1150 | ||
1151 | ASSERT(xfs_buf_islocked(bp)); | |
1152 | ||
1153 | bp->b_flags |= XBF_WRITE; | |
1154 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q); | |
1155 | ||
1156 | xfs_bdstrat_cb(bp); | |
1157 | ||
1158 | error = xfs_buf_iowait(bp); | |
1159 | if (error) { | |
1160 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1161 | SHUTDOWN_META_IO_ERROR); | |
1162 | } | |
1163 | return error; | |
1164 | } | |
1165 | ||
4e23471a CH |
1166 | /* |
1167 | * Wrapper around bdstrat so that we can stop data from going to disk in case | |
1168 | * we are shutting down the filesystem. Typically user data goes thru this | |
1169 | * path; one of the exceptions is the superblock. | |
1170 | */ | |
1171 | void | |
1172 | xfsbdstrat( | |
1173 | struct xfs_mount *mp, | |
1174 | struct xfs_buf *bp) | |
1175 | { | |
1176 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1177 | trace_xfs_bdstrat_shut(bp, _RET_IP_); | |
1178 | xfs_bioerror_relse(bp); | |
1179 | return; | |
1180 | } | |
1181 | ||
1182 | xfs_buf_iorequest(bp); | |
1183 | } | |
1184 | ||
b8f82a4a | 1185 | STATIC void |
ce8e922c NS |
1186 | _xfs_buf_ioend( |
1187 | xfs_buf_t *bp, | |
1da177e4 LT |
1188 | int schedule) |
1189 | { | |
0e6e847f | 1190 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
ce8e922c | 1191 | xfs_buf_ioend(bp, schedule); |
1da177e4 LT |
1192 | } |
1193 | ||
782e3b3b | 1194 | STATIC void |
ce8e922c | 1195 | xfs_buf_bio_end_io( |
1da177e4 | 1196 | struct bio *bio, |
1da177e4 LT |
1197 | int error) |
1198 | { | |
ce8e922c | 1199 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1da177e4 | 1200 | |
37eb17e6 DC |
1201 | /* |
1202 | * don't overwrite existing errors - otherwise we can lose errors on | |
1203 | * buffers that require multiple bios to complete. | |
1204 | */ | |
1205 | if (!bp->b_error) | |
1206 | xfs_buf_ioerror(bp, -error); | |
1da177e4 | 1207 | |
37eb17e6 | 1208 | if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
73c77e2c JB |
1209 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); |
1210 | ||
ce8e922c | 1211 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1212 | bio_put(bio); |
1da177e4 LT |
1213 | } |
1214 | ||
3e85c868 DC |
1215 | static void |
1216 | xfs_buf_ioapply_map( | |
1217 | struct xfs_buf *bp, | |
1218 | int map, | |
1219 | int *buf_offset, | |
1220 | int *count, | |
1221 | int rw) | |
1da177e4 | 1222 | { |
3e85c868 DC |
1223 | int page_index; |
1224 | int total_nr_pages = bp->b_page_count; | |
1225 | int nr_pages; | |
1226 | struct bio *bio; | |
1227 | sector_t sector = bp->b_maps[map].bm_bn; | |
1228 | int size; | |
1229 | int offset; | |
1da177e4 | 1230 | |
ce8e922c | 1231 | total_nr_pages = bp->b_page_count; |
1da177e4 | 1232 | |
3e85c868 DC |
1233 | /* skip the pages in the buffer before the start offset */ |
1234 | page_index = 0; | |
1235 | offset = *buf_offset; | |
1236 | while (offset >= PAGE_SIZE) { | |
1237 | page_index++; | |
1238 | offset -= PAGE_SIZE; | |
f538d4da CH |
1239 | } |
1240 | ||
3e85c868 DC |
1241 | /* |
1242 | * Limit the IO size to the length of the current vector, and update the | |
1243 | * remaining IO count for the next time around. | |
1244 | */ | |
1245 | size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count); | |
1246 | *count -= size; | |
1247 | *buf_offset += size; | |
34951f5c | 1248 | |
1da177e4 | 1249 | next_chunk: |
