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; |
a6867a68 | 45 | STATIC int xfsbufd(void *); |
23ea4032 | 46 | |
7989cb8e | 47 | static struct workqueue_struct *xfslogd_workqueue; |
1da177e4 | 48 | |
ce8e922c NS |
49 | #ifdef XFS_BUF_LOCK_TRACKING |
50 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
51 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
52 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 53 | #else |
ce8e922c NS |
54 | # define XB_SET_OWNER(bp) do { } while (0) |
55 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
56 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
57 | #endif |
58 | ||
ce8e922c NS |
59 | #define xb_to_gfp(flags) \ |
60 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \ | |
61 | ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN) | |
1da177e4 | 62 | |
ce8e922c NS |
63 | #define xb_to_km(flags) \ |
64 | (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP) | |
1da177e4 | 65 | |
1da177e4 | 66 | |
73c77e2c JB |
67 | static inline int |
68 | xfs_buf_is_vmapped( | |
69 | struct xfs_buf *bp) | |
70 | { | |
71 | /* | |
72 | * Return true if the buffer is vmapped. | |
73 | * | |
74 | * The XBF_MAPPED flag is set if the buffer should be mapped, but the | |
75 | * code is clever enough to know it doesn't have to map a single page, | |
76 | * so the check has to be both for XBF_MAPPED and bp->b_page_count > 1. | |
77 | */ | |
78 | return (bp->b_flags & XBF_MAPPED) && bp->b_page_count > 1; | |
79 | } | |
80 | ||
81 | static inline int | |
82 | xfs_buf_vmap_len( | |
83 | struct xfs_buf *bp) | |
84 | { | |
85 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
86 | } | |
87 | ||
1da177e4 | 88 | /* |
430cbeb8 DC |
89 | * xfs_buf_lru_add - add a buffer to the LRU. |
90 | * | |
91 | * The LRU takes a new reference to the buffer so that it will only be freed | |
92 | * once the shrinker takes the buffer off the LRU. | |
93 | */ | |
94 | STATIC void | |
95 | xfs_buf_lru_add( | |
96 | struct xfs_buf *bp) | |
97 | { | |
98 | struct xfs_buftarg *btp = bp->b_target; | |
99 | ||
100 | spin_lock(&btp->bt_lru_lock); | |
101 | if (list_empty(&bp->b_lru)) { | |
102 | atomic_inc(&bp->b_hold); | |
103 | list_add_tail(&bp->b_lru, &btp->bt_lru); | |
104 | btp->bt_lru_nr++; | |
105 | } | |
106 | spin_unlock(&btp->bt_lru_lock); | |
107 | } | |
108 | ||
109 | /* | |
110 | * xfs_buf_lru_del - remove a buffer from the LRU | |
111 | * | |
112 | * The unlocked check is safe here because it only occurs when there are not | |
113 | * b_lru_ref counts left on the inode under the pag->pag_buf_lock. it is there | |
114 | * to optimise the shrinker removing the buffer from the LRU and calling | |
25985edc | 115 | * xfs_buf_free(). i.e. it removes an unnecessary round trip on the |
430cbeb8 | 116 | * bt_lru_lock. |
1da177e4 | 117 | */ |
430cbeb8 DC |
118 | STATIC void |
119 | xfs_buf_lru_del( | |
120 | struct xfs_buf *bp) | |
121 | { | |
122 | struct xfs_buftarg *btp = bp->b_target; | |
123 | ||
124 | if (list_empty(&bp->b_lru)) | |
125 | return; | |
126 | ||
127 | spin_lock(&btp->bt_lru_lock); | |
128 | if (!list_empty(&bp->b_lru)) { | |
129 | list_del_init(&bp->b_lru); | |
130 | btp->bt_lru_nr--; | |
131 | } | |
132 | spin_unlock(&btp->bt_lru_lock); | |
133 | } | |
134 | ||
135 | /* | |
136 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
137 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
138 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
139 | * to remove the reference that LRU holds on the buffer. | |
140 | * | |
141 | * This prevents build-up of stale buffers on the LRU. | |
142 | */ | |
143 | void | |
144 | xfs_buf_stale( | |
145 | struct xfs_buf *bp) | |
146 | { | |
147 | bp->b_flags |= XBF_STALE; | |
af5c4bee | 148 | xfs_buf_delwri_dequeue(bp); |
430cbeb8 DC |
149 | atomic_set(&(bp)->b_lru_ref, 0); |
150 | if (!list_empty(&bp->b_lru)) { | |
151 | struct xfs_buftarg *btp = bp->b_target; | |
152 | ||
153 | spin_lock(&btp->bt_lru_lock); | |
154 | if (!list_empty(&bp->b_lru)) { | |
155 | list_del_init(&bp->b_lru); | |
156 | btp->bt_lru_nr--; | |
157 | atomic_dec(&bp->b_hold); | |
158 | } | |
159 | spin_unlock(&btp->bt_lru_lock); | |
160 | } | |
161 | ASSERT(atomic_read(&bp->b_hold) >= 1); | |
162 | } | |
1da177e4 | 163 | |
4347b9d7 CH |
164 | struct xfs_buf * |
165 | xfs_buf_alloc( | |
166 | struct xfs_buftarg *target, | |
204ab25f | 167 | xfs_off_t range_base, |
1da177e4 | 168 | size_t range_length, |
ce8e922c | 169 | xfs_buf_flags_t flags) |
1da177e4 | 170 | { |
4347b9d7 CH |
171 | struct xfs_buf *bp; |
172 | ||
173 | bp = kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags)); | |
174 | if (unlikely(!bp)) | |
175 | return NULL; | |
176 | ||
1da177e4 | 177 | /* |
ce8e922c | 178 | * We don't want certain flags to appear in b_flags. |
1da177e4 | 179 | */ |
ce8e922c NS |
180 | flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD); |
181 | ||
182 | memset(bp, 0, sizeof(xfs_buf_t)); | |
183 | atomic_set(&bp->b_hold, 1); | |
430cbeb8 | 184 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 185 | init_completion(&bp->b_iowait); |
430cbeb8 | 186 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 187 | INIT_LIST_HEAD(&bp->b_list); |
74f75a0c | 188 | RB_CLEAR_NODE(&bp->b_rbnode); |
a731cd11 | 189 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
ce8e922c NS |
190 | XB_SET_OWNER(bp); |
191 | bp->b_target = target; | |
192 | bp->b_file_offset = range_base; | |
1da177e4 LT |
193 | /* |
194 | * Set buffer_length and count_desired to the same value initially. | |
195 | * I/O routines should use count_desired, which will be the same in | |
196 | * most cases but may be reset (e.g. XFS recovery). | |
197 | */ | |
ce8e922c NS |
198 | bp->b_buffer_length = bp->b_count_desired = range_length; |
199 | bp->b_flags = flags; | |
200 | bp->b_bn = XFS_BUF_DADDR_NULL; | |
201 | atomic_set(&bp->b_pin_count, 0); | |
202 | init_waitqueue_head(&bp->b_waiters); | |
203 | ||
204 | XFS_STATS_INC(xb_create); | |
0b1b213f | 205 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 CH |
206 | |
207 | return bp; | |
1da177e4 LT |
208 | } |
209 | ||
210 | /* | |
ce8e922c NS |
211 | * Allocate a page array capable of holding a specified number |
212 | * of pages, and point the page buf at it. | |
1da177e4 LT |
213 | */ |
214 | STATIC int | |
ce8e922c NS |
215 | _xfs_buf_get_pages( |
216 | xfs_buf_t *bp, | |
1da177e4 | 217 | int page_count, |
ce8e922c | 218 | xfs_buf_flags_t flags) |
1da177e4 LT |
219 | { |
220 | /* Make sure that we have a page list */ | |
ce8e922c NS |
221 | if (bp->b_pages == NULL) { |
222 | bp->b_offset = xfs_buf_poff(bp->b_file_offset); | |
223 | bp->b_page_count = page_count; | |
224 | if (page_count <= XB_PAGES) { | |
225 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 226 | } else { |
ce8e922c NS |
227 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
228 | page_count, xb_to_km(flags)); | |
229 | if (bp->b_pages == NULL) | |
1da177e4 LT |
230 | return -ENOMEM; |
231 | } | |
ce8e922c | 232 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
233 | } |
234 | return 0; | |
235 | } | |
236 | ||
237 | /* | |
ce8e922c | 238 | * Frees b_pages if it was allocated. |
1da177e4 LT |
239 | */ |
240 | STATIC void | |
ce8e922c | 241 | _xfs_buf_free_pages( |
1da177e4 LT |
242 | xfs_buf_t *bp) |
243 | { | |
ce8e922c | 244 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 245 | kmem_free(bp->b_pages); |
3fc98b1a | 246 | bp->b_pages = NULL; |
1da177e4 LT |
247 | } |
248 | } | |
249 | ||
250 | /* | |
251 | * Releases the specified buffer. | |
252 | * | |
253 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 254 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
255 | * hashed and refcounted buffers |
256 | */ | |
257 | void | |
ce8e922c | 258 | xfs_buf_free( |
1da177e4 LT |
259 | xfs_buf_t *bp) |
260 | { | |
0b1b213f | 261 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 262 | |
430cbeb8 DC |
263 | ASSERT(list_empty(&bp->b_lru)); |
264 | ||
