]>
Commit | Line | Data |
---|---|---|
32a7627c N |
1 | /* |
2 | * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003 | |
3 | * | |
4 | * bitmap_create - sets up the bitmap structure | |
5 | * bitmap_destroy - destroys the bitmap structure | |
6 | * | |
7 | * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.: | |
8 | * - added disk storage for bitmap | |
9 | * - changes to allow various bitmap chunk sizes | |
10 | * - added bitmap daemon (to asynchronously clear bitmap bits from disk) | |
11 | */ | |
12 | ||
13 | /* | |
14 | * Still to do: | |
15 | * | |
16 | * flush after percent set rather than just time based. (maybe both). | |
17 | * wait if count gets too high, wake when it drops to half. | |
18 | * allow bitmap to be mirrored with superblock (before or after...) | |
19 | * allow hot-add to re-instate a current device. | |
20 | * allow hot-add of bitmap after quiessing device | |
21 | */ | |
22 | ||
23 | #include <linux/module.h> | |
32a7627c N |
24 | #include <linux/errno.h> |
25 | #include <linux/slab.h> | |
26 | #include <linux/init.h> | |
27 | #include <linux/config.h> | |
28 | #include <linux/timer.h> | |
29 | #include <linux/sched.h> | |
30 | #include <linux/list.h> | |
31 | #include <linux/file.h> | |
32 | #include <linux/mount.h> | |
33 | #include <linux/buffer_head.h> | |
34 | #include <linux/raid/md.h> | |
35 | #include <linux/raid/bitmap.h> | |
36 | ||
37 | /* debug macros */ | |
38 | ||
39 | #define DEBUG 0 | |
40 | ||
41 | #if DEBUG | |
42 | /* these are for debugging purposes only! */ | |
43 | ||
44 | /* define one and only one of these */ | |
45 | #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */ | |
46 | #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/ | |
47 | #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */ | |
48 | #define INJECT_FAULTS_4 0 /* undef */ | |
49 | #define INJECT_FAULTS_5 0 /* undef */ | |
50 | #define INJECT_FAULTS_6 0 | |
51 | ||
52 | /* if these are defined, the driver will fail! debug only */ | |
53 | #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */ | |
54 | #define INJECT_FATAL_FAULT_2 0 /* undef */ | |
55 | #define INJECT_FATAL_FAULT_3 0 /* undef */ | |
56 | #endif | |
57 | ||
58 | //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */ | |
59 | #define DPRINTK(x...) do { } while(0) | |
60 | ||
61 | #ifndef PRINTK | |
62 | # if DEBUG > 0 | |
63 | # define PRINTK(x...) printk(KERN_DEBUG x) | |
64 | # else | |
65 | # define PRINTK(x...) | |
66 | # endif | |
67 | #endif | |
68 | ||
69 | static inline char * bmname(struct bitmap *bitmap) | |
70 | { | |
71 | return bitmap->mddev ? mdname(bitmap->mddev) : "mdX"; | |
72 | } | |
73 | ||
74 | ||
75 | /* | |
76 | * test if the bitmap is active | |
77 | */ | |
78 | int bitmap_active(struct bitmap *bitmap) | |
79 | { | |
80 | unsigned long flags; | |
81 | int res = 0; | |
82 | ||
83 | if (!bitmap) | |
84 | return res; | |
85 | spin_lock_irqsave(&bitmap->lock, flags); | |
86 | res = bitmap->flags & BITMAP_ACTIVE; | |
87 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
88 | return res; | |
89 | } | |
90 | ||
91 | #define WRITE_POOL_SIZE 256 | |
92 | /* mempool for queueing pending writes on the bitmap file */ | |
b4e3ca1a | 93 | static void *write_pool_alloc(gfp_t gfp_flags, void *data) |
32a7627c N |
94 | { |
95 | return kmalloc(sizeof(struct page_list), gfp_flags); | |
96 | } | |
97 | ||
98 | static void write_pool_free(void *ptr, void *data) | |
99 | { | |
100 | kfree(ptr); | |
101 | } | |
102 | ||
103 | /* | |
104 | * just a placeholder - calls kmalloc for bitmap pages | |
105 | */ | |
106 | static unsigned char *bitmap_alloc_page(struct bitmap *bitmap) | |
