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 CH |
37 | #include "xfs_sb.h" |
38 | #include "xfs_inum.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 NS |
58 | #define xb_to_gfp(flags) \ |
59 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \ | |
60 | ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN) | |
1da177e4 | 61 | |
ce8e922c NS |
62 | #define xb_to_km(flags) \ |
63 | (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP) | |
1da177e4 | 64 | |
1da177e4 | 65 | |
73c77e2c JB |
66 | static inline int |
67 | xfs_buf_is_vmapped( | |
68 | struct xfs_buf *bp) | |
69 | { | |
70 | /* | |
71 | * Return true if the buffer is vmapped. | |
72 | * | |
73 | * The XBF_MAPPED flag is set if the buffer should be mapped, but the | |
74 | * code is clever enough to know it doesn't have to map a single page, | |
75 | * so the check has to be both for XBF_MAPPED and bp->b_page_count > 1. | |
76 | */ | |
77 | return (bp->b_flags & XBF_MAPPED) && bp->b_page_count > 1; | |
78 | } | |
79 | ||
80 | static inline int | |
81 | xfs_buf_vmap_len( | |
82 | struct xfs_buf *bp) | |
83 | { | |
84 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
85 | } | |
86 | ||
1da177e4 | 87 | /* |
430cbeb8 DC |
88 | * xfs_buf_lru_add - add a buffer to the LRU. |
89 | * | |
90 | * The LRU takes a new reference to the buffer so that it will only be freed | |
91 | * once the shrinker takes the buffer off the LRU. | |
92 | */ | |
93 | STATIC void | |
94 | xfs_buf_lru_add( | |
95 | struct xfs_buf *bp) | |
96 | { | |
97 | struct xfs_buftarg *btp = bp->b_target; | |
98 | ||
99 | spin_lock(&btp->bt_lru_lock); | |
100 | if (list_empty(&bp->b_lru)) { | |
101 | atomic_inc(&bp->b_hold); | |
102 | list_add_tail(&bp->b_lru, &btp->bt_lru); | |
103 | btp->bt_lru_nr++; | |
104 | } | |
105 | spin_unlock(&btp->bt_lru_lock); | |
106 | } | |
107 | ||
108 | /* | |
109 | * xfs_buf_lru_del - remove a buffer from the LRU | |
110 | * | |
111 | * The unlocked check is safe here because it only occurs when there are not | |
112 | * b_lru_ref counts left on the inode under the pag->pag_buf_lock. it is there | |
113 | * to optimise the shrinker removing the buffer from the LRU and calling | |
25985edc | 114 | * xfs_buf_free(). i.e. it removes an unnecessary round trip on the |
430cbeb8 | 115 | * bt_lru_lock. |
1da177e4 | 116 | */ |
430cbeb8 DC |
117 | STATIC void |
118 | xfs_buf_lru_del( | |
119 | struct xfs_buf *bp) | |
120 | { | |
121 | struct xfs_buftarg *btp = bp->b_target; | |
122 | ||
123 | if (list_empty(&bp->b_lru)) | |
124 | return; | |
125 | ||
126 | spin_lock(&btp->bt_lru_lock); | |
127 | if (!list_empty(&bp->b_lru)) { | |
128 | list_del_init(&bp->b_lru); | |
129 | btp->bt_lru_nr--; | |
130 | } | |
131 | spin_unlock(&btp->bt_lru_lock); | |
132 | } | |
133 | ||
134 | /* | |
135 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
136 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
137 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
138 | * to remove the reference that LRU holds on the buffer. | |
139 | * | |
140 | * This prevents build-up of stale buffers on the LRU. | |
141 | */ | |
142 | void | |
143 | xfs_buf_stale( | |
144 | struct xfs_buf *bp) | |
145 | { | |
43ff2122 CH |
146 | ASSERT(xfs_buf_islocked(bp)); |
147 | ||
430cbeb8 | 148 | bp->b_flags |= XBF_STALE; |
43ff2122 CH |
149 | |
150 | /* | |
151 | * Clear the delwri status so that a delwri queue walker will not | |
152 | * flush this buffer to disk now that it is stale. The delwri queue has | |
153 | * a reference to the buffer, so this is safe to do. | |
154 | */ | |
155 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
156 | ||
430cbeb8 DC |
157 | atomic_set(&(bp)->b_lru_ref, 0); |
158 | if (!list_empty(&bp->b_lru)) { | |
159 | struct xfs_buftarg *btp = bp->b_target; | |
160 | ||
161 | spin_lock(&btp->bt_lru_lock); | |
162 | if (!list_empty(&bp->b_lru)) { | |
163 | list_del_init(&bp->b_lru); | |
164 | btp->bt_lru_nr--; | |
165 | atomic_dec(&bp->b_hold); | |
166 | } | |
167 | spin_unlock(&btp->bt_lru_lock); | |
168 | } | |
169 | ASSERT(atomic_read(&bp->b_hold) >= 1); | |
170 | } | |
1da177e4 | 171 | |
4347b9d7 CH |
172 | struct xfs_buf * |
173 | xfs_buf_alloc( | |
174 | struct xfs_buftarg *target, | |
e70b73f8 DC |
175 | xfs_daddr_t blkno, |
176 | size_t numblks, | |
ce8e922c | 177 | xfs_buf_flags_t flags) |
1da177e4 | 178 | { |
4347b9d7 CH |
179 | struct xfs_buf *bp; |
180 | ||
bf813cdd | 181 | bp = kmem_zone_zalloc(xfs_buf_zone, xb_to_km(flags)); |
4347b9d7 CH |
182 | if (unlikely(!bp)) |
183 | return NULL; | |
184 | ||
1da177e4 | 185 | /* |
ce8e922c | 186 | * We don't want certain flags to appear in b_flags. |
1da177e4 | 187 | */ |
ce8e922c NS |
188 | flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD); |
189 | ||
ce8e922c | 190 | atomic_set(&bp->b_hold, 1); |
430cbeb8 | 191 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 192 | init_completion(&bp->b_iowait); |
430cbeb8 | 193 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 194 | INIT_LIST_HEAD(&bp->b_list); |
74f75a0c | 195 | RB_CLEAR_NODE(&bp->b_rbnode); |
a731cd11 | 196 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
ce8e922c NS |
197 | XB_SET_OWNER(bp); |
198 | bp->b_target = target; | |
de1cbee4 | 199 | |
1da177e4 | 200 | /* |
4e94b71b | 201 | * Set length and count_desired to the same value initially. |
1da177e4 LT |
202 | * I/O routines should use count_desired, which will be the same in |
203 | * most cases but may be reset (e.g. XFS recovery). | |
204 | */ | |
4e94b71b DC |
205 | bp->b_length = numblks; |
206 | bp->b_count_desired = numblks << BBSHIFT; | |
ce8e922c | 207 | bp->b_flags = flags; |
e70b73f8 DC |
208 | |
209 | /* | |
210 | * We do not set the block number here in the buffer because we have not | |
211 | * finished initialising the buffer. We insert the buffer into the cache | |
212 | * in this state, so this ensures that we are unable to do IO on a | |
213 | * buffer that hasn't been fully initialised. | |
214 | */ | |
ce8e922c NS |
215 | bp->b_bn = XFS_BUF_DADDR_NULL; |
216 | atomic_set(&bp->b_pin_count, 0); | |
217 | init_waitqueue_head(&bp->b_waiters); | |
218 | ||
219 | XFS_STATS_INC(xb_create); | |
0b1b213f | 220 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 CH |
221 | |
222 | return bp; | |
1da177e4 LT |
223 | } |
224 | ||
225 | /* | |
ce8e922c NS |
226 | * Allocate a page array capable of holding a specified number |
227 | * of pages, and point the page buf at it. | |
1da177e4 LT |
228 | */ |
229 | STATIC int | |
ce8e922c NS |
230 | _xfs_buf_get_pages( |
231 | xfs_buf_t *bp, | |
1da177e4 | 232 | int page_count, |
ce8e922c | 233 | xfs_buf_flags_t flags) |
1da177e4 LT |
234 | { |
235 | /* Make sure that we have a page list */ | |
ce8e922c | 236 | if (bp->b_pages == NULL) { |
ce8e922c NS |
237 | bp->b_page_count = page_count; |
238 | if (page_count <= XB_PAGES) { | |
239 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 240 | } else { |
ce8e922c NS |
241 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
242 | page_count, xb_to_km(flags)); | |
243 | if (bp->b_pages == NULL) | |
1da177e4 LT |
244 | return -ENOMEM; |
245 | } | |
ce8e922c | 246 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
247 | } |
248 | return 0; | |
249 | } | |
250 | ||
251 | /* | |
ce8e922c | 252 | * Frees b_pages if it was allocated. |
1da177e4 LT |
253 | */ |
254 | STATIC void | |
ce8e922c | 255 | _xfs_buf_free_pages( |
