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" |
ed3b4d6c | 38 | #include "xfs_log.h" |
b7963133 | 39 | #include "xfs_ag.h" |
b7963133 | 40 | #include "xfs_mount.h" |
0b1b213f | 41 | #include "xfs_trace.h" |
b7963133 | 42 | |
7989cb8e | 43 | static kmem_zone_t *xfs_buf_zone; |
23ea4032 | 44 | |
7989cb8e | 45 | static struct workqueue_struct *xfslogd_workqueue; |
1da177e4 | 46 | |
ce8e922c NS |
47 | #ifdef XFS_BUF_LOCK_TRACKING |
48 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
49 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
50 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 51 | #else |
ce8e922c NS |
52 | # define XB_SET_OWNER(bp) do { } while (0) |
53 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
54 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
55 | #endif |
56 | ||
ce8e922c | 57 | #define xb_to_gfp(flags) \ |
aa5c158e | 58 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN) |
1da177e4 | 59 | |
1da177e4 | 60 | |
73c77e2c JB |
61 | static inline int |
62 | xfs_buf_is_vmapped( | |
63 | struct xfs_buf *bp) | |
64 | { | |
65 | /* | |
66 | * Return true if the buffer is vmapped. | |
67 | * | |
611c9946 DC |
68 | * b_addr is null if the buffer is not mapped, but the code is clever |
69 | * enough to know it doesn't have to map a single page, so the check has | |
70 | * to be both for b_addr and bp->b_page_count > 1. | |
73c77e2c | 71 | */ |
611c9946 | 72 | return bp->b_addr && bp->b_page_count > 1; |
73c77e2c JB |
73 | } |
74 | ||
75 | static inline int | |
76 | xfs_buf_vmap_len( | |
77 | struct xfs_buf *bp) | |
78 | { | |
79 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
80 | } | |
81 | ||
1da177e4 | 82 | /* |
430cbeb8 DC |
83 | * xfs_buf_lru_add - add a buffer to the LRU. |
84 | * | |
85 | * The LRU takes a new reference to the buffer so that it will only be freed | |
86 | * once the shrinker takes the buffer off the LRU. | |
87 | */ | |
88 | STATIC void | |
89 | xfs_buf_lru_add( | |
90 | struct xfs_buf *bp) | |
91 | { | |
92 | struct xfs_buftarg *btp = bp->b_target; | |
93 | ||
94 | spin_lock(&btp->bt_lru_lock); | |
95 | if (list_empty(&bp->b_lru)) { | |
96 | atomic_inc(&bp->b_hold); | |
97 | list_add_tail(&bp->b_lru, &btp->bt_lru); | |
98 | btp->bt_lru_nr++; | |
99 | } | |
100 | spin_unlock(&btp->bt_lru_lock); | |
101 | } | |
102 | ||
103 | /* | |
104 | * xfs_buf_lru_del - remove a buffer from the LRU | |
105 | * | |
106 | * The unlocked check is safe here because it only occurs when there are not | |
107 | * b_lru_ref counts left on the inode under the pag->pag_buf_lock. it is there | |
108 | * to optimise the shrinker removing the buffer from the LRU and calling | |
25985edc | 109 | * xfs_buf_free(). i.e. it removes an unnecessary round trip on the |
430cbeb8 | 110 | * bt_lru_lock. |
1da177e4 | 111 | */ |
430cbeb8 DC |
112 | STATIC void |
113 | xfs_buf_lru_del( | |
114 | struct xfs_buf *bp) | |
115 | { | |
116 | struct xfs_buftarg *btp = bp->b_target; | |
117 | ||
118 | if (list_empty(&bp->b_lru)) | |
119 | return; | |
120 | ||
121 | spin_lock(&btp->bt_lru_lock); | |
122 | if (!list_empty(&bp->b_lru)) { | |
123 | list_del_init(&bp->b_lru); | |
124 | btp->bt_lru_nr--; | |
125 | } | |
126 | spin_unlock(&btp->bt_lru_lock); | |
127 | } | |
128 | ||
129 | /* | |
130 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
131 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
132 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
133 | * to remove the reference that LRU holds on the buffer. | |
134 | * | |
135 | * This prevents build-up of stale buffers on the LRU. | |
136 | */ | |
137 | void | |
138 | xfs_buf_stale( | |
139 | struct xfs_buf *bp) | |
140 | { | |
43ff2122 CH |
141 | ASSERT(xfs_buf_islocked(bp)); |
142 | ||
430cbeb8 | 143 | bp->b_flags |= XBF_STALE; |
43ff2122 CH |
144 | |
145 | /* | |
146 | * Clear the delwri status so that a delwri queue walker will not | |
147 | * flush this buffer to disk now that it is stale. The delwri queue has | |
148 | * a reference to the buffer, so this is safe to do. | |
149 | */ | |
150 | bp->b_flags &= ~_XBF_DELWRI_Q; | |
151 | ||
430cbeb8 DC |
152 | atomic_set(&(bp)->b_lru_ref, 0); |
153 | if (!list_empty(&bp->b_lru)) { | |
154 | struct xfs_buftarg *btp = bp->b_target; | |
155 | ||
156 | spin_lock(&btp->bt_lru_lock); | |
157 | if (!list_empty(&bp->b_lru)) { | |
158 | list_del_init(&bp->b_lru); | |
159 | btp->bt_lru_nr--; | |
160 | atomic_dec(&bp->b_hold); | |
161 | } | |
162 | spin_unlock(&btp->bt_lru_lock); | |
163 | } | |
164 | ASSERT(atomic_read(&bp->b_hold) >= 1); | |
165 | } | |
1da177e4 | 166 | |
4347b9d7 CH |
167 | struct xfs_buf * |
168 | xfs_buf_alloc( | |
169 | struct xfs_buftarg *target, | |
e70b73f8 DC |
170 | xfs_daddr_t blkno, |
171 | size_t numblks, | |
ce8e922c | 172 | xfs_buf_flags_t flags) |
1da177e4 | 173 | { |
4347b9d7 CH |
174 | struct xfs_buf *bp; |
175 | ||
aa5c158e | 176 | bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS); |
4347b9d7 CH |
177 | if (unlikely(!bp)) |
178 | return NULL; | |
179 | ||
1da177e4 | 180 | /* |
12bcb3f7 DC |
181 | * We don't want certain flags to appear in b_flags unless they are |
182 | * specifically set by later operations on the buffer. | |
1da177e4 | 183 | */ |
611c9946 | 184 | flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD); |
ce8e922c | 185 | |
ce8e922c | 186 | atomic_set(&bp->b_hold, 1); |
430cbeb8 | 187 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 188 | init_completion(&bp->b_iowait); |
430cbeb8 | 189 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 190 | INIT_LIST_HEAD(&bp->b_list); |
74f75a0c | 191 | RB_CLEAR_NODE(&bp->b_rbnode); |
a731cd11 | 192 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
ce8e922c NS |
193 | XB_SET_OWNER(bp); |
194 | bp->b_target = target; | |
de1cbee4 | 195 | |
1da177e4 | 196 | /* |
aa0e8833 DC |
197 | * Set length and io_length to the same value initially. |
198 | * I/O routines should use io_length, which will be the same in | |
1da177e4 LT |
199 | * most cases but may be reset (e.g. XFS recovery). |
200 | */ | |
4e94b71b | 201 | bp->b_length = numblks; |
aa0e8833 | 202 | bp->b_io_length = numblks; |
ce8e922c | 203 | bp->b_flags = flags; |
59c84ed0 | 204 | bp->b_bn = blkno; |
ce8e922c NS |
205 | atomic_set(&bp->b_pin_count, 0); |
206 | init_waitqueue_head(&bp->b_waiters); | |
207 | ||
208 | XFS_STATS_INC(xb_create); | |
0b1b213f | 209 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 CH |
210 | |
211 | return bp; | |
1da177e4 LT |
212 | } |
213 | ||
214 | /* | |
ce8e922c NS |
215 | * Allocate a page array capable of holding a specified number |
216 | * of pages, and point the page buf at it. | |
1da177e4 LT |
217 | */ |
218 | STATIC int | |
ce8e922c NS |
219 | _xfs_buf_get_pages( |
220 | xfs_buf_t *bp, | |
1da177e4 | 221 | int page_count, |
ce8e922c | 222 | xfs_buf_flags_t flags) |
1da177e4 LT |
223 | { |
224 | /* Make sure that we have a page list */ | |
ce8e922c | 225 | if (bp->b_pages == NULL) { |
ce8e922c NS |
226 | bp->b_page_count = page_count; |
227 | if (page_count <= XB_PAGES) { | |
228 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 229 | } else { |
ce8e922c | 230 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
aa5c158e | 231 | page_count, KM_NOFS); |
ce8e922c | 232 | if (bp->b_pages == NULL) |
1da177e4 LT |
233 | return -ENOMEM; |
234 | } | |
ce8e922c | 235 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
236 | } |
237 | return 0; | |
238 | } | |
239 | ||
240 | /* | |
