]>
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; |
0829c360 | 48 | struct workqueue_struct *xfsdatad_workqueue; |
c626d174 | 49 | struct workqueue_struct *xfsconvertd_workqueue; |
1da177e4 | 50 | |
ce8e922c NS |
51 | #ifdef XFS_BUF_LOCK_TRACKING |
52 | # define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid) | |
53 | # define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1) | |
54 | # define XB_GET_OWNER(bp) ((bp)->b_last_holder) | |
1da177e4 | 55 | #else |
ce8e922c NS |
56 | # define XB_SET_OWNER(bp) do { } while (0) |
57 | # define XB_CLEAR_OWNER(bp) do { } while (0) | |
58 | # define XB_GET_OWNER(bp) do { } while (0) | |
1da177e4 LT |
59 | #endif |
60 | ||
ce8e922c NS |
61 | #define xb_to_gfp(flags) \ |
62 | ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \ | |
63 | ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN) | |
1da177e4 | 64 | |
ce8e922c NS |
65 | #define xb_to_km(flags) \ |
66 | (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP) | |
1da177e4 | 67 | |
1da177e4 | 68 | |
73c77e2c JB |
69 | static inline int |
70 | xfs_buf_is_vmapped( | |
71 | struct xfs_buf *bp) | |
72 | { | |
73 | /* | |
74 | * Return true if the buffer is vmapped. | |
75 | * | |
76 | * The XBF_MAPPED flag is set if the buffer should be mapped, but the | |
77 | * code is clever enough to know it doesn't have to map a single page, | |
78 | * so the check has to be both for XBF_MAPPED and bp->b_page_count > 1. | |
79 | */ | |
80 | return (bp->b_flags & XBF_MAPPED) && bp->b_page_count > 1; | |
81 | } | |
82 | ||
83 | static inline int | |
84 | xfs_buf_vmap_len( | |
85 | struct xfs_buf *bp) | |
86 | { | |
87 | return (bp->b_page_count * PAGE_SIZE) - bp->b_offset; | |
88 | } | |
89 | ||
1da177e4 | 90 | /* |
430cbeb8 DC |
91 | * xfs_buf_lru_add - add a buffer to the LRU. |
92 | * | |
93 | * The LRU takes a new reference to the buffer so that it will only be freed | |
94 | * once the shrinker takes the buffer off the LRU. | |
95 | */ | |
96 | STATIC void | |
97 | xfs_buf_lru_add( | |
98 | struct xfs_buf *bp) | |
99 | { | |
100 | struct xfs_buftarg *btp = bp->b_target; | |
101 | ||
102 | spin_lock(&btp->bt_lru_lock); | |
103 | if (list_empty(&bp->b_lru)) { | |
104 | atomic_inc(&bp->b_hold); | |
105 | list_add_tail(&bp->b_lru, &btp->bt_lru); | |
106 | btp->bt_lru_nr++; | |
107 | } | |
108 | spin_unlock(&btp->bt_lru_lock); | |
109 | } | |
110 | ||
111 | /* | |
112 | * xfs_buf_lru_del - remove a buffer from the LRU | |
113 | * | |
114 | * The unlocked check is safe here because it only occurs when there are not | |
115 | * b_lru_ref counts left on the inode under the pag->pag_buf_lock. it is there | |
116 | * to optimise the shrinker removing the buffer from the LRU and calling | |
25985edc | 117 | * xfs_buf_free(). i.e. it removes an unnecessary round trip on the |
430cbeb8 | 118 | * bt_lru_lock. |
1da177e4 | 119 | */ |
430cbeb8 DC |
120 | STATIC void |
121 | xfs_buf_lru_del( | |
122 | struct xfs_buf *bp) | |
123 | { | |
124 | struct xfs_buftarg *btp = bp->b_target; | |
125 | ||
126 | if (list_empty(&bp->b_lru)) | |
127 | return; | |
128 | ||
129 | spin_lock(&btp->bt_lru_lock); | |
130 | if (!list_empty(&bp->b_lru)) { | |
131 | list_del_init(&bp->b_lru); | |
132 | btp->bt_lru_nr--; | |
133 | } | |
134 | spin_unlock(&btp->bt_lru_lock); | |
135 | } | |
136 | ||
137 | /* | |
138 | * When we mark a buffer stale, we remove the buffer from the LRU and clear the | |
139 | * b_lru_ref count so that the buffer is freed immediately when the buffer | |
140 | * reference count falls to zero. If the buffer is already on the LRU, we need | |
141 | * to remove the reference that LRU holds on the buffer. | |
142 | * | |
143 | * This prevents build-up of stale buffers on the LRU. | |
144 | */ | |
145 | void | |
146 | xfs_buf_stale( | |
147 | struct xfs_buf *bp) | |
148 | { | |
149 | bp->b_flags |= XBF_STALE; | |
af5c4bee | 150 | xfs_buf_delwri_dequeue(bp); |
430cbeb8 DC |
151 | atomic_set(&(bp)->b_lru_ref, 0); |
152 | if (!list_empty(&bp->b_lru)) { | |
153 | struct xfs_buftarg *btp = bp->b_target; | |
154 | ||
155 | spin_lock(&btp->bt_lru_lock); | |
156 | if (!list_empty(&bp->b_lru)) { | |
157 | list_del_init(&bp->b_lru); | |
158 | btp->bt_lru_nr--; | |
159 | atomic_dec(&bp->b_hold); | |
160 | } | |
161 | spin_unlock(&btp->bt_lru_lock); | |
162 | } | |
163 | ASSERT(atomic_read(&bp->b_hold) >= 1); | |
164 | } | |
1da177e4 | 165 | |
4347b9d7 CH |
166 | struct xfs_buf * |
167 | xfs_buf_alloc( | |
168 | struct xfs_buftarg *target, | |
204ab25f | 169 | xfs_off_t range_base, |
1da177e4 | 170 | size_t range_length, |
ce8e922c | 171 | xfs_buf_flags_t flags) |
1da177e4 | 172 | { |
4347b9d7 CH |
173 | struct xfs_buf *bp; |
174 | ||
175 | bp = kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags)); | |
176 | if (unlikely(!bp)) | |
177 | return NULL; | |
178 | ||
1da177e4 | 179 | /* |
ce8e922c | 180 | * We don't want certain flags to appear in b_flags. |
1da177e4 | 181 | */ |
ce8e922c NS |
182 | flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD); |
183 | ||
184 | memset(bp, 0, sizeof(xfs_buf_t)); | |
185 | atomic_set(&bp->b_hold, 1); | |
430cbeb8 | 186 | atomic_set(&bp->b_lru_ref, 1); |
b4dd330b | 187 | init_completion(&bp->b_iowait); |
430cbeb8 | 188 | INIT_LIST_HEAD(&bp->b_lru); |
ce8e922c | 189 | INIT_LIST_HEAD(&bp->b_list); |
74f75a0c | 190 | RB_CLEAR_NODE(&bp->b_rbnode); |
a731cd11 | 191 | sema_init(&bp->b_sema, 0); /* held, no waiters */ |
ce8e922c NS |
192 | XB_SET_OWNER(bp); |
193 | bp->b_target = target; | |
194 | bp->b_file_offset = range_base; | |
1da177e4 LT |
195 | /* |
196 | * Set buffer_length and count_desired to the same value initially. | |
197 | * I/O routines should use count_desired, which will be the same in | |
198 | * most cases but may be reset (e.g. XFS recovery). | |
199 | */ | |
ce8e922c NS |
200 | bp->b_buffer_length = bp->b_count_desired = range_length; |
201 | bp->b_flags = flags; | |
202 | bp->b_bn = XFS_BUF_DADDR_NULL; | |
203 | atomic_set(&bp->b_pin_count, 0); | |
204 | init_waitqueue_head(&bp->b_waiters); | |
205 | ||
206 | XFS_STATS_INC(xb_create); | |
0b1b213f | 207 | trace_xfs_buf_init(bp, _RET_IP_); |
4347b9d7 CH |
208 | |
209 | return bp; | |
1da177e4 LT |
210 | } |
211 | ||
212 | /* | |
ce8e922c NS |
213 | * Allocate a page array capable of holding a specified number |
214 | * of pages, and point the page buf at it. | |
1da177e4 LT |
215 | */ |
216 | STATIC int | |
ce8e922c NS |
217 | _xfs_buf_get_pages( |
218 | xfs_buf_t *bp, | |
1da177e4 | 219 | int page_count, |
ce8e922c | 220 | xfs_buf_flags_t flags) |
1da177e4 LT |
221 | { |
222 | /* Make sure that we have a page list */ | |
ce8e922c NS |
223 | if (bp->b_pages == NULL) { |
224 | bp->b_offset = xfs_buf_poff(bp->b_file_offset); | |
225 | bp->b_page_count = page_count; | |
226 | if (page_count <= XB_PAGES) { | |
227 | bp->b_pages = bp->b_page_array; | |
1da177e4 | 228 | } else { |
ce8e922c NS |
229 | bp->b_pages = kmem_alloc(sizeof(struct page *) * |
230 | page_count, xb_to_km(flags)); | |
231 | if (bp->b_pages == NULL) | |
1da177e4 LT |
232 | return -ENOMEM; |
233 | } | |
ce8e922c | 234 | memset(bp->b_pages, 0, sizeof(struct page *) * page_count); |
1da177e4 LT |
235 | } |
236 | return 0; | |
237 | } | |
238 | ||
239 | /* | |
ce8e922c | 240 | * Frees b_pages if it was allocated. |
1da177e4 LT |
241 | */ |
242 | STATIC void | |
ce8e922c | 243 | _xfs_buf_free_pages( |
1da177e4 LT |
244 | xfs_buf_t *bp) |
245 | { | |
ce8e922c | 246 | if (bp->b_pages != bp->b_page_array) { |
f0e2d93c | 247 | kmem_free(bp->b_pages); |