0e6e847f | 265 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
266 | uint i; |
267 | ||
73c77e2c | 268 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
269 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
270 | bp->b_page_count); | |
1da177e4 | 271 | |
948ecdb4 NS |
272 | for (i = 0; i < bp->b_page_count; i++) { |
273 | struct page *page = bp->b_pages[i]; | |
274 | ||
0e6e847f | 275 | __free_page(page); |
948ecdb4 | 276 | } |
0e6e847f DC |
277 | } else if (bp->b_flags & _XBF_KMEM) |
278 | kmem_free(bp->b_addr); | |
3fc98b1a | 279 | _xfs_buf_free_pages(bp); |
4347b9d7 | 280 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
281 | } |
282 | ||
283 | /* | |
0e6e847f | 284 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
285 | */ |
286 | STATIC int | |
0e6e847f | 287 | xfs_buf_allocate_memory( |
1da177e4 LT |
288 | xfs_buf_t *bp, |
289 | uint flags) | |
290 | { | |
ce8e922c | 291 | size_t size = bp->b_count_desired; |
1da177e4 | 292 | size_t nbytes, offset; |
ce8e922c | 293 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 294 | unsigned short page_count, i; |
204ab25f | 295 | xfs_off_t end; |
1da177e4 LT |
296 | int error; |
297 | ||
0e6e847f DC |
298 | /* |
299 | * for buffers that are contained within a single page, just allocate | |
300 | * the memory from the heap - there's no need for the complexity of | |
301 | * page arrays to keep allocation down to order 0. | |
302 | */ | |
303 | if (bp->b_buffer_length < PAGE_SIZE) { | |
304 | bp->b_addr = kmem_alloc(bp->b_buffer_length, xb_to_km(flags)); | |
305 | if (!bp->b_addr) { | |
306 | /* low memory - use alloc_page loop instead */ | |
307 | goto use_alloc_page; | |
308 | } | |
309 | ||
310 | if (((unsigned long)(bp->b_addr + bp->b_buffer_length - 1) & | |
311 | PAGE_MASK) != | |
312 | ((unsigned long)bp->b_addr & PAGE_MASK)) { | |
313 | /* b_addr spans two pages - use alloc_page instead */ | |
314 | kmem_free(bp->b_addr); | |
315 | bp->b_addr = NULL; | |
316 | goto use_alloc_page; | |
317 | } | |
318 | bp->b_offset = offset_in_page(bp->b_addr); | |
319 | bp->b_pages = bp->b_page_array; | |
320 | bp->b_pages[0] = virt_to_page(bp->b_addr); | |
321 | bp->b_page_count = 1; | |
322 | bp->b_flags |= XBF_MAPPED | _XBF_KMEM; | |
323 | return 0; | |
324 | } | |
325 | ||
326 | use_alloc_page: | |
ce8e922c NS |
327 | end = bp->b_file_offset + bp->b_buffer_length; |
328 | page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset); | |
ce8e922c | 329 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
330 | if (unlikely(error)) |
331 | return error; | |
1da177e4 | 332 | |
ce8e922c | 333 | offset = bp->b_offset; |
0e6e847f | 334 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 335 | |
ce8e922c | 336 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
337 | struct page *page; |
338 | uint retries = 0; | |
0e6e847f DC |
339 | retry: |
340 | page = alloc_page(gfp_mask); | |
1da177e4 | 341 | if (unlikely(page == NULL)) { |
ce8e922c NS |
342 | if (flags & XBF_READ_AHEAD) { |
343 | bp->b_page_count = i; | |
0e6e847f DC |
344 | error = ENOMEM; |
345 | goto out_free_pages; | |
1da177e4 LT |
346 | } |
347 | ||
348 | /* | |
349 | * This could deadlock. | |
350 | * | |
351 | * But until all the XFS lowlevel code is revamped to | |
352 | * handle buffer allocation failures we can't do much. | |
353 | */ | |
354 | if (!(++retries % 100)) | |
4f10700a DC |
355 | xfs_err(NULL, |
356 | "possible memory allocation deadlock in %s (mode:0x%x)", | |
34a622b2 | 357 | __func__, gfp_mask); |
1da177e4 | 358 | |
ce8e922c | 359 | XFS_STATS_INC(xb_page_retries); |
8aa7e847 | 360 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
361 | goto retry; |
362 | } | |
363 | ||
ce8e922c | 364 | XFS_STATS_INC(xb_page_found); |
1da177e4 | 365 | |
0e6e847f | 366 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 367 | size -= nbytes; |
ce8e922c | 368 | bp->b_pages[i] = page; |
1da177e4 LT |
369 | offset = 0; |
370 | } | |
0e6e847f | 371 | return 0; |
1da177e4 | 372 | |
0e6e847f DC |
373 | out_free_pages: |
374 | for (i = 0; i < bp->b_page_count; i++) | |
375 | __free_page(bp->b_pages[i]); | |
1da177e4 LT |
376 | return error; |
377 | } | |
378 | ||
379 | /* | |
25985edc | 380 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
381 | */ |
382 | STATIC int | |
ce8e922c | 383 | _xfs_buf_map_pages( |
1da177e4 LT |
384 | xfs_buf_t *bp, |
385 | uint flags) | |
386 | { | |
0e6e847f | 387 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 388 | if (bp->b_page_count == 1) { |
0e6e847f | 389 | /* A single page buffer is always mappable */ |
ce8e922c NS |
390 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
391 | bp->b_flags |= XBF_MAPPED; | |
392 | } else if (flags & XBF_MAPPED) { | |
a19fb380 DC |
393 | int retried = 0; |
394 | ||
395 | do { | |
396 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
397 | -1, PAGE_KERNEL); | |
398 | if (bp->b_addr) | |
399 | break; | |
400 | vm_unmap_aliases(); | |
401 | } while (retried++ <= 1); | |
402 | ||
403 | if (!bp->b_addr) | |
1da177e4 | 404 | return -ENOMEM; |
ce8e922c NS |
405 | bp->b_addr += bp->b_offset; |
406 | bp->b_flags |= XBF_MAPPED; | |
1da177e4 LT |
407 | } |
408 | ||
409 | return 0; | |
410 | } | |
411 | ||
412 | /* | |
413 | * Finding and Reading Buffers | |
414 | */ | |
415 | ||
416 | /* | |
ce8e922c | 417 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 | 418 | * a given range of an inode. The buffer is returned |
eabbaf11 | 419 | * locked. No I/O is implied by this call. |
1da177e4 LT |
420 | */ |
421 | xfs_buf_t * | |
ce8e922c | 422 | _xfs_buf_find( |
1da177e4 | 423 | xfs_buftarg_t *btp, /* block device target */ |
204ab25f | 424 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 425 | size_t isize, /* length of range */ |
ce8e922c NS |
426 | xfs_buf_flags_t flags, |
427 | xfs_buf_t *new_bp) | |
1da177e4 | 428 | { |
204ab25f | 429 | xfs_off_t range_base; |
1da177e4 | 430 | size_t range_length; |
74f75a0c DC |
431 | struct xfs_perag *pag; |
432 | struct rb_node **rbp; | |
433 | struct rb_node *parent; | |
434 | xfs_buf_t *bp; | |
1da177e4 LT |
435 | |
436 | range_base = (ioff << BBSHIFT); | |
437 | range_length = (isize << BBSHIFT); | |
438 | ||
439 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
ce8e922c | 440 | ASSERT(!(range_length < (1 << btp->bt_sshift))); |
204ab25f | 441 | ASSERT(!(range_base & (xfs_off_t)btp->bt_smask)); |
1da177e4 | 442 | |
74f75a0c DC |
443 | /* get tree root */ |
444 | pag = xfs_perag_get(btp->bt_mount, | |
445 | xfs_daddr_to_agno(btp->bt_mount, ioff)); | |
446 | ||
447 | /* walk tree */ | |
448 | spin_lock(&pag->pag_buf_lock); | |
449 | rbp = &pag->pag_buf_tree.rb_node; | |
450 | parent = NULL; | |
451 | bp = NULL; | |
452 | while (*rbp) { | |
453 | parent = *rbp; | |
454 | bp = rb_entry(parent, struct xfs_buf, b_rbnode); | |
455 | ||
456 | if (range_base < bp->b_file_offset) | |
457 | rbp = &(*rbp)->rb_left; | |
458 | else if (range_base > bp->b_file_offset) | |
459 | rbp = &(*rbp)->rb_right; | |
460 | else { | |
461 | /* | |
462 | * found a block offset match. If the range doesn't | |
463 | * match, the only way this is allowed is if the buffer | |
464 | * in the cache is stale and the transaction that made | |
465 | * it stale has not yet committed. i.e. we are | |
466 | * reallocating a busy extent. Skip this buffer and | |
467 | * continue searching to the right for an exact match. | |
468 | */ | |
469 | if (bp->b_buffer_length != range_length) { | |
470 | ASSERT(bp->b_flags & XBF_STALE); | |
471 | rbp = &(*rbp)->rb_right; | |
472 | continue; | |
473 | } | |
ce8e922c | 474 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
475 | goto found; |
476 | } | |
477 | } | |
478 | ||
479 | /* No match found */ | |
ce8e922c | 480 | if (new_bp) { |
74f75a0c DC |
481 | rb_link_node(&new_bp->b_rbnode, parent, rbp); |
482 | rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree); | |
483 | /* the buffer keeps the perag reference until it is freed */ | |
484 | new_bp->b_pag = pag; | |