107 | { | |
108 | unsigned char *page; | |
109 | ||
44456d37 | 110 | #ifdef INJECT_FAULTS_1 |
32a7627c N |
111 | page = NULL; |
112 | #else | |
113 | page = kmalloc(PAGE_SIZE, GFP_NOIO); | |
114 | #endif | |
115 | if (!page) | |
116 | printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap)); | |
117 | else | |
a654b9d8 | 118 | PRINTK("%s: bitmap_alloc_page: allocated page at %p\n", |
32a7627c N |
119 | bmname(bitmap), page); |
120 | return page; | |
121 | } | |
122 | ||
123 | /* | |
124 | * for now just a placeholder -- just calls kfree for bitmap pages | |
125 | */ | |
126 | static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page) | |
127 | { | |
128 | PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page); | |
129 | kfree(page); | |
130 | } | |
131 | ||
132 | /* | |
133 | * check a page and, if necessary, allocate it (or hijack it if the alloc fails) | |
134 | * | |
135 | * 1) check to see if this page is allocated, if it's not then try to alloc | |
136 | * 2) if the alloc fails, set the page's hijacked flag so we'll use the | |
137 | * page pointer directly as a counter | |
138 | * | |
139 | * if we find our page, we increment the page's refcount so that it stays | |
140 | * allocated while we're using it | |
141 | */ | |
142 | static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create) | |
143 | { | |
144 | unsigned char *mappage; | |
145 | ||
146 | if (page >= bitmap->pages) { | |
147 | printk(KERN_ALERT | |
148 | "%s: invalid bitmap page request: %lu (> %lu)\n", | |
149 | bmname(bitmap), page, bitmap->pages-1); | |
150 | return -EINVAL; | |
151 | } | |
152 | ||
153 | ||
154 | if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ | |
155 | return 0; | |
156 | ||
157 | if (bitmap->bp[page].map) /* page is already allocated, just return */ | |
158 | return 0; | |
159 | ||
160 | if (!create) | |
161 | return -ENOENT; | |
162 | ||
163 | spin_unlock_irq(&bitmap->lock); | |
164 | ||
165 | /* this page has not been allocated yet */ | |
166 | ||
167 | if ((mappage = bitmap_alloc_page(bitmap)) == NULL) { | |
168 | PRINTK("%s: bitmap map page allocation failed, hijacking\n", | |
169 | bmname(bitmap)); | |
170 | /* failed - set the hijacked flag so that we can use the | |
171 | * pointer as a counter */ | |
172 | spin_lock_irq(&bitmap->lock); | |
173 | if (!bitmap->bp[page].map) | |
174 | bitmap->bp[page].hijacked = 1; | |
175 | goto out; | |
176 | } | |
177 | ||
178 | /* got a page */ | |
179 | ||
180 | spin_lock_irq(&bitmap->lock); | |
181 | ||
182 | /* recheck the page */ | |
183 | ||
184 | if (bitmap->bp[page].map || bitmap->bp[page].hijacked) { | |
185 | /* somebody beat us to getting the page */ | |
186 | bitmap_free_page(bitmap, mappage); | |
187 | return 0; | |
188 | } | |
189 | ||
190 | /* no page was in place and we have one, so install it */ | |
191 | ||
192 | memset(mappage, 0, PAGE_SIZE); | |
193 | bitmap->bp[page].map = mappage; | |
194 | bitmap->missing_pages--; | |
195 | out: | |
196 | return 0; | |
197 | } | |
198 | ||
199 | ||
200 | /* if page is completely empty, put it back on the free list, or dealloc it */ | |
201 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
202 | /* Note: lock should be held when calling this */ | |
858119e1 | 203 | static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page) |
32a7627c N |
204 | { |
205 | char *ptr; | |
206 | ||
207 | if (bitmap->bp[page].count) /* page is still busy */ | |
208 | return; | |
209 | ||
210 | /* page is no longer in use, it can be released */ | |
211 | ||
212 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