ce8e922c | 241 | * Frees b_pages if it was allocated. |
1da177e4 LT |
242 | */ |
243 | STATIC void | |
ce8e922c | 244 | _xfs_buf_free_pages( |
1da177e4 LT |
245 | xfs_buf_t *bp) |
246 | { | |
ce8e922c | 247 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 248 | kmem_free(bp->b_pages); |
3fc98b1a | 249 | bp->b_pages = NULL; |
1da177e4 LT |
250 | } |
251 | } | |
252 | ||
253 | /* | |
254 | * Releases the specified buffer. | |
255 | * | |
256 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 257 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
258 | * hashed and refcounted buffers |
259 | */ | |
260 | void | |
ce8e922c | 261 | xfs_buf_free( |
1da177e4 LT |
262 | xfs_buf_t *bp) |
263 | { | |
0b1b213f | 264 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 265 | |
430cbeb8 DC |
266 | ASSERT(list_empty(&bp->b_lru)); |
267 | ||
0e6e847f | 268 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
269 | uint i; |
270 | ||
73c77e2c | 271 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
272 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
273 | bp->b_page_count); | |
1da177e4 | 274 | |
948ecdb4 NS |
275 | for (i = 0; i < bp->b_page_count; i++) { |
276 | struct page *page = bp->b_pages[i]; | |
277 | ||
0e6e847f | 278 | __free_page(page); |
948ecdb4 | 279 | } |
0e6e847f DC |
280 | } else if (bp->b_flags & _XBF_KMEM) |
281 | kmem_free(bp->b_addr); | |
3fc98b1a | 282 | _xfs_buf_free_pages(bp); |
4347b9d7 | 283 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
284 | } |
285 | ||
286 | /* | |
0e6e847f | 287 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
288 | */ |
289 | STATIC int | |
0e6e847f | 290 | xfs_buf_allocate_memory( |
1da177e4 LT |
291 | xfs_buf_t *bp, |
292 | uint flags) | |
293 | { | |
aa0e8833 | 294 | size_t size; |
1da177e4 | 295 | size_t nbytes, offset; |
ce8e922c | 296 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 297 | unsigned short page_count, i; |
795cac72 | 298 | xfs_off_t start, end; |
1da177e4 LT |
299 | int error; |
300 | ||
0e6e847f DC |
301 | /* |
302 | * for buffers that are contained within a single page, just allocate | |
303 | * the memory from the heap - there's no need for the complexity of | |
304 | * page arrays to keep allocation down to order 0. | |
305 | */ | |
795cac72 DC |
306 | size = BBTOB(bp->b_length); |
307 | if (size < PAGE_SIZE) { | |
aa5c158e | 308 | bp->b_addr = kmem_alloc(size, KM_NOFS); |
0e6e847f DC |
309 | if (!bp->b_addr) { |
310 | /* low memory - use alloc_page loop instead */ | |
311 | goto use_alloc_page; | |
312 | } | |
313 | ||
795cac72 | 314 | if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) != |
0e6e847f DC |
315 | ((unsigned long)bp->b_addr & PAGE_MASK)) { |
316 | /* b_addr spans two pages - use alloc_page instead */ | |
317 | kmem_free(bp->b_addr); | |
318 | bp->b_addr = NULL; | |
319 | goto use_alloc_page; | |
320 | } | |
321 | bp->b_offset = offset_in_page(bp->b_addr); | |
322 | bp->b_pages = bp->b_page_array; | |
323 | bp->b_pages[0] = virt_to_page(bp->b_addr); | |
324 | bp->b_page_count = 1; | |
611c9946 | 325 | bp->b_flags |= _XBF_KMEM; |
0e6e847f DC |
326 | return 0; |
327 | } | |
328 | ||
329 | use_alloc_page: | |
795cac72 DC |
330 | start = BBTOB(bp->b_bn) >> PAGE_SHIFT; |
331 | end = (BBTOB(bp->b_bn + bp->b_length) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
332 | page_count = end - start; | |
ce8e922c | 333 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
334 | if (unlikely(error)) |
335 | return error; | |
1da177e4 | 336 | |
ce8e922c | 337 | offset = bp->b_offset; |
0e6e847f | 338 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 339 | |
ce8e922c | 340 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
341 | struct page *page; |
342 | uint retries = 0; | |
0e6e847f DC |
343 | retry: |
344 | page = alloc_page(gfp_mask); | |
1da177e4 | 345 | if (unlikely(page == NULL)) { |
ce8e922c NS |
346 | if (flags & XBF_READ_AHEAD) { |
347 | bp->b_page_count = i; | |
0e6e847f DC |
348 | error = ENOMEM; |
349 | goto out_free_pages; | |
1da177e4 LT |
350 | } |
351 | ||
352 | /* | |
353 | * This could deadlock. | |
354 | * | |
355 | * But until all the XFS lowlevel code is revamped to | |
356 | * handle buffer allocation failures we can't do much. | |
357 | */ | |
358 | if (!(++retries % 100)) | |
4f10700a DC |
359 | xfs_err(NULL, |
360 | "possible memory allocation deadlock in %s (mode:0x%x)", | |
34a622b2 | 361 | __func__, gfp_mask); |
1da177e4 | 362 | |
ce8e922c | 363 | XFS_STATS_INC(xb_page_retries); |
8aa7e847 | 364 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
365 | goto retry; |
366 | } | |
367 | ||
ce8e922c | 368 | XFS_STATS_INC(xb_page_found); |
1da177e4 | 369 | |
0e6e847f | 370 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 371 | size -= nbytes; |
ce8e922c | 372 | bp->b_pages[i] = page; |
1da177e4 LT |
373 | offset = 0; |
374 | } | |
0e6e847f | 375 | return 0; |
1da177e4 | 376 | |
0e6e847f DC |
377 | out_free_pages: |
378 | for (i = 0; i < bp->b_page_count; i++) | |
379 | __free_page(bp->b_pages[i]); | |
1da177e4 LT |
380 | return error; |
381 | } | |
382 | ||
383 | /* | |
25985edc | 384 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
385 | */ |
386 | STATIC int | |
ce8e922c | 387 | _xfs_buf_map_pages( |
1da177e4 LT |
388 | xfs_buf_t *bp, |
389 | uint flags) | |
390 | { | |
0e6e847f | 391 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 392 | if (bp->b_page_count == 1) { |
0e6e847f | 393 | /* A single page buffer is always mappable */ |
ce8e922c | 394 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
611c9946 DC |
395 | } else if (flags & XBF_UNMAPPED) { |
396 | bp->b_addr = NULL; | |
397 | } else { | |
a19fb380 DC |
398 | int retried = 0; |
399 | ||
400 | do { | |
401 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
402 | -1, PAGE_KERNEL); | |
403 | if (bp->b_addr) | |
404 | break; | |
405 | vm_unmap_aliases(); | |
406 | } while (retried++ <= 1); | |
407 | ||
408 | if (!bp->b_addr) | |
1da177e4 | 409 | return -ENOMEM; |
ce8e922c | 410 | bp->b_addr += bp->b_offset; |
1da177e4 LT |
411 | } |
412 | ||
413 | return 0; | |
414 | } | |
415 | ||
416 | /* | |
417 | * Finding and Reading Buffers | |
418 | */ | |
419 | ||
420 | /* | |
ce8e922c | 421 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 | 422 | * a given range of an inode. The buffer is returned |
eabbaf11 | 423 | * locked. No I/O is implied by this call. |
1da177e4 LT |
424 | */ |
425 | xfs_buf_t * | |
ce8e922c | 426 | _xfs_buf_find( |
e70b73f8 DC |
427 | struct xfs_buftarg *btp, |
428 | xfs_daddr_t blkno, | |
429 | size_t numblks, | |
ce8e922c NS |
430 | xfs_buf_flags_t flags, |
431 | xfs_buf_t *new_bp) | |
1da177e4 | 432 | { |
e70b73f8 | 433 | size_t numbytes; |
74f75a0c DC |
434 | struct xfs_perag *pag; |
435 | struct rb_node **rbp; | |
436 | struct rb_node *parent; | |
437 | xfs_buf_t *bp; | |
1da177e4 | 438 | |
e70b73f8 | 439 | numbytes = BBTOB(numblks); |
1da177e4 LT |
440 | |
441 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
e70b73f8 | 442 | ASSERT(!(numbytes < (1 << btp->bt_sshift))); |
de1cbee4 | 443 | ASSERT(!(BBTOB(blkno) & (xfs_off_t)btp->bt_smask)); |
1da177e4 | 444 | |
74f75a0c DC |
445 | /* get tree root */ |
446 | pag = xfs_perag_get(btp->bt_mount, | |
e70b73f8 | 447 | xfs_daddr_to_agno(btp->bt_mount, blkno)); |
74f75a0c DC |
448 | |
449 | /* walk tree */ | |
450 | spin_lock(&pag->pag_buf_lock); | |
451 | rbp = &pag->pag_buf_tree.rb_node; | |
452 | parent = NULL; | |