3fc98b1a | 248 | bp->b_pages = NULL; |
1da177e4 LT |
249 | } |
250 | } | |
251 | ||
252 | /* | |
253 | * Releases the specified buffer. | |
254 | * | |
255 | * The modification state of any associated pages is left unchanged. | |
ce8e922c | 256 | * The buffer most not be on any hash - use xfs_buf_rele instead for |
1da177e4 LT |
257 | * hashed and refcounted buffers |
258 | */ | |
259 | void | |
ce8e922c | 260 | xfs_buf_free( |
1da177e4 LT |
261 | xfs_buf_t *bp) |
262 | { | |
0b1b213f | 263 | trace_xfs_buf_free(bp, _RET_IP_); |
1da177e4 | 264 | |
430cbeb8 DC |
265 | ASSERT(list_empty(&bp->b_lru)); |
266 | ||
0e6e847f | 267 | if (bp->b_flags & _XBF_PAGES) { |
1da177e4 LT |
268 | uint i; |
269 | ||
73c77e2c | 270 | if (xfs_buf_is_vmapped(bp)) |
8a262e57 AE |
271 | vm_unmap_ram(bp->b_addr - bp->b_offset, |
272 | bp->b_page_count); | |
1da177e4 | 273 | |
948ecdb4 NS |
274 | for (i = 0; i < bp->b_page_count; i++) { |
275 | struct page *page = bp->b_pages[i]; | |
276 | ||
0e6e847f | 277 | __free_page(page); |
948ecdb4 | 278 | } |
0e6e847f DC |
279 | } else if (bp->b_flags & _XBF_KMEM) |
280 | kmem_free(bp->b_addr); | |
3fc98b1a | 281 | _xfs_buf_free_pages(bp); |
4347b9d7 | 282 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
283 | } |
284 | ||
285 | /* | |
0e6e847f | 286 | * Allocates all the pages for buffer in question and builds it's page list. |
1da177e4 LT |
287 | */ |
288 | STATIC int | |
0e6e847f | 289 | xfs_buf_allocate_memory( |
1da177e4 LT |
290 | xfs_buf_t *bp, |
291 | uint flags) | |
292 | { | |
ce8e922c | 293 | size_t size = bp->b_count_desired; |
1da177e4 | 294 | size_t nbytes, offset; |
ce8e922c | 295 | gfp_t gfp_mask = xb_to_gfp(flags); |
1da177e4 | 296 | unsigned short page_count, i; |
204ab25f | 297 | xfs_off_t end; |
1da177e4 LT |
298 | int error; |
299 | ||
0e6e847f DC |
300 | /* |
301 | * for buffers that are contained within a single page, just allocate | |
302 | * the memory from the heap - there's no need for the complexity of | |
303 | * page arrays to keep allocation down to order 0. | |
304 | */ | |
305 | if (bp->b_buffer_length < PAGE_SIZE) { | |
306 | bp->b_addr = kmem_alloc(bp->b_buffer_length, xb_to_km(flags)); | |
307 | if (!bp->b_addr) { | |
308 | /* low memory - use alloc_page loop instead */ | |
309 | goto use_alloc_page; | |
310 | } | |
311 | ||
312 | if (((unsigned long)(bp->b_addr + bp->b_buffer_length - 1) & | |
313 | PAGE_MASK) != | |
314 | ((unsigned long)bp->b_addr & PAGE_MASK)) { | |
315 | /* b_addr spans two pages - use alloc_page instead */ | |
316 | kmem_free(bp->b_addr); | |
317 | bp->b_addr = NULL; | |
318 | goto use_alloc_page; | |
319 | } | |
320 | bp->b_offset = offset_in_page(bp->b_addr); | |
321 | bp->b_pages = bp->b_page_array; | |
322 | bp->b_pages[0] = virt_to_page(bp->b_addr); | |
323 | bp->b_page_count = 1; | |
324 | bp->b_flags |= XBF_MAPPED | _XBF_KMEM; | |
325 | return 0; | |
326 | } | |
327 | ||
328 | use_alloc_page: | |
ce8e922c NS |
329 | end = bp->b_file_offset + bp->b_buffer_length; |
330 | page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset); | |
ce8e922c | 331 | error = _xfs_buf_get_pages(bp, page_count, flags); |
1da177e4 LT |
332 | if (unlikely(error)) |
333 | return error; | |
1da177e4 | 334 | |
ce8e922c | 335 | offset = bp->b_offset; |
0e6e847f | 336 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 337 | |
ce8e922c | 338 | for (i = 0; i < bp->b_page_count; i++) { |
1da177e4 LT |
339 | struct page *page; |
340 | uint retries = 0; | |
0e6e847f DC |
341 | retry: |
342 | page = alloc_page(gfp_mask); | |
1da177e4 | 343 | if (unlikely(page == NULL)) { |
ce8e922c NS |
344 | if (flags & XBF_READ_AHEAD) { |
345 | bp->b_page_count = i; | |
0e6e847f DC |
346 | error = ENOMEM; |
347 | goto out_free_pages; | |
1da177e4 LT |
348 | } |
349 | ||
350 | /* | |
351 | * This could deadlock. | |
352 | * | |
353 | * But until all the XFS lowlevel code is revamped to | |
354 | * handle buffer allocation failures we can't do much. | |
355 | */ | |
356 | if (!(++retries % 100)) | |
4f10700a DC |
357 | xfs_err(NULL, |
358 | "possible memory allocation deadlock in %s (mode:0x%x)", | |
34a622b2 | 359 | __func__, gfp_mask); |
1da177e4 | 360 | |
ce8e922c | 361 | XFS_STATS_INC(xb_page_retries); |
8aa7e847 | 362 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
363 | goto retry; |
364 | } | |
365 | ||
ce8e922c | 366 | XFS_STATS_INC(xb_page_found); |
1da177e4 | 367 | |
0e6e847f | 368 | nbytes = min_t(size_t, size, PAGE_SIZE - offset); |
1da177e4 | 369 | size -= nbytes; |
ce8e922c | 370 | bp->b_pages[i] = page; |
1da177e4 LT |
371 | offset = 0; |
372 | } | |
0e6e847f | 373 | return 0; |
1da177e4 | 374 | |
0e6e847f DC |
375 | out_free_pages: |
376 | for (i = 0; i < bp->b_page_count; i++) | |
377 | __free_page(bp->b_pages[i]); | |
1da177e4 LT |
378 | return error; |
379 | } | |
380 | ||
381 | /* | |
25985edc | 382 | * Map buffer into kernel address-space if necessary. |
1da177e4 LT |
383 | */ |
384 | STATIC int | |
ce8e922c | 385 | _xfs_buf_map_pages( |
1da177e4 LT |
386 | xfs_buf_t *bp, |
387 | uint flags) | |
388 | { | |
0e6e847f | 389 | ASSERT(bp->b_flags & _XBF_PAGES); |
ce8e922c | 390 | if (bp->b_page_count == 1) { |
0e6e847f | 391 | /* A single page buffer is always mappable */ |
ce8e922c NS |
392 | bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset; |
393 | bp->b_flags |= XBF_MAPPED; | |
394 | } else if (flags & XBF_MAPPED) { | |
a19fb380 DC |
395 | int retried = 0; |
396 | ||
397 | do { | |
398 | bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count, | |
399 | -1, PAGE_KERNEL); | |
400 | if (bp->b_addr) | |
401 | break; | |
402 | vm_unmap_aliases(); | |
403 | } while (retried++ <= 1); | |
404 | ||
405 | if (!bp->b_addr) | |
1da177e4 | 406 | return -ENOMEM; |
ce8e922c NS |
407 | bp->b_addr += bp->b_offset; |
408 | bp->b_flags |= XBF_MAPPED; | |
1da177e4 LT |
409 | } |
410 | ||
411 | return 0; | |
412 | } | |
413 | ||
414 | /* | |
415 | * Finding and Reading Buffers | |
416 | */ | |
417 | ||
418 | /* | |
ce8e922c | 419 | * Look up, and creates if absent, a lockable buffer for |
1da177e4 | 420 | * a given range of an inode. The buffer is returned |
eabbaf11 | 421 | * locked. No I/O is implied by this call. |
1da177e4 LT |
422 | */ |
423 | xfs_buf_t * | |
ce8e922c | 424 | _xfs_buf_find( |
1da177e4 | 425 | xfs_buftarg_t *btp, /* block device target */ |
204ab25f | 426 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 427 | size_t isize, /* length of range */ |
ce8e922c NS |
428 | xfs_buf_flags_t flags, |
429 | xfs_buf_t *new_bp) | |
1da177e4 | 430 | { |
204ab25f | 431 | xfs_off_t range_base; |
1da177e4 | 432 | size_t range_length; |
74f75a0c DC |
433 | struct xfs_perag *pag; |
434 | struct rb_node **rbp; | |
435 | struct rb_node *parent; | |
436 | xfs_buf_t *bp; | |
1da177e4 LT |
437 | |
438 | range_base = (ioff << BBSHIFT); | |
439 | range_length = (isize << BBSHIFT); | |
440 | ||
441 | /* Check for IOs smaller than the sector size / not sector aligned */ | |
ce8e922c | 442 | ASSERT(!(range_length < (1 << btp->bt_sshift))); |
204ab25f | 443 | ASSERT(!(range_base & (xfs_off_t)btp->bt_smask)); |
1da177e4 | 444 | |
74f75a0c DC |
445 | /* get tree root */ |
446 | pag = xfs_perag_get(btp->bt_mount, | |
447 | xfs_daddr_to_agno(btp->bt_mount, ioff)); | |
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 | ||
458 | if (range_base < bp->b_file_offset) | |
459 | rbp = &(*rbp)->rb_left; | |
460 | else if (range_base > bp->b_file_offset) | |
461 | rbp = &(*rbp)->rb_right; | |
462 | else { | |
463 | /* | |
464 | * found a block offset match. If the range doesn't | |