213 | bitmap->bp[page].hijacked = 0; | |
214 | bitmap->bp[page].map = NULL; | |
215 | return; | |
216 | } | |
217 | ||
218 | /* normal case, free the page */ | |
219 | ||
220 | #if 0 | |
221 | /* actually ... let's not. We will probably need the page again exactly when | |
222 | * memory is tight and we are flusing to disk | |
223 | */ | |
224 | return; | |
225 | #else | |
226 | ptr = bitmap->bp[page].map; | |
227 | bitmap->bp[page].map = NULL; | |
228 | bitmap->missing_pages++; | |
229 | bitmap_free_page(bitmap, ptr); | |
230 | return; | |
231 | #endif | |
232 | } | |
233 | ||
234 | ||
235 | /* | |
236 | * bitmap file handling - read and write the bitmap file and its superblock | |
237 | */ | |
238 | ||
239 | /* copy the pathname of a file to a buffer */ | |
240 | char *file_path(struct file *file, char *buf, int count) | |
241 | { | |
242 | struct dentry *d; | |
243 | struct vfsmount *v; | |
244 | ||
245 | if (!buf) | |
246 | return NULL; | |
247 | ||
248 | d = file->f_dentry; | |
249 | v = file->f_vfsmnt; | |
250 | ||
251 | buf = d_path(d, v, buf, count); | |
252 | ||
253 | return IS_ERR(buf) ? NULL : buf; | |
254 | } | |
255 | ||
256 | /* | |
257 | * basic page I/O operations | |
258 | */ | |
259 | ||
a654b9d8 N |
260 | /* IO operations when bitmap is stored near all superblocks */ |
261 | static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index) | |
262 | { | |
263 | /* choose a good rdev and read the page from there */ | |
264 | ||
265 | mdk_rdev_t *rdev; | |
266 | struct list_head *tmp; | |
267 | struct page *page = alloc_page(GFP_KERNEL); | |
268 | sector_t target; | |
269 | ||
270 | if (!page) | |
271 | return ERR_PTR(-ENOMEM); | |
a654b9d8 | 272 | |
ab904d63 | 273 | ITERATE_RDEV(mddev, rdev, tmp) { |
b2d444d7 N |
274 | if (! test_bit(In_sync, &rdev->flags) |
275 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
276 | continue; |
277 | ||
a654b9d8 N |
278 | target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512); |
279 | ||
ab904d63 N |
280 | if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) { |
281 | page->index = index; | |
282 | return page; | |
283 | } | |
284 | } | |
285 | return ERR_PTR(-EIO); | |
a654b9d8 | 286 | |
a654b9d8 N |
287 | } |
288 | ||
289 | static int write_sb_page(mddev_t *mddev, long offset, struct page *page, int wait) | |
290 | { | |
291 | mdk_rdev_t *rdev; | |
292 | struct list_head *tmp; | |
293 | ||
294 | ITERATE_RDEV(mddev, rdev, tmp) | |
b2d444d7 N |
295 | if (test_bit(In_sync, &rdev->flags) |
296 | && !test_bit(Faulty, &rdev->flags)) | |
a654b9d8 N |
297 | md_super_write(mddev, rdev, |
298 | (rdev->sb_offset<<1) + offset | |
299 | + page->index * (PAGE_SIZE/512), | |
300 | PAGE_SIZE, | |
301 | page); | |
302 | ||
303 | if (wait) | |
a9701a30 | 304 | md_super_wait(mddev); |
a654b9d8 N |
305 | return 0; |
306 | } | |
307 | ||
32a7627c | 308 | /* |
a654b9d8 | 309 | * write out a page to a file |
32a7627c | 310 | */ |
77ad4bc7 | 311 | static int write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c N |
312 | { |
313 | int ret = -ENOMEM; | |
314 | ||
a654b9d8 N |
315 | if (bitmap->file == NULL) |
316 | return write_sb_page(bitmap->mddev, bitmap->offset, page, wait); | |
317 | ||
c708443c N |
318 | flush_dcache_page(page); /* make sure visible to anyone reading the file */ |
319 | ||
8a5e9cf1 N |
320 | if (wait) |
321 | lock_page(page); | |
322 | else { | |
323 | if (TestSetPageLocked(page)) | |
324 | return -EAGAIN; /* already locked */ | |
325 | if (PageWriteback(page)) { | |
326 | unlock_page(page); | |
327 | return -EAGAIN; | |
328 | } | |
329 | } | |
32a7627c | 330 | |