1da177e4 LT |
256 | xfs_buf_t *bp) |
257 | { | |
ce8e922c | 258 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 259 | kmem_free(bp->b_pages); |
3fc98b1a | 260 | bp->b_pages = NULL; |
1da177e4 LT |
261 | } |
262 | } | |
263 | ||
264 | /* | |
265 | * Releases the specified buffer. | |
266 | * | |
267 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 268 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
269 | * hashed and refcounted buffers |
270 | */ | |
271 | void | |
ce8e922c | 272 | xfs_buf_free( |
1da177e4 LT |
273 | xfs_buf_t *bp) |
274 | { | |
0b1b213f | 275 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 276 | |
430cbeb8 DC |
277 | ASSERT(list_empty(&bp->b_lru)); |
278 | ||
0e6e847f | 279 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
280 | uint i; |
281 | ||
73c77e2c | 282 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
283 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
284 | bp->b_page_count); | |
1da177e4 | 285 | |
948ecdb4 NS |
286 | for (i = 0; i < bp->b_page_count; i++) { |
287 | struct page *page = bp->b_pages[i]; | |
288 | ||
0e6e847f | 289 | __free_page(page); |
948ecdb4 | 290 | } |
0e6e847f DC |
291 | } else if (bp->b_flags & _XBF_KMEM) |
292 | kmem_free(bp->b_addr); | |
3fc98b1a | 293 | _xfs_buf_free_pages(bp); |
4347b9d7 | 294 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
295 | } |
296 | ||
297 | /* | |
0e6e847f | 298 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
299 | */ |
300 | STATIC int | |
0e6e847f | 301 | xfs_buf_allocate_memory( |
1da177e4 LT |
302 | xfs_buf_t *bp, |
303 | uint flags) | |
304 | { | |
ce8e922c | 305 | size_t size = bp->b_count_desired; |
1da177e4 | 306 | size_t nbytes, offset; |
ce8e922c | 307 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 308 | unsigned short page_count, i; |
204ab25f | 309 | xfs_off_t end; |
1da177e4 LT |
310 | int error; |
311 | ||
0e6e847f DC |
312 | /* |
313 | * for buffers that are contained within a single page, just allocate | |
314 | * the memory from the heap - there's no need for the complexity of | |
315 | * page arrays to keep allocation down to order 0. | |
316 | */ | |
4e94b71b DC |
317 | if (bp->b_length < BTOBB(PAGE_SIZE)) { |
318 | bp->b_addr = kmem_alloc(BBTOB(bp->b_length), xb_to_km(flags)); | |
0e6e847f DC |
319 | if (!bp->b_addr) { |
320 | /* low memory - use alloc_page loop instead */ | |
321 | goto use_alloc_page; | |
322 | } | |
323 | ||
4e94b71b | 324 | if (((unsigned long)(bp->b_addr + BBTOB(bp->b_length) - 1) & |
0e6e847f DC |
325 | PAGE_MASK) != |
326 | ((unsigned long)bp->b_addr & PAGE_MASK)) { | |
327 | /* b_addr spans two pages - use alloc_page instead */ | |
328 | kmem_free(bp->b_addr); | |
329 | bp->b_addr = NULL; | |
330 | goto use_alloc_page; | |
331 | } | |
332 | bp->b_offset = offset_in_page(bp->b_addr); | |
333 | bp->b_pages = bp->b_page_array; | |
334 | bp->b_pages[0] = virt_to_page(bp->b_addr); | |
335 | bp->b_page_count = 1; | |
336 | bp->b_flags |= XBF_MAPPED | _XBF_KMEM; | |
337 | return 0; | |
338 | } | |
339 | ||
340 | use_alloc_page: | |
4e94b71b | 341 | end = BBTOB(bp->b_bn + bp->b_length); |
de1cbee4 | 342 | page_count = xfs_buf_btoc(end) - xfs_buf_btoct(BBTOB(bp->b_bn)); |
ce8e922c | 343 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
344 | if (unlikely(error)) |
345 | return error; | |
1da177e4 | 346 | |
ce8e922c | 347 | offset = bp->b_offset; |
0e6e847f | 348 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 349 | |
ce8e922c | 350 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
351 | struct page *page; |
352 | uint retries = 0; | |
0e6e847f DC |
353 | retry: |
354 | page = alloc_page(gfp_mask); | |
1da177e4 | 355 | if (unlikely(page == NULL)) { |
ce8e922c NS |
356 | if (flags & XBF_READ_AHEAD) { |
357 | bp->b_page_count = i; | |
0e6e847f DC |
358 | error = ENOMEM; |
359 | goto out_free_pages; | |
1da177e4 LT |
360 | } |
361 | ||
362 | /* | |
363 | * This could deadlock. | |
364 | * | |
365 | * But until all the XFS lowlevel code is revamped to | |
366 | * handle buffer allocation failures we can't do much. | |
367 | */ | |
368 | if (!(++retries % 100)) | |
4f10700a DC |
369 | xfs_err(NULL, |
370 | "possible memory allocation deadlock in %s (mode:0x%x)", | |
34a622b2 | 371 | __func__, gfp_mask); |
1da177e4 | 372 | |
ce8e922c | 373 | XFS_STATS_INC(xb_page_retries); |
8aa7e847 | 374 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
375 | goto retry; |
376 | } | |
377 | ||
ce8e922c | 378 | XFS_STATS_INC(xb_page_found); |
1da177e4 | 379 | |
0e6e847f | 380 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 381 | size -= nbytes; |
ce8e922c | 382 | bp->b_pages[i] = page; |
1da177e4 LT |
383 | offset = 0; |
384 | } | |
0e6e847f | 385 | return 0; |
1da177e4 | 386 | |
0e6e847f DC |
387 | out_free_pages: |
388 | for (i = 0; i < bp->b_page_count; i++) | |
389 | __free_page(bp->b_pages[i]); | |
1da177e4 LT |
390 | return error; |
391 | } | |
392 | ||
393 | /* | |
25985edc | 394 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
395 | */ |
396 | STATIC int | |
ce8e922c | 397 | _xfs_buf_map_pages( |
1da177e4 LT |
398 | xfs_buf_t *bp, |
399 | uint flags) | |
400 | { | |
0e6e847f | 401 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 402 | if (bp->b_page_count == 1) { |
0e6e847f | 403 | /* A single page buffer is always mappable */ |
ce8e922c NS |
404 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
405 | bp->b_flags |= XBF_MAPPED; | |
406 | } else if (flags & XBF_MAPPED) { | |
a19fb380 DC |
407 | int retried = 0; |
408 | ||
409 | do { | |
410 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
411 | -1, PAGE_KERNEL); | |
412 | if (bp->b_addr) | |
413 | break; | |
414 | vm_unmap_aliases(); | |
415 | } while (retried++ <= 1); | |
416 | ||
417 | if (!bp->b_addr) | |
1da177e4 | 418 | return -ENOMEM; |
ce8e922c NS |
419 | bp->b_addr += bp->b_offset; |
420 | bp->b_flags |= XBF_MAPPED; | |
1da177e4 LT |
421 | } |
422 | ||
423 | return 0; | |
424 | } | |
425 | ||
426 | /* | |
427 | * Finding and Reading Buffers | |
428 | */ | |
429 | ||
430 | /* | |
ce8e922c | 431 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 | 432 | * a given range of an inode. The buffer is returned |
eabbaf11 | 433 | * locked. No I/O is implied by this call. |
1da177e4 LT |
434 | */ |
435 | xfs_buf_t * | |
ce8e922c | 436 | _xfs_buf_find( |
e70b73f8 DC |
437 | struct xfs_buftarg *btp, |
438 | xfs_daddr_t blkno, | |
439 | size_t numblks, | |
ce8e922c NS |
440 | xfs_buf_flags_t flags, |
441 | xfs_buf_t *new_bp) | |
1da177e4 | 442 | { |
e70b73f8 | 443 | size_t numbytes; |
74f75a0c DC |
444 | struct xfs_perag *pag; |
445 | struct rb_node **rbp; | |
446 | struct rb_node *parent; | |
447 | xfs_buf_t *bp; | |
1da177e4 | 448 | |
e70b73f8 | 449 | numbytes = BBTOB(numblks); |
1da177e4 LT |
450 | |
451 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
e70b73f8 | 452 | ASSERT(!(numbytes < (1 << btp->bt_sshift))); |
de1cbee4 | 453 | ASSERT(!(BBTOB(blkno) & (xfs_off_t)btp->bt_smask)); |
1da177e4 | 454 | |
74f75a0c DC |
455 | /* get tree root */ |
456 | pag = xfs_perag_get(btp->bt_mount, | |
e70b73f8 | 457 | xfs_daddr_to_agno(btp->bt_mount, blkno)); |
74f75a0c DC |
458 | |
459 | /* walk tree */ | |
460 | spin_lock(&pag->pag_buf_lock); | |
461 | rbp = &pag->pag_buf_tree.rb_node; | |
462 | parent = NULL; | |
463 | bp = NULL; | |
464 | while (*rbp) { | |
465 | parent = *rbp; | |