453 | bp = NULL; | |
454 | while (*rbp) { | |
455 | parent = *rbp; | |
456 | bp = rb_entry(parent, struct xfs_buf, b_rbnode); | |
457 | ||
de1cbee4 | 458 | if (blkno < bp->b_bn) |
74f75a0c | 459 | rbp = &(*rbp)->rb_left; |
de1cbee4 | 460 | else if (blkno > bp->b_bn) |
74f75a0c DC |
461 | rbp = &(*rbp)->rb_right; |
462 | else { | |
463 | /* | |
de1cbee4 | 464 | * found a block number match. If the range doesn't |
74f75a0c DC |
465 | * match, the only way this is allowed is if the buffer |
466 | * in the cache is stale and the transaction that made | |
467 | * it stale has not yet committed. i.e. we are | |
468 | * reallocating a busy extent. Skip this buffer and | |
469 | * continue searching to the right for an exact match. | |
470 | */ | |
4e94b71b | 471 | if (bp->b_length != numblks) { |
74f75a0c DC |
472 | ASSERT(bp->b_flags & XBF_STALE); |
473 | rbp = &(*rbp)->rb_right; | |
474 | continue; | |
475 | } | |
ce8e922c | 476 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
477 | goto found; |
478 | } | |
479 | } | |
480 | ||
481 | /* No match found */ | |
ce8e922c | 482 | if (new_bp) { |
74f75a0c DC |
483 | rb_link_node(&new_bp->b_rbnode, parent, rbp); |
484 | rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree); | |
485 | /* the buffer keeps the perag reference until it is freed */ | |
486 | new_bp->b_pag = pag; | |
487 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 488 | } else { |
ce8e922c | 489 | XFS_STATS_INC(xb_miss_locked); |
74f75a0c DC |
490 | spin_unlock(&pag->pag_buf_lock); |
491 | xfs_perag_put(pag); | |
1da177e4 | 492 | } |
ce8e922c | 493 | return new_bp; |
1da177e4 LT |
494 | |
495 | found: | |
74f75a0c DC |
496 | spin_unlock(&pag->pag_buf_lock); |
497 | xfs_perag_put(pag); | |
1da177e4 | 498 | |
0c842ad4 CH |
499 | if (!xfs_buf_trylock(bp)) { |
500 | if (flags & XBF_TRYLOCK) { | |
ce8e922c NS |
501 | xfs_buf_rele(bp); |
502 | XFS_STATS_INC(xb_busy_locked); | |
503 | return NULL; | |
1da177e4 | 504 | } |
0c842ad4 CH |
505 | xfs_buf_lock(bp); |
506 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
507 | } |
508 | ||
0e6e847f DC |
509 | /* |
510 | * if the buffer is stale, clear all the external state associated with | |
511 | * it. We need to keep flags such as how we allocated the buffer memory | |
512 | * intact here. | |
513 | */ | |
ce8e922c NS |
514 | if (bp->b_flags & XBF_STALE) { |
515 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
611c9946 | 516 | bp->b_flags &= _XBF_KMEM | _XBF_PAGES; |
2f926587 | 517 | } |
0b1b213f CH |
518 | |
519 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ce8e922c NS |
520 | XFS_STATS_INC(xb_get_locked); |
521 | return bp; | |
1da177e4 LT |
522 | } |
523 | ||
524 | /* | |
3815832a DC |
525 | * Assembles a buffer covering the specified range. The code is optimised for |
526 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
527 | * more hits than misses. | |
1da177e4 | 528 | */ |
3815832a | 529 | struct xfs_buf * |
6ad112bf | 530 | xfs_buf_get( |
e70b73f8 DC |
531 | xfs_buftarg_t *target, |
532 | xfs_daddr_t blkno, | |
533 | size_t numblks, | |
ce8e922c | 534 | xfs_buf_flags_t flags) |
1da177e4 | 535 | { |
3815832a DC |
536 | struct xfs_buf *bp; |
537 | struct xfs_buf *new_bp; | |
0e6e847f | 538 | int error = 0; |
1da177e4 | 539 | |
e70b73f8 | 540 | bp = _xfs_buf_find(target, blkno, numblks, flags, NULL); |
3815832a DC |
541 | if (likely(bp)) |
542 | goto found; | |
543 | ||
e70b73f8 | 544 | new_bp = xfs_buf_alloc(target, blkno, numblks, flags); |
ce8e922c | 545 | if (unlikely(!new_bp)) |
1da177e4 LT |
546 | return NULL; |
547 | ||
fe2429b0 DC |
548 | error = xfs_buf_allocate_memory(new_bp, flags); |
549 | if (error) { | |
550 | kmem_zone_free(xfs_buf_zone, new_bp); | |
551 | return NULL; | |
552 | } | |
553 | ||
e70b73f8 | 554 | bp = _xfs_buf_find(target, blkno, numblks, flags, new_bp); |
3815832a | 555 | if (!bp) { |
fe2429b0 | 556 | xfs_buf_free(new_bp); |
3815832a DC |
557 | return NULL; |
558 | } | |
559 | ||
fe2429b0 DC |
560 | if (bp != new_bp) |
561 | xfs_buf_free(new_bp); | |
1da177e4 | 562 | |
aa0e8833 | 563 | bp->b_io_length = bp->b_length; |
3815832a DC |
564 | |
565 | found: | |
611c9946 | 566 | if (!bp->b_addr) { |
ce8e922c | 567 | error = _xfs_buf_map_pages(bp, flags); |
1da177e4 | 568 | if (unlikely(error)) { |
4f10700a DC |
569 | xfs_warn(target->bt_mount, |
570 | "%s: failed to map pages\n", __func__); | |
a8acad70 DC |
571 | xfs_buf_relse(bp); |
572 | return NULL; | |
1da177e4 LT |
573 | } |
574 | } | |
575 | ||
ce8e922c | 576 | XFS_STATS_INC(xb_get); |
0b1b213f | 577 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 578 | return bp; |
1da177e4 LT |
579 | } |
580 | ||
5d765b97 CH |
581 | STATIC int |
582 | _xfs_buf_read( | |
583 | xfs_buf_t *bp, | |
584 | xfs_buf_flags_t flags) | |
585 | { | |
43ff2122 | 586 | ASSERT(!(flags & XBF_WRITE)); |
5d765b97 CH |
587 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); |
588 | ||
43ff2122 | 589 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD); |
1d5ae5df | 590 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); |
5d765b97 | 591 | |
0e95f19a DC |
592 | xfs_buf_iorequest(bp); |
593 | if (flags & XBF_ASYNC) | |
594 | return 0; | |
ec53d1db | 595 | return xfs_buf_iowait(bp); |
5d765b97 CH |
596 | } |
597 | ||
1da177e4 | 598 | xfs_buf_t * |
6ad112bf | 599 | xfs_buf_read( |
1da177e4 | 600 | xfs_buftarg_t *target, |
e70b73f8 DC |
601 | xfs_daddr_t blkno, |
602 | size_t numblks, | |
ce8e922c | 603 | xfs_buf_flags_t flags) |
1da177e4 | 604 | { |
ce8e922c NS |
605 | xfs_buf_t *bp; |
606 | ||
607 | flags |= XBF_READ; | |
608 | ||
e70b73f8 | 609 | bp = xfs_buf_get(target, blkno, numblks, flags); |
ce8e922c | 610 | if (bp) { |
0b1b213f CH |
611 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
612 | ||
ce8e922c | 613 | if (!XFS_BUF_ISDONE(bp)) { |
ce8e922c | 614 | XFS_STATS_INC(xb_get_read); |
5d765b97 | 615 | _xfs_buf_read(bp, flags); |
ce8e922c | 616 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
617 | /* |
618 | * Read ahead call which is already satisfied, | |
619 | * drop the buffer | |
620 | */ | |
a8acad70 DC |
621 | xfs_buf_relse(bp); |
622 | return NULL; | |
1da177e4 | 623 | } else { |
1da177e4 | 624 | /* We do not want read in the flags */ |
ce8e922c | 625 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
626 | } |
627 | } | |
628 | ||
ce8e922c | 629 | return bp; |
1da177e4 LT |
630 | } |
631 | ||
1da177e4 | 632 | /* |
ce8e922c NS |
633 | * If we are not low on memory then do the readahead in a deadlock |
634 | * safe manner. | |
1da177e4 LT |
635 | */ |
636 | void | |
ce8e922c | 637 | xfs_buf_readahead( |
1da177e4 | 638 | xfs_buftarg_t *target, |
e70b73f8 DC |
639 | xfs_daddr_t blkno, |
640 | size_t numblks) | |
1da177e4 | 641 | { |
0e6e847f | 642 | if (bdi_read_congested(target->bt_bdi)) |
1da177e4 LT |
643 | return; |
644 | ||
e70b73f8 | 645 | xfs_buf_read(target, blkno, numblks, |
aa5c158e | 646 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD); |
1da177e4 LT |
647 | } |
648 | ||
5adc94c2 DC |
649 | /* |
650 | * Read an uncached buffer from disk. Allocates and returns a locked | |
651 | * buffer containing the disk contents or nothing. | |
652 | */ | |
653 | struct xfs_buf * | |
654 | xfs_buf_read_uncached( | |
5adc94c2 DC |
655 | struct xfs_buftarg *target, |
656 | xfs_daddr_t daddr, | |
e70b73f8 | 657 | size_t numblks, |
5adc94c2 DC |
658 | int flags) |
659 | { | |
660 | xfs_buf_t *bp; | |
661 | int error; | |
662 | ||