34ef75f0 | 331 | ret = page->mapping->a_ops->prepare_write(bitmap->file, page, 0, PAGE_SIZE); |
32a7627c | 332 | if (!ret) |
34ef75f0 | 333 | ret = page->mapping->a_ops->commit_write(bitmap->file, page, 0, |
32a7627c N |
334 | PAGE_SIZE); |
335 | if (ret) { | |
32a7627c N |
336 | unlock_page(page); |
337 | return ret; | |
338 | } | |
339 | ||
340 | set_page_dirty(page); /* force it to be written out */ | |
77ad4bc7 N |
341 | |
342 | if (!wait) { | |
343 | /* add to list to be waited for by daemon */ | |
344 | struct page_list *item = mempool_alloc(bitmap->write_pool, GFP_NOIO); | |
345 | item->page = page; | |
2d1f3b5d | 346 | get_page(page); |
77ad4bc7 N |
347 | spin_lock(&bitmap->write_lock); |
348 | list_add(&item->list, &bitmap->complete_pages); | |
349 | spin_unlock(&bitmap->write_lock); | |
350 | md_wakeup_thread(bitmap->writeback_daemon); | |
351 | } | |
32a7627c N |
352 | return write_one_page(page, wait); |
353 | } | |
354 | ||
355 | /* read a page from a file, pinning it into cache, and return bytes_read */ | |
356 | static struct page *read_page(struct file *file, unsigned long index, | |
357 | unsigned long *bytes_read) | |
358 | { | |
359 | struct inode *inode = file->f_mapping->host; | |
360 | struct page *page = NULL; | |
361 | loff_t isize = i_size_read(inode); | |
2d1f3b5d | 362 | unsigned long end_index = isize >> PAGE_SHIFT; |
32a7627c | 363 | |
2d1f3b5d N |
364 | PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE, |
365 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c N |
366 | |
367 | page = read_cache_page(inode->i_mapping, index, | |
368 | (filler_t *)inode->i_mapping->a_ops->readpage, file); | |
369 | if (IS_ERR(page)) | |
370 | goto out; | |
371 | wait_on_page_locked(page); | |
372 | if (!PageUptodate(page) || PageError(page)) { | |
2d1f3b5d | 373 | put_page(page); |
32a7627c N |
374 | page = ERR_PTR(-EIO); |
375 | goto out; | |
376 | } | |
377 | ||
378 | if (index > end_index) /* we have read beyond EOF */ | |
379 | *bytes_read = 0; | |
380 | else if (index == end_index) /* possible short read */ | |
2d1f3b5d | 381 | *bytes_read = isize & ~PAGE_MASK; |
32a7627c | 382 | else |
2d1f3b5d | 383 | *bytes_read = PAGE_SIZE; /* got a full page */ |
32a7627c N |
384 | out: |
385 | if (IS_ERR(page)) | |
386 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n", | |
2d1f3b5d N |
387 | (int)PAGE_SIZE, |
388 | (unsigned long long)index << PAGE_SHIFT, | |
32a7627c N |
389 | PTR_ERR(page)); |
390 | return page; | |
391 | } | |
392 | ||
393 | /* | |
394 | * bitmap file superblock operations | |
395 | */ | |
396 | ||
397 | /* update the event counter and sync the superblock to disk */ | |
398 | int bitmap_update_sb(struct bitmap *bitmap) | |
399 | { | |
400 | bitmap_super_t *sb; | |
401 | unsigned long flags; | |
402 | ||
403 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
404 | return 0; | |
405 | spin_lock_irqsave(&bitmap->lock, flags); | |
406 | if (!bitmap->sb_page) { /* no superblock */ | |
407 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
408 | return 0; | |
409 | } | |
32a7627c | 410 | spin_unlock_irqrestore(&bitmap->lock, flags); |
ea03aff9 | 411 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c N |
412 | sb->events = cpu_to_le64(bitmap->mddev->events); |
413 | if (!bitmap->mddev->degraded) | |
414 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); | |
ea03aff9 | 415 | kunmap_atomic(sb, KM_USER0); |
8a5e9cf1 | 416 | return write_page(bitmap, bitmap->sb_page, 1); |
32a7627c N |
417 | } |
418 | ||
419 | /* print out the bitmap file superblock */ | |
420 | void bitmap_print_sb(struct bitmap *bitmap) | |
421 | { | |
422 | bitmap_super_t *sb; | |