466 | bp = rb_entry(parent, struct xfs_buf, b_rbnode); | |
467 | ||
de1cbee4 | 468 | if (blkno < bp->b_bn) |
74f75a0c | 469 | rbp = &(*rbp)->rb_left; |
de1cbee4 | 470 | else if (blkno > bp->b_bn) |
74f75a0c DC |
471 | rbp = &(*rbp)->rb_right; |
472 | else { | |
473 | /* | |
de1cbee4 | 474 | * found a block number match. If the range doesn't |
74f75a0c DC |
475 | * match, the only way this is allowed is if the buffer |
476 | * in the cache is stale and the transaction that made | |
477 | * it stale has not yet committed. i.e. we are | |
478 | * reallocating a busy extent. Skip this buffer and | |
479 | * continue searching to the right for an exact match. | |
480 | */ | |
4e94b71b | 481 | if (bp->b_length != numblks) { |
74f75a0c DC |
482 | ASSERT(bp->b_flags & XBF_STALE); |
483 | rbp = &(*rbp)->rb_right; | |
484 | continue; | |
485 | } | |
ce8e922c | 486 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
487 | goto found; |
488 | } | |
489 | } | |
490 | ||
491 | /* No match found */ | |
ce8e922c | 492 | if (new_bp) { |
74f75a0c DC |
493 | rb_link_node(&new_bp->b_rbnode, parent, rbp); |
494 | rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree); | |
495 | /* the buffer keeps the perag reference until it is freed */ | |
496 | new_bp->b_pag = pag; | |
497 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 498 | } else { |
ce8e922c | 499 | XFS_STATS_INC(xb_miss_locked); |
74f75a0c DC |
500 | spin_unlock(&pag->pag_buf_lock); |
501 | xfs_perag_put(pag); | |
1da177e4 | 502 | } |
ce8e922c | 503 | return new_bp; |
1da177e4 LT |
504 | |
505 | found: | |
74f75a0c DC |
506 | spin_unlock(&pag->pag_buf_lock); |
507 | xfs_perag_put(pag); | |
1da177e4 | 508 | |
0c842ad4 CH |
509 | if (!xfs_buf_trylock(bp)) { |
510 | if (flags & XBF_TRYLOCK) { | |
ce8e922c NS |
511 | xfs_buf_rele(bp); |
512 | XFS_STATS_INC(xb_busy_locked); | |
513 | return NULL; | |
1da177e4 | 514 | } |
0c842ad4 CH |
515 | xfs_buf_lock(bp); |
516 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
517 | } |
518 | ||
0e6e847f DC |
519 | /* |
520 | * if the buffer is stale, clear all the external state associated with | |
521 | * it. We need to keep flags such as how we allocated the buffer memory | |
522 | * intact here. | |
523 | */ | |
ce8e922c NS |
524 | if (bp->b_flags & XBF_STALE) { |
525 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
0e6e847f | 526 | bp->b_flags &= XBF_MAPPED | _XBF_KMEM | _XBF_PAGES; |
2f926587 | 527 | } |
0b1b213f CH |
528 | |
529 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ce8e922c NS |
530 | XFS_STATS_INC(xb_get_locked); |
531 | return bp; | |
1da177e4 LT |
532 | } |
533 | ||
534 | /* | |
3815832a DC |
535 | * Assembles a buffer covering the specified range. The code is optimised for |
536 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
537 | * more hits than misses. | |
1da177e4 | 538 | */ |
3815832a | 539 | struct xfs_buf * |
6ad112bf | 540 | xfs_buf_get( |
e70b73f8 DC |
541 | xfs_buftarg_t *target, |
542 | xfs_daddr_t blkno, | |
543 | size_t numblks, | |
ce8e922c | 544 | xfs_buf_flags_t flags) |
1da177e4 | 545 | { |
3815832a DC |
546 | struct xfs_buf *bp; |
547 | struct xfs_buf *new_bp; | |
0e6e847f | 548 | int error = 0; |
1da177e4 | 549 | |
e70b73f8 | 550 | bp = _xfs_buf_find(target, blkno, numblks, flags, NULL); |
3815832a DC |
551 | if (likely(bp)) |
552 | goto found; | |
553 | ||
e70b73f8 | 554 | new_bp = xfs_buf_alloc(target, blkno, numblks, flags); |
ce8e922c | 555 | if (unlikely(!new_bp)) |
1da177e4 LT |
556 | return NULL; |
557 | ||
fe2429b0 DC |
558 | error = xfs_buf_allocate_memory(new_bp, flags); |
559 | if (error) { | |
560 | kmem_zone_free(xfs_buf_zone, new_bp); | |
561 | return NULL; | |
562 | } | |
563 | ||
e70b73f8 | 564 | bp = _xfs_buf_find(target, blkno, numblks, flags, new_bp); |
3815832a | 565 | if (!bp) { |
fe2429b0 | 566 | xfs_buf_free(new_bp); |
3815832a DC |
567 | return NULL; |
568 | } | |
569 | ||
fe2429b0 DC |
570 | if (bp != new_bp) |
571 | xfs_buf_free(new_bp); | |
1da177e4 | 572 | |
3815832a DC |
573 | /* |
574 | * Now we have a workable buffer, fill in the block number so | |
575 | * that we can do IO on it. | |
576 | */ | |
e70b73f8 | 577 | bp->b_bn = blkno; |
4e94b71b | 578 | bp->b_count_desired = BBTOB(bp->b_length); |
3815832a DC |
579 | |
580 | found: | |
ce8e922c NS |
581 | if (!(bp->b_flags & XBF_MAPPED)) { |
582 | error = _xfs_buf_map_pages(bp, flags); | |
1da177e4 | 583 | if (unlikely(error)) { |
4f10700a DC |
584 | xfs_warn(target->bt_mount, |
585 | "%s: failed to map pages\n", __func__); | |
1da177e4 LT |
586 | goto no_buffer; |
587 | } | |
588 | } | |
589 | ||
ce8e922c | 590 | XFS_STATS_INC(xb_get); |
0b1b213f | 591 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 592 | return bp; |
1da177e4 | 593 | |
3815832a | 594 | no_buffer: |
ce8e922c NS |
595 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
596 | xfs_buf_unlock(bp); | |
597 | xfs_buf_rele(bp); | |
1da177e4 LT |
598 | return NULL; |
599 | } | |
600 | ||
5d765b97 CH |
601 | STATIC int |
602 | _xfs_buf_read( | |
603 | xfs_buf_t *bp, | |
604 | xfs_buf_flags_t flags) | |
605 | { | |
43ff2122 | 606 | ASSERT(!(flags & XBF_WRITE)); |
5d765b97 CH |
607 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); |
608 | ||
43ff2122 | 609 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD); |
1d5ae5df | 610 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 611 | |
0e95f19a DC |
612 | xfs_buf_iorequest(bp); |
613 | if (flags & XBF_ASYNC) | |
614 | return 0; | |
ec53d1db | 615 | return xfs_buf_iowait(bp); |
5d765b97 CH |
616 | } |
617 | ||
1da177e4 | 618 | xfs_buf_t * |
6ad112bf | 619 | xfs_buf_read( |
1da177e4 | 620 | xfs_buftarg_t *target, |
e70b73f8 DC |
621 | xfs_daddr_t blkno, |
622 | size_t numblks, | |
ce8e922c | 623 | xfs_buf_flags_t flags) |
1da177e4 | 624 | { |
ce8e922c NS |
625 | xfs_buf_t *bp; |
626 | ||
627 | flags |= XBF_READ; | |
628 | ||
e70b73f8 | 629 | bp = xfs_buf_get(target, blkno, numblks, flags); |
ce8e922c | 630 | if (bp) { |
0b1b213f CH |
631 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
632 | ||
ce8e922c | 633 | if (!XFS_BUF_ISDONE(bp)) { |
ce8e922c | 634 | XFS_STATS_INC(xb_get_read); |
5d765b97 | 635 | _xfs_buf_read(bp, flags); |
ce8e922c | 636 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
637 | /* |
638 | * Read ahead call which is already satisfied, | |
639 | * drop the buffer | |
640 | */ | |
641 | goto no_buffer; | |
642 | } else { | |
1da177e4 | 643 | /* We do not want read in the flags */ |
ce8e922c | 644 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
645 | } |
646 | } | |
647 | ||
ce8e922c | 648 | return bp; |
1da177e4 LT |
649 | |
650 | no_buffer: | |
ce8e922c NS |
651 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
652 | xfs_buf_unlock(bp); | |
653 | xfs_buf_rele(bp); | |
1da177e4 LT |
654 | return NULL; |
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 | |
ce8e922c | 662 | xfs_buf_readahead( |
1da177e4 | 663 | xfs_buftarg_t *target, |
e70b73f8 DC |
664 | xfs_daddr_t blkno, |
665 | size_t numblks) | |
1da177e4 | 666 | { |
0e6e847f | 667 | if (bdi_read_congested(target->bt_bdi)) |
1da177e4 LT |
668 | return; |
669 | ||
e70b73f8 | 670 | xfs_buf_read(target, blkno, numblks, |
1a1a3e97 | 671 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD|XBF_DONT_BLOCK); |
1da177e4 LT |
672 | } |
673 | ||
5adc94c2 DC |
674 | /* |
675 | * Read an uncached buffer from disk. Allocates and returns a locked | |