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 | */ | |
471 | if (bp->b_buffer_length != range_length) { | |
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); | |
0e6e847f | 516 | bp->b_flags &= XBF_MAPPED | _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( |
1da177e4 | 531 | xfs_buftarg_t *target,/* target for buffer */ |
204ab25f | 532 | xfs_off_t ioff, /* starting offset of range */ |
1da177e4 | 533 | size_t isize, /* length of range */ |
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 | |
3815832a DC |
540 | bp = _xfs_buf_find(target, ioff, isize, flags, NULL); |
541 | if (likely(bp)) | |
542 | goto found; | |
543 | ||
4347b9d7 CH |
544 | new_bp = xfs_buf_alloc(target, ioff << BBSHIFT, isize << BBSHIFT, |
545 | flags); | |
ce8e922c | 546 | if (unlikely(!new_bp)) |
1da177e4 LT |
547 | return NULL; |
548 | ||
ce8e922c | 549 | bp = _xfs_buf_find(target, ioff, isize, flags, new_bp); |
3815832a | 550 | if (!bp) { |
4347b9d7 | 551 | kmem_zone_free(xfs_buf_zone, new_bp); |
3815832a DC |
552 | return NULL; |
553 | } | |
554 | ||
ce8e922c | 555 | if (bp == new_bp) { |
0e6e847f | 556 | error = xfs_buf_allocate_memory(bp, flags); |
1da177e4 LT |
557 | if (error) |
558 | goto no_buffer; | |
3815832a | 559 | } else |
4347b9d7 | 560 | kmem_zone_free(xfs_buf_zone, new_bp); |
1da177e4 | 561 | |
3815832a DC |
562 | /* |
563 | * Now we have a workable buffer, fill in the block number so | |
564 | * that we can do IO on it. | |
565 | */ | |
566 | bp->b_bn = ioff; | |
567 | bp->b_count_desired = bp->b_buffer_length; | |
568 | ||
569 | found: | |
ce8e922c NS |
570 | if (!(bp->b_flags & XBF_MAPPED)) { |
571 | error = _xfs_buf_map_pages(bp, flags); | |
1da177e4 | 572 | if (unlikely(error)) { |
4f10700a DC |
573 | xfs_warn(target->bt_mount, |
574 | "%s: failed to map pages\n", __func__); | |
1da177e4 LT |
575 | goto no_buffer; |
576 | } | |
577 | } | |
578 | ||
ce8e922c | 579 | XFS_STATS_INC(xb_get); |
0b1b213f | 580 | trace_xfs_buf_get(bp, flags, _RET_IP_); |
ce8e922c | 581 | return bp; |
1da177e4 | 582 | |
3815832a | 583 | no_buffer: |
ce8e922c NS |
584 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
585 | xfs_buf_unlock(bp); | |
586 | xfs_buf_rele(bp); | |
1da177e4 LT |
587 | return NULL; |
588 | } | |
589 | ||
5d765b97 CH |
590 | STATIC int |
591 | _xfs_buf_read( | |
592 | xfs_buf_t *bp, | |
593 | xfs_buf_flags_t flags) | |
594 | { | |
595 | int status; | |
596 | ||
5d765b97 CH |
597 | ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE))); |
598 | ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL); | |
599 | ||
1d5ae5df CH |
600 | bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | XBF_READ_AHEAD); |
601 | bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD); | |
5d765b97 CH |
602 | |
603 | status = xfs_buf_iorequest(bp); | |
5a52c2a5 | 604 | if (status || bp->b_error || (flags & XBF_ASYNC)) |
ec53d1db DC |
605 | return status; |
606 | return xfs_buf_iowait(bp); | |
5d765b97 CH |
607 | } |
608 | ||
1da177e4 | 609 | xfs_buf_t * |
6ad112bf | 610 | xfs_buf_read( |
1da177e4 | 611 | xfs_buftarg_t *target, |
204ab25f | 612 | xfs_off_t ioff, |
1da177e4 | 613 | size_t isize, |
ce8e922c | 614 | xfs_buf_flags_t flags) |
1da177e4 | 615 | { |
ce8e922c NS |
616 | xfs_buf_t *bp; |
617 | ||
618 | flags |= XBF_READ; | |
619 | ||
6ad112bf | 620 | bp = xfs_buf_get(target, ioff, isize, flags); |
ce8e922c | 621 | if (bp) { |
0b1b213f CH |
622 | trace_xfs_buf_read(bp, flags, _RET_IP_); |
623 | ||
ce8e922c | 624 | if (!XFS_BUF_ISDONE(bp)) { |
ce8e922c | 625 | XFS_STATS_INC(xb_get_read); |
5d765b97 | 626 | _xfs_buf_read(bp, flags); |
ce8e922c | 627 | } else if (flags & XBF_ASYNC) { |
1da177e4 LT |
628 | /* |
629 | * Read ahead call which is already satisfied, | |
630 | * drop the buffer | |
631 | */ | |
632 | goto no_buffer; | |
633 | } else { | |
1da177e4 | 634 | /* We do not want read in the flags */ |
ce8e922c | 635 | bp->b_flags &= ~XBF_READ; |
1da177e4 LT |
636 | } |
637 | } | |
638 | ||
ce8e922c | 639 | return bp; |
1da177e4 LT |
640 | |
641 | no_buffer: | |
ce8e922c NS |
642 | if (flags & (XBF_LOCK | XBF_TRYLOCK)) |
643 | xfs_buf_unlock(bp); | |
644 | xfs_buf_rele(bp); | |
1da177e4 LT |
645 | return NULL; |
646 | } | |
647 | ||
1da177e4 | 648 | /* |
ce8e922c NS |
649 | * If we are not low on memory then do the readahead in a deadlock |
650 | * safe manner. | |
1da177e4 LT |
651 | */ |
652 | void | |
ce8e922c | 653 | xfs_buf_readahead( |
1da177e4 | 654 | xfs_buftarg_t *target, |
204ab25f | 655 | xfs_off_t ioff, |
1a1a3e97 | 656 | size_t isize) |
1da177e4 | 657 | { |
0e6e847f | 658 | if (bdi_read_congested(target->bt_bdi)) |
1da177e4 LT |
659 | return; |
660 | ||
1a1a3e97 CH |
661 | xfs_buf_read(target, ioff, isize, |
662 | XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD|XBF_DONT_BLOCK); | |
1da177e4 LT |
663 | } |
664 | ||
5adc94c2 DC |
665 | /* |
666 | * Read an uncached buffer from disk. Allocates and returns a locked | |
667 | * buffer containing the disk contents or nothing. | |
668 | */ | |
669 | struct xfs_buf * | |
670 | xfs_buf_read_uncached( | |
671 | struct xfs_mount *mp, | |
672 | struct xfs_buftarg *target, | |
673 | xfs_daddr_t daddr, | |
674 | size_t length, | |
675 | int flags) | |
676 | { | |
677 | xfs_buf_t *bp; | |
678 | int error; | |
679 | ||
680 | bp = xfs_buf_get_uncached(target, length, flags); | |
681 | if (!bp) | |
682 | return NULL; | |
683 | ||
684 | /* set up the buffer for a read IO */ | |
5adc94c2 DC |
685 | XFS_BUF_SET_ADDR(bp, daddr); |
686 | XFS_BUF_READ(bp); | |
5adc94c2 DC |
687 | |
688 | xfsbdstrat(mp, bp); | |
1a1a3e97 | 689 | error = xfs_buf_iowait(bp); |
5adc94c2 DC |
690 | if (error || bp->b_error) { |
691 | xfs_buf_relse(bp); | |
692 | return NULL; | |
693 | } | |
694 | return bp; | |
1da177e4 LT |
695 | } |
696 | ||
44396476 DC |
697 | /* |
698 | * Return a buffer allocated as an empty buffer and associated to external | |
699 | * memory via xfs_buf_associate_memory() back to it's empty state. | |
700 | */ | |
701 | void | |
702 | xfs_buf_set_empty( | |
703 | struct xfs_buf *bp, | |
704 | size_t len) | |
705 | { | |
706 | if (bp->b_pages) | |
707 | _xfs_buf_free_pages(bp); | |
708 | ||
709 | bp->b_pages = NULL; | |
710 | bp->b_page_count = 0; | |
711 | bp->b_addr = NULL; | |
712 | bp->b_file_offset = 0; | |
713 | bp->b_buffer_length = bp->b_count_desired = len; | |
714 | bp->b_bn = XFS_BUF_DADDR_NULL; | |
715 | bp->b_flags &= ~XBF_MAPPED; | |
716 | } | |
717 | ||
1da177e4 LT |
718 | static inline struct page * |
719 | mem_to_page( | |
720 | void *addr) | |
721 | { | |
9e2779fa | 722 | if ((!is_vmalloc_addr(addr))) { |
1da177e4 LT |
723 | return virt_to_page(addr); |
724 | } else { | |
725 | return vmalloc_to_page(addr); | |
726 | } | |
727 | } | |
728 | ||
729 | int | |
ce8e922c NS |
730 | xfs_buf_associate_memory( |
731 | xfs_buf_t *bp, | |
1da177e4 LT |
732 | void *mem, |
733 | size_t len) | |
734 | { | |
735 | int rval; | |
736 | int i = 0; | |
d1afb678 LM |
737 | unsigned long pageaddr; |
738 | unsigned long offset; | |
739 | size_t buflen; | |
1da177e4 LT |
740 | int page_count; |
741 | ||
0e6e847f | 742 | pageaddr = (unsigned long)mem & PAGE_MASK; |
d1afb678 | 743 | offset = (unsigned long)mem - pageaddr; |
0e6e847f DC |
744 | buflen = PAGE_ALIGN(len + offset); |
745 | page_count = buflen >> PAGE_SHIFT; | |
1da177e4 LT |
746 | |
747 | /* Free any previous set of page pointers */ | |
ce8e922c NS |
748 | if (bp->b_pages) |
749 | _xfs_buf_free_pages(bp); | |
1da177e4 | 750 | |
ce8e922c NS |
751 | bp->b_pages = NULL; |