ce8e922c | 1250 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1251 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1252 | if (nr_pages > total_nr_pages) | |
1253 | nr_pages = total_nr_pages; | |
1254 | ||
1255 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1256 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1257 | bio->bi_sector = sector; |
ce8e922c NS |
1258 | bio->bi_end_io = xfs_buf_bio_end_io; |
1259 | bio->bi_private = bp; | |
1da177e4 | 1260 | |
0e6e847f | 1261 | |
3e85c868 | 1262 | for (; size && nr_pages; nr_pages--, page_index++) { |
0e6e847f | 1263 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1264 | |
1265 | if (nbytes > size) | |
1266 | nbytes = size; | |
1267 | ||
3e85c868 DC |
1268 | rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes, |
1269 | offset); | |
ce8e922c | 1270 | if (rbytes < nbytes) |
1da177e4 LT |
1271 | break; |
1272 | ||
1273 | offset = 0; | |
aa0e8833 | 1274 | sector += BTOBB(nbytes); |
1da177e4 LT |
1275 | size -= nbytes; |
1276 | total_nr_pages--; | |
1277 | } | |
1278 | ||
1da177e4 | 1279 | if (likely(bio->bi_size)) { |
73c77e2c JB |
1280 | if (xfs_buf_is_vmapped(bp)) { |
1281 | flush_kernel_vmap_range(bp->b_addr, | |
1282 | xfs_buf_vmap_len(bp)); | |
1283 | } | |
1da177e4 LT |
1284 | submit_bio(rw, bio); |
1285 | if (size) | |
1286 | goto next_chunk; | |
1287 | } else { | |
37eb17e6 DC |
1288 | /* |
1289 | * This is guaranteed not to be the last io reference count | |
1290 | * because the caller (xfs_buf_iorequest) holds a count itself. | |
1291 | */ | |
1292 | atomic_dec(&bp->b_io_remaining); | |
ce8e922c | 1293 | xfs_buf_ioerror(bp, EIO); |
ec53d1db | 1294 | bio_put(bio); |
1da177e4 | 1295 | } |
3e85c868 DC |
1296 | |
1297 | } | |
1298 | ||
1299 | STATIC void | |
1300 | _xfs_buf_ioapply( | |
1301 | struct xfs_buf *bp) | |
1302 | { | |
1303 | struct blk_plug plug; | |
1304 | int rw; | |
1305 | int offset; | |
1306 | int size; | |
1307 | int i; | |
1308 | ||
c163f9a1 DC |
1309 | /* |
1310 | * Make sure we capture only current IO errors rather than stale errors | |
1311 | * left over from previous use of the buffer (e.g. failed readahead). | |
1312 | */ | |
1313 | bp->b_error = 0; | |
1314 | ||
3e85c868 DC |
1315 | if (bp->b_flags & XBF_WRITE) { |
1316 | if (bp->b_flags & XBF_SYNCIO) | |
1317 | rw = WRITE_SYNC; | |
1318 | else | |
1319 | rw = WRITE; | |
1320 | if (bp->b_flags & XBF_FUA) | |
1321 | rw |= REQ_FUA; | |
1322 | if (bp->b_flags & XBF_FLUSH) | |
1323 | rw |= REQ_FLUSH; | |
1813dd64 DC |
1324 | |
1325 | /* | |
1326 | * Run the write verifier callback function if it exists. If | |
1327 | * this function fails it will mark the buffer with an error and | |
1328 | * the IO should not be dispatched. | |
1329 | */ | |
1330 | if (bp->b_ops) { | |
1331 | bp->b_ops->verify_write(bp); | |
1332 | if (bp->b_error) { | |
1333 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1334 | SHUTDOWN_CORRUPT_INCORE); | |
1335 | return; | |
1336 | } | |
1337 | } | |
3e85c868 DC |
1338 | } else if (bp->b_flags & XBF_READ_AHEAD) { |
1339 | rw = READA; | |
1340 | } else { | |
1341 | rw = READ; | |
1342 | } | |
1343 | ||
1344 | /* we only use the buffer cache for meta-data */ | |
1345 | rw |= REQ_META; | |
1346 | ||
1347 | /* | |
1348 | * Walk all the vectors issuing IO on them. Set up the initial offset | |