e70b73f8 | 663 | bp = xfs_buf_get_uncached(target, numblks, flags); |
5adc94c2 DC |
664 | if (!bp) |
665 | return NULL; | |
666 | ||
667 | /* set up the buffer for a read IO */ | |
5adc94c2 DC |
668 | XFS_BUF_SET_ADDR(bp, daddr); |
669 | XFS_BUF_READ(bp); | |
5adc94c2 | 670 | |
e70b73f8 | 671 | xfsbdstrat(target->bt_mount, bp); |
1a1a3e97 | 672 | error = xfs_buf_iowait(bp); |
0e95f19a | 673 | if (error) { |
5adc94c2 DC |
674 | xfs_buf_relse(bp); |
675 | return NULL; | |
676 | } | |
677 | return bp; | |
1da177e4 LT |
678 | } |
679 | ||
44396476 DC |
680 | /* |
681 | * Return a buffer allocated as an empty buffer and associated to external | |
682 | * memory via xfs_buf_associate_memory() back to it's empty state. | |
683 | */ | |
684 | void | |
685 | xfs_buf_set_empty( | |
686 | struct xfs_buf *bp, | |
e70b73f8 | 687 | size_t numblks) |
44396476 DC |
688 | { |
689 | if (bp->b_pages) | |
690 | _xfs_buf_free_pages(bp); | |
691 | ||
692 | bp->b_pages = NULL; | |
693 | bp->b_page_count = 0; | |
694 | bp->b_addr = NULL; | |
4e94b71b | 695 | bp->b_length = numblks; |
aa0e8833 | 696 | bp->b_io_length = numblks; |
44396476 | 697 | bp->b_bn = XFS_BUF_DADDR_NULL; |
44396476 DC |
698 | } |
699 | ||
1da177e4 LT |
700 | static inline struct page * |
701 | mem_to_page( | |
702 | void *addr) | |
703 | { | |
9e2779fa | 704 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
705 | return virt_to_page(addr); |
706 | } else { | |
707 | return vmalloc_to_page(addr); | |
708 | } | |
709 | } | |
710 | ||
711 | int | |
ce8e922c NS |
712 | xfs_buf_associate_memory( |
713 | xfs_buf_t *bp, | |
1da177e4 LT |
714 | void *mem, |
715 | size_t len) | |
716 | { | |
717 | int rval; | |
718 | int i = 0; | |
d1afb678 LM |
719 | unsigned long pageaddr; |
720 | unsigned long offset; | |
721 | size_t buflen; | |
1da177e4 LT |
722 | int page_count; |
723 | ||
0e6e847f | 724 | pageaddr = (unsigned long)mem & PAGE_MASK; |
d1afb678 | 725 | offset = (unsigned long)mem - pageaddr; |
0e6e847f DC |
726 | buflen = PAGE_ALIGN(len + offset); |
727 | page_count = buflen >> PAGE_SHIFT; | |
1da177e4 LT |
728 | |
729 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
730 | if (bp->b_pages) |
731 | _xfs_buf_free_pages(bp); | |
1da177e4 | 732 | |
ce8e922c NS |
733 | bp->b_pages = NULL; |
734 | bp->b_addr = mem; | |
1da177e4 | 735 | |
aa5c158e | 736 | rval = _xfs_buf_get_pages(bp, page_count, 0); |
1da177e4 LT |
737 | if (rval) |
738 | return rval; | |
739 | ||
ce8e922c | 740 | bp->b_offset = offset; |
d1afb678 LM |
741 | |
742 | for (i = 0; i < bp->b_page_count; i++) { | |
743 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
0e6e847f | 744 | pageaddr += PAGE_SIZE; |
1da177e4 | 745 | } |
1da177e4 | 746 | |
aa0e8833 | 747 | bp->b_io_length = BTOBB(len); |
4e94b71b | 748 | bp->b_length = BTOBB(buflen); |
1da177e4 LT |
749 | |
750 | return 0; | |
751 | } | |
752 | ||
753 | xfs_buf_t * | |
686865f7 DC |
754 | xfs_buf_get_uncached( |
755 | struct xfs_buftarg *target, | |
e70b73f8 | 756 | size_t numblks, |
686865f7 | 757 | int flags) |
1da177e4 | 758 | { |
e70b73f8 | 759 | unsigned long page_count; |
1fa40b01 | 760 | int error, i; |
1da177e4 | 761 | xfs_buf_t *bp; |
1da177e4 | 762 | |
59c84ed0 | 763 | bp = xfs_buf_alloc(target, XFS_BUF_DADDR_NULL, numblks, 0); |
1da177e4 LT |
764 | if (unlikely(bp == NULL)) |
765 | goto fail; | |
1da177e4 | 766 | |
e70b73f8 | 767 | page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT; |
1fa40b01 CH |
768 | error = _xfs_buf_get_pages(bp, page_count, 0); |
769 | if (error) | |
1da177e4 LT |
770 | goto fail_free_buf; |
771 | ||
1fa40b01 | 772 | for (i = 0; i < page_count; i++) { |
686865f7 | 773 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
774 | if (!bp->b_pages[i]) |
775 | goto fail_free_mem; | |
1da177e4 | 776 | } |
1fa40b01 | 777 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 778 | |
611c9946 | 779 | error = _xfs_buf_map_pages(bp, 0); |
1fa40b01 | 780 | if (unlikely(error)) { |
4f10700a DC |
781 | xfs_warn(target->bt_mount, |
782 | "%s: failed to map pages\n", __func__); | |
1da177e4 | 783 | goto fail_free_mem; |
1fa40b01 | 784 | } |
1da177e4 | 785 | |
686865f7 | 786 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 787 | return bp; |
1fa40b01 | 788 | |
1da177e4 | 789 | fail_free_mem: |
1fa40b01 CH |
790 | while (--i >= 0) |
791 | __free_page(bp->b_pages[i]); | |
ca165b88 | 792 | _xfs_buf_free_pages(bp); |
1da177e4 | 793 | fail_free_buf: |
4347b9d7 | 794 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
795 | fail: |
796 | return NULL; | |
797 | } | |
798 | ||
799 | /* | |
1da177e4 LT |
800 | * Increment reference count on buffer, to hold the buffer concurrently |
801 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
802 | * Must hold the buffer already to call this function. |
803 | */ | |
804 | void | |
ce8e922c NS |
805 | xfs_buf_hold( |
806 | xfs_buf_t *bp) | |
1da177e4 | 807 | { |
0b1b213f | 808 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 809 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
810 | } |
811 | ||
812 | /* | |
ce8e922c NS |
813 | * Releases a hold on the specified buffer. If the |
814 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
815 | */ |
816 | void | |
ce8e922c NS |
817 | xfs_buf_rele( |
818 | xfs_buf_t *bp) | |
1da177e4 | 819 | { |
74f75a0c | 820 | struct xfs_perag *pag = bp->b_pag; |
1da177e4 | 821 | |
0b1b213f | 822 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 823 | |
74f75a0c | 824 | if (!pag) { |
430cbeb8 | 825 | ASSERT(list_empty(&bp->b_lru)); |
74f75a0c | 826 | ASSERT(RB_EMPTY_NODE(&bp->b_rbnode)); |
fad3aa1e NS |
827 | if (atomic_dec_and_test(&bp->b_hold)) |
828 | xfs_buf_free(bp); | |
829 | return; | |
830 | } | |
831 | ||
74f75a0c | 832 | ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode)); |
430cbeb8 | 833 | |
3790689f | 834 | ASSERT(atomic_read(&bp->b_hold) > 0); |
74f75a0c | 835 | if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) { |
bfc60177 | 836 | if (!(bp->b_flags & XBF_STALE) && |
430cbeb8 DC |
837 | atomic_read(&bp->b_lru_ref)) { |
838 | xfs_buf_lru_add(bp); | |
839 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 840 | } else { |
430cbeb8 | 841 | xfs_buf_lru_del(bp); |
43ff2122 | 842 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
74f75a0c DC |
843 | rb_erase(&bp->b_rbnode, &pag->pag_buf_tree); |
844 | spin_unlock(&pag->pag_buf_lock); | |
845 | xfs_perag_put(pag); | |
ce8e922c | 846 | xfs_buf_free(bp); |
1da177e4 LT |
847 | } |
848 | } | |
849 | } | |
850 | ||
851 | ||
852 | /* | |
0e6e847f | 853 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
854 | * |
855 | * If we come across a stale, pinned, locked buffer, we know that we are | |
856 | * being asked to lock a buffer that has been reallocated. Because it is | |
857 | * pinned, we know that the log has not been pushed to disk and hence it | |
858 | * will still be locked. Rather than continuing to have trylock attempts | |
859 | * fail until someone else pushes the log, push it ourselves before | |
860 | * returning. This means that the xfsaild will not get stuck trying | |
861 | * to push on stale inode buffers. | |
1da177e4 LT |
862 | */ |
863 | int | |
0c842ad4 CH |
864 | xfs_buf_trylock( |
865 | struct xfs_buf *bp) | |
1da177e4 LT |
866 | { |
867 | int locked; | |
868 | ||
ce8e922c | 869 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 870 | if (locked) |
ce8e922c | 871 | XB_SET_OWNER(bp); |
90810b9e DC |
872 | else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