752 | bp->b_addr = mem; | |
1da177e4 | 753 | |
36fae17a | 754 | rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK); |
1da177e4 LT |
755 | if (rval) |
756 | return rval; | |
757 | ||
ce8e922c | 758 | bp->b_offset = offset; |
d1afb678 LM |
759 | |
760 | for (i = 0; i < bp->b_page_count; i++) { | |
761 | bp->b_pages[i] = mem_to_page((void *)pageaddr); | |
0e6e847f | 762 | pageaddr += PAGE_SIZE; |
1da177e4 | 763 | } |
1da177e4 | 764 | |
d1afb678 LM |
765 | bp->b_count_desired = len; |
766 | bp->b_buffer_length = buflen; | |
ce8e922c | 767 | bp->b_flags |= XBF_MAPPED; |
1da177e4 LT |
768 | |
769 | return 0; | |
770 | } | |
771 | ||
772 | xfs_buf_t * | |
686865f7 DC |
773 | xfs_buf_get_uncached( |
774 | struct xfs_buftarg *target, | |
1da177e4 | 775 | size_t len, |
686865f7 | 776 | int flags) |
1da177e4 | 777 | { |
1fa40b01 CH |
778 | unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT; |
779 | int error, i; | |
1da177e4 | 780 | xfs_buf_t *bp; |
1da177e4 | 781 | |
4347b9d7 | 782 | bp = xfs_buf_alloc(target, 0, len, 0); |
1da177e4 LT |
783 | if (unlikely(bp == NULL)) |
784 | goto fail; | |
1da177e4 | 785 | |
1fa40b01 CH |
786 | error = _xfs_buf_get_pages(bp, page_count, 0); |
787 | if (error) | |
1da177e4 LT |
788 | goto fail_free_buf; |
789 | ||
1fa40b01 | 790 | for (i = 0; i < page_count; i++) { |
686865f7 | 791 | bp->b_pages[i] = alloc_page(xb_to_gfp(flags)); |
1fa40b01 CH |
792 | if (!bp->b_pages[i]) |
793 | goto fail_free_mem; | |
1da177e4 | 794 | } |
1fa40b01 | 795 | bp->b_flags |= _XBF_PAGES; |
1da177e4 | 796 | |
1fa40b01 CH |
797 | error = _xfs_buf_map_pages(bp, XBF_MAPPED); |
798 | if (unlikely(error)) { | |
4f10700a DC |
799 | xfs_warn(target->bt_mount, |
800 | "%s: failed to map pages\n", __func__); | |
1da177e4 | 801 | goto fail_free_mem; |
1fa40b01 | 802 | } |
1da177e4 | 803 | |
686865f7 | 804 | trace_xfs_buf_get_uncached(bp, _RET_IP_); |
1da177e4 | 805 | return bp; |
1fa40b01 | 806 | |
1da177e4 | 807 | fail_free_mem: |
1fa40b01 CH |
808 | while (--i >= 0) |
809 | __free_page(bp->b_pages[i]); | |
ca165b88 | 810 | _xfs_buf_free_pages(bp); |
1da177e4 | 811 | fail_free_buf: |
4347b9d7 | 812 | kmem_zone_free(xfs_buf_zone, bp); |
1da177e4 LT |
813 | fail: |
814 | return NULL; | |
815 | } | |
816 | ||
817 | /* | |
1da177e4 LT |
818 | * Increment reference count on buffer, to hold the buffer concurrently |
819 | * with another thread which may release (free) the buffer asynchronously. | |
1da177e4 LT |
820 | * Must hold the buffer already to call this function. |
821 | */ | |
822 | void | |
ce8e922c NS |
823 | xfs_buf_hold( |
824 | xfs_buf_t *bp) | |
1da177e4 | 825 | { |
0b1b213f | 826 | trace_xfs_buf_hold(bp, _RET_IP_); |
ce8e922c | 827 | atomic_inc(&bp->b_hold); |
1da177e4 LT |
828 | } |
829 | ||
830 | /* | |
ce8e922c NS |
831 | * Releases a hold on the specified buffer. If the |
832 | * the hold count is 1, calls xfs_buf_free. | |
1da177e4 LT |
833 | */ |
834 | void | |
ce8e922c NS |
835 | xfs_buf_rele( |
836 | xfs_buf_t *bp) | |
1da177e4 | 837 | { |
74f75a0c | 838 | struct xfs_perag *pag = bp->b_pag; |
1da177e4 | 839 | |
0b1b213f | 840 | trace_xfs_buf_rele(bp, _RET_IP_); |
1da177e4 | 841 | |
74f75a0c | 842 | if (!pag) { |
430cbeb8 | 843 | ASSERT(list_empty(&bp->b_lru)); |
74f75a0c | 844 | ASSERT(RB_EMPTY_NODE(&bp->b_rbnode)); |
fad3aa1e NS |
845 | if (atomic_dec_and_test(&bp->b_hold)) |
846 | xfs_buf_free(bp); | |
847 | return; | |
848 | } | |
849 | ||
74f75a0c | 850 | ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode)); |
430cbeb8 | 851 | |
3790689f | 852 | ASSERT(atomic_read(&bp->b_hold) > 0); |
74f75a0c | 853 | if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) { |
bfc60177 | 854 | if (!(bp->b_flags & XBF_STALE) && |
430cbeb8 DC |
855 | atomic_read(&bp->b_lru_ref)) { |
856 | xfs_buf_lru_add(bp); | |
857 | spin_unlock(&pag->pag_buf_lock); | |
1da177e4 | 858 | } else { |
430cbeb8 | 859 | xfs_buf_lru_del(bp); |
ce8e922c | 860 | ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q))); |
74f75a0c DC |
861 | rb_erase(&bp->b_rbnode, &pag->pag_buf_tree); |
862 | spin_unlock(&pag->pag_buf_lock); | |
863 | xfs_perag_put(pag); | |
ce8e922c | 864 | xfs_buf_free(bp); |
1da177e4 LT |
865 | } |
866 | } | |
867 | } | |
868 | ||
869 | ||
870 | /* | |
0e6e847f | 871 | * Lock a buffer object, if it is not already locked. |
90810b9e DC |
872 | * |
873 | * If we come across a stale, pinned, locked buffer, we know that we are | |
874 | * being asked to lock a buffer that has been reallocated. Because it is | |
875 | * pinned, we know that the log has not been pushed to disk and hence it | |
876 | * will still be locked. Rather than continuing to have trylock attempts | |
877 | * fail until someone else pushes the log, push it ourselves before | |
878 | * returning. This means that the xfsaild will not get stuck trying | |
879 | * to push on stale inode buffers. | |
1da177e4 LT |
880 | */ |
881 | int | |
0c842ad4 CH |
882 | xfs_buf_trylock( |
883 | struct xfs_buf *bp) | |
1da177e4 LT |
884 | { |
885 | int locked; | |
886 | ||
ce8e922c | 887 | locked = down_trylock(&bp->b_sema) == 0; |
0b1b213f | 888 | if (locked) |
ce8e922c | 889 | XB_SET_OWNER(bp); |
90810b9e DC |
890 | else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
891 | xfs_log_force(bp->b_target->bt_mount, 0); | |
0b1b213f | 892 | |
0c842ad4 CH |
893 | trace_xfs_buf_trylock(bp, _RET_IP_); |
894 | return locked; | |
1da177e4 | 895 | } |
1da177e4 LT |
896 | |
897 | /* | |
0e6e847f | 898 | * Lock a buffer object. |
ed3b4d6c DC |
899 | * |
900 | * If we come across a stale, pinned, locked buffer, we know that we | |
901 | * are being asked to lock a buffer that has been reallocated. Because | |
902 | * it is pinned, we know that the log has not been pushed to disk and | |
903 | * hence it will still be locked. Rather than sleeping until someone | |
904 | * else pushes the log, push it ourselves before trying to get the lock. | |
1da177e4 | 905 | */ |
ce8e922c NS |
906 | void |
907 | xfs_buf_lock( | |
0c842ad4 | 908 | struct xfs_buf *bp) |
1da177e4 | 909 | { |
0b1b213f CH |
910 | trace_xfs_buf_lock(bp, _RET_IP_); |
911 | ||
ed3b4d6c | 912 | if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE)) |
ebad861b | 913 | xfs_log_force(bp->b_target->bt_mount, 0); |
ce8e922c NS |
914 | down(&bp->b_sema); |
915 | XB_SET_OWNER(bp); | |
0b1b213f CH |
916 | |
917 | trace_xfs_buf_lock_done(bp, _RET_IP_); | |
1da177e4 LT |
918 | } |
919 | ||
920 | /* | |
ce8e922c | 921 | * Releases the lock on the buffer object. |
2f926587 | 922 | * If the buffer is marked delwri but is not queued, do so before we |
ce8e922c | 923 | * unlock the buffer as we need to set flags correctly. We also need to |
2f926587 DC |
924 | * take a reference for the delwri queue because the unlocker is going to |
925 | * drop their's and they don't know we just queued it. | |
1da177e4 LT |
926 | */ |
927 | void | |
ce8e922c | 928 | xfs_buf_unlock( |
0c842ad4 | 929 | struct xfs_buf *bp) |
1da177e4 | 930 | { |
ce8e922c NS |
931 | XB_CLEAR_OWNER(bp); |
932 | up(&bp->b_sema); | |
0b1b213f CH |
933 | |
934 | trace_xfs_buf_unlock(bp, _RET_IP_); | |
1da177e4 LT |
935 | } |
936 | ||
ce8e922c NS |
937 | STATIC void |
938 | xfs_buf_wait_unpin( | |
939 | xfs_buf_t *bp) | |
1da177e4 LT |
940 | { |
941 | DECLARE_WAITQUEUE (wait, current); | |
942 | ||
ce8e922c | 943 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 LT |
944 | return; |
945 | ||
ce8e922c | 946 | add_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
947 | for (;;) { |
948 | set_current_state(TASK_UNINTERRUPTIBLE); | |