676 | * buffer containing the disk contents or nothing. | |
677 | */ | |
678 | struct xfs_buf * | |
679 | xfs_buf_read_uncached( | |
5adc94c2 DC |
680 | struct xfs_buftarg *target, |
681 | xfs_daddr_t daddr, | |
e70b73f8 | 682 | size_t numblks, |
5adc94c2 DC |
683 | int flags) |
684 | { | |
685 | xfs_buf_t *bp; | |
686 | int error; | |
687 | ||
e70b73f8 | 688 | bp = xfs_buf_get_uncached(target, numblks, flags); |
5adc94c2 DC |
689 | if (!bp) |
690 | return NULL; | |
691 | ||
692 | /* set up the buffer for a read IO */ | |
5adc94c2 DC |
693 | XFS_BUF_SET_ADDR(bp, daddr); |
694 | XFS_BUF_READ(bp); | |
5adc94c2 | 695 | |
e70b73f8 | 696 | xfsbdstrat(target->bt_mount, bp); |
1a1a3e97 | 697 | error = xfs_buf_iowait(bp); |
0e95f19a | 698 | if (error) { |
5adc94c2 DC |
699 | xfs_buf_relse(bp); |
700 | return NULL; | |
701 | } | |
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 DC |
720 | bp->b_length = numblks; |
721 | bp->b_count_desired = numblks << BBSHIFT; | |
44396476 DC |
722 | bp->b_bn = XFS_BUF_DADDR_NULL; |
723 | bp->b_flags &= ~XBF_MAPPED; | |
724 | } | |
725 | ||
1da177e4 LT |
726 | static inline struct page * |
727 | mem_to_page( | |
728 | void *addr) | |
729 | { | |
9e2779fa | 730 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
731 | return virt_to_page(addr); |
732 | } else { | |
733 | return vmalloc_to_page(addr); | |
734 | } | |
735 | } | |
736 | ||
737 | int | |
ce8e922c NS |
738 | xfs_buf_associate_memory( |
739 | xfs_buf_t *bp, | |
1da177e4 LT |
740 | void *mem, |
741 | size_t len) | |
742 | { | |
743 | int rval; | |
744 | int i = 0; | |
d1afb678 LM |
745 | unsigned long pageaddr; |
746 | unsigned long offset; | |
747 | size_t buflen; | |
1da177e4 LT |
748 | int page_count; |
749 | ||
0e6e847f | 750 | pageaddr = (unsigned long)mem & PAGE_MASK; |
d1afb678 | 751 | offset = (unsigned long)mem - pageaddr; |
0e6e847f DC |
752 | buflen = PAGE_ALIGN(len + offset); |
753 | page_count = buflen >> PAGE_SHIFT; | |
1da177e4 LT |
754 | |
755 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
756 | if (bp->b_pages) |
757 | _xfs_buf_free_pages(bp); | |
1da177e4 | 758 | |
ce8e922c NS |
759 | bp->b_pages = NULL; |
760 | bp->b_addr = mem; | |
1da177e4 | 761 | |
36fae17a | 762 | rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK); |
1da177e4 LT |
763 | if (rval) |
764 | return rval; | |
765 | ||
ce8e922c | 766 | bp->b_offset = offset; |
d1afb678 LM |
767 | |
768 | for (i = 0; i < bp->b_page_count; i++) { | |
769 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
0e6e847f | 770 | pageaddr += PAGE_SIZE; |
1da177e4 | 771 | } |
1da177e4 | 772 | |
d1afb678 | 773 | bp->b_count_desired = len; |
4e94b71b | 774 | bp->b_length = BTOBB(buflen); |
ce8e922c | 775 | bp->b_flags |= XBF_MAPPED; |
1da177e4 LT |
776 | |
777 | return 0; | |
778 | } | |
779 | ||
780 | xfs_buf_t * | |
686865f7 DC |
781 | xfs_buf_get_uncached( |
782 | struct xfs_buftarg *target, | |
e70b73f8 | 783 | size_t numblks, |
686865f7 | 784 | int flags) |
1da177e4 | 785 | { |
e70b73f8 | 786 | unsigned long page_count; |
1fa40b01 | 787 | int error, i; |
1da177e4 | 788 | xfs_buf_t *bp; |
1da177e4 | 789 | |
e70b73f8 | 790 | bp = xfs_buf_alloc(target, 0, numblks, 0); |
1da177e4 LT |
791 | if (unlikely(bp == NULL)) |
792 | goto fail; | |
1da177e4 | 793 | |
e70b73f8 | 794 | page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT; |
1fa40b01 CH |
795 | error = _xfs_buf_get_pages(bp, page_count, 0); |
796 | if (error) | |
1da177e4 LT |
797 | goto fail_free_buf; |
798 | ||
1fa40b01 | 799 | for (i = 0; i < page_count; i++) { |
686865f7 | 800 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
801 | if (!bp->b_pages[i]) |
802 | goto fail_free_mem; | |
1da177e4 | 803 | } |
1fa40b01 | 804 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 805 | |
1fa40b01 CH |
806 | error = _xfs_buf_map_pages(bp, XBF_MAPPED); |
807 | if (unlikely(error)) { | |
4f10700a DC |
808 | xfs_warn(target->bt_mount, |
809 | "%s: failed to map pages\n", __func__); | |
1da177e4 | 810 | goto fail_free_mem; |
1fa40b01 | 811 | } |
1da177e4 | 812 | |
686865f7 | 813 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 814 | return bp; |
1fa40b01 | 815 | |
1da177e4 | 816 | fail_free_mem: |
1fa40b01 CH |
817 | while (--i >= 0) |
818 | __free_page(bp->b_pages[i]); | |
ca165b88 | 819 | _xfs_buf_free_pages(bp); |
1da177e4 | 820 | fail_free_buf: |
4347b9d7 | 821 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
822 | fail: |
823 | return NULL; | |
824 | } | |
825 | ||
826 | /* | |
1da177e4 LT |
827 | * Increment reference count on buffer, to hold the buffer concurrently |
828 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
829 | * Must hold the buffer already to call this function. |
830 | */ | |
831 | void | |
ce8e922c NS |
832 | xfs_buf_hold( |
833 | xfs_buf_t *bp) | |
1da177e4 | 834 | { |
0b1b213f | 835 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 836 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
837 | } |
838 | ||
839 | /* | |
ce8e922c NS |
840 | * Releases a hold on the specified buffer. If the |
841 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
842 | */ |
843 | void | |
ce8e922c NS |
844 | xfs_buf_rele( |
845 | xfs_buf_t *bp) | |
1da177e4 | 846 | { |
74f75a0c | 847 | struct xfs_perag *pag = bp->b_pag; |
1da177e4 | 848 | |
0b1b213f | 849 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 850 | |
74f75a0c | 851 | if (!pag) { |
430cbeb8 | 852 | ASSERT(list_empty(&bp->b_lru)); |
74f75a0c | 853 | ASSERT(RB_EMPTY_NODE(&bp->b_rbnode)); |
fad3aa1e NS |
854 | if (atomic_dec_and_test(&bp->b_hold)) |
855 | xfs_buf_free(bp); | |
856 | return; | |
857 | } | |
858 | ||
74f75a0c | 859 | ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode)); |
430cbeb8 | 860 | |
3790689f | 861 | ASSERT(atomic_read(&bp->b_hold) > 0); |
74f75a0c | 862 | if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) { |
bfc60177 | 863 | if (!(bp->b_flags & XBF_STALE) && |
430cbeb8 DC |
864 | atomic_read(&bp->b_lru_ref)) { |
865 | xfs_buf_lru_add(bp); | |
866 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 867 | } else { |
430cbeb8 | 868 | xfs_buf_lru_del(bp); |
43ff2122 | 869 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
74f75a0c DC |
870 | rb_erase(&bp->b_rbnode, &pag->pag_buf_tree); |
871 | spin_unlock(&pag->pag_buf_lock); | |
872 | xfs_perag_put(pag); | |
ce8e922c | 873 | xfs_buf_free(bp); |
1da177e4 LT |
874 | } |
875 | } | |
876 | } | |
877 | ||
878 | ||
879 | /* | |
0e6e847f | 880 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
881 | * |
882 | * If we come across a stale, pinned, locked buffer, we know that we are | |
883 | * being asked to lock a buffer that has been reallocated. Because it is | |
884 | * pinned, we know that the log has not been pushed to disk and hence it | |
885 | * will still be locked. Rather than continuing to have trylock attempts | |
886 | * fail until someone else pushes the log, push it ourselves before | |
887 | * returning. This means that the xfsaild will not get stuck trying | |
888 | * to push on stale inode buffers. | |
1da177e4 LT |
889 | */ |
890 | int | |
0c842ad4 CH |
891 | xfs_buf_trylock( |
892 | struct xfs_buf *bp) | |
1da177e4 LT |
893 | { |
894 | int locked; | |
895 | ||
ce8e922c | 896 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 897 | if (locked) |
ce8e922c | 898 | XB_SET_OWNER(bp); |
90810b9e DC |
899 | else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