423 | ||
424 | if (!bitmap || !bitmap->sb_page) | |
425 | return; | |
ea03aff9 | 426 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 427 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
428 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
429 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
430 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
431 | *(__u32 *)(sb->uuid+0), |
432 | *(__u32 *)(sb->uuid+4), | |
433 | *(__u32 *)(sb->uuid+8), | |
434 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 435 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 436 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 437 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 438 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
439 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
440 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
441 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
442 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
443 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 444 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
ea03aff9 | 445 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
446 | } |
447 | ||
448 | /* read the superblock from the bitmap file and initialize some bitmap fields */ | |
449 | static int bitmap_read_sb(struct bitmap *bitmap) | |
450 | { | |
451 | char *reason = NULL; | |
452 | bitmap_super_t *sb; | |
4b6d287f | 453 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c N |
454 | unsigned long bytes_read; |
455 | unsigned long long events; | |
456 | int err = -EINVAL; | |
457 | ||
458 | /* page 0 is the superblock, read it... */ | |
a654b9d8 N |
459 | if (bitmap->file) |
460 | bitmap->sb_page = read_page(bitmap->file, 0, &bytes_read); | |
461 | else { | |
462 | bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0); | |
463 | bytes_read = PAGE_SIZE; | |
464 | } | |
32a7627c N |
465 | if (IS_ERR(bitmap->sb_page)) { |
466 | err = PTR_ERR(bitmap->sb_page); | |
467 | bitmap->sb_page = NULL; | |
468 | return err; | |
469 | } | |
470 | ||
ea03aff9 | 471 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c N |
472 | |
473 | if (bytes_read < sizeof(*sb)) { /* short read */ | |
474 | printk(KERN_INFO "%s: bitmap file superblock truncated\n", | |
475 | bmname(bitmap)); | |
476 | err = -ENOSPC; | |
477 | goto out; | |
478 | } | |
479 | ||
480 | chunksize = le32_to_cpu(sb->chunksize); | |
481 | daemon_sleep = le32_to_cpu(sb->daemon_sleep); | |
4b6d287f | 482 | write_behind = le32_to_cpu(sb->write_behind); |
32a7627c N |
483 | |
484 | /* verify that the bitmap-specific fields are valid */ | |
485 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
486 | reason = "bad magic"; | |
bd926c63 N |
487 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
488 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 489 | reason = "unrecognized superblock version"; |
7dd5d34c N |
490 | else if (chunksize < PAGE_SIZE) |
491 | reason = "bitmap chunksize too small"; | |
32a7627c N |
492 | else if ((1 << ffz(~chunksize)) != chunksize) |
493 | reason = "bitmap chunksize not a power of 2"; | |
7dd5d34c N |
494 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ) |
495 | reason = "daemon sleep period out of range"; | |
4b6d287f N |
496 | else if (write_behind > COUNTER_MAX) |
497 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
498 | if (reason) { |
499 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
500 | bmname(bitmap), reason); | |
501 | goto out; | |
502 | } | |
503 | ||
504 | /* keep the array size field of the bitmap superblock up to date */ | |