1349 | * into the buffer and the desired IO size before we start - | |
1350 | * _xfs_buf_ioapply_vec() will modify them appropriately for each | |
1351 | * subsequent call. | |
1352 | */ | |
1353 | offset = bp->b_offset; | |
1354 | size = BBTOB(bp->b_io_length); | |
1355 | blk_start_plug(&plug); | |
1356 | for (i = 0; i < bp->b_map_count; i++) { | |
1357 | xfs_buf_ioapply_map(bp, i, &offset, &size, rw); | |
1358 | if (bp->b_error) | |
1359 | break; | |
1360 | if (size <= 0) | |
1361 | break; /* all done */ | |
1362 | } | |
1363 | blk_finish_plug(&plug); | |
1da177e4 LT |
1364 | } |
1365 | ||
0e95f19a | 1366 | void |
ce8e922c NS |
1367 | xfs_buf_iorequest( |
1368 | xfs_buf_t *bp) | |
1da177e4 | 1369 | { |
0b1b213f | 1370 | trace_xfs_buf_iorequest(bp, _RET_IP_); |
1da177e4 | 1371 | |
43ff2122 | 1372 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
1da177e4 | 1373 | |
375ec69d | 1374 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1375 | xfs_buf_wait_unpin(bp); |
ce8e922c | 1376 | xfs_buf_hold(bp); |
1da177e4 LT |
1377 | |
1378 | /* Set the count to 1 initially, this will stop an I/O | |
1379 | * completion callout which happens before we have started | |
ce8e922c | 1380 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1381 | */ |
ce8e922c NS |
1382 | atomic_set(&bp->b_io_remaining, 1); |
1383 | _xfs_buf_ioapply(bp); | |
08023d6d | 1384 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1385 | |
ce8e922c | 1386 | xfs_buf_rele(bp); |
1da177e4 LT |
1387 | } |
1388 | ||
1389 | /* | |
0e95f19a DC |
1390 | * Waits for I/O to complete on the buffer supplied. It returns immediately if |
1391 | * no I/O is pending or there is already a pending error on the buffer. It | |
1392 | * returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1393 | */ |
1394 | int | |
ce8e922c NS |
1395 | xfs_buf_iowait( |
1396 | xfs_buf_t *bp) | |
1da177e4 | 1397 | { |
0b1b213f CH |
1398 | trace_xfs_buf_iowait(bp, _RET_IP_); |
1399 | ||
0e95f19a DC |
1400 | if (!bp->b_error) |
1401 | wait_for_completion(&bp->b_iowait); | |
0b1b213f CH |
1402 | |
1403 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
ce8e922c | 1404 | return bp->b_error; |
1da177e4 LT |
1405 | } |
1406 | ||
ce8e922c NS |
1407 | xfs_caddr_t |
1408 | xfs_buf_offset( | |
1409 | xfs_buf_t *bp, | |
1da177e4 LT |
1410 | size_t offset) |
1411 | { | |
1412 | struct page *page; | |
1413 | ||
611c9946 | 1414 | if (bp->b_addr) |
62926044 | 1415 | return bp->b_addr + offset; |
1da177e4 | 1416 | |
ce8e922c | 1417 | offset += bp->b_offset; |
0e6e847f DC |
1418 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
1419 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_SIZE-1)); | |
1da177e4 LT |
1420 | } |
1421 | ||
1422 | /* | |
1da177e4 LT |
1423 | * Move data into or out of a buffer. |
1424 | */ | |
1425 | void | |
ce8e922c NS |
1426 | xfs_buf_iomove( |
1427 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1428 | size_t boff, /* starting buffer offset */ |
1429 | size_t bsize, /* length to copy */ | |
b9c48649 | 1430 | void *data, /* data address */ |
ce8e922c | 1431 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 | 1432 | { |
795cac72 | 1433 | size_t bend; |
1da177e4 LT |
1434 | |
1435 | bend = boff + bsize; | |
1436 | while (boff < bend) { | |
795cac72 DC |
1437 | struct page *page; |
1438 | int page_index, page_offset, csize; | |
1439 | ||