900 | xfs_log_force(bp->b_target->bt_mount, 0); | |
0b1b213f | 901 | |
0c842ad4 CH |
902 | trace_xfs_buf_trylock(bp, _RET_IP_); |
903 | return locked; | |
1da177e4 | 904 | } |
1da177e4 LT |
905 | |
906 | /* | |
0e6e847f | 907 | * Lock a buffer object. |
ed3b4d6c DC |
908 | * |
909 | * If we come across a stale, pinned, locked buffer, we know that we | |
910 | * are being asked to lock a buffer that has been reallocated. Because | |
911 | * it is pinned, we know that the log has not been pushed to disk and | |
912 | * hence it will still be locked. Rather than sleeping until someone | |
913 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 914 | */ |
ce8e922c NS |
915 | void |
916 | xfs_buf_lock( | |
0c842ad4 | 917 | struct xfs_buf *bp) |
1da177e4 | 918 | { |
0b1b213f CH |
919 | trace_xfs_buf_lock(bp, _RET_IP_); |
920 | ||
ed3b4d6c | 921 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 922 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
923 | down(&bp->b_sema); |
924 | XB_SET_OWNER(bp); | |
0b1b213f CH |
925 | |
926 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
927 | } |
928 | ||
1da177e4 | 929 | void |
ce8e922c | 930 | xfs_buf_unlock( |
0c842ad4 | 931 | struct xfs_buf *bp) |
1da177e4 | 932 | { |
ce8e922c NS |
933 | XB_CLEAR_OWNER(bp); |
934 | up(&bp->b_sema); | |
0b1b213f CH |
935 | |
936 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
937 | } |
938 | ||
ce8e922c NS |
939 | STATIC void |
940 | xfs_buf_wait_unpin( | |
941 | xfs_buf_t *bp) | |
1da177e4 LT |
942 | { |
943 | DECLARE_WAITQUEUE (wait, current); | |
944 | ||
ce8e922c | 945 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
946 | return; |
947 | ||
ce8e922c | 948 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
949 | for (;;) { |
950 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 951 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 952 | break; |
7eaceacc | 953 | io_schedule(); |
1da177e4 | 954 | } |
ce8e922c | 955 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
956 | set_current_state(TASK_RUNNING); |
957 | } | |
958 | ||
959 | /* | |
960 | * Buffer Utility Routines | |
961 | */ | |
962 | ||
1da177e4 | 963 | STATIC void |
ce8e922c | 964 | xfs_buf_iodone_work( |
c4028958 | 965 | struct work_struct *work) |
1da177e4 | 966 | { |
c4028958 DH |
967 | xfs_buf_t *bp = |
968 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 969 | |
80f6c29d | 970 | if (bp->b_iodone) |
ce8e922c NS |
971 | (*(bp->b_iodone))(bp); |
972 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
973 | xfs_buf_relse(bp); |
974 | } | |
975 | ||
976 | void | |
ce8e922c NS |
977 | xfs_buf_ioend( |
978 | xfs_buf_t *bp, | |
1da177e4 LT |
979 | int schedule) |
980 | { | |
0b1b213f CH |
981 | trace_xfs_buf_iodone(bp, _RET_IP_); |
982 | ||
77be55a5 | 983 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
ce8e922c NS |
984 | if (bp->b_error == 0) |
985 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 986 | |
ce8e922c | 987 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 988 | if (schedule) { |
c4028958 | 989 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 990 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 991 | } else { |
c4028958 | 992 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
993 | } |
994 | } else { | |
b4dd330b | 995 | complete(&bp->b_iowait); |
1da177e4 LT |
996 | } |
997 | } | |
998 | ||
1da177e4 | 999 | void |
ce8e922c NS |
1000 | xfs_buf_ioerror( |
1001 | xfs_buf_t *bp, | |
1002 | int error) | |
1da177e4 LT |
1003 | { |
1004 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c | 1005 | bp->b_error = (unsigned short)error; |
0b1b213f | 1006 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1007 | } |
1008 | ||
901796af CH |
1009 | void |
1010 | xfs_buf_ioerror_alert( | |
1011 | struct xfs_buf *bp, | |
1012 | const char *func) | |
1013 | { | |
1014 | xfs_alert(bp->b_target->bt_mount, | |
1015 | "metadata I/O error: block 0x%llx (\"%s\") error %d buf count %zd", | |
1016 | (__uint64_t)XFS_BUF_ADDR(bp), func, | |
1017 | bp->b_error, XFS_BUF_COUNT(bp)); | |
1018 | } | |
1019 | ||
1da177e4 | 1020 | int |
64e0bc7d | 1021 | xfs_bwrite( |
5d765b97 | 1022 | struct xfs_buf *bp) |
1da177e4 | 1023 | { |
8c38366f | 1024 | int error; |
1da177e4 | 1025 | |
43ff2122 CH |
1026 | ASSERT(xfs_buf_islocked(bp)); |
1027 | ||
64e0bc7d | 1028 | bp->b_flags |= XBF_WRITE; |
43ff2122 | 1029 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q); |
1da177e4 | 1030 | |
939d723b | 1031 | xfs_bdstrat_cb(bp); |
1da177e4 | 1032 | |
8c38366f | 1033 | error = xfs_buf_iowait(bp); |
c2b006c1 CH |
1034 | if (error) { |
1035 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1036 | SHUTDOWN_META_IO_ERROR); | |
1037 | } | |
64e0bc7d | 1038 | return error; |
5d765b97 | 1039 | } |
1da177e4 | 1040 | |
4e23471a CH |
1041 | /* |
1042 | * Called when we want to stop a buffer from getting written or read. | |
1a1a3e97 | 1043 | * We attach the EIO error, muck with its flags, and call xfs_buf_ioend |
4e23471a CH |
1044 | * so that the proper iodone callbacks get called. |
1045 | */ | |
1046 | STATIC int | |
1047 | xfs_bioerror( | |
1048 | xfs_buf_t *bp) | |
1049 | { | |
1050 | #ifdef XFSERRORDEBUG | |
1051 | ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone); | |
1052 | #endif | |
1053 | ||
1054 | /* | |
1055 | * No need to wait until the buffer is unpinned, we aren't flushing it. | |
1056 | */ | |
5a52c2a5 | 1057 | xfs_buf_ioerror(bp, EIO); |
4e23471a CH |
1058 | |
1059 | /* | |
1a1a3e97 | 1060 | * We're calling xfs_buf_ioend, so delete XBF_DONE flag. |
4e23471a CH |
1061 | */ |
1062 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1063 | XFS_BUF_UNDONE(bp); |
c867cb61 | 1064 | xfs_buf_stale(bp); |
4e23471a | 1065 | |
1a1a3e97 | 1066 | xfs_buf_ioend(bp, 0); |
4e23471a CH |
1067 | |
1068 | return EIO; | |
1069 | } | |
1070 | ||
1071 | /* | |
1072 | * Same as xfs_bioerror, except that we are releasing the buffer | |
1a1a3e97 | 1073 | * here ourselves, and avoiding the xfs_buf_ioend call. |
4e23471a CH |
1074 | * This is meant for userdata errors; metadata bufs come with |
1075 | * iodone functions attached, so that we can track down errors. | |
1076 | */ | |
1077 | STATIC int | |
1078 | xfs_bioerror_relse( | |
1079 | struct xfs_buf *bp) | |
1080 | { | |
ed43233b | 1081 | int64_t fl = bp->b_flags; |
4e23471a CH |
1082 | /* |
1083 | * No need to wait until the buffer is unpinned. | |
1084 | * We aren't flushing it. | |
1085 | * | |
1086 | * chunkhold expects B_DONE to be set, whether | |
1087 | * we actually finish the I/O or not. We don't want to | |
1088 | * change that interface. | |
1089 | */ | |
1090 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1091 | XFS_BUF_DONE(bp); |
c867cb61 | 1092 | xfs_buf_stale(bp); |
cb669ca5 | 1093 | bp->b_iodone = NULL; |
0cadda1c | 1094 | if (!(fl & XBF_ASYNC)) { |
4e23471a CH |
1095 | /* |
1096 | * Mark b_error and B_ERROR _both_. | |
1097 | * Lot's of chunkcache code assumes that. | |
1098 | * There's no reason to mark error for | |
1099 | * ASYNC buffers. | |
1100 | */ | |
5a52c2a5 | 1101 | xfs_buf_ioerror(bp, EIO); |
5fde0326 | 1102 | complete(&bp->b_iowait); |
4e23471a CH |
1103 | } else { |
1104 | xfs_buf_relse(bp); | |
1105 | } | |
1106 | ||
1107 | return EIO; | |
1108 | } | |
1109 | ||
1110 | ||
1111 | /* | |
1112 | * All xfs metadata buffers except log state machine buffers | |