ce8e922c | 949 | if (atomic_read(&bp->b_pin_count) == 0) |
1da177e4 | 950 | break; |
7eaceacc | 951 | io_schedule(); |
1da177e4 | 952 | } |
ce8e922c | 953 | remove_wait_queue(&bp->b_waiters, &wait); |
1da177e4 LT |
954 | set_current_state(TASK_RUNNING); |
955 | } | |
956 | ||
957 | /* | |
958 | * Buffer Utility Routines | |
959 | */ | |
960 | ||
1da177e4 | 961 | STATIC void |
ce8e922c | 962 | xfs_buf_iodone_work( |
c4028958 | 963 | struct work_struct *work) |
1da177e4 | 964 | { |
c4028958 DH |
965 | xfs_buf_t *bp = |
966 | container_of(work, xfs_buf_t, b_iodone_work); | |
1da177e4 | 967 | |
80f6c29d | 968 | if (bp->b_iodone) |
ce8e922c NS |
969 | (*(bp->b_iodone))(bp); |
970 | else if (bp->b_flags & XBF_ASYNC) | |
1da177e4 LT |
971 | xfs_buf_relse(bp); |
972 | } | |
973 | ||
974 | void | |
ce8e922c NS |
975 | xfs_buf_ioend( |
976 | xfs_buf_t *bp, | |
1da177e4 LT |
977 | int schedule) |
978 | { | |
0b1b213f CH |
979 | trace_xfs_buf_iodone(bp, _RET_IP_); |
980 | ||
77be55a5 | 981 | bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD); |
ce8e922c NS |
982 | if (bp->b_error == 0) |
983 | bp->b_flags |= XBF_DONE; | |
1da177e4 | 984 | |
ce8e922c | 985 | if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) { |
1da177e4 | 986 | if (schedule) { |
c4028958 | 987 | INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work); |
ce8e922c | 988 | queue_work(xfslogd_workqueue, &bp->b_iodone_work); |
1da177e4 | 989 | } else { |
c4028958 | 990 | xfs_buf_iodone_work(&bp->b_iodone_work); |
1da177e4 LT |
991 | } |
992 | } else { | |
b4dd330b | 993 | complete(&bp->b_iowait); |
1da177e4 LT |
994 | } |
995 | } | |
996 | ||
1da177e4 | 997 | void |
ce8e922c NS |
998 | xfs_buf_ioerror( |
999 | xfs_buf_t *bp, | |
1000 | int error) | |
1da177e4 LT |
1001 | { |
1002 | ASSERT(error >= 0 && error <= 0xffff); | |
ce8e922c | 1003 | bp->b_error = (unsigned short)error; |
0b1b213f | 1004 | trace_xfs_buf_ioerror(bp, error, _RET_IP_); |
1da177e4 LT |
1005 | } |
1006 | ||
1da177e4 | 1007 | int |
64e0bc7d | 1008 | xfs_bwrite( |
5d765b97 | 1009 | struct xfs_buf *bp) |
1da177e4 | 1010 | { |
8c38366f | 1011 | int error; |
1da177e4 | 1012 | |
64e0bc7d | 1013 | bp->b_flags |= XBF_WRITE; |
8c38366f | 1014 | bp->b_flags &= ~(XBF_ASYNC | XBF_READ); |
1da177e4 | 1015 | |
5d765b97 | 1016 | xfs_buf_delwri_dequeue(bp); |
939d723b | 1017 | xfs_bdstrat_cb(bp); |
1da177e4 | 1018 | |
8c38366f | 1019 | error = xfs_buf_iowait(bp); |
c2b006c1 CH |
1020 | if (error) { |
1021 | xfs_force_shutdown(bp->b_target->bt_mount, | |
1022 | SHUTDOWN_META_IO_ERROR); | |
1023 | } | |
64e0bc7d | 1024 | return error; |
5d765b97 | 1025 | } |
1da177e4 | 1026 | |
4e23471a CH |
1027 | /* |
1028 | * Called when we want to stop a buffer from getting written or read. | |
1a1a3e97 | 1029 | * We attach the EIO error, muck with its flags, and call xfs_buf_ioend |
4e23471a CH |
1030 | * so that the proper iodone callbacks get called. |
1031 | */ | |
1032 | STATIC int | |
1033 | xfs_bioerror( | |
1034 | xfs_buf_t *bp) | |
1035 | { | |
1036 | #ifdef XFSERRORDEBUG | |
1037 | ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone); | |
1038 | #endif | |
1039 | ||
1040 | /* | |
1041 | * No need to wait until the buffer is unpinned, we aren't flushing it. | |
1042 | */ | |
5a52c2a5 | 1043 | xfs_buf_ioerror(bp, EIO); |
4e23471a CH |
1044 | |
1045 | /* | |
1a1a3e97 | 1046 | * We're calling xfs_buf_ioend, so delete XBF_DONE flag. |
4e23471a CH |
1047 | */ |
1048 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1049 | XFS_BUF_UNDONE(bp); |
c867cb61 | 1050 | xfs_buf_stale(bp); |
4e23471a | 1051 | |
1a1a3e97 | 1052 | xfs_buf_ioend(bp, 0); |
4e23471a CH |
1053 | |
1054 | return EIO; | |
1055 | } | |
1056 | ||
1057 | /* | |
1058 | * Same as xfs_bioerror, except that we are releasing the buffer | |
1a1a3e97 | 1059 | * here ourselves, and avoiding the xfs_buf_ioend call. |
4e23471a CH |
1060 | * This is meant for userdata errors; metadata bufs come with |
1061 | * iodone functions attached, so that we can track down errors. | |
1062 | */ | |
1063 | STATIC int | |
1064 | xfs_bioerror_relse( | |
1065 | struct xfs_buf *bp) | |
1066 | { | |
ed43233b | 1067 | int64_t fl = bp->b_flags; |
4e23471a CH |
1068 | /* |
1069 | * No need to wait until the buffer is unpinned. | |
1070 | * We aren't flushing it. | |
1071 | * | |
1072 | * chunkhold expects B_DONE to be set, whether | |
1073 | * we actually finish the I/O or not. We don't want to | |
1074 | * change that interface. | |
1075 | */ | |
1076 | XFS_BUF_UNREAD(bp); | |
4e23471a | 1077 | XFS_BUF_DONE(bp); |
c867cb61 | 1078 | xfs_buf_stale(bp); |
cb669ca5 | 1079 | bp->b_iodone = NULL; |
0cadda1c | 1080 | if (!(fl & XBF_ASYNC)) { |
4e23471a CH |
1081 | /* |
1082 | * Mark b_error and B_ERROR _both_. | |
1083 | * Lot's of chunkcache code assumes that. | |
1084 | * There's no reason to mark error for | |
1085 | * ASYNC buffers. | |
1086 | */ | |
5a52c2a5 | 1087 | xfs_buf_ioerror(bp, EIO); |
5fde0326 | 1088 | complete(&bp->b_iowait); |
4e23471a CH |
1089 | } else { |
1090 | xfs_buf_relse(bp); | |
1091 | } | |
1092 | ||
1093 | return EIO; | |
1094 | } | |
1095 | ||
1096 | ||
1097 | /* | |
1098 | * All xfs metadata buffers except log state machine buffers | |
1099 | * get this attached as their b_bdstrat callback function. | |
1100 | * This is so that we can catch a buffer | |
1101 | * after prematurely unpinning it to forcibly shutdown the filesystem. | |
1102 | */ | |
1103 | int | |
1104 | xfs_bdstrat_cb( | |
1105 | struct xfs_buf *bp) | |
1106 | { | |
ebad861b | 1107 | if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
4e23471a CH |
1108 | trace_xfs_bdstrat_shut(bp, _RET_IP_); |
1109 | /* | |
1110 | * Metadata write that didn't get logged but | |
1111 | * written delayed anyway. These aren't associated | |
1112 | * with a transaction, and can be ignored. | |
1113 | */ | |
1114 | if (!bp->b_iodone && !XFS_BUF_ISREAD(bp)) | |
1115 | return xfs_bioerror_relse(bp); | |
1116 | else | |
1117 | return xfs_bioerror(bp); | |
1118 | } | |
1119 | ||
1120 | xfs_buf_iorequest(bp); | |
1121 | return 0; | |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * Wrapper around bdstrat so that we can stop data from going to disk in case | |
1126 | * we are shutting down the filesystem. Typically user data goes thru this | |
1127 | * path; one of the exceptions is the superblock. | |
1128 | */ | |
1129 | void | |
1130 | xfsbdstrat( | |
1131 | struct xfs_mount *mp, | |
1132 | struct xfs_buf *bp) | |
1133 | { | |
1134 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1135 | trace_xfs_bdstrat_shut(bp, _RET_IP_); | |
1136 | xfs_bioerror_relse(bp); | |
1137 | return; | |
1138 | } | |
1139 | ||
1140 | xfs_buf_iorequest(bp); | |
1141 | } | |
1142 | ||
b8f82a4a | 1143 | STATIC void |
ce8e922c NS |
1144 | _xfs_buf_ioend( |
1145 | xfs_buf_t *bp, | |
1da177e4 LT |
1146 | int schedule) |
1147 | { | |
0e6e847f | 1148 | if (atomic_dec_and_test(&bp->b_io_remaining) == 1) |
ce8e922c | 1149 | xfs_buf_ioend(bp, schedule); |
1da177e4 LT |
1150 | } |
1151 | ||
782e3b3b | 1152 | STATIC void |
ce8e922c | 1153 | xfs_buf_bio_end_io( |
1da177e4 | 1154 | struct bio *bio, |
1da177e4 LT |
1155 | int error) |
1156 | { | |
ce8e922c | 1157 | xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private; |
1da177e4 | 1158 | |
cfbe5267 | 1159 | xfs_buf_ioerror(bp, -error); |
1da177e4 | 1160 | |
73c77e2c JB |
1161 | if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ)) |
1162 | invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp)); | |
1163 | ||
ce8e922c | 1164 | _xfs_buf_ioend(bp, 1); |