1440 | page_index = (boff + bp->b_offset) >> PAGE_SHIFT; | |
1441 | page_offset = (boff + bp->b_offset) & ~PAGE_MASK; | |
1442 | page = bp->b_pages[page_index]; | |
1443 | csize = min_t(size_t, PAGE_SIZE - page_offset, | |
1444 | BBTOB(bp->b_io_length) - boff); | |
1da177e4 | 1445 | |
795cac72 | 1446 | ASSERT((csize + page_offset) <= PAGE_SIZE); |
1da177e4 LT |
1447 | |
1448 | switch (mode) { | |
ce8e922c | 1449 | case XBRW_ZERO: |
795cac72 | 1450 | memset(page_address(page) + page_offset, 0, csize); |
1da177e4 | 1451 | break; |
ce8e922c | 1452 | case XBRW_READ: |
795cac72 | 1453 | memcpy(data, page_address(page) + page_offset, csize); |
1da177e4 | 1454 | break; |
ce8e922c | 1455 | case XBRW_WRITE: |
795cac72 | 1456 | memcpy(page_address(page) + page_offset, data, csize); |
1da177e4 LT |
1457 | } |
1458 | ||
1459 | boff += csize; | |
1460 | data += csize; | |
1461 | } | |
1462 | } | |
1463 | ||
1464 | /* | |
ce8e922c | 1465 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1466 | */ |
1467 | ||
1468 | /* | |
430cbeb8 DC |
1469 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1470 | * returned. These buffers will have an elevated hold count, so wait on those | |
1471 | * while freeing all the buffers only held by the LRU. | |
1da177e4 | 1472 | */ |
e80dfa19 DC |
1473 | static enum lru_status |
1474 | xfs_buftarg_wait_rele( | |
1475 | struct list_head *item, | |
1476 | spinlock_t *lru_lock, | |
1477 | void *arg) | |
1478 | ||
1da177e4 | 1479 | { |
e80dfa19 | 1480 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); |
a4082357 | 1481 | struct list_head *dispose = arg; |
430cbeb8 | 1482 | |
e80dfa19 | 1483 | if (atomic_read(&bp->b_hold) > 1) { |
a4082357 | 1484 | /* need to wait, so skip it this pass */ |
e80dfa19 | 1485 | trace_xfs_buf_wait_buftarg(bp, _RET_IP_); |
a4082357 | 1486 | return LRU_SKIP; |
1da177e4 | 1487 | } |
a4082357 DC |
1488 | if (!spin_trylock(&bp->b_lock)) |
1489 | return LRU_SKIP; | |
e80dfa19 | 1490 | |
a4082357 DC |
1491 | /* |
1492 | * clear the LRU reference count so the buffer doesn't get | |
1493 | * ignored in xfs_buf_rele(). | |
1494 | */ | |
1495 | atomic_set(&bp->b_lru_ref, 0); | |
1496 | bp->b_state |= XFS_BSTATE_DISPOSE; | |
1497 | list_move(item, dispose); | |
1498 | spin_unlock(&bp->b_lock); | |
1499 | return LRU_REMOVED; | |
1da177e4 LT |
1500 | } |
1501 | ||
e80dfa19 DC |
1502 | void |
1503 | xfs_wait_buftarg( | |
1504 | struct xfs_buftarg *btp) | |
1505 | { | |
a4082357 DC |
1506 | LIST_HEAD(dispose); |
1507 | int loop = 0; | |
1508 | ||
1509 | /* loop until there is nothing left on the lru list. */ | |
1510 | while (list_lru_count(&btp->bt_lru)) { | |
e80dfa19 | 1511 | list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele, |
a4082357 DC |
1512 | &dispose, LONG_MAX); |
1513 | ||
1514 | while (!list_empty(&dispose)) { | |
1515 | struct xfs_buf *bp; | |
1516 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1517 | list_del_init(&bp->b_lru); | |
1518 | xfs_buf_rele(bp); | |
1519 | } | |
1520 | if (loop++ != 0) | |
1521 | delay(100); | |
1522 | } | |
e80dfa19 DC |
1523 | } |
1524 | ||
1525 | static enum lru_status | |
1526 | xfs_buftarg_isolate( | |
1527 | struct list_head *item, | |
1528 | spinlock_t *lru_lock, | |
1529 | void *arg) | |
1530 | { | |
1531 | struct xfs_buf *bp = container_of(item, struct xfs_buf, b_lru); | |