873 | xfs_log_force(bp->b_target->bt_mount, 0); | |
0b1b213f | 874 | |
0c842ad4 CH |
875 | trace_xfs_buf_trylock(bp, _RET_IP_); |
876 | return locked; | |
1da177e4 | 877 | } |
1da177e4 LT |
878 | |
879 | /* | |
0e6e847f | 880 | * Lock a buffer object. |
ed3b4d6c DC |
881 | * |
882 | * If we come across a stale, pinned, locked buffer, we know that we | |
883 | * are being asked to lock a buffer that has been reallocated. Because | |
884 | * it is pinned, we know that the log has not been pushed to disk and | |
885 | * hence it will still be locked. Rather than sleeping until someone | |
886 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 887 | */ |
ce8e922c NS |
888 | void |
889 | xfs_buf_lock( | |
0c842ad4 | 890 | struct xfs_buf *bp) |
1da177e4 | 891 | { |
0b1b213f CH |
892 | trace_xfs_buf_lock(bp, _RET_IP_); |
893 | ||
ed3b4d6c | 894 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 895 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
896 | down(&bp->b_sema); |
897 | XB_SET_OWNER(bp); | |
0b1b213f CH |
898 | |
899 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
900 | } |
901 | ||
1da177e4 | 902 | void |
ce8e922c | 903 | xfs_buf_unlock( |
0c842ad4 | 904 | struct xfs_buf *bp) |
1da177e4 | 905 | { |
ce8e922c NS |
906 | XB_CLEAR_OWNER(bp); |
907 | up(&bp->b_sema); | |
0b1b213f CH |
908 | |
909 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
910 | } |
911 | ||
ce8e922c NS |
912 | STATIC void |
913 | xfs_buf_wait_unpin( | |
914 | xfs_buf_t *bp) | |
1da177e4 LT |
915 | { |
916 | DECLARE_WAITQUEUE (wait, current); | |
917 | ||
ce8e922c | 918 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
919 | return; |
920 | ||
ce8e922c | 921 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
922 | for (;;) { |
923 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 924 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 925 | break; |
7eaceacc | 926 | io_schedule(); |
1da177e4 | 927 | } |
ce8e922c | 928 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
929 | set_current_state(TASK_RUNNING); |
930 | } | |
931 | ||
932 | /* | |
933 | * Buffer Utility Routines | |
934 | */ | |
935 | ||
1da177e4 | 936 | STATIC void |
ce8e922c | 937 | xfs_buf_iodone_work( |
c4028958 | 938 | struct work_struct *work) |
1da177e4 | 939 | { |
c4028958 DH |
940 | xfs_buf_t *bp = |
941 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 942 | |
80f6c29d | 943 | if (bp->b_iodone) |
ce8e922c NS |
944 | (*(bp->b_iodone))(bp); |
945 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
946 | xfs_buf_relse(bp); |
947 | } | |
948 | ||
949 | void | |
ce8e922c NS |
950 | xfs_buf_ioend( |
951 | xfs_buf_t *bp, | |
1da177e4 LT |
952 | int schedule) |
953 | { | |
0b1b213f CH |
954 | trace_xfs_buf_iodone(bp, _RET_IP_); |
955 | ||
77be55a5 | 956 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
ce8e922c NS |
957 | if (bp->b_error == 0) |
958 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 959 | |
ce8e922c | 960 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 961 | if (schedule) { |
c4028958 | 962 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 963 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 964 | } else { |
c4028958 | 965 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
966 | } |
967 | } else { | |
b4dd330b | 968 | complete(&bp->b_iowait); |
1da177e4 LT |
969 | } |
970 | } | |
971 | ||
1da177e4 | 972 | void |
ce8e922c NS |
973 | xfs_buf_ioerror( |
974 | xfs_buf_t *bp, | |
975 | int error) | |
1da177e4 LT |
976 | { |
977 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c | 978 | bp->b_error = (unsigned short)error; |
0b1b213f | 979 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
980 | } |
981 | ||
901796af CH |
982 | void |
983 | xfs_buf_ioerror_alert( | |
984 | struct xfs_buf *bp, | |
985 | const char *func) | |
986 | { | |
987 | xfs_alert(bp->b_target->bt_mount, | |
aa0e8833 DC |
988 | "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d", |
989 | (__uint64_t)XFS_BUF_ADDR(bp), func, bp->b_error, bp->b_length); | |
901796af CH |
990 | } |
991 | ||
1da177e4 | 992 | int |
64e0bc7d | 993 | xfs_bwrite( |
5d765b97 | 994 | struct xfs_buf *bp) |
1da177e4 | 995 | { |
8c38366f | 996 | int error; |
1da177e4 | 997 | |
43ff2122 CH |
998 | ASSERT(xfs_buf_islocked(bp)); |
999 | ||
64e0bc7d | 1000 | bp->b_flags |= XBF_WRITE; |
43ff2122 | 1001 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q); |
1da177e4 | 1002 | |
939d723b | 1003 | xfs_bdstrat_cb(bp); |
1da177e4 | 1004 | |
8c38366f | 1005 | error = xfs_buf_iowait(bp); |
c2b006c1 CH |
1006 | if (error) { |
1007 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1008 | SHUTDOWN_META_IO_ERROR); | |
1009 | } | |
64e0bc7d | 1010 | return error; |
5d765b97 | 1011 | } |
1da177e4 | 1012 | |
4e23471a CH |
1013 | /* |
1014 | * Called when we want to stop a buffer from getting written or read. | |
1a1a3e97 | 1015 | * We attach the EIO error, muck with its flags, and call xfs_buf_ioend |
4e23471a CH |
1016 | * so that the proper iodone callbacks get called. |
1017 | */ | |
1018 | STATIC int | |
1019 | xfs_bioerror( | |
1020 | xfs_buf_t *bp) | |
1021 | { | |
1022 | #ifdef XFSERRORDEBUG | |
1023 | ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone); | |
1024 | #endif | |
1025 | ||
1026 | /* | |
1027 | * No need to wait until the buffer is unpinned, we aren't flushing it. | |
1028 | */ | |
5a52c2a5 | 1029 | xfs_buf_ioerror(bp, EIO); |
4e23471a CH |
1030 | |
1031 | /* | |
1a1a3e97 | 1032 | * We're calling xfs_buf_ioend, so delete XBF_DONE flag. |
4e23471a CH |
1033 | */ |
1034 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1035 | XFS_BUF_UNDONE(bp); |
c867cb61 | 1036 | xfs_buf_stale(bp); |
4e23471a | 1037 | |
1a1a3e97 | 1038 | xfs_buf_ioend(bp, 0); |
4e23471a CH |
1039 | |
1040 | return EIO; | |
1041 | } | |
1042 | ||
1043 | /* | |
1044 | * Same as xfs_bioerror, except that we are releasing the buffer | |
1a1a3e97 | 1045 | * here ourselves, and avoiding the xfs_buf_ioend call. |
4e23471a CH |
1046 | * This is meant for userdata errors; metadata bufs come with |
1047 | * iodone functions attached, so that we can track down errors. | |
1048 | */ | |
1049 | STATIC int | |
1050 | xfs_bioerror_relse( | |
1051 | struct xfs_buf *bp) | |
1052 | { | |
ed43233b | 1053 | int64_t fl = bp->b_flags; |
4e23471a CH |
1054 | /* |
1055 | * No need to wait until the buffer is unpinned. | |
1056 | * We aren't flushing it. | |
1057 | * | |
1058 | * chunkhold expects B_DONE to be set, whether | |
1059 | * we actually finish the I/O or not. We don't want to | |
1060 | * change that interface. | |
1061 | */ | |
1062 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1063 | XFS_BUF_DONE(bp); |
c867cb61 | 1064 | xfs_buf_stale(bp); |
cb669ca5 | 1065 | bp->b_iodone = NULL; |
0cadda1c | 1066 | if (!(fl & XBF_ASYNC)) { |
4e23471a CH |
1067 | /* |
1068 | * Mark b_error and B_ERROR _both_. | |
1069 | * Lot's of chunkcache code assumes that. | |
1070 | * There's no reason to mark error for | |
1071 | * ASYNC buffers. | |
1072 | */ | |
5a52c2a5 | 1073 | xfs_buf_ioerror(bp, EIO); |
5fde0326 | 1074 | complete(&bp->b_iowait); |
4e23471a CH |
1075 | } else { |
1076 | xfs_buf_relse(bp); | |
1077 | } | |
1078 | ||
1079 | return EIO; | |
1080 | } | |
1081 | ||
1082 | ||
1083 | /* | |
1084 | * All xfs metadata buffers except log state machine buffers | |
1085 | * get this attached as their b_bdstrat callback function. | |
1086 | * This is so that we can catch a buffer | |