1da177e4 | 1165 | bio_put(bio); |
1da177e4 LT |
1166 | } |
1167 | ||
1168 | STATIC void | |
ce8e922c NS |
1169 | _xfs_buf_ioapply( |
1170 | xfs_buf_t *bp) | |
1da177e4 | 1171 | { |
a9759f2d | 1172 | int rw, map_i, total_nr_pages, nr_pages; |
1da177e4 | 1173 | struct bio *bio; |
ce8e922c NS |
1174 | int offset = bp->b_offset; |
1175 | int size = bp->b_count_desired; | |
1176 | sector_t sector = bp->b_bn; | |
1da177e4 | 1177 | |
ce8e922c | 1178 | total_nr_pages = bp->b_page_count; |
1da177e4 LT |
1179 | map_i = 0; |
1180 | ||
1d5ae5df CH |
1181 | if (bp->b_flags & XBF_WRITE) { |
1182 | if (bp->b_flags & XBF_SYNCIO) | |
1183 | rw = WRITE_SYNC; | |
1184 | else | |
1185 | rw = WRITE; | |
1186 | if (bp->b_flags & XBF_FUA) | |
1187 | rw |= REQ_FUA; | |
1188 | if (bp->b_flags & XBF_FLUSH) | |
1189 | rw |= REQ_FLUSH; | |
1190 | } else if (bp->b_flags & XBF_READ_AHEAD) { | |
1191 | rw = READA; | |
51bdd706 | 1192 | } else { |
1d5ae5df | 1193 | rw = READ; |
f538d4da CH |
1194 | } |
1195 | ||
34951f5c CH |
1196 | /* we only use the buffer cache for meta-data */ |
1197 | rw |= REQ_META; | |
1198 | ||
1da177e4 | 1199 | next_chunk: |
ce8e922c | 1200 | atomic_inc(&bp->b_io_remaining); |
1da177e4 LT |
1201 | nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT); |
1202 | if (nr_pages > total_nr_pages) | |
1203 | nr_pages = total_nr_pages; | |
1204 | ||
1205 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
ce8e922c | 1206 | bio->bi_bdev = bp->b_target->bt_bdev; |
1da177e4 | 1207 | bio->bi_sector = sector; |
ce8e922c NS |
1208 | bio->bi_end_io = xfs_buf_bio_end_io; |
1209 | bio->bi_private = bp; | |
1da177e4 | 1210 | |
0e6e847f | 1211 | |
1da177e4 | 1212 | for (; size && nr_pages; nr_pages--, map_i++) { |
0e6e847f | 1213 | int rbytes, nbytes = PAGE_SIZE - offset; |
1da177e4 LT |
1214 | |
1215 | if (nbytes > size) | |
1216 | nbytes = size; | |
1217 | ||
ce8e922c NS |
1218 | rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset); |
1219 | if (rbytes < nbytes) | |
1da177e4 LT |
1220 | break; |
1221 | ||
1222 | offset = 0; | |
1223 | sector += nbytes >> BBSHIFT; | |
1224 | size -= nbytes; | |
1225 | total_nr_pages--; | |
1226 | } | |
1227 | ||
1da177e4 | 1228 | if (likely(bio->bi_size)) { |
73c77e2c JB |
1229 | if (xfs_buf_is_vmapped(bp)) { |
1230 | flush_kernel_vmap_range(bp->b_addr, | |
1231 | xfs_buf_vmap_len(bp)); | |
1232 | } | |
1da177e4 LT |
1233 | submit_bio(rw, bio); |
1234 | if (size) | |
1235 | goto next_chunk; | |
1236 | } else { | |
ce8e922c | 1237 | xfs_buf_ioerror(bp, EIO); |
ec53d1db | 1238 | bio_put(bio); |
1da177e4 LT |
1239 | } |
1240 | } | |
1241 | ||
1da177e4 | 1242 | int |
ce8e922c NS |
1243 | xfs_buf_iorequest( |
1244 | xfs_buf_t *bp) | |
1da177e4 | 1245 | { |
0b1b213f | 1246 | trace_xfs_buf_iorequest(bp, _RET_IP_); |
1da177e4 | 1247 | |
375ec69d | 1248 | ASSERT(!(bp->b_flags & XBF_DELWRI)); |
1da177e4 | 1249 | |
375ec69d | 1250 | if (bp->b_flags & XBF_WRITE) |
ce8e922c | 1251 | xfs_buf_wait_unpin(bp); |
ce8e922c | 1252 | xfs_buf_hold(bp); |
1da177e4 LT |
1253 | |
1254 | /* Set the count to 1 initially, this will stop an I/O | |
1255 | * completion callout which happens before we have started | |
ce8e922c | 1256 | * all the I/O from calling xfs_buf_ioend too early. |
1da177e4 | 1257 | */ |
ce8e922c NS |
1258 | atomic_set(&bp->b_io_remaining, 1); |
1259 | _xfs_buf_ioapply(bp); | |
1260 | _xfs_buf_ioend(bp, 0); | |
1da177e4 | 1261 | |
ce8e922c | 1262 | xfs_buf_rele(bp); |
1da177e4 LT |
1263 | return 0; |
1264 | } | |
1265 | ||
1266 | /* | |
ce8e922c NS |
1267 | * Waits for I/O to complete on the buffer supplied. |
1268 | * It returns immediately if no I/O is pending. | |
1269 | * It returns the I/O error code, if any, or 0 if there was no error. | |
1da177e4 LT |
1270 | */ |
1271 | int | |
ce8e922c NS |
1272 | xfs_buf_iowait( |
1273 | xfs_buf_t *bp) | |
1da177e4 | 1274 | { |
0b1b213f CH |
1275 | trace_xfs_buf_iowait(bp, _RET_IP_); |
1276 | ||
b4dd330b | 1277 | wait_for_completion(&bp->b_iowait); |
0b1b213f CH |
1278 | |
1279 | trace_xfs_buf_iowait_done(bp, _RET_IP_); | |
ce8e922c | 1280 | return bp->b_error; |
1da177e4 LT |
1281 | } |
1282 | ||
ce8e922c NS |
1283 | xfs_caddr_t |
1284 | xfs_buf_offset( | |
1285 | xfs_buf_t *bp, | |
1da177e4 LT |
1286 | size_t offset) |
1287 | { | |
1288 | struct page *page; | |
1289 | ||
ce8e922c | 1290 | if (bp->b_flags & XBF_MAPPED) |
62926044 | 1291 | return bp->b_addr + offset; |
1da177e4 | 1292 | |
ce8e922c | 1293 | offset += bp->b_offset; |
0e6e847f DC |
1294 | page = bp->b_pages[offset >> PAGE_SHIFT]; |
1295 | return (xfs_caddr_t)page_address(page) + (offset & (PAGE_SIZE-1)); | |
1da177e4 LT |
1296 | } |
1297 | ||
1298 | /* | |
1da177e4 LT |
1299 | * Move data into or out of a buffer. |
1300 | */ | |
1301 | void | |
ce8e922c NS |
1302 | xfs_buf_iomove( |
1303 | xfs_buf_t *bp, /* buffer to process */ | |
1da177e4 LT |
1304 | size_t boff, /* starting buffer offset */ |
1305 | size_t bsize, /* length to copy */ | |
b9c48649 | 1306 | void *data, /* data address */ |
ce8e922c | 1307 | xfs_buf_rw_t mode) /* read/write/zero flag */ |
1da177e4 LT |
1308 | { |
1309 | size_t bend, cpoff, csize; | |
1310 | struct page *page; | |
1311 | ||
1312 | bend = boff + bsize; | |
1313 | while (boff < bend) { | |
ce8e922c NS |
1314 | page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)]; |
1315 | cpoff = xfs_buf_poff(boff + bp->b_offset); | |
1da177e4 | 1316 | csize = min_t(size_t, |
0e6e847f | 1317 | PAGE_SIZE-cpoff, bp->b_count_desired-boff); |
1da177e4 | 1318 | |
0e6e847f | 1319 | ASSERT(((csize + cpoff) <= PAGE_SIZE)); |
1da177e4 LT |
1320 | |
1321 | switch (mode) { | |
ce8e922c | 1322 | case XBRW_ZERO: |
1da177e4 LT |
1323 | memset(page_address(page) + cpoff, 0, csize); |
1324 | break; | |
ce8e922c | 1325 | case XBRW_READ: |
1da177e4 LT |
1326 | memcpy(data, page_address(page) + cpoff, csize); |
1327 | break; | |
ce8e922c | 1328 | case XBRW_WRITE: |
1da177e4 LT |
1329 | memcpy(page_address(page) + cpoff, data, csize); |
1330 | } | |
1331 | ||
1332 | boff += csize; | |
1333 | data += csize; | |
1334 | } | |
1335 | } | |
1336 | ||
1337 | /* | |
ce8e922c | 1338 | * Handling of buffer targets (buftargs). |
1da177e4 LT |
1339 | */ |
1340 | ||
1341 | /* | |
430cbeb8 DC |
1342 | * Wait for any bufs with callbacks that have been submitted but have not yet |
1343 | * returned. These buffers will have an elevated hold count, so wait on those | |
1344 | * while freeing all the buffers only held by the LRU. | |
1da177e4 LT |
1345 | */ |
1346 | void | |
1347 | xfs_wait_buftarg( | |
74f75a0c | 1348 | struct xfs_buftarg *btp) |
1da177e4 | 1349 | { |
430cbeb8 DC |
1350 | struct xfs_buf *bp; |
1351 | ||
1352 | restart: | |
1353 | spin_lock(&btp->bt_lru_lock); | |
1354 | while (!list_empty(&btp->bt_lru)) { | |
1355 | bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru); | |
1356 | if (atomic_read(&bp->b_hold) > 1) { | |
1357 | spin_unlock(&btp->bt_lru_lock); | |
26af6552 | 1358 | delay(100); |
430cbeb8 | 1359 | goto restart; |
1da177e4 | 1360 | } |
430cbeb8 DC |
1361 | /* |
1362 | * clear the LRU reference count so the bufer doesn't get | |
1363 | * ignored in xfs_buf_rele(). | |
1364 | */ | |
1365 | atomic_set(&bp->b_lru_ref, 0); | |
1366 | spin_unlock(&btp->bt_lru_lock); | |
1367 | xfs_buf_rele(bp); | |
1368 | spin_lock(&btp->bt_lru_lock); | |
1da177e4 | 1369 | } |
430cbeb8 | 1370 | spin_unlock(&btp->bt_lru_lock); |
1da177e4 LT |
1371 | } |
1372 | ||
ff57ab21 DC |
1373 | int |
1374 | xfs_buftarg_shrink( | |
1375 | struct shrinker *shrink, | |
1495f230 | 1376 | struct shrink_control *sc) |