1532 | struct list_head *dispose = arg; | |
1533 | ||
a4082357 DC |
1534 | /* |
1535 | * we are inverting the lru lock/bp->b_lock here, so use a trylock. | |
1536 | * If we fail to get the lock, just skip it. | |
1537 | */ | |
1538 | if (!spin_trylock(&bp->b_lock)) | |
1539 | return LRU_SKIP; | |
e80dfa19 DC |
1540 | /* |
1541 | * Decrement the b_lru_ref count unless the value is already | |
1542 | * zero. If the value is already zero, we need to reclaim the | |
1543 | * buffer, otherwise it gets another trip through the LRU. | |
1544 | */ | |
a4082357 DC |
1545 | if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) { |
1546 | spin_unlock(&bp->b_lock); | |
e80dfa19 | 1547 | return LRU_ROTATE; |
a4082357 | 1548 | } |
e80dfa19 | 1549 | |
a4082357 | 1550 | bp->b_state |= XFS_BSTATE_DISPOSE; |
e80dfa19 | 1551 | list_move(item, dispose); |
a4082357 | 1552 | spin_unlock(&bp->b_lock); |
e80dfa19 DC |
1553 | return LRU_REMOVED; |
1554 | } | |
1555 | ||
addbda40 | 1556 | static unsigned long |
e80dfa19 | 1557 | xfs_buftarg_shrink_scan( |
ff57ab21 | 1558 | struct shrinker *shrink, |
1495f230 | 1559 | struct shrink_control *sc) |
a6867a68 | 1560 | { |
ff57ab21 DC |
1561 | struct xfs_buftarg *btp = container_of(shrink, |
1562 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1563 | LIST_HEAD(dispose); |
addbda40 | 1564 | unsigned long freed; |
e80dfa19 | 1565 | unsigned long nr_to_scan = sc->nr_to_scan; |
430cbeb8 | 1566 | |
e80dfa19 DC |
1567 | freed = list_lru_walk_node(&btp->bt_lru, sc->nid, xfs_buftarg_isolate, |
1568 | &dispose, &nr_to_scan); | |
430cbeb8 DC |
1569 | |
1570 | while (!list_empty(&dispose)) { | |
e80dfa19 | 1571 | struct xfs_buf *bp; |
430cbeb8 DC |
1572 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); |
1573 | list_del_init(&bp->b_lru); | |
1574 | xfs_buf_rele(bp); | |
1575 | } | |
1576 | ||
e80dfa19 DC |
1577 | return freed; |
1578 | } | |
1579 | ||
addbda40 | 1580 | static unsigned long |
e80dfa19 DC |
1581 | xfs_buftarg_shrink_count( |
1582 | struct shrinker *shrink, | |
1583 | struct shrink_control *sc) | |
1584 | { | |
1585 | struct xfs_buftarg *btp = container_of(shrink, | |
1586 | struct xfs_buftarg, bt_shrinker); | |
1587 | return list_lru_count_node(&btp->bt_lru, sc->nid); | |
a6867a68 DC |
1588 | } |
1589 | ||
1da177e4 LT |
1590 | void |
1591 | xfs_free_buftarg( | |
b7963133 CH |
1592 | struct xfs_mount *mp, |
1593 | struct xfs_buftarg *btp) | |
1da177e4 | 1594 | { |
ff57ab21 | 1595 | unregister_shrinker(&btp->bt_shrinker); |
f5e1dd34 | 1596 | list_lru_destroy(&btp->bt_lru); |
ff57ab21 | 1597 | |
b7963133 CH |
1598 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1599 | xfs_blkdev_issue_flush(btp); | |
a6867a68 | 1600 | |
f0e2d93c | 1601 | kmem_free(btp); |
1da177e4 LT |
1602 | } |
1603 | ||
1da177e4 LT |
1604 | STATIC int |
1605 | xfs_setsize_buftarg_flags( | |
1606 | xfs_buftarg_t *btp, | |
1607 | unsigned int blocksize, | |
1608 | unsigned int sectorsize, | |
1609 | int verbose) | |
1610 | { | |
ce8e922c NS |
1611 | btp->bt_bsize = blocksize; |
1612 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1613 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1614 | |
ce8e922c | 1615 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
02b102df CH |
1616 | char name[BDEVNAME_SIZE]; |
1617 | ||
1618 | bdevname(btp->bt_bdev, name); | |
1619 | ||
4f10700a DC |