505 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
506 | ||
507 | if (!bitmap->mddev->persistent) | |
508 | goto success; | |
509 | ||
510 | /* | |
511 | * if we have a persistent array superblock, compare the | |
512 | * bitmap's UUID and event counter to the mddev's | |
513 | */ | |
514 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
515 | printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n", | |
516 | bmname(bitmap)); | |
517 | goto out; | |
518 | } | |
519 | events = le64_to_cpu(sb->events); | |
520 | if (events < bitmap->mddev->events) { | |
521 | printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) " | |
522 | "-- forcing full recovery\n", bmname(bitmap), events, | |
523 | (unsigned long long) bitmap->mddev->events); | |
524 | sb->state |= BITMAP_STALE; | |
525 | } | |
526 | success: | |
527 | /* assign fields using values from superblock */ | |
528 | bitmap->chunksize = chunksize; | |
529 | bitmap->daemon_sleep = daemon_sleep; | |
585f0dd5 | 530 | bitmap->daemon_lastrun = jiffies; |
4b6d287f | 531 | bitmap->max_write_behind = write_behind; |
32a7627c | 532 | bitmap->flags |= sb->state; |
bd926c63 N |
533 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
534 | bitmap->flags |= BITMAP_HOSTENDIAN; | |
32a7627c | 535 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
6a07997f N |
536 | if (sb->state & BITMAP_STALE) |
537 | bitmap->events_cleared = bitmap->mddev->events; | |
32a7627c N |
538 | err = 0; |
539 | out: | |
ea03aff9 | 540 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
541 | if (err) |
542 | bitmap_print_sb(bitmap); | |
543 | return err; | |
544 | } | |
545 | ||
546 | enum bitmap_mask_op { | |
547 | MASK_SET, | |
548 | MASK_UNSET | |
549 | }; | |
550 | ||
551 | /* record the state of the bitmap in the superblock */ | |
552 | static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits, | |
553 | enum bitmap_mask_op op) | |
554 | { | |
555 | bitmap_super_t *sb; | |
556 | unsigned long flags; | |
557 | ||
558 | spin_lock_irqsave(&bitmap->lock, flags); | |
559 | if (!bitmap || !bitmap->sb_page) { /* can't set the state */ | |
560 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
561 | return; | |
562 | } | |
2d1f3b5d | 563 | get_page(bitmap->sb_page); |
32a7627c | 564 | spin_unlock_irqrestore(&bitmap->lock, flags); |
ea03aff9 | 565 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c N |
566 | switch (op) { |
567 | case MASK_SET: sb->state |= bits; | |
568 | break; | |
569 | case MASK_UNSET: sb->state &= ~bits; | |
570 | break; | |
571 | default: BUG(); | |
572 | } | |
ea03aff9 | 573 | kunmap_atomic(sb, KM_USER0); |
2d1f3b5d | 574 | put_page(bitmap->sb_page); |
32a7627c N |
575 | } |
576 | ||
577 | /* | |
578 | * general bitmap file operations | |
579 | */ | |
580 | ||
581 | /* calculate the index of the page that contains this bit */ | |
582 | static inline unsigned long file_page_index(unsigned long chunk) | |
583 | { | |
584 | return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT; | |
585 | } | |
586 | ||
587 | /* calculate the (bit) offset of this bit within a page */ | |
588 | static inline unsigned long file_page_offset(unsigned long chunk) | |
589 | { | |
590 | return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1); | |
591 | } | |
592 | ||
593 | /* | |
594 | * return a pointer to the page in the filemap that contains the given bit | |
595 | * | |
596 | * this lookup is complicated by the fact that the bitmap sb might be exactly | |
597 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page | |
598 | * 0 or page 1 | |
599 | */ | |
600 | static inline struct page *filemap_get_page(struct bitmap *bitmap, | |