1113 | * get this attached as their b_bdstrat callback function. | |
1114 | * This is so that we can catch a buffer | |
1115 | * after prematurely unpinning it to forcibly shutdown the filesystem. | |
1116 | */ | |
1117 | int | |
1118 | xfs_bdstrat_cb( | |
1119 | struct xfs_buf *bp) | |
1120 | { | |
ebad861b | 1121 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
4e23471a CH |
1122 | trace_xfs_bdstrat_shut(bp, _RET_IP_); |
1123 | /* | |
1124 | * Metadata write that didn't get logged but | |
1125 | * written delayed anyway. These aren't associated | |
1126 | * with a transaction, and can be ignored. | |
1127 | */ | |
1128 | if (!bp->b_iodone && !XFS_BUF_ISREAD(bp)) | |
1129 | return xfs_bioerror_relse(bp); | |
1130 | else | |
1131 | return xfs_bioerror(bp); | |
1132 | } | |
1133 | ||
1134 | xfs_buf_iorequest(bp); | |
1135 | return 0; | |
1136 | } | |
1137 | ||
1138 | /* | |
1139 | * Wrapper around bdstrat so that we can stop data from going to disk in case | |
1140 | * we are shutting down the filesystem. Typically user data goes thru this | |
1141 | * path; one of the exceptions is the superblock. | |
1142 | */ | |
1143 | void | |
1144 | xfsbdstrat( | |
1145 | struct xfs_mount *mp, | |
1146 | struct xfs_buf *bp) | |
1147 | { | |
1148 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1149 | trace_xfs_bdstrat_shut(bp, _RET_IP_); | |
1150 | xfs_bioerror_relse(bp); | |
1151 | return; | |
1152 | } | |
1153 | ||
1154 | xfs_buf_iorequest(bp); | |
1155 | } | |
1156 | ||
b8f82a4a | 1157 | STATIC void |
ce8e922c NS |
1158 | _xfs_buf_ioend( |
1159 | xfs_buf_t *bp, | |
1da177e4 LT |
1160 | int schedule) |
1161 | { | |
0e6e847f | 1162 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
ce8e922c | 1163 | xfs_buf_ioend(bp, schedule); |
1da177e4 LT |
1164 | } |
1165 | ||
782e3b3b | 1166 | STATIC void |
ce8e922c | 1167 | xfs_buf_bio_end_io( |
1da177e4 | 1168 | struct bio *bio, |
1da177e4 LT |
1169 | int error) |
1170 | { | |
ce8e922c | 1171 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1da177e4 | 1172 | |
cfbe5267 | 1173 | xfs_buf_ioerror(bp, -error); |
1da177e4 | 1174 | |
73c77e2c JB |
1175 | if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
1176 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); | |
1177 | ||
ce8e922c | 1178 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1179 | bio_put(bio); |
1da177e4 LT |
1180 | } |
1181 | ||
1182 | STATIC void | |
ce8e922c NS |
1183 | _xfs_buf_ioapply( |
1184 | xfs_buf_t *bp) | |
1da177e4 | 1185 | { |
a9759f2d | 1186 | int rw, map_i, total_nr_pages, nr_pages; |
1da177e4 | 1187 | struct bio *bio; |
ce8e922c NS |
1188 | int offset = bp->b_offset; |
1189 | int size = bp->b_count_desired; | |
1190 | sector_t sector = bp->b_bn; | |
1da177e4 | 1191 | |
ce8e922c | 1192 | total_nr_pages = bp->b_page_count; |
1da177e4 LT |
1193 | map_i = 0; |
1194 | ||
1d5ae5df CH |
1195 | if (bp->b_flags & XBF_WRITE) { |
1196 | if (bp->b_flags & XBF_SYNCIO) | |
1197 | rw = WRITE_SYNC; | |
1198 | else | |
1199 | rw = WRITE; | |
1200 | if (bp->b_flags & XBF_FUA) | |
1201 | rw |= REQ_FUA; | |
1202 | if (bp->b_flags & XBF_FLUSH) | |
1203 | rw |= REQ_FLUSH; | |
1204 | } else if (bp->b_flags & XBF_READ_AHEAD) { | |
1205 | rw = READA; | |
51bdd706 | 1206 | } else { |
1d5ae5df | 1207 | rw = READ; |
f538d4da CH |
1208 | } |
1209 | ||
34951f5c CH |
1210 | /* we only use the buffer cache for meta-data */ |
1211 | rw |= REQ_META; | |
1212 | ||
1da177e4 | 1213 | next_chunk: |
ce8e922c | 1214 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1215 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1216 | if (nr_pages > total_nr_pages) | |
1217 | nr_pages = total_nr_pages; | |
1218 | ||
1219 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1220 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1221 | bio->bi_sector = sector; |
ce8e922c NS |
1222 | bio->bi_end_io = xfs_buf_bio_end_io; |
1223 | bio->bi_private = bp; | |
1da177e4 | 1224 | |
0e6e847f | 1225 | |
1da177e4 | 1226 | for (; size && nr_pages; nr_pages--, map_i++) { |
0e6e847f | 1227 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1228 | |
1229 | if (nbytes > size) | |
1230 | nbytes = size; | |
1231 | ||
ce8e922c NS |
1232 | rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset); |
1233 | if (rbytes < nbytes) | |
1da177e4 LT |
1234 | break; |
1235 | ||
1236 | offset = 0; | |
1237 | sector += nbytes >> BBSHIFT; | |
1238 | size -= nbytes; | |
1239 | total_nr_pages--; | |
1240 | } | |
1241 | ||
1da177e4 | 1242 | if (likely(bio->bi_size)) { |
73c77e2c JB |
1243 | if (xfs_buf_is_vmapped(bp)) { |
1244 | flush_kernel_vmap_range(bp->b_addr, | |
1245 | xfs_buf_vmap_len(bp)); | |
1246 | } | |
1da177e4 LT |
1247 | submit_bio(rw, bio); |
1248 | if (size) | |
1249 | goto next_chunk; | |
1250 | } else { | |
ce8e922c | 1251 | xfs_buf_ioerror(bp, EIO); |
ec53d1db | 1252 | bio_put(bio); |
1da177e4 LT |
1253 | } |
1254 | } | |
1255 | ||
0e95f19a | 1256 | void |
ce8e922c NS |
1257 | xfs_buf_iorequest( |
1258 | xfs_buf_t *bp) | |
1da177e4 | 1259 | { |
0b1b213f | 1260 | trace_xfs_buf_iorequest(bp, _RET_IP_); |
1da177e4 | 1261 | |
43ff2122 | 1262 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
1da177e4 | 1263 | |
375ec69d | 1264 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1265 | xfs_buf_wait_unpin(bp); |
ce8e922c | 1266 | xfs_buf_hold(bp); |
1da177e4 LT |
1267 | |
1268 | /* Set the count to 1 initially, this will stop an I/O | |
1269 | * completion callout which happens before we have started | |
ce8e922c | 1270 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1271 | */ |
ce8e922c NS |
1272 | atomic_set(&bp->b_io_remaining, 1); |
1273 | _xfs_buf_ioapply(bp); | |
1274 | _xfs_buf_ioend(bp, 0); | |
1da177e4 | 1275 | |
ce8e922c | 1276 | xfs_buf_rele(bp); |
1da177e4 LT |
1277 | } |
1278 | ||
1279 | /* | |
0e95f19a DC |
1280 | * Waits for I/O to complete on the buffer supplied. It returns immediately if |
1281 | * no I/O is pending or there is already a pending error on the buffer. It | |
1282 | * returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1283 | */ |
1284 | int | |
ce8e922c NS |
1285 | xfs_buf_iowait( |
1286 | xfs_buf_t *bp) | |
1da177e4 | 1287 | { |
0b1b213f CH |
1288 | trace_xfs_buf_iowait(bp, _RET_IP_); |
1289 | ||
0e95f19a DC |
1290 | if (!bp->b_error) |
1291 | wait_for_completion(&bp->b_iowait); | |
0b1b213f CH |
1292 | |
1293 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
ce8e922c | 1294 | return bp->b_error; |
1da177e4 LT |
1295 | } |
1296 | ||
ce8e922c NS |
1297 | xfs_caddr_t |
1298 | xfs_buf_offset( | |
1299 | xfs_buf_t *bp, | |
1da177e4 LT |
1300 | size_t offset) |
1301 | { | |
1302 | struct page *page; | |
1303 | ||
ce8e922c | 1304 | if (bp->b_flags & XBF_MAPPED) |
62926044 | 1305 | return bp->b_addr + offset; |
1da177e4 | 1306 | |
ce8e922c | 1307 | offset += bp->b_offset; |
0e6e847f DC |
1308 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
1309 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_SIZE-1)); | |
1da177e4 LT |
1310 | } |
1311 | ||
1312 | /* | |
1da177e4 LT |
1313 | * Move data into or out of a buffer. |
1314 | */ | |
1315 | void | |
ce8e922c NS |
1316 | xfs_buf_iomove( |
1317 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1318 | size_t boff, /* starting buffer offset */ |
1319 | size_t bsize, /* length to copy */ | |
b9c48649 | 1320 | void *data, /* data address */ |
ce8e922c | 1321 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 LT |