485 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 486 | } else { |
ce8e922c | 487 | XFS_STATS_INC(xb_miss_locked); |
74f75a0c DC |
488 | spin_unlock(&pag->pag_buf_lock); |
489 | xfs_perag_put(pag); | |
1da177e4 | 490 | } |
ce8e922c | 491 | return new_bp; |
1da177e4 LT |
492 | |
493 | found: | |
74f75a0c DC |
494 | spin_unlock(&pag->pag_buf_lock); |
495 | xfs_perag_put(pag); | |
1da177e4 | 496 | |
0c842ad4 CH |
497 | if (!xfs_buf_trylock(bp)) { |
498 | if (flags & XBF_TRYLOCK) { | |
ce8e922c NS |
499 | xfs_buf_rele(bp); |
500 | XFS_STATS_INC(xb_busy_locked); | |
501 | return NULL; | |
1da177e4 | 502 | } |
0c842ad4 CH |
503 | xfs_buf_lock(bp); |
504 | XFS_STATS_INC(xb_get_locked_waited); | |
1da177e4 LT |
505 | } |
506 | ||
0e6e847f DC |
507 | /* |
508 | * if the buffer is stale, clear all the external state associated with | |
509 | * it. We need to keep flags such as how we allocated the buffer memory | |
510 | * intact here. | |
511 | */ | |
ce8e922c NS |
512 | if (bp->b_flags & XBF_STALE) { |
513 | ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0); | |
0e6e847f | 514 | bp->b_flags &= XBF_MAPPED | _XBF_KMEM | _XBF_PAGES; |
2f926587 | 515 | } |
0b1b213f CH |
516 | |
517 | trace_xfs_buf_find(bp, flags, _RET_IP_); | |
ce8e922c NS |
518 | XFS_STATS_INC(xb_get_locked); |
519 | return bp; | |
1da177e4 LT |
520 | } |
521 | ||
522 | /* | |
3815832a DC |
523 | * Assembles a buffer covering the specified range. The code is optimised for |
524 | * cache hits, as metadata intensive workloads will see 3 orders of magnitude | |
525 | * more hits than misses. | |
1da177e4 | 526 | */ |
3815832a | 527 | struct xfs_buf * |
6ad112bf | 528 | xfs_buf_get( |
1da177e4 | 529 | xfs_buftarg_t *target,/* target for buffer */ |
204ab25f | 530 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 531 | size_t isize, /* length of range */ |
ce8e922c | 532 | xfs_buf_flags_t flags) |
1da177e4 | 533 | { |
3815832a DC |
534 | struct xfs_buf *bp; |
535 | struct xfs_buf *new_bp; | |
0e6e847f | 536 | int error = 0; |
1da177e4 | 537 | |
3815832a DC |
538 | bp = _xfs_buf_find(target, ioff, isize, flags, NULL); |
539 | if (likely(bp)) | |
540 | goto found; | |
541 | ||
4347b9d7 CH |
542 | new_bp = xfs_buf_alloc(target, ioff << BBSHIFT, isize << BBSHIFT, |
543 | flags); | |
ce8e922c | 544 | if (unlikely(!new_bp)) |
1da177e4 LT |
545 | return NULL; |
546 | ||
ce8e922c | 547 | bp = _xfs_buf_find(target, ioff, isize, flags, new_bp); |
3815832a | 548 | if (!bp) { |
4347b9d7 | 549 | kmem_zone_free(xfs_buf_zone, new_bp); |
3815832a DC |
550 | return NULL; |
551 | } | |
552 | ||
ce8e922c | 553 | if (bp == new_bp) { |
0e6e847f | 554 | error = xfs_buf_allocate_memory(bp, flags); |
1da177e4 LT |
555 | if (error) |
556 | goto no_buffer; | |
3815832a | 557 | } else |
4347b9d7 | 558 | kmem_zone_free(xfs_buf_zone, new_bp); |
1da177e4 | 559 | |
3815832a DC |
560 | /* |
561 | * Now we have a workable buffer, fill in the block number so | |
562 | * that we can do IO on it. | |
563 | */ | |
564 | bp->b_bn = ioff; | |
565 | bp->b_count_desired = bp->b_buffer_length; | |
566 | ||
567 | found: | |
ce8e922c NS |
568 | if (!(bp->b_flags & XBF_MAPPED)) { |
569 | error = _xfs_buf_map_pages(bp, flags); | |
1da177e4 | 570 | if (unlikely(error)) { |
4f10700a DC |
571 | xfs_warn(target->bt_mount, |
572 | "%s: failed to map pages\n", __func__); | |
1da177e4 LT |
573 | goto no_buffer; |
574 | } | |
575 | } | |
576 | ||
ce8e922c | 577 | XFS_STATS_INC(xb_get); |
0b1b213f | 578 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 579 | return bp; |
1da177e4 | 580 | |
3815832a | 581 | no_buffer: |
ce8e922c NS |
582 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
583 | xfs_buf_unlock(bp); | |
584 | xfs_buf_rele(bp); | |
1da177e4 LT |
585 | return NULL; |
586 | } | |
587 | ||
5d765b97 CH |
588 | STATIC int |
589 | _xfs_buf_read( | |
590 | xfs_buf_t *bp, | |
591 | xfs_buf_flags_t flags) | |
592 | { | |
593 | int status; | |
594 | ||
5d765b97 CH |
595 | ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE))); |
596 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); | |
597 | ||
1d5ae5df CH |
598 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | XBF_READ_AHEAD); |
599 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); | |
5d765b97 CH |
600 | |
601 | status = xfs_buf_iorequest(bp); | |
5a52c2a5 | 602 | if (status || bp->b_error || (flags & XBF_ASYNC)) |
ec53d1db DC |
603 | return status; |
604 | return xfs_buf_iowait(bp); | |
5d765b97 CH |
605 | } |
606 | ||
1da177e4 | 607 | xfs_buf_t * |
6ad112bf | 608 | xfs_buf_read( |
1da177e4 | 609 | xfs_buftarg_t *target, |
204ab25f | 610 | xfs_off_t ioff, |
1da177e4 | 611 | size_t isize, |
ce8e922c | 612 | xfs_buf_flags_t flags) |
1da177e4 | 613 | { |
ce8e922c NS |
614 | xfs_buf_t *bp; |
615 | ||
616 | flags |= XBF_READ; | |
617 | ||
6ad112bf | 618 | bp = xfs_buf_get(target, ioff, isize, flags); |
ce8e922c | 619 | if (bp) { |
0b1b213f CH |
620 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
621 | ||
ce8e922c | 622 | if (!XFS_BUF_ISDONE(bp)) { |
ce8e922c | 623 | XFS_STATS_INC(xb_get_read); |
5d765b97 | 624 | _xfs_buf_read(bp, flags); |
ce8e922c | 625 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
626 | /* |
627 | * Read ahead call which is already satisfied, | |
628 | * drop the buffer | |
629 | */ | |
630 | goto no_buffer; | |
631 | } else { | |
1da177e4 | 632 | /* We do not want read in the flags */ |
ce8e922c | 633 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
634 | } |
635 | } | |
636 | ||
ce8e922c | 637 | return bp; |
1da177e4 LT |
638 | |
639 | no_buffer: | |
ce8e922c NS |
640 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
641 | xfs_buf_unlock(bp); | |
642 | xfs_buf_rele(bp); | |
1da177e4 LT |
643 | return NULL; |
644 | } | |
645 | ||
1da177e4 | 646 | /* |
ce8e922c NS |
647 | * If we are not low on memory then do the readahead in a deadlock |
648 | * safe manner. | |
1da177e4 LT |
649 | */ |
650 | void | |
ce8e922c | 651 | xfs_buf_readahead( |
1da177e4 | 652 | xfs_buftarg_t *target, |
204ab25f | 653 | xfs_off_t ioff, |
1a1a3e97 | 654 | size_t isize) |
1da177e4 | 655 | { |
0e6e847f | 656 | if (bdi_read_congested(target->bt_bdi)) |
1da177e4 LT |
657 | return; |
658 | ||
1a1a3e97 CH |
659 | xfs_buf_read(target, ioff, isize, |
660 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD|XBF_DONT_BLOCK); | |
1da177e4 LT |
661 | } |
662 | ||
5adc94c2 DC |
663 | /* |
664 | * Read an uncached buffer from disk. Allocates and returns a locked | |
665 | * buffer containing the disk contents or nothing. | |
666 | */ | |
667 | struct xfs_buf * | |
668 | xfs_buf_read_uncached( | |
669 | struct xfs_mount *mp, | |
670 | struct xfs_buftarg *target, | |
671 | xfs_daddr_t daddr, | |
672 | size_t length, | |
673 | int flags) | |
674 | { | |
675 | xfs_buf_t *bp; | |
676 | int error; | |
677 | ||
678 | bp = xfs_buf_get_uncached(target, length, flags); | |
679 | if (!bp) | |
680 | return NULL; | |
681 | ||
682 | /* set up the buffer for a read IO */ | |
5adc94c2 DC |
683 | XFS_BUF_SET_ADDR(bp, daddr); |
684 | XFS_BUF_READ(bp); | |
5adc94c2 DC |
685 | |
686 | xfsbdstrat(mp, bp); | |
1a1a3e97 | 687 | error = xfs_buf_iowait(bp); |
5adc94c2 DC |
688 | if (error || bp->b_error) { |
689 | xfs_buf_relse(bp); | |
690 | return NULL; | |
691 | } | |
692 | return bp; | |
1da177e4 LT |
693 | } |
694 | ||
44396476 DC |
695 | /* |
696 | * Return a buffer allocated as an empty buffer and associated to external | |
697 | * memory via xfs_buf_associate_memory() back to it's empty state. | |
698 | */ | |
699 | void | |
700 | xfs_buf_set_empty( | |
701 | struct xfs_buf *bp, | |
702 | size_t len) | |
703 | { | |
704 | if (bp->b_pages) | |
705 | _xfs_buf_free_pages(bp); | |
706 | ||
707 | bp->b_pages = NULL; | |
708 | bp->b_page_count = 0; | |
709 | bp->b_addr = NULL; | |
710 | bp->b_file_offset = 0; | |
711 | bp->b_buffer_length = bp->b_count_desired = len; | |
712 | bp->b_bn = XFS_BUF_DADDR_NULL; | |
713 | bp->b_flags &= ~XBF_MAPPED; | |