1087 | * after prematurely unpinning it to forcibly shutdown the filesystem. | |
1088 | */ | |
1089 | int | |
1090 | xfs_bdstrat_cb( | |
1091 | struct xfs_buf *bp) | |
1092 | { | |
ebad861b | 1093 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
4e23471a CH |
1094 | trace_xfs_bdstrat_shut(bp, _RET_IP_); |
1095 | /* | |
1096 | * Metadata write that didn't get logged but | |
1097 | * written delayed anyway. These aren't associated | |
1098 | * with a transaction, and can be ignored. | |
1099 | */ | |
1100 | if (!bp->b_iodone && !XFS_BUF_ISREAD(bp)) | |
1101 | return xfs_bioerror_relse(bp); | |
1102 | else | |
1103 | return xfs_bioerror(bp); | |
1104 | } | |
1105 | ||
1106 | xfs_buf_iorequest(bp); | |
1107 | return 0; | |
1108 | } | |
1109 | ||
1110 | /* | |
1111 | * Wrapper around bdstrat so that we can stop data from going to disk in case | |
1112 | * we are shutting down the filesystem. Typically user data goes thru this | |
1113 | * path; one of the exceptions is the superblock. | |
1114 | */ | |
1115 | void | |
1116 | xfsbdstrat( | |
1117 | struct xfs_mount *mp, | |
1118 | struct xfs_buf *bp) | |
1119 | { | |
1120 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1121 | trace_xfs_bdstrat_shut(bp, _RET_IP_); | |
1122 | xfs_bioerror_relse(bp); | |
1123 | return; | |
1124 | } | |
1125 | ||
1126 | xfs_buf_iorequest(bp); | |
1127 | } | |
1128 | ||
b8f82a4a | 1129 | STATIC void |
ce8e922c NS |
1130 | _xfs_buf_ioend( |
1131 | xfs_buf_t *bp, | |
1da177e4 LT |
1132 | int schedule) |
1133 | { | |
0e6e847f | 1134 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
ce8e922c | 1135 | xfs_buf_ioend(bp, schedule); |
1da177e4 LT |
1136 | } |
1137 | ||
782e3b3b | 1138 | STATIC void |
ce8e922c | 1139 | xfs_buf_bio_end_io( |
1da177e4 | 1140 | struct bio *bio, |
1da177e4 LT |
1141 | int error) |
1142 | { | |
ce8e922c | 1143 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1da177e4 | 1144 | |
cfbe5267 | 1145 | xfs_buf_ioerror(bp, -error); |
1da177e4 | 1146 | |
73c77e2c JB |
1147 | if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
1148 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); | |
1149 | ||
ce8e922c | 1150 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1151 | bio_put(bio); |
1da177e4 LT |
1152 | } |
1153 | ||
1154 | STATIC void | |
ce8e922c NS |
1155 | _xfs_buf_ioapply( |
1156 | xfs_buf_t *bp) | |
1da177e4 | 1157 | { |
a9759f2d | 1158 | int rw, map_i, total_nr_pages, nr_pages; |
1da177e4 | 1159 | struct bio *bio; |
ce8e922c | 1160 | int offset = bp->b_offset; |
aa0e8833 | 1161 | int size = BBTOB(bp->b_io_length); |
ce8e922c | 1162 | sector_t sector = bp->b_bn; |
1da177e4 | 1163 | |
ce8e922c | 1164 | total_nr_pages = bp->b_page_count; |
1da177e4 LT |
1165 | map_i = 0; |
1166 | ||
1d5ae5df CH |
1167 | if (bp->b_flags & XBF_WRITE) { |
1168 | if (bp->b_flags & XBF_SYNCIO) | |
1169 | rw = WRITE_SYNC; | |
1170 | else | |
1171 | rw = WRITE; | |
1172 | if (bp->b_flags & XBF_FUA) | |
1173 | rw |= REQ_FUA; | |
1174 | if (bp->b_flags & XBF_FLUSH) | |
1175 | rw |= REQ_FLUSH; | |
1176 | } else if (bp->b_flags & XBF_READ_AHEAD) { | |
1177 | rw = READA; | |
51bdd706 | 1178 | } else { |
1d5ae5df | 1179 | rw = READ; |
f538d4da CH |
1180 | } |
1181 | ||
34951f5c CH |
1182 | /* we only use the buffer cache for meta-data */ |
1183 | rw |= REQ_META; | |
1184 | ||
1da177e4 | 1185 | next_chunk: |
ce8e922c | 1186 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1187 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1188 | if (nr_pages > total_nr_pages) | |
1189 | nr_pages = total_nr_pages; | |
1190 | ||
1191 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1192 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1193 | bio->bi_sector = sector; |
ce8e922c NS |
1194 | bio->bi_end_io = xfs_buf_bio_end_io; |
1195 | bio->bi_private = bp; | |
1da177e4 | 1196 | |
0e6e847f | 1197 | |
1da177e4 | 1198 | for (; size && nr_pages; nr_pages--, map_i++) { |
0e6e847f | 1199 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1200 | |
1201 | if (nbytes > size) | |
1202 | nbytes = size; | |
1203 | ||
ce8e922c NS |
1204 | rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset); |
1205 | if (rbytes < nbytes) | |
1da177e4 LT |
1206 | break; |
1207 | ||
1208 | offset = 0; | |
aa0e8833 | 1209 | sector += BTOBB(nbytes); |
1da177e4 LT |
1210 | size -= nbytes; |
1211 | total_nr_pages--; | |
1212 | } | |
1213 | ||
1da177e4 | 1214 | if (likely(bio->bi_size)) { |
73c77e2c JB |
1215 | if (xfs_buf_is_vmapped(bp)) { |
1216 | flush_kernel_vmap_range(bp->b_addr, | |
1217 | xfs_buf_vmap_len(bp)); | |
1218 | } | |
1da177e4 LT |
1219 | submit_bio(rw, bio); |
1220 | if (size) | |
1221 | goto next_chunk; | |
1222 | } else { | |
ce8e922c | 1223 | xfs_buf_ioerror(bp, EIO); |
ec53d1db | 1224 | bio_put(bio); |
1da177e4 LT |
1225 | } |
1226 | } | |
1227 | ||
0e95f19a | 1228 | void |
ce8e922c NS |
1229 | xfs_buf_iorequest( |
1230 | xfs_buf_t *bp) | |
1da177e4 | 1231 | { |
0b1b213f | 1232 | trace_xfs_buf_iorequest(bp, _RET_IP_); |
1da177e4 | 1233 | |
43ff2122 | 1234 | ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); |
1da177e4 | 1235 | |
375ec69d | 1236 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1237 | xfs_buf_wait_unpin(bp); |
ce8e922c | 1238 | xfs_buf_hold(bp); |
1da177e4 LT |
1239 | |
1240 | /* Set the count to 1 initially, this will stop an I/O | |
1241 | * completion callout which happens before we have started | |
ce8e922c | 1242 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1243 | */ |
ce8e922c NS |
1244 | atomic_set(&bp->b_io_remaining, 1); |
1245 | _xfs_buf_ioapply(bp); | |
1246 | _xfs_buf_ioend(bp, 0); | |
1da177e4 | 1247 | |
ce8e922c | 1248 | xfs_buf_rele(bp); |
1da177e4 LT |
1249 | } |
1250 | ||
1251 | /* | |
0e95f19a DC |
1252 | * Waits for I/O to complete on the buffer supplied. It returns immediately if |
1253 | * no I/O is pending or there is already a pending error on the buffer. It | |
1254 | * returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1255 | */ |
1256 | int | |
ce8e922c NS |
1257 | xfs_buf_iowait( |
1258 | xfs_buf_t *bp) | |
1da177e4 | 1259 | { |
0b1b213f CH |
1260 | trace_xfs_buf_iowait(bp, _RET_IP_); |
1261 | ||
0e95f19a DC |
1262 | if (!bp->b_error) |
1263 | wait_for_completion(&bp->b_iowait); | |
0b1b213f CH |
1264 | |
1265 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
ce8e922c | 1266 | return bp->b_error; |
1da177e4 LT |
1267 | } |
1268 | ||
ce8e922c NS |
1269 | xfs_caddr_t |
1270 | xfs_buf_offset( | |
1271 | xfs_buf_t *bp, | |
1da177e4 LT |
1272 | size_t offset) |
1273 | { | |
1274 | struct page *page; | |
1275 | ||
611c9946 | 1276 | if (bp->b_addr) |
62926044 | 1277 | return bp->b_addr + offset; |
1da177e4 | 1278 | |
ce8e922c | 1279 | offset += bp->b_offset; |
0e6e847f DC |
1280 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
1281 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_SIZE-1)); | |
1da177e4 LT |
1282 | } |
1283 | ||
1284 | /* | |
1da177e4 LT |
1285 | * Move data into or out of a buffer. |
1286 | */ | |
1287 | void | |
ce8e922c NS |
1288 | xfs_buf_iomove( |
1289 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1290 | size_t boff, /* starting buffer offset */ |
1291 | size_t bsize, /* length to copy */ | |
b9c48649 | 1292 | void *data, /* data address */ |
ce8e922c | 1293 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 | 1294 | { |
795cac72 | 1295 | size_t bend; |
1da177e4 LT |
1296 | |
1297 | bend = boff + bsize; | |
1298 | while (boff < bend) { | |