1322 | { |
1323 | size_t bend, cpoff, csize; | |
1324 | struct page *page; | |
1325 | ||
1326 | bend = boff + bsize; | |
1327 | while (boff < bend) { | |
ce8e922c NS |
1328 | page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)]; |
1329 | cpoff = xfs_buf_poff(boff + bp->b_offset); | |
1da177e4 | 1330 | csize = min_t(size_t, |
0e6e847f | 1331 | PAGE_SIZE-cpoff, bp->b_count_desired-boff); |
1da177e4 | 1332 | |
0e6e847f | 1333 | ASSERT(((csize + cpoff) <= PAGE_SIZE)); |
1da177e4 LT |
1334 | |
1335 | switch (mode) { | |
ce8e922c | 1336 | case XBRW_ZERO: |
1da177e4 LT |
1337 | memset(page_address(page) + cpoff, 0, csize); |
1338 | break; | |
ce8e922c | 1339 | case XBRW_READ: |
1da177e4 LT |
1340 | memcpy(data, page_address(page) + cpoff, csize); |
1341 | break; | |
ce8e922c | 1342 | case XBRW_WRITE: |
1da177e4 LT |
1343 | memcpy(page_address(page) + cpoff, data, csize); |
1344 | } | |
1345 | ||
1346 | boff += csize; | |
1347 | data += csize; | |
1348 | } | |
1349 | } | |
1350 | ||
1351 | /* | |
ce8e922c | 1352 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1353 | */ |
1354 | ||
1355 | /* | |
430cbeb8 DC |
1356 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1357 | * returned. These buffers will have an elevated hold count, so wait on those | |
1358 | * while freeing all the buffers only held by the LRU. | |
1da177e4 LT |
1359 | */ |
1360 | void | |
1361 | xfs_wait_buftarg( | |
74f75a0c | 1362 | struct xfs_buftarg *btp) |
1da177e4 | 1363 | { |
430cbeb8 DC |
1364 | struct xfs_buf *bp; |
1365 | ||
1366 | restart: | |
1367 | spin_lock(&btp->bt_lru_lock); | |
1368 | while (!list_empty(&btp->bt_lru)) { | |
1369 | bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru); | |
1370 | if (atomic_read(&bp->b_hold) > 1) { | |
1371 | spin_unlock(&btp->bt_lru_lock); | |
26af6552 | 1372 | delay(100); |
430cbeb8 | 1373 | goto restart; |
1da177e4 | 1374 | } |
430cbeb8 | 1375 | /* |
90802ed9 | 1376 | * clear the LRU reference count so the buffer doesn't get |
430cbeb8 DC |
1377 | * ignored in xfs_buf_rele(). |
1378 | */ | |
1379 | atomic_set(&bp->b_lru_ref, 0); | |
1380 | spin_unlock(&btp->bt_lru_lock); | |
1381 | xfs_buf_rele(bp); | |
1382 | spin_lock(&btp->bt_lru_lock); | |
1da177e4 | 1383 | } |
430cbeb8 | 1384 | spin_unlock(&btp->bt_lru_lock); |
1da177e4 LT |
1385 | } |
1386 | ||
ff57ab21 DC |
1387 | int |
1388 | xfs_buftarg_shrink( | |
1389 | struct shrinker *shrink, | |
1495f230 | 1390 | struct shrink_control *sc) |
a6867a68 | 1391 | { |
ff57ab21 DC |
1392 | struct xfs_buftarg *btp = container_of(shrink, |
1393 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1394 | struct xfs_buf *bp; |
1495f230 | 1395 | int nr_to_scan = sc->nr_to_scan; |
430cbeb8 DC |
1396 | LIST_HEAD(dispose); |
1397 | ||
1398 | if (!nr_to_scan) | |
1399 | return btp->bt_lru_nr; | |
1400 | ||
1401 | spin_lock(&btp->bt_lru_lock); | |
1402 | while (!list_empty(&btp->bt_lru)) { | |
1403 | if (nr_to_scan-- <= 0) | |
1404 | break; | |
1405 | ||
1406 | bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru); | |
1407 | ||
1408 | /* | |
1409 | * Decrement the b_lru_ref count unless the value is already | |
1410 | * zero. If the value is already zero, we need to reclaim the | |
1411 | * buffer, otherwise it gets another trip through the LRU. | |
1412 | */ | |
1413 | if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) { | |
1414 | list_move_tail(&bp->b_lru, &btp->bt_lru); | |
1415 | continue; | |
1416 | } | |
1417 | ||
1418 | /* | |
1419 | * remove the buffer from the LRU now to avoid needing another | |
1420 | * lock round trip inside xfs_buf_rele(). | |
1421 | */ | |
1422 | list_move(&bp->b_lru, &dispose); | |
1423 | btp->bt_lru_nr--; | |
ff57ab21 | 1424 | } |
430cbeb8 DC |
1425 | spin_unlock(&btp->bt_lru_lock); |
1426 | ||
1427 | while (!list_empty(&dispose)) { | |
1428 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1429 | list_del_init(&bp->b_lru); | |
1430 | xfs_buf_rele(bp); | |
1431 | } | |
1432 | ||
1433 | return btp->bt_lru_nr; | |
a6867a68 DC |
1434 | } |
1435 | ||
1da177e4 LT |
1436 | void |
1437 | xfs_free_buftarg( | |
b7963133 CH |
1438 | struct xfs_mount *mp, |
1439 | struct xfs_buftarg *btp) | |
1da177e4 | 1440 | { |
ff57ab21 DC |
1441 | unregister_shrinker(&btp->bt_shrinker); |
1442 | ||
b7963133 CH |
1443 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1444 | xfs_blkdev_issue_flush(btp); | |
a6867a68 | 1445 | |
f0e2d93c | 1446 | kmem_free(btp); |
1da177e4 LT |
1447 | } |
1448 | ||
1da177e4 LT |
1449 | STATIC int |
1450 | xfs_setsize_buftarg_flags( | |
1451 | xfs_buftarg_t *btp, | |
1452 | unsigned int blocksize, | |
1453 | unsigned int sectorsize, | |
1454 | int verbose) | |
1455 | { | |
ce8e922c NS |
1456 | btp->bt_bsize = blocksize; |
1457 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1458 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1459 | |
ce8e922c | 1460 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
02b102df CH |
1461 | char name[BDEVNAME_SIZE]; |
1462 | ||
1463 | bdevname(btp->bt_bdev, name); | |
1464 | ||
4f10700a DC |
1465 | xfs_warn(btp->bt_mount, |
1466 | "Cannot set_blocksize to %u on device %s\n", | |
02b102df | 1467 | sectorsize, name); |
1da177e4 LT |
1468 | return EINVAL; |
1469 | } | |
1470 | ||
1da177e4 LT |
1471 | return 0; |
1472 | } | |
1473 | ||
1474 | /* | |
ce8e922c NS |
1475 | * When allocating the initial buffer target we have not yet |
1476 | * read in the superblock, so don't know what sized sectors | |
1477 | * are being used is at this early stage. Play safe. | |
1478 | */ | |
1da177e4 LT |
1479 | STATIC int |
1480 | xfs_setsize_buftarg_early( | |
1481 | xfs_buftarg_t *btp, | |
1482 | struct block_device *bdev) | |
1483 | { | |
1484 | return xfs_setsize_buftarg_flags(btp, | |
0e6e847f | 1485 | PAGE_SIZE, bdev_logical_block_size(bdev), 0); |
1da177e4 LT |
1486 | } |
1487 | ||
1488 | int | |
1489 | xfs_setsize_buftarg( | |
1490 | xfs_buftarg_t *btp, | |
1491 | unsigned int blocksize, | |
1492 | unsigned int sectorsize) | |
1493 | { | |
1494 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1495 | } | |
1496 | ||
1da177e4 LT |
1497 | xfs_buftarg_t * |
1498 | xfs_alloc_buftarg( | |
ebad861b | 1499 | struct xfs_mount *mp, |
1da177e4 | 1500 | struct block_device *bdev, |
e2a07812 JE |
1501 | int external, |
1502 | const char *fsname) | |
1da177e4 LT |
1503 | { |
1504 | xfs_buftarg_t *btp; | |
1505 | ||
1506 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1507 | ||
ebad861b | 1508 | btp->bt_mount = mp; |
ce8e922c NS |
1509 | btp->bt_dev = bdev->bd_dev; |
1510 | btp->bt_bdev = bdev; | |
0e6e847f DC |
1511 | btp->bt_bdi = blk_get_backing_dev_info(bdev); |
1512 | if (!btp->bt_bdi) | |
1513 | goto error; | |
1514 | ||
430cbeb8 DC |
1515 | INIT_LIST_HEAD(&btp->bt_lru); |
1516 | spin_lock_init(&btp->bt_lru_lock); | |
1da177e4 LT |
1517 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1518 | goto error; | |
ff57ab21 DC |
1519 | btp->bt_shrinker.shrink = xfs_buftarg_shrink; |
1520 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; | |
1521 | register_shrinker(&btp->bt_shrinker); | |
1da177e4 LT |
1522 | return btp; |
1523 | ||
1524 | error: | |
f0e2d93c | 1525 | kmem_free(btp); |
1da177e4 LT |
1526 | return NULL; |
1527 | } | |
1528 | ||
1da177e4 | 1529 | /* |
43ff2122 CH |
1530 | * Add a buffer to the delayed write list. |
1531 | * | |
1532 | * This queues a buffer for writeout if it hasn't already been. Note that | |