714 | } | |
715 | ||
1da177e4 LT |
716 | static inline struct page * |
717 | mem_to_page( | |
718 | void *addr) | |
719 | { | |
9e2779fa | 720 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
721 | return virt_to_page(addr); |
722 | } else { | |
723 | return vmalloc_to_page(addr); | |
724 | } | |
725 | } | |
726 | ||
727 | int | |
ce8e922c NS |
728 | xfs_buf_associate_memory( |
729 | xfs_buf_t *bp, | |
1da177e4 LT |
730 | void *mem, |
731 | size_t len) | |
732 | { | |
733 | int rval; | |
734 | int i = 0; | |
d1afb678 LM |
735 | unsigned long pageaddr; |
736 | unsigned long offset; | |
737 | size_t buflen; | |
1da177e4 LT |
738 | int page_count; |
739 | ||
0e6e847f | 740 | pageaddr = (unsigned long)mem & PAGE_MASK; |
d1afb678 | 741 | offset = (unsigned long)mem - pageaddr; |
0e6e847f DC |
742 | buflen = PAGE_ALIGN(len + offset); |
743 | page_count = buflen >> PAGE_SHIFT; | |
1da177e4 LT |
744 | |
745 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
746 | if (bp->b_pages) |
747 | _xfs_buf_free_pages(bp); | |
1da177e4 | 748 | |
ce8e922c NS |
749 | bp->b_pages = NULL; |
750 | bp->b_addr = mem; | |
1da177e4 | 751 | |
36fae17a | 752 | rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK); |
1da177e4 LT |
753 | if (rval) |
754 | return rval; | |
755 | ||
ce8e922c | 756 | bp->b_offset = offset; |
d1afb678 LM |
757 | |
758 | for (i = 0; i < bp->b_page_count; i++) { | |
759 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
0e6e847f | 760 | pageaddr += PAGE_SIZE; |
1da177e4 | 761 | } |
1da177e4 | 762 | |
d1afb678 LM |
763 | bp->b_count_desired = len; |
764 | bp->b_buffer_length = buflen; | |
ce8e922c | 765 | bp->b_flags |= XBF_MAPPED; |
1da177e4 LT |
766 | |
767 | return 0; | |
768 | } | |
769 | ||
770 | xfs_buf_t * | |
686865f7 DC |
771 | xfs_buf_get_uncached( |
772 | struct xfs_buftarg *target, | |
1da177e4 | 773 | size_t len, |
686865f7 | 774 | int flags) |
1da177e4 | 775 | { |
1fa40b01 CH |
776 | unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT; |
777 | int error, i; | |
1da177e4 | 778 | xfs_buf_t *bp; |
1da177e4 | 779 | |
4347b9d7 | 780 | bp = xfs_buf_alloc(target, 0, len, 0); |
1da177e4 LT |
781 | if (unlikely(bp == NULL)) |
782 | goto fail; | |
1da177e4 | 783 | |
1fa40b01 CH |
784 | error = _xfs_buf_get_pages(bp, page_count, 0); |
785 | if (error) | |
1da177e4 LT |
786 | goto fail_free_buf; |
787 | ||
1fa40b01 | 788 | for (i = 0; i < page_count; i++) { |
686865f7 | 789 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
790 | if (!bp->b_pages[i]) |
791 | goto fail_free_mem; | |
1da177e4 | 792 | } |
1fa40b01 | 793 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 794 | |
1fa40b01 CH |
795 | error = _xfs_buf_map_pages(bp, XBF_MAPPED); |
796 | if (unlikely(error)) { | |
4f10700a DC |
797 | xfs_warn(target->bt_mount, |
798 | "%s: failed to map pages\n", __func__); | |
1da177e4 | 799 | goto fail_free_mem; |
1fa40b01 | 800 | } |
1da177e4 | 801 | |
686865f7 | 802 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 803 | return bp; |
1fa40b01 | 804 | |
1da177e4 | 805 | fail_free_mem: |
1fa40b01 CH |
806 | while (--i >= 0) |
807 | __free_page(bp->b_pages[i]); | |
ca165b88 | 808 | _xfs_buf_free_pages(bp); |
1da177e4 | 809 | fail_free_buf: |
4347b9d7 | 810 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
811 | fail: |
812 | return NULL; | |
813 | } | |
814 | ||
815 | /* | |
1da177e4 LT |
816 | * Increment reference count on buffer, to hold the buffer concurrently |
817 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
818 | * Must hold the buffer already to call this function. |
819 | */ | |
820 | void | |
ce8e922c NS |
821 | xfs_buf_hold( |
822 | xfs_buf_t *bp) | |
1da177e4 | 823 | { |
0b1b213f | 824 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 825 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
826 | } |
827 | ||
828 | /* | |
ce8e922c NS |
829 | * Releases a hold on the specified buffer. If the |
830 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
831 | */ |
832 | void | |
ce8e922c NS |
833 | xfs_buf_rele( |
834 | xfs_buf_t *bp) | |
1da177e4 | 835 | { |
74f75a0c | 836 | struct xfs_perag *pag = bp->b_pag; |
1da177e4 | 837 | |
0b1b213f | 838 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 839 | |
74f75a0c | 840 | if (!pag) { |
430cbeb8 | 841 | ASSERT(list_empty(&bp->b_lru)); |
74f75a0c | 842 | ASSERT(RB_EMPTY_NODE(&bp->b_rbnode)); |
fad3aa1e NS |
843 | if (atomic_dec_and_test(&bp->b_hold)) |
844 | xfs_buf_free(bp); | |
845 | return; | |
846 | } | |
847 | ||
74f75a0c | 848 | ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode)); |
430cbeb8 | 849 | |
3790689f | 850 | ASSERT(atomic_read(&bp->b_hold) > 0); |
74f75a0c | 851 | if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) { |
bfc60177 | 852 | if (!(bp->b_flags & XBF_STALE) && |
430cbeb8 DC |
853 | atomic_read(&bp->b_lru_ref)) { |
854 | xfs_buf_lru_add(bp); | |
855 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 856 | } else { |
430cbeb8 | 857 | xfs_buf_lru_del(bp); |
ce8e922c | 858 | ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q))); |
74f75a0c DC |
859 | rb_erase(&bp->b_rbnode, &pag->pag_buf_tree); |
860 | spin_unlock(&pag->pag_buf_lock); | |
861 | xfs_perag_put(pag); | |
ce8e922c | 862 | xfs_buf_free(bp); |
1da177e4 LT |
863 | } |
864 | } | |
865 | } | |
866 | ||
867 | ||
868 | /* | |
0e6e847f | 869 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
870 | * |
871 | * If we come across a stale, pinned, locked buffer, we know that we are | |
872 | * being asked to lock a buffer that has been reallocated. Because it is | |
873 | * pinned, we know that the log has not been pushed to disk and hence it | |
874 | * will still be locked. Rather than continuing to have trylock attempts | |
875 | * fail until someone else pushes the log, push it ourselves before | |
876 | * returning. This means that the xfsaild will not get stuck trying | |
877 | * to push on stale inode buffers. | |
1da177e4 LT |
878 | */ |
879 | int | |
0c842ad4 CH |
880 | xfs_buf_trylock( |
881 | struct xfs_buf *bp) | |
1da177e4 LT |
882 | { |
883 | int locked; | |
884 | ||
ce8e922c | 885 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 886 | if (locked) |
ce8e922c | 887 | XB_SET_OWNER(bp); |
90810b9e DC |
888 | else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
889 | xfs_log_force(bp->b_target->bt_mount, 0); | |
0b1b213f | 890 | |
0c842ad4 CH |
891 | trace_xfs_buf_trylock(bp, _RET_IP_); |
892 | return locked; | |
1da177e4 | 893 | } |
1da177e4 LT |
894 | |
895 | /* | |
0e6e847f | 896 | * Lock a buffer object. |
ed3b4d6c DC |
897 | * |
898 | * If we come across a stale, pinned, locked buffer, we know that we | |
899 | * are being asked to lock a buffer that has been reallocated. Because | |
900 | * it is pinned, we know that the log has not been pushed to disk and | |
901 | * hence it will still be locked. Rather than sleeping until someone | |
902 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 903 | */ |
ce8e922c NS |
904 | void |
905 | xfs_buf_lock( | |
0c842ad4 | 906 | struct xfs_buf *bp) |
1da177e4 | 907 | { |
0b1b213f CH |
908 | trace_xfs_buf_lock(bp, _RET_IP_); |
909 | ||
ed3b4d6c | 910 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 911 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
912 | down(&bp->b_sema); |
913 | XB_SET_OWNER(bp); | |
0b1b213f CH |
914 | |
915 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
916 | } |
917 | ||
918 | /* | |
ce8e922c | 919 | * Releases the lock on the buffer object. |
2f926587 | 920 | * If the buffer is marked delwri but is not queued, do so before we |
ce8e922c | 921 | * unlock the buffer as we need to set flags correctly. We also need to |
2f926587 DC |
922 | * take a reference for the delwri queue because the unlocker is going to |
923 | * drop their's and they don't know we just queued it. | |
1da177e4 LT |
924 | */ |
925 | void | |
ce8e922c | 926 | xfs_buf_unlock( |
0c842ad4 | 927 | struct xfs_buf *bp) |
1da177e4 | 928 | { |
ce8e922c NS |