601 | unsigned long chunk) | |
602 | { | |
603 | return bitmap->filemap[file_page_index(chunk) - file_page_index(0)]; | |
604 | } | |
605 | ||
606 | ||
607 | static void bitmap_file_unmap(struct bitmap *bitmap) | |
608 | { | |
609 | struct page **map, *sb_page; | |
610 | unsigned long *attr; | |
611 | int pages; | |
612 | unsigned long flags; | |
613 | ||
614 | spin_lock_irqsave(&bitmap->lock, flags); | |
615 | map = bitmap->filemap; | |
616 | bitmap->filemap = NULL; | |
617 | attr = bitmap->filemap_attr; | |
618 | bitmap->filemap_attr = NULL; | |
619 | pages = bitmap->file_pages; | |
620 | bitmap->file_pages = 0; | |
621 | sb_page = bitmap->sb_page; | |
622 | bitmap->sb_page = NULL; | |
623 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
624 | ||
625 | while (pages--) | |
626 | if (map[pages]->index != 0) /* 0 is sb_page, release it below */ | |
2d1f3b5d | 627 | put_page(map[pages]); |
32a7627c N |
628 | kfree(map); |
629 | kfree(attr); | |
630 | ||
1345b1d8 | 631 | safe_put_page(sb_page); |
32a7627c N |
632 | } |
633 | ||
500af87a | 634 | static void bitmap_stop_daemon(struct bitmap *bitmap); |
32a7627c N |
635 | |
636 | /* dequeue the next item in a page list -- don't call from irq context */ | |
77ad4bc7 | 637 | static struct page_list *dequeue_page(struct bitmap *bitmap) |
32a7627c N |
638 | { |
639 | struct page_list *item = NULL; | |
77ad4bc7 | 640 | struct list_head *head = &bitmap->complete_pages; |
32a7627c N |
641 | |
642 | spin_lock(&bitmap->write_lock); | |
643 | if (list_empty(head)) | |
644 | goto out; | |
645 | item = list_entry(head->prev, struct page_list, list); | |
646 | list_del(head->prev); | |
647 | out: | |
648 | spin_unlock(&bitmap->write_lock); | |
649 | return item; | |
650 | } | |
651 | ||
652 | static void drain_write_queues(struct bitmap *bitmap) | |
653 | { | |
32a7627c | 654 | struct page_list *item; |
32a7627c | 655 | |
77ad4bc7 N |
656 | while ((item = dequeue_page(bitmap))) { |
657 | /* don't bother to wait */ | |
2d1f3b5d | 658 | put_page(item->page); |
77ad4bc7 | 659 | mempool_free(item, bitmap->write_pool); |
32a7627c N |
660 | } |
661 | ||
32a7627c | 662 | wake_up(&bitmap->write_wait); |
32a7627c N |
663 | } |
664 | ||
665 | static void bitmap_file_put(struct bitmap *bitmap) | |
666 | { | |
667 | struct file *file; | |
668 | struct inode *inode; | |
669 | unsigned long flags; | |
670 | ||
671 | spin_lock_irqsave(&bitmap->lock, flags); | |
672 | file = bitmap->file; | |
673 | bitmap->file = NULL; | |
674 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
675 | ||
500af87a | 676 | bitmap_stop_daemon(bitmap); |
32a7627c N |
677 | |
678 | drain_write_queues(bitmap); | |
679 | ||
680 | bitmap_file_unmap(bitmap); | |
681 | ||
682 | if (file) { | |
683 | inode = file->f_mapping->host; | |
684 | spin_lock(&inode->i_lock); | |
685 | atomic_set(&inode->i_writecount, 1); /* allow writes again */ | |
686 | spin_unlock(&inode->i_lock); | |
687 | fput(file); | |
688 | } | |
689 | } | |
690 | ||
691 | ||
692 | /* | |
693 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
694 | * then it is no longer reliable, so we stop using it and we mark the file | |
695 | * as failed in the superblock | |
696 | */ | |
697 | static void bitmap_file_kick(struct bitmap *bitmap) | |
698 | { | |
699 | char *path, *ptr = NULL; | |
700 | ||
701 | bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET); | |
702 | bitmap_update_sb(bitmap); | |
703 | ||
a654b9d8 N |
704 | if (bitmap->file) { |
705 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
706 | if (path) | |
707 | ptr = file_path(bitmap->file, path, PAGE_SIZE); | |
32a7627c | 708 | |