1533 | * neither this routine nor the buffer list submission functions perform | |
1534 | * any internal synchronization. It is expected that the lists are thread-local | |
1535 | * to the callers. | |
1536 | * | |
1537 | * Returns true if we queued up the buffer, or false if it already had | |
1538 | * been on the buffer list. | |
1da177e4 | 1539 | */ |
43ff2122 | 1540 | bool |
ce8e922c | 1541 | xfs_buf_delwri_queue( |
43ff2122 CH |
1542 | struct xfs_buf *bp, |
1543 | struct list_head *list) | |
1da177e4 | 1544 | { |
43ff2122 | 1545 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 1546 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1547 | |
43ff2122 CH |
1548 | /* |
1549 | * If the buffer is already marked delwri it already is queued up | |
1550 | * by someone else for imediate writeout. Just ignore it in that | |
1551 | * case. | |
1552 | */ | |
1553 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
1554 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
1555 | return false; | |
1da177e4 | 1556 | } |
1da177e4 | 1557 | |
43ff2122 | 1558 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
1559 | |
1560 | /* | |
43ff2122 CH |
1561 | * If a buffer gets written out synchronously or marked stale while it |
1562 | * is on a delwri list we lazily remove it. To do this, the other party | |
1563 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
1564 | * It remains referenced and on the list. In a rare corner case it | |
1565 | * might get readded to a delwri list after the synchronous writeout, in | |
1566 | * which case we need just need to re-add the flag here. | |
d808f617 | 1567 | */ |
43ff2122 CH |
1568 | bp->b_flags |= _XBF_DELWRI_Q; |
1569 | if (list_empty(&bp->b_list)) { | |
1570 | atomic_inc(&bp->b_hold); | |
1571 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 1572 | } |
585e6d88 | 1573 | |
43ff2122 | 1574 | return true; |
585e6d88 DC |
1575 | } |
1576 | ||
089716aa DC |
1577 | /* |
1578 | * Compare function is more complex than it needs to be because | |
1579 | * the return value is only 32 bits and we are doing comparisons | |
1580 | * on 64 bit values | |
1581 | */ | |
1582 | static int | |
1583 | xfs_buf_cmp( | |
1584 | void *priv, | |
1585 | struct list_head *a, | |
1586 | struct list_head *b) | |
1587 | { | |
1588 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1589 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1590 | xfs_daddr_t diff; | |
1591 | ||
1592 | diff = ap->b_bn - bp->b_bn; | |
1593 | if (diff < 0) | |
1594 | return -1; | |
1595 | if (diff > 0) | |
1596 | return 1; | |
1597 | return 0; | |
1598 | } | |
1599 | ||
43ff2122 CH |
1600 | static int |
1601 | __xfs_buf_delwri_submit( | |
1602 | struct list_head *buffer_list, | |
1603 | struct list_head *io_list, | |
1604 | bool wait) | |
1da177e4 | 1605 | { |
43ff2122 CH |
1606 | struct blk_plug plug; |
1607 | struct xfs_buf *bp, *n; | |
1608 | int pinned = 0; | |
1609 | ||
1610 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { | |
1611 | if (!wait) { | |
1612 | if (xfs_buf_ispinned(bp)) { | |
1613 | pinned++; | |
1614 | continue; | |
1615 | } | |
1616 | if (!xfs_buf_trylock(bp)) | |
1617 | continue; | |
1618 | } else { | |
1619 | xfs_buf_lock(bp); | |
1620 | } | |
978c7b2f | 1621 | |
43ff2122 CH |
1622 | /* |
1623 | * Someone else might have written the buffer synchronously or | |
1624 | * marked it stale in the meantime. In that case only the | |
1625 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
1626 | * reference and remove it from the list here. | |
1627 | */ | |
1628 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
1629 | list_del_init(&bp->b_list); | |
1630 | xfs_buf_relse(bp); | |
1631 | continue; | |
1632 | } | |
c9c12971 | 1633 | |
43ff2122 CH |
1634 | list_move_tail(&bp->b_list, io_list); |
1635 | trace_xfs_buf_delwri_split(bp, _RET_IP_); | |
1636 | } | |
1da177e4 | 1637 | |
43ff2122 | 1638 | list_sort(NULL, io_list, xfs_buf_cmp); |
1da177e4 | 1639 | |
43ff2122 CH |
1640 | blk_start_plug(&plug); |
1641 | list_for_each_entry_safe(bp, n, io_list, b_list) { | |
1642 | bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_ASYNC); | |
1643 | bp->b_flags |= XBF_WRITE; | |
a1b7ea5d | 1644 | |
43ff2122 CH |
1645 | if (!wait) { |
1646 | bp->b_flags |= XBF_ASYNC; | |
ce8e922c | 1647 | list_del_init(&bp->b_list); |
1da177e4 | 1648 | } |
43ff2122 CH |
1649 | xfs_bdstrat_cb(bp); |
1650 | } | |
1651 | blk_finish_plug(&plug); | |
1da177e4 | 1652 | |
43ff2122 | 1653 | return pinned; |
1da177e4 LT |
1654 | } |
1655 | ||
1656 | /* | |
43ff2122 CH |
1657 | * Write out a buffer list asynchronously. |
1658 | * | |
1659 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
1660 | * out and not wait for I/O completion on any of the buffers. This interface | |
1661 | * is only safely useable for callers that can track I/O completion by higher | |
1662 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
1663 | * function. | |
1da177e4 LT |
1664 | */ |
1665 | int | |
43ff2122 CH |
1666 | xfs_buf_delwri_submit_nowait( |
1667 | struct list_head *buffer_list) | |
1da177e4 | 1668 | { |
43ff2122 CH |
1669 | LIST_HEAD (io_list); |
1670 | return __xfs_buf_delwri_submit(buffer_list, &io_list, false); | |
1671 | } | |
1da177e4 | 1672 | |
43ff2122 CH |
1673 | /* |
1674 | * Write out a buffer list synchronously. | |
1675 | * | |
1676 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
1677 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
1678 | * so callers must have some other way of tracking buffers if they require such | |
1679 | * functionality. | |
1680 | */ | |
1681 | int | |
1682 | xfs_buf_delwri_submit( | |
1683 | struct list_head *buffer_list) | |
1684 | { | |
1685 | LIST_HEAD (io_list); | |
1686 | int error = 0, error2; | |
1687 | struct xfs_buf *bp; | |
1da177e4 | 1688 | |
43ff2122 | 1689 | __xfs_buf_delwri_submit(buffer_list, &io_list, true); |
1da177e4 | 1690 | |
43ff2122 CH |
1691 | /* Wait for IO to complete. */ |
1692 | while (!list_empty(&io_list)) { | |
1693 | bp = list_first_entry(&io_list, struct xfs_buf, b_list); | |
a1b7ea5d | 1694 | |
089716aa | 1695 | list_del_init(&bp->b_list); |
43ff2122 CH |
1696 | error2 = xfs_buf_iowait(bp); |
1697 | xfs_buf_relse(bp); | |
1698 | if (!error) | |
1699 | error = error2; | |
1da177e4 LT |
1700 | } |
1701 | ||
43ff2122 | 1702 | return error; |
1da177e4 LT |
1703 | } |
1704 | ||
04d8b284 | 1705 | int __init |
ce8e922c | 1706 | xfs_buf_init(void) |
1da177e4 | 1707 | { |
8758280f NS |
1708 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1709 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1710 | if (!xfs_buf_zone) |
0b1b213f | 1711 | goto out; |
04d8b284 | 1712 | |
51749e47 | 1713 | xfslogd_workqueue = alloc_workqueue("xfslogd", |
6370a6ad | 1714 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 1); |
23ea4032 | 1715 | if (!xfslogd_workqueue) |
04d8b284 | 1716 | goto out_free_buf_zone; |
1da177e4 | 1717 | |
23ea4032 | 1718 | return 0; |
1da177e4 | 1719 | |
23ea4032 | 1720 | out_free_buf_zone: |
ce8e922c | 1721 | kmem_zone_destroy(xfs_buf_zone); |
0b1b213f | 1722 | out: |
8758280f | 1723 | return -ENOMEM; |
1da177e4 LT |
1724 | } |
1725 | ||
1da177e4 | 1726 | void |
ce8e922c | 1727 | xfs_buf_terminate(void) |
1da177e4 | 1728 | { |
04d8b284 | 1729 | destroy_workqueue(xfslogd_workqueue); |
ce8e922c | 1730 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 1731 | } |