929 | XB_CLEAR_OWNER(bp); |
930 | up(&bp->b_sema); | |
0b1b213f CH |
931 | |
932 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
933 | } |
934 | ||
ce8e922c NS |
935 | STATIC void |
936 | xfs_buf_wait_unpin( | |
937 | xfs_buf_t *bp) | |
1da177e4 LT |
938 | { |
939 | DECLARE_WAITQUEUE (wait, current); | |
940 | ||
ce8e922c | 941 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
942 | return; |
943 | ||
ce8e922c | 944 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
945 | for (;;) { |
946 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 947 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 948 | break; |
7eaceacc | 949 | io_schedule(); |
1da177e4 | 950 | } |
ce8e922c | 951 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
952 | set_current_state(TASK_RUNNING); |
953 | } | |
954 | ||
955 | /* | |
956 | * Buffer Utility Routines | |
957 | */ | |
958 | ||
1da177e4 | 959 | STATIC void |
ce8e922c | 960 | xfs_buf_iodone_work( |
c4028958 | 961 | struct work_struct *work) |
1da177e4 | 962 | { |
c4028958 DH |
963 | xfs_buf_t *bp = |
964 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 965 | |
80f6c29d | 966 | if (bp->b_iodone) |
ce8e922c NS |
967 | (*(bp->b_iodone))(bp); |
968 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
969 | xfs_buf_relse(bp); |
970 | } | |
971 | ||
972 | void | |
ce8e922c NS |
973 | xfs_buf_ioend( |
974 | xfs_buf_t *bp, | |
1da177e4 LT |
975 | int schedule) |
976 | { | |
0b1b213f CH |
977 | trace_xfs_buf_iodone(bp, _RET_IP_); |
978 | ||
77be55a5 | 979 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
ce8e922c NS |
980 | if (bp->b_error == 0) |
981 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 982 | |
ce8e922c | 983 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 984 | if (schedule) { |
c4028958 | 985 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 986 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 987 | } else { |
c4028958 | 988 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
989 | } |
990 | } else { | |
b4dd330b | 991 | complete(&bp->b_iowait); |
1da177e4 LT |
992 | } |
993 | } | |
994 | ||
1da177e4 | 995 | void |
ce8e922c NS |
996 | xfs_buf_ioerror( |
997 | xfs_buf_t *bp, | |
998 | int error) | |
1da177e4 LT |
999 | { |
1000 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c | 1001 | bp->b_error = (unsigned short)error; |
0b1b213f | 1002 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1003 | } |
1004 | ||
901796af CH |
1005 | void |
1006 | xfs_buf_ioerror_alert( | |
1007 | struct xfs_buf *bp, | |
1008 | const char *func) | |
1009 | { | |
1010 | xfs_alert(bp->b_target->bt_mount, | |
1011 | "metadata I/O error: block 0x%llx (\"%s\") error %d buf count %zd", | |
1012 | (__uint64_t)XFS_BUF_ADDR(bp), func, | |
1013 | bp->b_error, XFS_BUF_COUNT(bp)); | |
1014 | } | |
1015 | ||
1da177e4 | 1016 | int |
64e0bc7d | 1017 | xfs_bwrite( |
5d765b97 | 1018 | struct xfs_buf *bp) |
1da177e4 | 1019 | { |
8c38366f | 1020 | int error; |
1da177e4 | 1021 | |
64e0bc7d | 1022 | bp->b_flags |= XBF_WRITE; |
8c38366f | 1023 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ); |
1da177e4 | 1024 | |
5d765b97 | 1025 | xfs_buf_delwri_dequeue(bp); |
939d723b | 1026 | xfs_bdstrat_cb(bp); |
1da177e4 | 1027 | |
8c38366f | 1028 | error = xfs_buf_iowait(bp); |
c2b006c1 CH |
1029 | if (error) { |
1030 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1031 | SHUTDOWN_META_IO_ERROR); | |
1032 | } | |
64e0bc7d | 1033 | return error; |
5d765b97 | 1034 | } |
1da177e4 | 1035 | |
4e23471a CH |
1036 | /* |
1037 | * Called when we want to stop a buffer from getting written or read. | |
1a1a3e97 | 1038 | * We attach the EIO error, muck with its flags, and call xfs_buf_ioend |
4e23471a CH |
1039 | * so that the proper iodone callbacks get called. |
1040 | */ | |
1041 | STATIC int | |
1042 | xfs_bioerror( | |
1043 | xfs_buf_t *bp) | |
1044 | { | |
1045 | #ifdef XFSERRORDEBUG | |
1046 | ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone); | |
1047 | #endif | |
1048 | ||
1049 | /* | |
1050 | * No need to wait until the buffer is unpinned, we aren't flushing it. | |
1051 | */ | |
5a52c2a5 | 1052 | xfs_buf_ioerror(bp, EIO); |
4e23471a CH |
1053 | |
1054 | /* | |
1a1a3e97 | 1055 | * We're calling xfs_buf_ioend, so delete XBF_DONE flag. |
4e23471a CH |
1056 | */ |
1057 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1058 | XFS_BUF_UNDONE(bp); |
c867cb61 | 1059 | xfs_buf_stale(bp); |
4e23471a | 1060 | |
1a1a3e97 | 1061 | xfs_buf_ioend(bp, 0); |
4e23471a CH |
1062 | |
1063 | return EIO; | |
1064 | } | |
1065 | ||
1066 | /* | |
1067 | * Same as xfs_bioerror, except that we are releasing the buffer | |
1a1a3e97 | 1068 | * here ourselves, and avoiding the xfs_buf_ioend call. |
4e23471a CH |
1069 | * This is meant for userdata errors; metadata bufs come with |
1070 | * iodone functions attached, so that we can track down errors. | |
1071 | */ | |
1072 | STATIC int | |
1073 | xfs_bioerror_relse( | |
1074 | struct xfs_buf *bp) | |
1075 | { | |
ed43233b | 1076 | int64_t fl = bp->b_flags; |
4e23471a CH |
1077 | /* |
1078 | * No need to wait until the buffer is unpinned. | |
1079 | * We aren't flushing it. | |
1080 | * | |
1081 | * chunkhold expects B_DONE to be set, whether | |
1082 | * we actually finish the I/O or not. We don't want to | |
1083 | * change that interface. | |
1084 | */ | |
1085 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1086 | XFS_BUF_DONE(bp); |
c867cb61 | 1087 | xfs_buf_stale(bp); |
cb669ca5 | 1088 | bp->b_iodone = NULL; |
0cadda1c | 1089 | if (!(fl & XBF_ASYNC)) { |
4e23471a CH |
1090 | /* |
1091 | * Mark b_error and B_ERROR _both_. | |
1092 | * Lot's of chunkcache code assumes that. | |
1093 | * There's no reason to mark error for | |
1094 | * ASYNC buffers. | |
1095 | */ | |
5a52c2a5 | 1096 | xfs_buf_ioerror(bp, EIO); |
5fde0326 | 1097 | complete(&bp->b_iowait); |
4e23471a CH |
1098 | } else { |
1099 | xfs_buf_relse(bp); | |
1100 | } | |
1101 | ||
1102 | return EIO; | |
1103 | } | |
1104 | ||
1105 | ||
1106 | /* | |
1107 | * All xfs metadata buffers except log state machine buffers | |
1108 | * get this attached as their b_bdstrat callback function. | |
1109 | * This is so that we can catch a buffer | |
1110 | * after prematurely unpinning it to forcibly shutdown the filesystem. | |
1111 | */ | |
1112 | int | |
1113 | xfs_bdstrat_cb( | |
1114 | struct xfs_buf *bp) | |
1115 | { | |
ebad861b | 1116 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
4e23471a CH |
1117 | trace_xfs_bdstrat_shut(bp, _RET_IP_); |
1118 | /* | |
1119 | * Metadata write that didn't get logged but | |
1120 | * written delayed anyway. These aren't associated | |
1121 | * with a transaction, and can be ignored. | |
1122 | */ | |
1123 | if (!bp->b_iodone && !XFS_BUF_ISREAD(bp)) | |
1124 | return xfs_bioerror_relse(bp); | |
1125 | else | |
1126 | return xfs_bioerror(bp); | |
1127 | } | |
1128 | ||
1129 | xfs_buf_iorequest(bp); | |
1130 | return 0; | |
1131 | } | |
1132 | ||
1133 | /* | |
1134 | * Wrapper around bdstrat so that we can stop data from going to disk in case | |
1135 | * we are shutting down the filesystem. Typically user data goes thru this | |
1136 | * path; one of the exceptions is the superblock. | |
1137 | */ | |
1138 | void | |
1139 | xfsbdstrat( | |
1140 | struct xfs_mount *mp, | |
1141 | struct xfs_buf *bp) | |
1142 | { | |
1143 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1144 | trace_xfs_bdstrat_shut(bp, _RET_IP_); | |
1145 | xfs_bioerror_relse(bp); | |
1146 | return; | |
1147 | } | |
1148 | ||
1149 | xfs_buf_iorequest(bp); | |
1150 | } | |
1151 | ||
b8f82a4a | 1152 | STATIC void |
ce8e922c NS |
1153 | _xfs_buf_ioend( |
1154 | xfs_buf_t *bp, | |
1da177e4 LT |
1155 | int schedule) |
1156 | { | |
0e6e847f | 1157 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
ce8e922c | 1158 | xfs_buf_ioend(bp, schedule); |
1da177e4 LT |
1159 | } |
1160 | ||
782e3b3b | 1161 | STATIC void |
ce8e922c | 1162 | xfs_buf_bio_end_io( |