795cac72 DC |
1299 | struct page *page; |
1300 | int page_index, page_offset, csize; | |
1301 | ||
1302 | page_index = (boff + bp->b_offset) >> PAGE_SHIFT; | |
1303 | page_offset = (boff + bp->b_offset) & ~PAGE_MASK; | |
1304 | page = bp->b_pages[page_index]; | |
1305 | csize = min_t(size_t, PAGE_SIZE - page_offset, | |
1306 | BBTOB(bp->b_io_length) - boff); | |
1da177e4 | 1307 | |
795cac72 | 1308 | ASSERT((csize + page_offset) <= PAGE_SIZE); |
1da177e4 LT |
1309 | |
1310 | switch (mode) { | |
ce8e922c | 1311 | case XBRW_ZERO: |
795cac72 | 1312 | memset(page_address(page) + page_offset, 0, csize); |
1da177e4 | 1313 | break; |
ce8e922c | 1314 | case XBRW_READ: |
795cac72 | 1315 | memcpy(data, page_address(page) + page_offset, csize); |
1da177e4 | 1316 | break; |
ce8e922c | 1317 | case XBRW_WRITE: |
795cac72 | 1318 | memcpy(page_address(page) + page_offset, data, csize); |
1da177e4 LT |
1319 | } |
1320 | ||
1321 | boff += csize; | |
1322 | data += csize; | |
1323 | } | |
1324 | } | |
1325 | ||
1326 | /* | |
ce8e922c | 1327 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1328 | */ |
1329 | ||
1330 | /* | |
430cbeb8 DC |
1331 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1332 | * returned. These buffers will have an elevated hold count, so wait on those | |
1333 | * while freeing all the buffers only held by the LRU. | |
1da177e4 LT |
1334 | */ |
1335 | void | |
1336 | xfs_wait_buftarg( | |
74f75a0c | 1337 | struct xfs_buftarg *btp) |
1da177e4 | 1338 | { |
430cbeb8 DC |
1339 | struct xfs_buf *bp; |
1340 | ||
1341 | restart: | |
1342 | spin_lock(&btp->bt_lru_lock); | |
1343 | while (!list_empty(&btp->bt_lru)) { | |
1344 | bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru); | |
1345 | if (atomic_read(&bp->b_hold) > 1) { | |
1346 | spin_unlock(&btp->bt_lru_lock); | |
26af6552 | 1347 | delay(100); |
430cbeb8 | 1348 | goto restart; |
1da177e4 | 1349 | } |
430cbeb8 | 1350 | /* |
90802ed9 | 1351 | * clear the LRU reference count so the buffer doesn't get |
430cbeb8 DC |
1352 | * ignored in xfs_buf_rele(). |
1353 | */ | |
1354 | atomic_set(&bp->b_lru_ref, 0); | |
1355 | spin_unlock(&btp->bt_lru_lock); | |
1356 | xfs_buf_rele(bp); | |
1357 | spin_lock(&btp->bt_lru_lock); | |
1da177e4 | 1358 | } |
430cbeb8 | 1359 | spin_unlock(&btp->bt_lru_lock); |
1da177e4 LT |
1360 | } |
1361 | ||
ff57ab21 DC |
1362 | int |
1363 | xfs_buftarg_shrink( | |
1364 | struct shrinker *shrink, | |
1495f230 | 1365 | struct shrink_control *sc) |
a6867a68 | 1366 | { |
ff57ab21 DC |
1367 | struct xfs_buftarg *btp = container_of(shrink, |
1368 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1369 | struct xfs_buf *bp; |
1495f230 | 1370 | int nr_to_scan = sc->nr_to_scan; |
430cbeb8 DC |
1371 | LIST_HEAD(dispose); |
1372 | ||
1373 | if (!nr_to_scan) | |
1374 | return btp->bt_lru_nr; | |
1375 | ||
1376 | spin_lock(&btp->bt_lru_lock); | |
1377 | while (!list_empty(&btp->bt_lru)) { | |
1378 | if (nr_to_scan-- <= 0) | |
1379 | break; | |
1380 | ||
1381 | bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru); | |
1382 | ||
1383 | /* | |
1384 | * Decrement the b_lru_ref count unless the value is already | |
1385 | * zero. If the value is already zero, we need to reclaim the | |
1386 | * buffer, otherwise it gets another trip through the LRU. | |
1387 | */ | |
1388 | if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) { | |
1389 | list_move_tail(&bp->b_lru, &btp->bt_lru); | |
1390 | continue; | |
1391 | } | |
1392 | ||
1393 | /* | |
1394 | * remove the buffer from the LRU now to avoid needing another | |
1395 | * lock round trip inside xfs_buf_rele(). | |
1396 | */ | |
1397 | list_move(&bp->b_lru, &dispose); | |
1398 | btp->bt_lru_nr--; | |
ff57ab21 | 1399 | } |
430cbeb8 DC |
1400 | spin_unlock(&btp->bt_lru_lock); |
1401 | ||
1402 | while (!list_empty(&dispose)) { | |
1403 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1404 | list_del_init(&bp->b_lru); | |
1405 | xfs_buf_rele(bp); | |
1406 | } | |
1407 | ||
1408 | return btp->bt_lru_nr; | |
a6867a68 DC |
1409 | } |
1410 | ||
1da177e4 LT |
1411 | void |
1412 | xfs_free_buftarg( | |
b7963133 CH |
1413 | struct xfs_mount *mp, |
1414 | struct xfs_buftarg *btp) | |
1da177e4 | 1415 | { |
ff57ab21 DC |
1416 | unregister_shrinker(&btp->bt_shrinker); |
1417 | ||
b7963133 CH |
1418 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1419 | xfs_blkdev_issue_flush(btp); | |
a6867a68 | 1420 | |
f0e2d93c | 1421 | kmem_free(btp); |
1da177e4 LT |
1422 | } |
1423 | ||
1da177e4 LT |
1424 | STATIC int |
1425 | xfs_setsize_buftarg_flags( | |
1426 | xfs_buftarg_t *btp, | |
1427 | unsigned int blocksize, | |
1428 | unsigned int sectorsize, | |
1429 | int verbose) | |
1430 | { | |
ce8e922c NS |
1431 | btp->bt_bsize = blocksize; |
1432 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1433 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1434 | |
ce8e922c | 1435 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
02b102df CH |
1436 | char name[BDEVNAME_SIZE]; |
1437 | ||
1438 | bdevname(btp->bt_bdev, name); | |
1439 | ||
4f10700a DC |
1440 | xfs_warn(btp->bt_mount, |
1441 | "Cannot set_blocksize to %u on device %s\n", | |
02b102df | 1442 | sectorsize, name); |
1da177e4 LT |
1443 | return EINVAL; |
1444 | } | |
1445 | ||
1da177e4 LT |
1446 | return 0; |
1447 | } | |
1448 | ||
1449 | /* | |
ce8e922c NS |
1450 | * When allocating the initial buffer target we have not yet |
1451 | * read in the superblock, so don't know what sized sectors | |
1452 | * are being used is at this early stage. Play safe. | |
1453 | */ | |
1da177e4 LT |
1454 | STATIC int |
1455 | xfs_setsize_buftarg_early( | |
1456 | xfs_buftarg_t *btp, | |
1457 | struct block_device *bdev) | |
1458 | { | |
1459 | return xfs_setsize_buftarg_flags(btp, | |
0e6e847f | 1460 | PAGE_SIZE, bdev_logical_block_size(bdev), 0); |
1da177e4 LT |
1461 | } |
1462 | ||
1463 | int | |
1464 | xfs_setsize_buftarg( | |
1465 | xfs_buftarg_t *btp, | |
1466 | unsigned int blocksize, | |
1467 | unsigned int sectorsize) | |
1468 | { | |
1469 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1470 | } | |
1471 | ||
1da177e4 LT |
1472 | xfs_buftarg_t * |
1473 | xfs_alloc_buftarg( | |
ebad861b | 1474 | struct xfs_mount *mp, |
1da177e4 | 1475 | struct block_device *bdev, |
e2a07812 JE |
1476 | int external, |
1477 | const char *fsname) | |
1da177e4 LT |
1478 | { |
1479 | xfs_buftarg_t *btp; | |
1480 | ||
1481 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1482 | ||
ebad861b | 1483 | btp->bt_mount = mp; |
ce8e922c NS |
1484 | btp->bt_dev = bdev->bd_dev; |
1485 | btp->bt_bdev = bdev; | |
0e6e847f DC |
1486 | btp->bt_bdi = blk_get_backing_dev_info(bdev); |
1487 | if (!btp->bt_bdi) | |
1488 | goto error; | |
1489 | ||
430cbeb8 DC |
1490 | INIT_LIST_HEAD(&btp->bt_lru); |
1491 | spin_lock_init(&btp->bt_lru_lock); | |
1da177e4 LT |
1492 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1493 | goto error; | |
ff57ab21 DC |
1494 | btp->bt_shrinker.shrink = xfs_buftarg_shrink; |
1495 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; | |
1496 | register_shrinker(&btp->bt_shrinker); | |
1da177e4 LT |
1497 | return btp; |
1498 | ||
1499 | error: | |
f0e2d93c | 1500 | kmem_free(btp); |
1da177e4 LT |
1501 | return NULL; |
1502 | } | |
1503 | ||
1da177e4 | 1504 | /* |
43ff2122 CH |
1505 | * Add a buffer to the delayed write list. |
1506 | * | |
1507 | * This queues a buffer for writeout if it hasn't already been. Note that | |