a6867a68 | 1377 | { |
ff57ab21 DC |
1378 | struct xfs_buftarg *btp = container_of(shrink, |
1379 | struct xfs_buftarg, bt_shrinker); | |
430cbeb8 | 1380 | struct xfs_buf *bp; |
1495f230 | 1381 | int nr_to_scan = sc->nr_to_scan; |
430cbeb8 DC |
1382 | LIST_HEAD(dispose); |
1383 | ||
1384 | if (!nr_to_scan) | |
1385 | return btp->bt_lru_nr; | |
1386 | ||
1387 | spin_lock(&btp->bt_lru_lock); | |
1388 | while (!list_empty(&btp->bt_lru)) { | |
1389 | if (nr_to_scan-- <= 0) | |
1390 | break; | |
1391 | ||
1392 | bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru); | |
1393 | ||
1394 | /* | |
1395 | * Decrement the b_lru_ref count unless the value is already | |
1396 | * zero. If the value is already zero, we need to reclaim the | |
1397 | * buffer, otherwise it gets another trip through the LRU. | |
1398 | */ | |
1399 | if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) { | |
1400 | list_move_tail(&bp->b_lru, &btp->bt_lru); | |
1401 | continue; | |
1402 | } | |
1403 | ||
1404 | /* | |
1405 | * remove the buffer from the LRU now to avoid needing another | |
1406 | * lock round trip inside xfs_buf_rele(). | |
1407 | */ | |
1408 | list_move(&bp->b_lru, &dispose); | |
1409 | btp->bt_lru_nr--; | |
ff57ab21 | 1410 | } |
430cbeb8 DC |
1411 | spin_unlock(&btp->bt_lru_lock); |
1412 | ||
1413 | while (!list_empty(&dispose)) { | |
1414 | bp = list_first_entry(&dispose, struct xfs_buf, b_lru); | |
1415 | list_del_init(&bp->b_lru); | |
1416 | xfs_buf_rele(bp); | |
1417 | } | |
1418 | ||
1419 | return btp->bt_lru_nr; | |
a6867a68 DC |
1420 | } |
1421 | ||
1da177e4 LT |
1422 | void |
1423 | xfs_free_buftarg( | |
b7963133 CH |
1424 | struct xfs_mount *mp, |
1425 | struct xfs_buftarg *btp) | |
1da177e4 | 1426 | { |
ff57ab21 DC |
1427 | unregister_shrinker(&btp->bt_shrinker); |
1428 | ||
1da177e4 | 1429 | xfs_flush_buftarg(btp, 1); |
b7963133 CH |
1430 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
1431 | xfs_blkdev_issue_flush(btp); | |
a6867a68 | 1432 | |
a6867a68 | 1433 | kthread_stop(btp->bt_task); |
f0e2d93c | 1434 | kmem_free(btp); |
1da177e4 LT |
1435 | } |
1436 | ||
1da177e4 LT |
1437 | STATIC int |
1438 | xfs_setsize_buftarg_flags( | |
1439 | xfs_buftarg_t *btp, | |
1440 | unsigned int blocksize, | |
1441 | unsigned int sectorsize, | |
1442 | int verbose) | |
1443 | { | |
ce8e922c NS |
1444 | btp->bt_bsize = blocksize; |
1445 | btp->bt_sshift = ffs(sectorsize) - 1; | |
1446 | btp->bt_smask = sectorsize - 1; | |
1da177e4 | 1447 | |
ce8e922c | 1448 | if (set_blocksize(btp->bt_bdev, sectorsize)) { |
4f10700a DC |
1449 | xfs_warn(btp->bt_mount, |
1450 | "Cannot set_blocksize to %u on device %s\n", | |
c35a549c | 1451 | sectorsize, xfs_buf_target_name(btp)); |
1da177e4 LT |
1452 | return EINVAL; |
1453 | } | |
1454 | ||
1da177e4 LT |
1455 | return 0; |
1456 | } | |
1457 | ||
1458 | /* | |
ce8e922c NS |
1459 | * When allocating the initial buffer target we have not yet |
1460 | * read in the superblock, so don't know what sized sectors | |
1461 | * are being used is at this early stage. Play safe. | |
1462 | */ | |
1da177e4 LT |
1463 | STATIC int |
1464 | xfs_setsize_buftarg_early( | |
1465 | xfs_buftarg_t *btp, | |
1466 | struct block_device *bdev) | |
1467 | { | |
1468 | return xfs_setsize_buftarg_flags(btp, | |
0e6e847f | 1469 | PAGE_SIZE, bdev_logical_block_size(bdev), 0); |
1da177e4 LT |
1470 | } |
1471 | ||
1472 | int | |
1473 | xfs_setsize_buftarg( | |
1474 | xfs_buftarg_t *btp, | |
1475 | unsigned int blocksize, | |
1476 | unsigned int sectorsize) | |
1477 | { | |
1478 | return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1); | |
1479 | } | |
1480 | ||
a6867a68 | 1481 | STATIC int |
c4e1c098 | 1482 | xfs_alloc_delwri_queue( |
e2a07812 JE |
1483 | xfs_buftarg_t *btp, |
1484 | const char *fsname) | |
a6867a68 | 1485 | { |
c4e1c098 CH |
1486 | INIT_LIST_HEAD(&btp->bt_delwri_queue); |
1487 | spin_lock_init(&btp->bt_delwri_lock); | |
a6867a68 | 1488 | btp->bt_flags = 0; |
e2a07812 | 1489 | btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd/%s", fsname); |
ff57ab21 DC |
1490 | if (IS_ERR(btp->bt_task)) |
1491 | return PTR_ERR(btp->bt_task); | |
1492 | return 0; | |
a6867a68 DC |
1493 | } |
1494 | ||
1da177e4 LT |
1495 | xfs_buftarg_t * |
1496 | xfs_alloc_buftarg( | |
ebad861b | 1497 | struct xfs_mount *mp, |
1da177e4 | 1498 | struct block_device *bdev, |
e2a07812 JE |
1499 | int external, |
1500 | const char *fsname) | |
1da177e4 LT |
1501 | { |
1502 | xfs_buftarg_t *btp; | |
1503 | ||
1504 | btp = kmem_zalloc(sizeof(*btp), KM_SLEEP); | |
1505 | ||
ebad861b | 1506 | btp->bt_mount = mp; |
ce8e922c NS |
1507 | btp->bt_dev = bdev->bd_dev; |
1508 | btp->bt_bdev = bdev; | |
0e6e847f DC |
1509 | btp->bt_bdi = blk_get_backing_dev_info(bdev); |
1510 | if (!btp->bt_bdi) | |
1511 | goto error; | |
1512 | ||
430cbeb8 DC |
1513 | INIT_LIST_HEAD(&btp->bt_lru); |
1514 | spin_lock_init(&btp->bt_lru_lock); | |
1da177e4 LT |
1515 | if (xfs_setsize_buftarg_early(btp, bdev)) |
1516 | goto error; | |
c4e1c098 | 1517 | if (xfs_alloc_delwri_queue(btp, fsname)) |
a6867a68 | 1518 | goto error; |
ff57ab21 DC |
1519 | btp->bt_shrinker.shrink = xfs_buftarg_shrink; |
1520 | btp->bt_shrinker.seeks = DEFAULT_SEEKS; | |
1521 | register_shrinker(&btp->bt_shrinker); | |
1da177e4 LT |
1522 | return btp; |
1523 | ||
1524 | error: | |
f0e2d93c | 1525 | kmem_free(btp); |
1da177e4 LT |
1526 | return NULL; |
1527 | } | |
1528 | ||
1529 | ||
1530 | /* | |
ce8e922c | 1531 | * Delayed write buffer handling |
1da177e4 | 1532 | */ |
61551f1e | 1533 | void |
ce8e922c | 1534 | xfs_buf_delwri_queue( |
527cfdf1 | 1535 | xfs_buf_t *bp) |
1da177e4 | 1536 | { |
c4e1c098 | 1537 | struct xfs_buftarg *btp = bp->b_target; |
a6867a68 | 1538 | |
0b1b213f CH |
1539 | trace_xfs_buf_delwri_queue(bp, _RET_IP_); |
1540 | ||
5a8ee6ba | 1541 | ASSERT(!(bp->b_flags & XBF_READ)); |
1da177e4 | 1542 | |
c4e1c098 | 1543 | spin_lock(&btp->bt_delwri_lock); |
ce8e922c | 1544 | if (!list_empty(&bp->b_list)) { |
5a8ee6ba | 1545 | /* if already in the queue, move it to the tail */ |
ce8e922c | 1546 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); |
c4e1c098 | 1547 | list_move_tail(&bp->b_list, &btp->bt_delwri_queue); |
5a8ee6ba | 1548 | } else { |
c9c12971 | 1549 | /* start xfsbufd as it is about to have something to do */ |
c4e1c098 | 1550 | if (list_empty(&btp->bt_delwri_queue)) |
5a8ee6ba | 1551 | wake_up_process(bp->b_target->bt_task); |
c9c12971 | 1552 | |
5a8ee6ba CH |
1553 | atomic_inc(&bp->b_hold); |
1554 | bp->b_flags |= XBF_DELWRI | _XBF_DELWRI_Q | XBF_ASYNC; | |
c4e1c098 | 1555 | list_add_tail(&bp->b_list, &btp->bt_delwri_queue); |
5a8ee6ba | 1556 | } |
ce8e922c | 1557 | bp->b_queuetime = jiffies; |
c4e1c098 | 1558 | spin_unlock(&btp->bt_delwri_lock); |
1da177e4 LT |
1559 | } |
1560 | ||
1561 | void | |
ce8e922c NS |
1562 | xfs_buf_delwri_dequeue( |
1563 | xfs_buf_t *bp) | |
1da177e4 LT |
1564 | { |
1565 | int dequeued = 0; | |
1566 | ||
c4e1c098 | 1567 | spin_lock(&bp->b_target->bt_delwri_lock); |
ce8e922c NS |
1568 | if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) { |
1569 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1570 | list_del_init(&bp->b_list); | |
1da177e4 LT |
1571 | dequeued = 1; |
1572 | } | |
ce8e922c | 1573 | bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q); |
c4e1c098 | 1574 | spin_unlock(&bp->b_target->bt_delwri_lock); |
1da177e4 LT |
1575 | |
1576 | if (dequeued) | |
ce8e922c | 1577 | xfs_buf_rele(bp); |
1da177e4 | 1578 | |
0b1b213f | 1579 | trace_xfs_buf_delwri_dequeue(bp, _RET_IP_); |
1da177e4 LT |
1580 | } |