1da177e4 | 1163 | struct bio *bio, |
1da177e4 LT |
1164 | int error) |
1165 | { | |
ce8e922c | 1166 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1da177e4 | 1167 | |
cfbe5267 | 1168 | xfs_buf_ioerror(bp, -error); |
1da177e4 | 1169 | |
73c77e2c JB |
1170 | if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
1171 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); | |
1172 | ||
ce8e922c | 1173 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1174 | bio_put(bio); |
1da177e4 LT |
1175 | } |
1176 | ||
1177 | STATIC void | |
ce8e922c NS |
1178 | _xfs_buf_ioapply( |
1179 | xfs_buf_t *bp) | |
1da177e4 | 1180 | { |
a9759f2d | 1181 | int rw, map_i, total_nr_pages, nr_pages; |
1da177e4 | 1182 | struct bio *bio; |
ce8e922c NS |
1183 | int offset = bp->b_offset; |
1184 | int size = bp->b_count_desired; | |
1185 | sector_t sector = bp->b_bn; | |
1da177e4 | 1186 | |
ce8e922c | 1187 | total_nr_pages = bp->b_page_count; |
1da177e4 LT |
1188 | map_i = 0; |
1189 | ||
1d5ae5df CH |
1190 | if (bp->b_flags & XBF_WRITE) { |
1191 | if (bp->b_flags & XBF_SYNCIO) | |
1192 | rw = WRITE_SYNC; | |
1193 | else | |
1194 | rw = WRITE; | |
1195 | if (bp->b_flags & XBF_FUA) | |
1196 | rw |= REQ_FUA; | |
1197 | if (bp->b_flags & XBF_FLUSH) | |
1198 | rw |= REQ_FLUSH; | |
1199 | } else if (bp->b_flags & XBF_READ_AHEAD) { | |
1200 | rw = READA; | |
51bdd706 | 1201 | } else { |
1d5ae5df | 1202 | rw = READ; |
f538d4da CH |
1203 | } |
1204 | ||
34951f5c CH |
1205 | /* we only use the buffer cache for meta-data */ |
1206 | rw |= REQ_META; | |
1207 | ||
1da177e4 | 1208 | next_chunk: |
ce8e922c | 1209 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1210 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1211 | if (nr_pages > total_nr_pages) | |
1212 | nr_pages = total_nr_pages; | |
1213 | ||
1214 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1215 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1216 | bio->bi_sector = sector; |
ce8e922c NS |
1217 | bio->bi_end_io = xfs_buf_bio_end_io; |
1218 | bio->bi_private = bp; | |
1da177e4 | 1219 | |
0e6e847f | 1220 | |
1da177e4 | 1221 | for (; size && nr_pages; nr_pages--, map_i++) { |
0e6e847f | 1222 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1223 | |
1224 | if (nbytes > size) | |
1225 | nbytes = size; | |
1226 | ||
ce8e922c NS |
1227 | rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset); |
1228 | if (rbytes < nbytes) | |
1da177e4 LT |
1229 | break; |
1230 | ||
1231 | offset = 0; | |
1232 | sector += nbytes >> BBSHIFT; | |
1233 | size -= nbytes; | |
1234 | total_nr_pages--; | |
1235 | } | |
1236 | ||
1da177e4 | 1237 | if (likely(bio->bi_size)) { |
73c77e2c JB |
1238 | if (xfs_buf_is_vmapped(bp)) { |
1239 | flush_kernel_vmap_range(bp->b_addr, | |
1240 | xfs_buf_vmap_len(bp)); | |
1241 | } | |
1da177e4 LT |
1242 | submit_bio(rw, bio); |
1243 | if (size) | |
1244 | goto next_chunk; | |
1245 | } else { | |
ce8e922c | 1246 | xfs_buf_ioerror(bp, EIO); |
ec53d1db | 1247 | bio_put(bio); |
1da177e4 LT |
1248 | } |
1249 | } | |
1250 | ||
1da177e4 | 1251 | int |
ce8e922c NS |
1252 | xfs_buf_iorequest( |
1253 | xfs_buf_t *bp) | |
1da177e4 | 1254 | { |
0b1b213f | 1255 | trace_xfs_buf_iorequest(bp, _RET_IP_); |
1da177e4 | 1256 | |
375ec69d | 1257 | ASSERT(!(bp->b_flags & XBF_DELWRI)); |
1da177e4 | 1258 | |
375ec69d | 1259 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1260 | xfs_buf_wait_unpin(bp); |
ce8e922c | 1261 | xfs_buf_hold(bp); |
1da177e4 LT |
1262 | |
1263 | /* Set the count to 1 initially, this will stop an I/O | |
1264 | * completion callout which happens before we have started | |
ce8e922c | 1265 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1266 | */ |
ce8e922c NS |
1267 | atomic_set(&bp->b_io_remaining, 1); |
1268 | _xfs_buf_ioapply(bp); | |
1269 | _xfs_buf_ioend(bp, 0); | |
1da177e4 | 1270 | |
ce8e922c | 1271 | xfs_buf_rele(bp); |
1da177e4 LT |
1272 | return 0; |
1273 | } | |
1274 | ||
1275 | /* | |
ce8e922c NS |
1276 | * Waits for I/O to complete on the buffer supplied. |
1277 | * It returns immediately if no I/O is pending. | |
1278 | * It returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1279 | */ |
1280 | int | |
ce8e922c NS |
1281 | xfs_buf_iowait( |
1282 | xfs_buf_t *bp) | |
1da177e4 | 1283 | { |
0b1b213f CH |
1284 | trace_xfs_buf_iowait(bp, _RET_IP_); |
1285 | ||
b4dd330b | 1286 | wait_for_completion(&bp->b_iowait); |
0b1b213f CH |
1287 | |
1288 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
ce8e922c | 1289 | return bp->b_error; |
1da177e4 LT |
1290 | } |
1291 | ||
ce8e922c NS |
1292 | xfs_caddr_t |
1293 | xfs_buf_offset( | |
1294 | xfs_buf_t *bp, | |
1da177e4 LT |
1295 | size_t offset) |
1296 | { | |
1297 | struct page *page; | |
1298 | ||
ce8e922c | 1299 | if (bp->b_flags & XBF_MAPPED) |
62926044 | 1300 | return bp->b_addr + offset; |
1da177e4 | 1301 | |
ce8e922c | 1302 | offset += bp->b_offset; |
0e6e847f DC |
1303 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
1304 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_SIZE-1)); | |
1da177e4 LT |
1305 | } |
1306 | ||
1307 | /* | |
1da177e4 LT |
1308 | * Move data into or out of a buffer. |
1309 | */ | |
1310 | void | |
ce8e922c NS |
1311 | xfs_buf_iomove( |
1312 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1313 | size_t boff, /* starting buffer offset */ |
1314 | size_t bsize, /* length to copy */ | |
b9c48649 | 1315 | void *data, /* data address */ |
ce8e922c | 1316 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 LT |
1317 | { |
1318 | size_t bend, cpoff, csize; | |
1319 | struct page *page; | |
1320 | ||
1321 | bend = boff + bsize; | |
1322 | while (boff < bend) { | |
ce8e922c NS |
1323 | page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)]; |
1324 | cpoff = xfs_buf_poff(boff + bp->b_offset); | |
1da177e4 | 1325 | csize = min_t(size_t, |
0e6e847f | 1326 | PAGE_SIZE-cpoff, bp->b_count_desired-boff); |
1da177e4 | 1327 | |
0e6e847f | 1328 | ASSERT(((csize + cpoff) <= PAGE_SIZE)); |
1da177e4 LT |
1329 | |
1330 | switch (mode) { | |
ce8e922c | 1331 | case XBRW_ZERO: |
1da177e4 LT |
1332 | memset(page_address(page) + cpoff, 0, csize); |
1333 | break; | |
ce8e922c | 1334 | case XBRW_READ: |
1da177e4 LT |
1335 | memcpy(data, page_address(page) + cpoff, csize); |
1336 | break; | |
ce8e922c | 1337 | case XBRW_WRITE: |
1da177e4 LT |
1338 | memcpy(page_address(page) + cpoff, data, csize); |
1339 | } | |
1340 | ||
1341 | boff += csize; | |
1342 | data += csize; | |
1343 | } | |
1344 | } | |
1345 | ||
1346 | /* | |
ce8e922c | 1347 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1348 | */ |
1349 | ||
1350 | /* | |
430cbeb8 DC |
1351 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1352 | * returned. These buffers will have an elevated hold count, so wait on those | |
1353 | * while freeing all the buffers only held by the LRU. | |
1da177e4 LT |
1354 | */ |
1355 | void | |
1356 | xfs_wait_buftarg( | |
74f75a0c | 1357 | struct xfs_buftarg *btp) |
1da177e4 | 1358 | { |
430cbeb8 DC |
1359 | struct xfs_buf *bp; |
1360 | ||
1361 | restart: | |
1362 | spin_lock(&btp->bt_lru_lock); | |
1363 | while (!list_empty(&btp->bt_lru)) { | |
1364 | bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru); | |
1365 | if (atomic_read(&bp->b_hold) > 1) { | |
1366 | spin_unlock(&btp->bt_lru_lock); | |
26af6552 | 1367 | delay(100); |
430cbeb8 | 1368 | goto restart; |
1da177e4 | 1369 | } |
430cbeb8 | 1370 | /* |
90802ed9 | 1371 | * clear the LRU reference count so the buffer doesn't get |
430cbeb8 DC |
1372 | * ignored in xfs_buf_rele(). |
1373 | */ | |
1374 | atomic_set(&bp->b_lru_ref, 0); | |
1375 | spin_unlock(&btp->bt_lru_lock); | |
1376 | xfs_buf_rele(bp); | |
1377 | spin_lock(&btp->bt_lru_lock); | |
1da177e4 | 1378 | } |
430cbeb8 | 1379 | spin_unlock(&btp->bt_lru_lock); |
1da177e4 LT |
1380 | } |
1381 | ||
ff57ab21 DC |
1382 | int |
1383 | xfs_buftarg_shrink( | |
1384 | struct shrinker *shrink, | |
1495f230 | 1385 | struct shrink_control *sc) |
a6867a68 | 1386 | { |
ff57ab21 DC |