1620 | xfs_warn(btp->bt_mount, |
1621 | "Cannot set_blocksize to %u on device %s\n", | |
02b102df | 1622 | sectorsize, name); |
1da177e4 LT |
1623 | return EINVAL; |
1624 | } | |
1625 | ||
1da177e4 LT |
1626 | return 0; |
1627 | } | |
1628 | ||
1629 | /* | |
ce8e922c NS |
1630 | * When allocating the initial buffer target we have not yet |
1631 | * read in the superblock, so don't know what sized sectors | |
8b4ad79c | 1632 | * are being used at this early stage. Play safe. |
ce8e922c | 1633 | */ |
1da177e4 LT |
1634 | STATIC int |
1635 | xfs_setsize_buftarg_early( | |
1636 | xfs_buftarg_t *btp, | |
1637 | struct block_device *bdev) | |
1638 | { | |
1639 | return xfs_setsize_buftarg_flags(btp, | |
0e6e847f | 1640 | PAGE_SIZE, bdev_logical_block_size(bdev), 0); |
1da177e4 LT |
1641 | } |
1642 | ||
1643 | int | |
1644 | xfs_setsize_buftarg( | |
1645 | xfs_buftarg_t *btp, | |
1646 | unsigned int blocksize, | |
1647 | unsigned int sectorsize) | |
1648 | { | |
1649 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1650 | } | |
1651 | ||
1da177e4 LT |
1652 | xfs_buftarg_t * |
1653 | xfs_alloc_buftarg( | |
ebad861b | 1654 | struct xfs_mount *mp, |
1da177e4 | 1655 | struct block_device *bdev, |
e2a07812 JE |
1656 | int external, |
1657 | const char *fsname) | |
1da177e4 LT |
1658 | { |
1659 | xfs_buftarg_t *btp; | |
1660 | ||
b17cb364 | 1661 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP | KM_NOFS); |
1da177e4 | 1662 | |
ebad861b | 1663 | btp->bt_mount = mp; |
ce8e922c NS |
1664 | btp->bt_dev = bdev->bd_dev; |
1665 | btp->bt_bdev = bdev; | |
0e6e847f DC |
1666 | btp->bt_bdi = blk_get_backing_dev_info(bdev); |
1667 | if (!btp->bt_bdi) | |
1668 | goto error; | |
1669 | ||
1da177e4 LT |
1670 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1671 | goto error; | |
5ca302c8 GC |
1672 | |
1673 | if (list_lru_init(&btp->bt_lru)) | |
1674 | goto error; | |
1675 | ||
e80dfa19 DC |
1676 | btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count; |
1677 | btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan; | |
ff57ab21 | 1678 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; |
e80dfa19 | 1679 | btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE; |
ff57ab21 | 1680 | register_shrinker(&btp->bt_shrinker); |
1da177e4 LT |
1681 | return btp; |
1682 | ||
1683 | error: | |
f0e2d93c | 1684 | kmem_free(btp); |
1da177e4 LT |
1685 | return NULL; |
1686 | } | |
1687 | ||
1da177e4 | 1688 | /* |
43ff2122 CH |
1689 | * Add a buffer to the delayed write list. |
1690 | * | |
1691 | * This queues a buffer for writeout if it hasn't already been. Note that | |
1692 | * neither this routine nor the buffer list submission functions perform | |
1693 | * any internal synchronization. It is expected that the lists are thread-local | |
1694 | * to the callers. | |
1695 | * | |
1696 | * Returns true if we queued up the buffer, or false if it already had | |
1697 | * been on the buffer list. | |
1da177e4 | 1698 | */ |
43ff2122 | 1699 | bool |
ce8e922c | 1700 | xfs_buf_delwri_queue( |
43ff2122 CH |
1701 | struct xfs_buf *bp, |
1702 | struct list_head *list) | |
1da177e4 | 1703 | { |
43ff2122 | 1704 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 1705 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1706 | |
43ff2122 CH |
1707 | /* |
1708 | * If the buffer is already marked delwri it already is queued up | |