1581 | ||
d808f617 DC |
1582 | /* |
1583 | * If a delwri buffer needs to be pushed before it has aged out, then promote | |
1584 | * it to the head of the delwri queue so that it will be flushed on the next | |
1585 | * xfsbufd run. We do this by resetting the queuetime of the buffer to be older | |
1586 | * than the age currently needed to flush the buffer. Hence the next time the | |
1587 | * xfsbufd sees it is guaranteed to be considered old enough to flush. | |
1588 | */ | |
1589 | void | |
1590 | xfs_buf_delwri_promote( | |
1591 | struct xfs_buf *bp) | |
1592 | { | |
1593 | struct xfs_buftarg *btp = bp->b_target; | |
1594 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1; | |
1595 | ||
1596 | ASSERT(bp->b_flags & XBF_DELWRI); | |
1597 | ASSERT(bp->b_flags & _XBF_DELWRI_Q); | |
1598 | ||
1599 | /* | |
1600 | * Check the buffer age before locking the delayed write queue as we | |
1601 | * don't need to promote buffers that are already past the flush age. | |
1602 | */ | |
1603 | if (bp->b_queuetime < jiffies - age) | |
1604 | return; | |
1605 | bp->b_queuetime = jiffies - age; | |
c4e1c098 CH |
1606 | spin_lock(&btp->bt_delwri_lock); |
1607 | list_move(&bp->b_list, &btp->bt_delwri_queue); | |
1608 | spin_unlock(&btp->bt_delwri_lock); | |
d808f617 DC |
1609 | } |
1610 | ||
1da177e4 | 1611 | STATIC void |
ce8e922c | 1612 | xfs_buf_runall_queues( |
1da177e4 LT |
1613 | struct workqueue_struct *queue) |
1614 | { | |
1615 | flush_workqueue(queue); | |
1616 | } | |
1617 | ||
585e6d88 DC |
1618 | /* |
1619 | * Move as many buffers as specified to the supplied list | |
1620 | * idicating if we skipped any buffers to prevent deadlocks. | |
1621 | */ | |
1622 | STATIC int | |
1623 | xfs_buf_delwri_split( | |
1624 | xfs_buftarg_t *target, | |
1625 | struct list_head *list, | |
5e6a07df | 1626 | unsigned long age) |
585e6d88 DC |
1627 | { |
1628 | xfs_buf_t *bp, *n; | |
585e6d88 | 1629 | int skipped = 0; |
5e6a07df | 1630 | int force; |
585e6d88 | 1631 | |
5e6a07df | 1632 | force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags); |
585e6d88 | 1633 | INIT_LIST_HEAD(list); |
c4e1c098 CH |
1634 | spin_lock(&target->bt_delwri_lock); |
1635 | list_for_each_entry_safe(bp, n, &target->bt_delwri_queue, b_list) { | |
585e6d88 DC |
1636 | ASSERT(bp->b_flags & XBF_DELWRI); |
1637 | ||
811e64c7 | 1638 | if (!xfs_buf_ispinned(bp) && xfs_buf_trylock(bp)) { |
5e6a07df | 1639 | if (!force && |
585e6d88 DC |
1640 | time_before(jiffies, bp->b_queuetime + age)) { |
1641 | xfs_buf_unlock(bp); | |
1642 | break; | |
1643 | } | |
1644 | ||
1d5ae5df | 1645 | bp->b_flags &= ~(XBF_DELWRI | _XBF_DELWRI_Q); |
585e6d88 DC |
1646 | bp->b_flags |= XBF_WRITE; |
1647 | list_move_tail(&bp->b_list, list); | |
bfe27419 | 1648 | trace_xfs_buf_delwri_split(bp, _RET_IP_); |
585e6d88 DC |
1649 | } else |
1650 | skipped++; | |
1651 | } | |
585e6d88 | 1652 | |
c4e1c098 | 1653 | spin_unlock(&target->bt_delwri_lock); |
585e6d88 | 1654 | return skipped; |
585e6d88 DC |
1655 | } |
1656 | ||
089716aa DC |
1657 | /* |
1658 | * Compare function is more complex than it needs to be because | |
1659 | * the return value is only 32 bits and we are doing comparisons | |
1660 | * on 64 bit values | |
1661 | */ | |
1662 | static int | |
1663 | xfs_buf_cmp( | |
1664 | void *priv, | |
1665 | struct list_head *a, | |
1666 | struct list_head *b) | |
1667 | { | |
1668 | struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list); | |
1669 | struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list); | |
1670 | xfs_daddr_t diff; | |
1671 | ||
1672 | diff = ap->b_bn - bp->b_bn; | |
1673 | if (diff < 0) | |
1674 | return -1; | |
1675 | if (diff > 0) | |
1676 | return 1; | |
1677 | return 0; | |
1678 | } | |
1679 | ||
1da177e4 | 1680 | STATIC int |
23ea4032 | 1681 | xfsbufd( |
585e6d88 | 1682 | void *data) |
1da177e4 | 1683 | { |
089716aa | 1684 | xfs_buftarg_t *target = (xfs_buftarg_t *)data; |
1da177e4 | 1685 | |
1da177e4 LT |
1686 | current->flags |= PF_MEMALLOC; |
1687 | ||
978c7b2f RW |
1688 | set_freezable(); |
1689 | ||
1da177e4 | 1690 | do { |
c9c12971 DC |
1691 | long age = xfs_buf_age_centisecs * msecs_to_jiffies(10); |
1692 | long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10); | |
089716aa | 1693 | struct list_head tmp; |
a1b7ea5d | 1694 | struct blk_plug plug; |
c9c12971 | 1695 | |
3e1d1d28 | 1696 | if (unlikely(freezing(current))) { |
ce8e922c | 1697 | set_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
3e1d1d28 | 1698 | refrigerator(); |
abd0cf7a | 1699 | } else { |
ce8e922c | 1700 | clear_bit(XBT_FORCE_SLEEP, &target->bt_flags); |
abd0cf7a | 1701 | } |
1da177e4 | 1702 | |
c9c12971 | 1703 | /* sleep for a long time if there is nothing to do. */ |
c4e1c098 | 1704 | if (list_empty(&target->bt_delwri_queue)) |
c9c12971 DC |
1705 | tout = MAX_SCHEDULE_TIMEOUT; |
1706 | schedule_timeout_interruptible(tout); | |
1da177e4 | 1707 | |
c9c12971 | 1708 | xfs_buf_delwri_split(target, &tmp, age); |
089716aa | 1709 | list_sort(NULL, &tmp, xfs_buf_cmp); |
a1b7ea5d CH |
1710 | |
1711 | blk_start_plug(&plug); | |
1da177e4 | 1712 | while (!list_empty(&tmp)) { |
089716aa DC |
1713 | struct xfs_buf *bp; |
1714 | bp = list_first_entry(&tmp, struct xfs_buf, b_list); | |
ce8e922c | 1715 | list_del_init(&bp->b_list); |
939d723b | 1716 | xfs_bdstrat_cb(bp); |
1da177e4 | 1717 | } |
a1b7ea5d | 1718 | blk_finish_plug(&plug); |
4df08c52 | 1719 | } while (!kthread_should_stop()); |
1da177e4 | 1720 | |
4df08c52 | 1721 | return 0; |
1da177e4 LT |
1722 | } |
1723 | ||
1724 | /* | |
ce8e922c NS |
1725 | * Go through all incore buffers, and release buffers if they belong to |
1726 | * the given device. This is used in filesystem error handling to | |
1727 | * preserve the consistency of its metadata. | |
1da177e4 LT |
1728 | */ |
1729 | int | |
1730 | xfs_flush_buftarg( | |
585e6d88 DC |
1731 | xfs_buftarg_t *target, |
1732 | int wait) | |
1da177e4 | 1733 | { |
089716aa | 1734 | xfs_buf_t *bp; |
585e6d88 | 1735 | int pincount = 0; |
089716aa DC |
1736 | LIST_HEAD(tmp_list); |
1737 | LIST_HEAD(wait_list); | |
a1b7ea5d | 1738 | struct blk_plug plug; |
1da177e4 | 1739 | |
c626d174 | 1740 | xfs_buf_runall_queues(xfsconvertd_workqueue); |
ce8e922c NS |
1741 | xfs_buf_runall_queues(xfsdatad_workqueue); |
1742 | xfs_buf_runall_queues(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 | |
83e75904 | 1794 | xfsdatad_workqueue = alloc_workqueue("xfsdatad", WQ_MEM_RECLAIM, 1); |
23ea4032 CH |
1795 | if (!xfsdatad_workqueue) |
1796 | goto out_destroy_xfslogd_workqueue; | |
1da177e4 | 1797 | |
83e75904 TH |
1798 | xfsconvertd_workqueue = alloc_workqueue("xfsconvertd", |
1799 | WQ_MEM_RECLAIM, 1); | |
c626d174 DC |
1800 | if (!xfsconvertd_workqueue) |
1801 | goto out_destroy_xfsdatad_workqueue; | |
1802 | ||
23ea4032 | 1803 | return 0; |
1da177e4 | 1804 | |
c626d174 DC |
1805 | out_destroy_xfsdatad_workqueue: |
1806 | destroy_workqueue(xfsdatad_workqueue); | |
23ea4032 CH |
1807 | out_destroy_xfslogd_workqueue: |
1808 | destroy_workqueue(xfslogd_workqueue); | |
23ea4032 | 1809 | out_free_buf_zone: |
ce8e922c | 1810 | kmem_zone_destroy(xfs_buf_zone); |
0b1b213f | 1811 | out: |
8758280f | 1812 | return -ENOMEM; |
1da177e4 LT |
1813 | } |
1814 | ||
1da177e4 | 1815 | void |
ce8e922c | 1816 | xfs_buf_terminate(void) |
1da177e4 | 1817 | { |
c626d174 | 1818 | destroy_workqueue(xfsconvertd_workqueue); |
04d8b284 CH |
1819 | destroy_workqueue(xfsdatad_workqueue); |
1820 | destroy_workqueue(xfslogd_workqueue); | |
ce8e922c | 1821 | kmem_zone_destroy(xfs_buf_zone); |
1da177e4 | 1822 | } |