1508 | * neither this routine nor the buffer list submission functions perform | |
1509 | * any internal synchronization. It is expected that the lists are thread-local | |
1510 | * to the callers. | |
1511 | * | |
1512 | * Returns true if we queued up the buffer, or false if it already had | |
1513 | * been on the buffer list. | |
1da177e4 | 1514 | */ |
43ff2122 | 1515 | bool |
ce8e922c | 1516 | xfs_buf_delwri_queue( |
43ff2122 CH |
1517 | struct xfs_buf *bp, |
1518 | struct list_head *list) | |
1da177e4 | 1519 | { |
43ff2122 | 1520 | ASSERT(xfs_buf_islocked(bp)); |
5a8ee6ba | 1521 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1522 | |
43ff2122 CH |
1523 | /* |
1524 | * If the buffer is already marked delwri it already is queued up | |
1525 | * by someone else for imediate writeout. Just ignore it in that | |
1526 | * case. | |
1527 | */ | |
1528 | if (bp->b_flags & _XBF_DELWRI_Q) { | |
1529 | trace_xfs_buf_delwri_queued(bp, _RET_IP_); | |
1530 | return false; | |
1da177e4 | 1531 | } |
1da177e4 | 1532 | |
43ff2122 | 1533 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
d808f617 DC |
1534 | |
1535 | /* | |
43ff2122 CH |
1536 | * If a buffer gets written out synchronously or marked stale while it |
1537 | * is on a delwri list we lazily remove it. To do this, the other party | |
1538 | * clears the _XBF_DELWRI_Q flag but otherwise leaves the buffer alone. | |
1539 | * It remains referenced and on the list. In a rare corner case it | |
1540 | * might get readded to a delwri list after the synchronous writeout, in | |
1541 | * which case we need just need to re-add the flag here. | |
d808f617 | 1542 | */ |
43ff2122 CH |
1543 | bp->b_flags |= _XBF_DELWRI_Q; |
1544 | if (list_empty(&bp->b_list)) { | |
1545 | atomic_inc(&bp->b_hold); | |
1546 | list_add_tail(&bp->b_list, list); | |
585e6d88 | 1547 | } |
585e6d88 | 1548 | |
43ff2122 | 1549 | return true; |
585e6d88 DC |
1550 | } |
1551 | ||
089716aa DC |
1552 | /* |
1553 | * Compare function is more complex than it needs to be because | |
1554 | * the return value is only 32 bits and we are doing comparisons | |
1555 | * on 64 bit values | |
1556 | */ | |
1557 | static int | |
1558 | xfs_buf_cmp( | |
1559 | void *priv, | |
1560 | struct list_head *a, | |
1561 | struct list_head *b) | |
1562 | { | |
1563 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1564 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1565 | xfs_daddr_t diff; | |
1566 | ||
1567 | diff = ap->b_bn - bp->b_bn; | |
1568 | if (diff < 0) | |
1569 | return -1; | |
1570 | if (diff > 0) | |
1571 | return 1; | |
1572 | return 0; | |
1573 | } | |
1574 | ||
43ff2122 CH |
1575 | static int |
1576 | __xfs_buf_delwri_submit( | |
1577 | struct list_head *buffer_list, | |
1578 | struct list_head *io_list, | |
1579 | bool wait) | |
1da177e4 | 1580 | { |
43ff2122 CH |
1581 | struct blk_plug plug; |
1582 | struct xfs_buf *bp, *n; | |
1583 | int pinned = 0; | |
1584 | ||
1585 | list_for_each_entry_safe(bp, n, buffer_list, b_list) { | |
1586 | if (!wait) { | |
1587 | if (xfs_buf_ispinned(bp)) { | |
1588 | pinned++; | |
1589 | continue; | |
1590 | } | |
1591 | if (!xfs_buf_trylock(bp)) | |
1592 | continue; | |
1593 | } else { | |
1594 | xfs_buf_lock(bp); | |
1595 | } | |
978c7b2f | 1596 | |
43ff2122 CH |
1597 | /* |
1598 | * Someone else might have written the buffer synchronously or | |
1599 | * marked it stale in the meantime. In that case only the | |
1600 | * _XBF_DELWRI_Q flag got cleared, and we have to drop the | |
1601 | * reference and remove it from the list here. | |
1602 | */ | |
1603 | if (!(bp->b_flags & _XBF_DELWRI_Q)) { | |
1604 | list_del_init(&bp->b_list); | |
1605 | xfs_buf_relse(bp); | |
1606 | continue; | |
1607 | } | |
c9c12971 | 1608 | |
43ff2122 CH |
1609 | list_move_tail(&bp->b_list, io_list); |
1610 | trace_xfs_buf_delwri_split(bp, _RET_IP_); | |
1611 | } | |
1da177e4 | 1612 | |
43ff2122 | 1613 | list_sort(NULL, io_list, xfs_buf_cmp); |
1da177e4 | 1614 | |
43ff2122 CH |
1615 | blk_start_plug(&plug); |
1616 | list_for_each_entry_safe(bp, n, io_list, b_list) { | |
1617 | bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_ASYNC); | |
1618 | bp->b_flags |= XBF_WRITE; | |
a1b7ea5d | 1619 | |
43ff2122 CH |
1620 | if (!wait) { |
1621 | bp->b_flags |= XBF_ASYNC; | |
ce8e922c | 1622 | list_del_init(&bp->b_list); |
1da177e4 | 1623 | } |
43ff2122 CH |
1624 | xfs_bdstrat_cb(bp); |
1625 | } | |
1626 | blk_finish_plug(&plug); | |
1da177e4 | 1627 | |
43ff2122 | 1628 | return pinned; |
1da177e4 LT |
1629 | } |
1630 | ||
1631 | /* | |
43ff2122 CH |
1632 | * Write out a buffer list asynchronously. |
1633 | * | |
1634 | * This will take the @buffer_list, write all non-locked and non-pinned buffers | |
1635 | * out and not wait for I/O completion on any of the buffers. This interface | |
1636 | * is only safely useable for callers that can track I/O completion by higher | |
1637 | * level means, e.g. AIL pushing as the @buffer_list is consumed in this | |
1638 | * function. | |
1da177e4 LT |
1639 | */ |
1640 | int | |
43ff2122 CH |
1641 | xfs_buf_delwri_submit_nowait( |
1642 | struct list_head *buffer_list) | |
1da177e4 | 1643 | { |
43ff2122 CH |
1644 | LIST_HEAD (io_list); |
1645 | return __xfs_buf_delwri_submit(buffer_list, &io_list, false); | |
1646 | } | |
1da177e4 | 1647 | |
43ff2122 CH |
1648 | /* |
1649 | * Write out a buffer list synchronously. | |
1650 | * | |
1651 | * This will take the @buffer_list, write all buffers out and wait for I/O | |
1652 | * completion on all of the buffers. @buffer_list is consumed by the function, | |
1653 | * so callers must have some other way of tracking buffers if they require such | |
1654 | * functionality. | |
1655 | */ | |
1656 | int | |
1657 | xfs_buf_delwri_submit( | |
1658 | struct list_head *buffer_list) | |
1659 | { | |
1660 | LIST_HEAD (io_list); | |
1661 | int error = 0, error2; | |
1662 | struct xfs_buf *bp; | |
1da177e4 | 1663 | |
43ff2122 | 1664 | __xfs_buf_delwri_submit(buffer_list, &io_list, true); |
1da177e4 | 1665 | |
43ff2122 CH |
1666 | /* Wait for IO to complete. */ |
1667 | while (!list_empty(&io_list)) { | |
1668 | bp = list_first_entry(&io_list, struct xfs_buf, b_list); | |
a1b7ea5d | 1669 | |
089716aa | 1670 | list_del_init(&bp->b_list); |
43ff2122 CH |
1671 | error2 = xfs_buf_iowait(bp); |
1672 | xfs_buf_relse(bp); | |
1673 | if (!error) | |
1674 | error = error2; | |
1da177e4 LT |
1675 | } |
1676 | ||
43ff2122 | 1677 | return error; |
1da177e4 LT |
1678 | } |
1679 | ||
04d8b284 | 1680 | int __init |
ce8e922c | 1681 | xfs_buf_init(void) |
1da177e4 | 1682 | { |
8758280f NS |
1683 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1684 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1685 | if (!xfs_buf_zone) |
0b1b213f | 1686 | goto out; |
04d8b284 | 1687 | |
51749e47 | 1688 | xfslogd_workqueue = alloc_workqueue("xfslogd", |
6370a6ad | 1689 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 1); |
23ea4032 | 1690 | if (!xfslogd_workqueue) |
04d8b284 | 1691 | goto out_free_buf_zone; |
1da177e4 | 1692 | |
23ea4032 | 1693 | return 0; |
1da177e4 | 1694 | |
23ea4032 | 1695 | out_free_buf_zone: |
ce8e922c | 1696 | kmem_zone_destroy(xfs_buf_zone); |
0b1b213f | 1697 | out: |
8758280f | 1698 | return -ENOMEM; |
1da177e4 LT |
1699 | } |
1700 | ||
1da177e4 | 1701 | void |
ce8e922c | 1702 | xfs_buf_terminate(void) |
1da177e4 | 1703 | { |
04d8b284 | 1704 | destroy_workqueue(xfslogd_workqueue); |
ce8e922c | 1705 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 1706 | } |