1709 | * by someone else for imediate writeout. Just ignore it in that | |
1710 | * case. | |
1711 | */ | |
1712 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
1713 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
1714 | return false; | |
1da177e4 | 1715 | } |
1da177e4 | 1716 | |
43ff2122 | 1717 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
1718 | |
1719 | /* | |
43ff2122 CH |
1720 | * If a buffer gets written out synchronously or marked stale while it |
1721 | * is on a delwri list we lazily remove it. To do this, the other party | |
1722 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
1723 | * It remains referenced and on the list. In a rare corner case it | |
1724 | * might get readded to a delwri list after the synchronous writeout, in | |
1725 | * which case we need just need to re-add the flag here. | |
d808f617 | 1726 | */ |
43ff2122 CH |
1727 | bp->b_flags |= _XBF_DELWRI_Q; |
1728 | if (list_empty(&bp->b_list)) { | |
1729 | atomic_inc(&bp->b_hold); | |
1730 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 1731 | } |
585e6d88 | 1732 | |
43ff2122 | 1733 | return true; |
585e6d88 DC |
1734 | } |
1735 | ||
089716aa DC |
1736 | /* |
1737 | * Compare function is more complex than it needs to be because | |
1738 | * the return value is only 32 bits and we are doing comparisons | |
1739 | * on 64 bit values | |
1740 | */ | |
1741 | static int | |
1742 | xfs_buf_cmp( | |
1743 | void *priv, | |
1744 | struct list_head *a, | |
1745 | struct list_head *b) | |
1746 | { | |
1747 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1748 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1749 | xfs_daddr_t diff; | |
1750 | ||
f4b42421 | 1751 | diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn; |
089716aa DC |
1752 | if (diff < 0) |
1753 | return -1; | |
1754 | if (diff > 0) | |
1755 | return 1; | |
1756 | return 0; | |
1757 | } | |
1758 | ||
43ff2122 CH |
1759 | static int |
1760 | __xfs_buf_delwri_submit( | |
1761 | struct list_head *buffer_list, | |
1762 | struct list_head *io_list, | |
1763 | bool wait) | |
1da177e4 | 1764 | { |
43ff2122 CH |
1765 | struct blk_plug plug; |
1766 | struct xfs_buf *bp, *n; | |
1767 | int pinned = 0; | |
1768 | ||
1769 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { | |
1770 | if (!wait) { | |
1771 | if (xfs_buf_ispinned(bp)) { | |
1772 | pinned++; | |
1773 | continue; | |
1774 | } | |
1775 | if (!xfs_buf_trylock(bp)) | |
1776 | continue; | |
1777 | } else { | |
1778 | xfs_buf_lock(bp); | |
1779 | } | |
978c7b2f | 1780 | |
43ff2122 CH |
1781 | /* |
1782 | * Someone else might have written the buffer synchronously or | |
1783 | * marked it stale in the meantime. In that case only the | |
1784 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
1785 | * reference and remove it from the list here. | |
1786 | */ | |
1787 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
1788 | list_del_init(&bp->b_list); | |
1789 | xfs_buf_relse(bp); | |
1790 | continue; | |
1791 | } | |
c9c12971 | 1792 | |
43ff2122 CH |
1793 | list_move_tail(&bp->b_list, io_list); |
1794 | trace_xfs_buf_delwri_split(bp, _RET_IP_); | |
1795 | } | |
1da177e4 | 1796 | |
43ff2122 | 1797 | list_sort(NULL, io_list, xfs_buf_cmp); |
1da177e4 | 1798 | |
43ff2122 CH |
1799 | blk_start_plug(&plug); |
1800 | list_for_each_entry_safe(bp, n, io_list, b_list) { | |