1387 | struct xfs_buftarg *btp = container_of(shrink, |
1388 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1389 | struct xfs_buf *bp; |
1495f230 | 1390 | int nr_to_scan = sc->nr_to_scan; |
430cbeb8 DC |
1391 | LIST_HEAD(dispose); |
1392 | ||
1393 | if (!nr_to_scan) | |
1394 | return btp->bt_lru_nr; | |
1395 | ||
1396 | spin_lock(&btp->bt_lru_lock); | |
1397 | while (!list_empty(&btp->bt_lru)) { | |
1398 | if (nr_to_scan-- <= 0) | |
1399 | break; | |
1400 | ||
1401 | bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru); | |
1402 | ||
1403 | /* | |
1404 | * Decrement the b_lru_ref count unless the value is already | |
1405 | * zero. If the value is already zero, we need to reclaim the | |
1406 | * buffer, otherwise it gets another trip through the LRU. | |
1407 | */ | |
1408 | if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) { | |
1409 | list_move_tail(&bp->b_lru, &btp->bt_lru); | |
1410 | continue; | |
1411 | } | |
1412 | ||
1413 | /* | |
1414 | * remove the buffer from the LRU now to avoid needing another | |
1415 | * lock round trip inside xfs_buf_rele(). | |
1416 | */ | |
1417 | list_move(&bp->b_lru, &dispose); | |
1418 | btp->bt_lru_nr--; | |
ff57ab21 | 1419 | } |
430cbeb8 DC |
1420 | spin_unlock(&btp->bt_lru_lock); |
1421 | ||
1422 | while (!list_empty(&dispose)) { | |
1423 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1424 | list_del_init(&bp->b_lru); | |
1425 | xfs_buf_rele(bp); | |
1426 | } | |
1427 | ||
1428 | return btp->bt_lru_nr; | |
a6867a68 DC |
1429 | } |
1430 | ||
1da177e4 LT |
1431 | void |
1432 | xfs_free_buftarg( | |
b7963133 CH |
1433 | struct xfs_mount *mp, |
1434 | struct xfs_buftarg *btp) | |
1da177e4 | 1435 | { |
ff57ab21 DC |
1436 | unregister_shrinker(&btp->bt_shrinker); |
1437 | ||
1da177e4 | 1438 | xfs_flush_buftarg(btp, 1); |
b7963133 CH |
1439 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1440 | xfs_blkdev_issue_flush(btp); | |
a6867a68 | 1441 | |
a6867a68 | 1442 | kthread_stop(btp->bt_task); |
f0e2d93c | 1443 | kmem_free(btp); |
1da177e4 LT |
1444 | } |
1445 | ||
1da177e4 LT |
1446 | STATIC int |
1447 | xfs_setsize_buftarg_flags( | |
1448 | xfs_buftarg_t *btp, | |
1449 | unsigned int blocksize, | |
1450 | unsigned int sectorsize, | |
1451 | int verbose) | |
1452 | { | |
ce8e922c NS |
1453 | btp->bt_bsize = blocksize; |
1454 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1455 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1456 | |
ce8e922c | 1457 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
02b102df CH |
1458 | char name[BDEVNAME_SIZE]; |
1459 | ||
1460 | bdevname(btp->bt_bdev, name); | |
1461 | ||
4f10700a DC |
1462 | xfs_warn(btp->bt_mount, |
1463 | "Cannot set_blocksize to %u on device %s\n", | |
02b102df | 1464 | sectorsize, name); |
1da177e4 LT |
1465 | return EINVAL; |
1466 | } | |
1467 | ||
1da177e4 LT |
1468 | return 0; |
1469 | } | |
1470 | ||
1471 | /* | |
ce8e922c NS |
1472 | * When allocating the initial buffer target we have not yet |
1473 | * read in the superblock, so don't know what sized sectors | |
1474 | * are being used is at this early stage. Play safe. | |
1475 | */ | |
1da177e4 LT |
1476 | STATIC int |
1477 | xfs_setsize_buftarg_early( | |
1478 | xfs_buftarg_t *btp, | |
1479 | struct block_device *bdev) | |
1480 | { | |
1481 | return xfs_setsize_buftarg_flags(btp, | |
0e6e847f | 1482 | PAGE_SIZE, bdev_logical_block_size(bdev), 0); |
1da177e4 LT |
1483 | } |
1484 | ||
1485 | int | |
1486 | xfs_setsize_buftarg( | |
1487 | xfs_buftarg_t *btp, | |
1488 | unsigned int blocksize, | |
1489 | unsigned int sectorsize) | |
1490 | { | |
1491 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1492 | } | |
1493 | ||
a6867a68 | 1494 | STATIC int |
c4e1c098 | 1495 | xfs_alloc_delwri_queue( |
e2a07812 JE |
1496 | xfs_buftarg_t *btp, |
1497 | const char *fsname) | |
a6867a68 | 1498 | { |
c4e1c098 CH |
1499 | INIT_LIST_HEAD(&btp->bt_delwri_queue); |
1500 | spin_lock_init(&btp->bt_delwri_lock); | |
a6867a68 | 1501 | btp->bt_flags = 0; |
e2a07812 | 1502 | btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd/%s", fsname); |
ff57ab21 DC |
1503 | if (IS_ERR(btp->bt_task)) |
1504 | return PTR_ERR(btp->bt_task); | |
1505 | return 0; | |
a6867a68 DC |
1506 | } |
1507 | ||
1da177e4 LT |
1508 | xfs_buftarg_t * |
1509 | xfs_alloc_buftarg( | |
ebad861b | 1510 | struct xfs_mount *mp, |
1da177e4 | 1511 | struct block_device *bdev, |
e2a07812 JE |
1512 | int external, |
1513 | const char *fsname) | |
1da177e4 LT |
1514 | { |
1515 | xfs_buftarg_t *btp; | |
1516 | ||
1517 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1518 | ||
ebad861b | 1519 | btp->bt_mount = mp; |
ce8e922c NS |
1520 | btp->bt_dev = bdev->bd_dev; |
1521 | btp->bt_bdev = bdev; | |
0e6e847f DC |
1522 | btp->bt_bdi = blk_get_backing_dev_info(bdev); |
1523 | if (!btp->bt_bdi) | |
1524 | goto error; | |
1525 | ||
430cbeb8 DC |
1526 | INIT_LIST_HEAD(&btp->bt_lru); |
1527 | spin_lock_init(&btp->bt_lru_lock); | |
1da177e4 LT |
1528 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1529 | goto error; | |
c4e1c098 | 1530 | if (xfs_alloc_delwri_queue(btp, fsname)) |
a6867a68 | 1531 | goto error; |
ff57ab21 DC |
1532 | btp->bt_shrinker.shrink = xfs_buftarg_shrink; |
1533 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; | |
1534 | register_shrinker(&btp->bt_shrinker); | |
1da177e4 LT |
1535 | return btp; |
1536 | ||
1537 | error: | |
f0e2d93c | 1538 | kmem_free(btp); |
1da177e4 LT |
1539 | return NULL; |
1540 | } | |
1541 | ||
1542 | ||
1543 | /* | |
ce8e922c | 1544 | * Delayed write buffer handling |
1da177e4 | 1545 | */ |
61551f1e | 1546 | void |
ce8e922c | 1547 | xfs_buf_delwri_queue( |
527cfdf1 | 1548 | xfs_buf_t *bp) |
1da177e4 | 1549 | { |
c4e1c098 | 1550 | struct xfs_buftarg *btp = bp->b_target; |
a6867a68 | 1551 | |
0b1b213f CH |
1552 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
1553 | ||
5a8ee6ba | 1554 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1555 | |
c4e1c098 | 1556 | spin_lock(&btp->bt_delwri_lock); |
ce8e922c | 1557 | if (!list_empty(&bp->b_list)) { |
5a8ee6ba | 1558 | /* if already in the queue, move it to the tail */ |
ce8e922c | 1559 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); |
c4e1c098 | 1560 | list_move_tail(&bp->b_list, &btp->bt_delwri_queue); |
5a8ee6ba | 1561 | } else { |
c9c12971 | 1562 | /* start xfsbufd as it is about to have something to do */ |
c4e1c098 | 1563 | if (list_empty(&btp->bt_delwri_queue)) |
5a8ee6ba | 1564 | wake_up_process(bp->b_target->bt_task); |
c9c12971 | 1565 | |
5a8ee6ba CH |
1566 | atomic_inc(&bp->b_hold); |
1567 | bp->b_flags |= XBF_DELWRI | _XBF_DELWRI_Q | XBF_ASYNC; | |
c4e1c098 | 1568 | list_add_tail(&bp->b_list, &btp->bt_delwri_queue); |
5a8ee6ba | 1569 | } |
ce8e922c | 1570 | bp->b_queuetime = jiffies; |
c4e1c098 | 1571 | spin_unlock(&btp->bt_delwri_lock); |
1da177e4 LT |
1572 | } |
1573 | ||
1574 | void | |
ce8e922c NS |
1575 | xfs_buf_delwri_dequeue( |
1576 | xfs_buf_t *bp) | |
1da177e4 LT |
1577 | { |
1578 | int dequeued = 0; | |
1579 | ||
c4e1c098 | 1580 | spin_lock(&bp->b_target->bt_delwri_lock); |
ce8e922c NS |
1581 | if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) { |
1582 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1583 | list_del_init(&bp->b_list); | |
1da177e4 LT |
1584 | dequeued = 1; |
1585 | } | |
ce8e922c | 1586 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q); |
c4e1c098 | 1587 | spin_unlock(&bp->b_target->bt_delwri_lock); |
1da177e4 LT |
1588 | |
1589 | if (dequeued) | |
ce8e922c | 1590 | xfs_buf_rele(bp); |
1da177e4 | 1591 | |
0b1b213f | 1592 | trace_xfs_buf_delwri_dequeue(bp, _RET_IP_); |
1da177e4 LT |
1593 | } |
1594 | ||
d808f617 DC |
1595 | /* |
1596 | * If a delwri buffer needs to be pushed before it has aged out, then promote | |
1597 | * it to the head of the delwri queue so that it will be flushed on the next | |
1598 | * xfsbufd run. We do this by resetting the queuetime of the buffer to be older | |