1801 | bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_ASYNC); | |
1802 | bp->b_flags |= XBF_WRITE; | |
a1b7ea5d | 1803 | |
43ff2122 CH |
1804 | if (!wait) { |
1805 | bp->b_flags |= XBF_ASYNC; | |
ce8e922c | 1806 | list_del_init(&bp->b_list); |
1da177e4 | 1807 | } |
43ff2122 CH |
1808 | xfs_bdstrat_cb(bp); |
1809 | } | |
1810 | blk_finish_plug(&plug); | |
1da177e4 | 1811 | |
43ff2122 | 1812 | return pinned; |
1da177e4 LT |
1813 | } |
1814 | ||
1815 | /* | |
43ff2122 CH |
1816 | * Write out a buffer list asynchronously. |
1817 | * | |
1818 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
1819 | * out and not wait for I/O completion on any of the buffers. This interface | |
1820 | * is only safely useable for callers that can track I/O completion by higher | |
1821 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
1822 | * function. | |
1da177e4 LT |
1823 | */ |
1824 | int | |
43ff2122 CH |
1825 | xfs_buf_delwri_submit_nowait( |
1826 | struct list_head *buffer_list) | |
1da177e4 | 1827 | { |
43ff2122 CH |
1828 | LIST_HEAD (io_list); |
1829 | return __xfs_buf_delwri_submit(buffer_list, &io_list, false); | |
1830 | } | |
1da177e4 | 1831 | |
43ff2122 CH |
1832 | /* |
1833 | * Write out a buffer list synchronously. | |
1834 | * | |
1835 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
1836 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
1837 | * so callers must have some other way of tracking buffers if they require such | |
1838 | * functionality. | |
1839 | */ | |
1840 | int | |
1841 | xfs_buf_delwri_submit( | |
1842 | struct list_head *buffer_list) | |
1843 | { | |
1844 | LIST_HEAD (io_list); | |
1845 | int error = 0, error2; | |
1846 | struct xfs_buf *bp; | |
1da177e4 | 1847 | |
43ff2122 | 1848 | __xfs_buf_delwri_submit(buffer_list, &io_list, true); |
1da177e4 | 1849 | |
43ff2122 CH |
1850 | /* Wait for IO to complete. */ |
1851 | while (!list_empty(&io_list)) { | |
1852 | bp = list_first_entry(&io_list, struct xfs_buf, b_list); | |
a1b7ea5d | 1853 | |
089716aa | 1854 | list_del_init(&bp->b_list); |
43ff2122 CH |
1855 | error2 = xfs_buf_iowait(bp); |
1856 | xfs_buf_relse(bp); | |
1857 | if (!error) | |
1858 | error = error2; | |
1da177e4 LT |
1859 | } |
1860 | ||
43ff2122 | 1861 | return error; |
1da177e4 LT |
1862 | } |
1863 | ||
04d8b284 | 1864 | int __init |
ce8e922c | 1865 | xfs_buf_init(void) |
1da177e4 | 1866 | { |
8758280f NS |
1867 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1868 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1869 | if (!xfs_buf_zone) |
0b1b213f | 1870 | goto out; |
04d8b284 | 1871 | |
51749e47 | 1872 | xfslogd_workqueue = alloc_workqueue("xfslogd", |
6370a6ad | 1873 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 1); |
23ea4032 | 1874 | if (!xfslogd_workqueue) |
04d8b284 | 1875 | goto out_free_buf_zone; |
1da177e4 | 1876 | |
23ea4032 | 1877 | return 0; |
1da177e4 | 1878 | |
23ea4032 | 1879 | out_free_buf_zone: |
ce8e922c | 1880 | kmem_zone_destroy(xfs_buf_zone); |
0b1b213f | 1881 | out: |
8758280f | 1882 | return -ENOMEM; |
1da177e4 LT |
1883 | } |
1884 | ||
1da177e4 | 1885 | void |
ce8e922c | 1886 | xfs_buf_terminate(void) |
1da177e4 | 1887 | { |
04d8b284 | 1888 | destroy_workqueue(xfslogd_workqueue); |
ce8e922c | 1889 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 1890 | } |