1599 | * than the age currently needed to flush the buffer. Hence the next time the | |
1600 | * xfsbufd sees it is guaranteed to be considered old enough to flush. | |
1601 | */ | |
1602 | void | |
1603 | xfs_buf_delwri_promote( | |
1604 | struct xfs_buf *bp) | |
1605 | { | |
1606 | struct xfs_buftarg *btp = bp->b_target; | |
1607 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1; | |
1608 | ||
1609 | ASSERT(bp->b_flags & XBF_DELWRI); | |
1610 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1611 | ||
1612 | /* | |
1613 | * Check the buffer age before locking the delayed write queue as we | |
1614 | * don't need to promote buffers that are already past the flush age. | |
1615 | */ | |
1616 | if (bp->b_queuetime < jiffies - age) | |
1617 | return; | |
1618 | bp->b_queuetime = jiffies - age; | |
c4e1c098 CH |
1619 | spin_lock(&btp->bt_delwri_lock); |
1620 | list_move(&bp->b_list, &btp->bt_delwri_queue); | |
1621 | spin_unlock(&btp->bt_delwri_lock); | |
d808f617 DC |
1622 | } |
1623 | ||
585e6d88 DC |
1624 | /* |
1625 | * Move as many buffers as specified to the supplied list | |
1626 | * idicating if we skipped any buffers to prevent deadlocks. | |
1627 | */ | |
1628 | STATIC int | |
1629 | xfs_buf_delwri_split( | |
1630 | xfs_buftarg_t *target, | |
1631 | struct list_head *list, | |
5e6a07df | 1632 | unsigned long age) |
585e6d88 DC |
1633 | { |
1634 | xfs_buf_t *bp, *n; | |
585e6d88 | 1635 | int skipped = 0; |
5e6a07df | 1636 | int force; |
585e6d88 | 1637 | |
5e6a07df | 1638 | force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
585e6d88 | 1639 | INIT_LIST_HEAD(list); |
c4e1c098 CH |
1640 | spin_lock(&target->bt_delwri_lock); |
1641 | list_for_each_entry_safe(bp, n, &target->bt_delwri_queue, b_list) { | |
585e6d88 DC |
1642 | ASSERT(bp->b_flags & XBF_DELWRI); |
1643 | ||
811e64c7 | 1644 | if (!xfs_buf_ispinned(bp) && xfs_buf_trylock(bp)) { |
5e6a07df | 1645 | if (!force && |
585e6d88 DC |
1646 | time_before(jiffies, bp->b_queuetime + age)) { |
1647 | xfs_buf_unlock(bp); | |
1648 | break; | |
1649 | } | |
1650 | ||
1d5ae5df | 1651 | bp->b_flags &= ~(XBF_DELWRI | _XBF_DELWRI_Q); |
585e6d88 DC |
1652 | bp->b_flags |= XBF_WRITE; |
1653 | list_move_tail(&bp->b_list, list); | |
bfe27419 | 1654 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
585e6d88 DC |
1655 | } else |
1656 | skipped++; | |
1657 | } | |
585e6d88 | 1658 | |
c4e1c098 | 1659 | spin_unlock(&target->bt_delwri_lock); |
585e6d88 | 1660 | return skipped; |
585e6d88 DC |
1661 | } |
1662 | ||
089716aa DC |
1663 | /* |
1664 | * Compare function is more complex than it needs to be because | |
1665 | * the return value is only 32 bits and we are doing comparisons | |
1666 | * on 64 bit values | |
1667 | */ | |
1668 | static int | |
1669 | xfs_buf_cmp( | |
1670 | void *priv, | |
1671 | struct list_head *a, | |
1672 | struct list_head *b) | |
1673 | { | |
1674 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1675 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1676 | xfs_daddr_t diff; | |
1677 | ||
1678 | diff = ap->b_bn - bp->b_bn; | |
1679 | if (diff < 0) | |
1680 | return -1; | |
1681 | if (diff > 0) | |
1682 | return 1; | |
1683 | return 0; | |
1684 | } | |
1685 | ||
1da177e4 | 1686 | STATIC int |
23ea4032 | 1687 | xfsbufd( |
585e6d88 | 1688 | void *data) |
1da177e4 | 1689 | { |
089716aa | 1690 | xfs_buftarg_t *target = (xfs_buftarg_t *)data; |
1da177e4 | 1691 | |
1da177e4 LT |
1692 | current->flags |= PF_MEMALLOC; |
1693 | ||
978c7b2f RW |
1694 | set_freezable(); |
1695 | ||
1da177e4 | 1696 | do { |
c9c12971 DC |
1697 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10); |
1698 | long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10); | |
089716aa | 1699 | struct list_head tmp; |
a1b7ea5d | 1700 | struct blk_plug plug; |
c9c12971 | 1701 | |
687d1c5e | 1702 | if (unlikely(freezing(current))) |
a0acae0e | 1703 | try_to_freeze(); |
1da177e4 | 1704 | |
c9c12971 | 1705 | /* sleep for a long time if there is nothing to do. */ |
c4e1c098 | 1706 | if (list_empty(&target->bt_delwri_queue)) |
c9c12971 DC |
1707 | tout = MAX_SCHEDULE_TIMEOUT; |
1708 | schedule_timeout_interruptible(tout); | |
1da177e4 | 1709 | |
c9c12971 | 1710 | xfs_buf_delwri_split(target, &tmp, age); |
089716aa | 1711 | list_sort(NULL, &tmp, xfs_buf_cmp); |
a1b7ea5d CH |
1712 | |
1713 | blk_start_plug(&plug); | |
1da177e4 | 1714 | while (!list_empty(&tmp)) { |
089716aa DC |
1715 | struct xfs_buf *bp; |
1716 | bp = list_first_entry(&tmp, struct xfs_buf, b_list); | |
ce8e922c | 1717 | list_del_init(&bp->b_list); |
939d723b | 1718 | xfs_bdstrat_cb(bp); |
1da177e4 | 1719 | } |
a1b7ea5d | 1720 | blk_finish_plug(&plug); |
4df08c52 | 1721 | } while (!kthread_should_stop()); |
1da177e4 | 1722 | |
4df08c52 | 1723 | return 0; |
1da177e4 LT |
1724 | } |
1725 | ||
1726 | /* | |
ce8e922c NS |
1727 | * Go through all incore buffers, and release buffers if they belong to |
1728 | * the given device. This is used in filesystem error handling to | |
1729 | * preserve the consistency of its metadata. | |
1da177e4 LT |
1730 | */ |
1731 | int | |
1732 | xfs_flush_buftarg( | |
585e6d88 DC |
1733 | xfs_buftarg_t *target, |
1734 | int wait) | |
1da177e4 | 1735 | { |
089716aa | 1736 | xfs_buf_t *bp; |
585e6d88 | 1737 | int pincount = 0; |
089716aa DC |
1738 | LIST_HEAD(tmp_list); |
1739 | LIST_HEAD(wait_list); | |
a1b7ea5d | 1740 | struct blk_plug plug; |
1da177e4 | 1741 | |
5a93a064 | 1742 | flush_workqueue(xfslogd_workqueue); |
1da177e4 | 1743 | |
5e6a07df | 1744 | set_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
089716aa | 1745 | pincount = xfs_buf_delwri_split(target, &tmp_list, 0); |
1da177e4 LT |
1746 | |
1747 | /* | |
089716aa DC |
1748 | * Dropped the delayed write list lock, now walk the temporary list. |
1749 | * All I/O is issued async and then if we need to wait for completion | |
1750 | * we do that after issuing all the IO. | |
1da177e4 | 1751 | */ |
089716aa | 1752 | list_sort(NULL, &tmp_list, xfs_buf_cmp); |
a1b7ea5d CH |
1753 | |
1754 | blk_start_plug(&plug); | |
089716aa DC |
1755 | while (!list_empty(&tmp_list)) { |
1756 | bp = list_first_entry(&tmp_list, struct xfs_buf, b_list); | |
585e6d88 | 1757 | ASSERT(target == bp->b_target); |
089716aa DC |
1758 | list_del_init(&bp->b_list); |
1759 | if (wait) { | |
ce8e922c | 1760 | bp->b_flags &= ~XBF_ASYNC; |
089716aa DC |
1761 | list_add(&bp->b_list, &wait_list); |
1762 | } | |
939d723b | 1763 | xfs_bdstrat_cb(bp); |
1da177e4 | 1764 | } |
a1b7ea5d | 1765 | blk_finish_plug(&plug); |
1da177e4 | 1766 | |
089716aa | 1767 | if (wait) { |
a1b7ea5d | 1768 | /* Wait for IO to complete. */ |
089716aa DC |
1769 | while (!list_empty(&wait_list)) { |
1770 | bp = list_first_entry(&wait_list, struct xfs_buf, b_list); | |
f07c2250 | 1771 | |
089716aa | 1772 | list_del_init(&bp->b_list); |
1a1a3e97 | 1773 | xfs_buf_iowait(bp); |
089716aa DC |
1774 | xfs_buf_relse(bp); |
1775 | } | |
1da177e4 LT |
1776 | } |
1777 | ||
1da177e4 LT |
1778 | return pincount; |
1779 | } | |
1780 | ||
04d8b284 | 1781 | int __init |
ce8e922c | 1782 | xfs_buf_init(void) |
1da177e4 | 1783 | { |
8758280f NS |
1784 | xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf", |
1785 | KM_ZONE_HWALIGN, NULL); | |
ce8e922c | 1786 | if (!xfs_buf_zone) |
0b1b213f | 1787 | goto out; |
04d8b284 | 1788 | |
51749e47 | 1789 | xfslogd_workqueue = alloc_workqueue("xfslogd", |
6370a6ad | 1790 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 1); |
23ea4032 | 1791 | if (!xfslogd_workqueue) |
04d8b284 | 1792 | goto out_free_buf_zone; |
1da177e4 | 1793 | |
23ea4032 | 1794 | return 0; |
1da177e4 | 1795 | |
23ea4032 | 1796 | out_free_buf_zone: |
ce8e922c | 1797 | kmem_zone_destroy(xfs_buf_zone); |
0b1b213f | 1798 | out: |
8758280f | 1799 | return -ENOMEM; |
1da177e4 LT |
1800 | } |
1801 | ||
1da177e4 | 1802 | void |
ce8e922c | 1803 | xfs_buf_terminate(void) |
1da177e4 | 1804 | { |
04d8b284 | 1805 | destroy_workqueue(xfslogd_workqueue); |
ce8e922c | 1806 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 1807 | } |