a654b9d8 N |
709 | printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n", |
710 | bmname(bitmap), ptr ? ptr : ""); | |
32a7627c | 711 | |
a654b9d8 N |
712 | kfree(path); |
713 | } | |
32a7627c N |
714 | |
715 | bitmap_file_put(bitmap); | |
716 | ||
717 | return; | |
718 | } | |
719 | ||
720 | enum bitmap_page_attr { | |
721 | BITMAP_PAGE_DIRTY = 1, // there are set bits that need to be synced | |
722 | BITMAP_PAGE_CLEAN = 2, // there are bits that might need to be cleared | |
723 | BITMAP_PAGE_NEEDWRITE=4, // there are cleared bits that need to be synced | |
724 | }; | |
725 | ||
726 | static inline void set_page_attr(struct bitmap *bitmap, struct page *page, | |
727 | enum bitmap_page_attr attr) | |
728 | { | |
729 | bitmap->filemap_attr[page->index] |= attr; | |
730 | } | |
731 | ||
732 | static inline void clear_page_attr(struct bitmap *bitmap, struct page *page, | |
733 | enum bitmap_page_attr attr) | |
734 | { | |
735 | bitmap->filemap_attr[page->index] &= ~attr; | |
736 | } | |
737 | ||
738 | static inline unsigned long get_page_attr(struct bitmap *bitmap, struct page *page) | |
739 | { | |
740 | return bitmap->filemap_attr[page->index]; | |
741 | } | |
742 | ||
743 | /* | |
744 | * bitmap_file_set_bit -- called before performing a write to the md device | |
745 | * to set (and eventually sync) a particular bit in the bitmap file | |
746 | * | |
747 | * we set the bit immediately, then we record the page number so that | |
748 | * when an unplug occurs, we can flush the dirty pages out to disk | |
749 | */ | |
750 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
751 | { | |
752 | unsigned long bit; | |
753 | struct page *page; | |
754 | void *kaddr; | |
755 | unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap); | |
756 | ||
a654b9d8 | 757 | if (!bitmap->filemap) { |
32a7627c N |
758 | return; |
759 | } | |
760 | ||
761 | page = filemap_get_page(bitmap, chunk); | |
762 | bit = file_page_offset(chunk); | |
763 | ||
764 | ||
765 | /* make sure the page stays cached until it gets written out */ | |
766 | if (! (get_page_attr(bitmap, page) & BITMAP_PAGE_DIRTY)) | |
2d1f3b5d | 767 | get_page(page); |
32a7627c N |
768 | |
769 | /* set the bit */ | |
770 | kaddr = kmap_atomic(page, KM_USER0); | |
bd926c63 N |
771 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
772 | set_bit(bit, kaddr); | |
773 | else | |
774 | ext2_set_bit(bit, kaddr); | |
32a7627c N |
775 | kunmap_atomic(kaddr, KM_USER0); |
776 | PRINTK("set file bit %lu page %lu\n", bit, page->index); | |
777 | ||
778 | /* record page number so it gets flushed to disk when unplug occurs */ | |
779 | set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); | |
780 | ||
781 | } | |
782 | ||
783 | /* this gets called when the md device is ready to unplug its underlying | |
784 | * (slave) device queues -- before we let any writes go down, we need to | |
785 | * sync the dirty pages of the bitmap file to disk */ | |
786 | int bitmap_unplug(struct bitmap *bitmap) | |
787 | { | |
788 | unsigned long i, attr, flags; | |
789 | struct page *page; | |
790 | int wait = 0; | |
8a5e9cf1 | 791 | int err; |
32a7627c N |
792 | |
793 | if (!bitmap) | |
794 | return 0; | |
795 | ||
796 | /* look at each page to see if there are any set bits that need to be | |
797 | * flushed out to disk */ | |
798 | for (i = 0; i < bitmap->file_pages; i++) { | |
799 | spin_lock_irqsave(&bitmap->lock, flags); | |
a654b9d8 | 800 | if (!bitmap->filemap) { |
32a7627c N |
801 | spin_unlock_irqrestore(&bitmap->lock, flags); |
802 | return 0; | |
803 | } | |
804 | page = bitmap->filemap[i]; | |
805 | attr = get_page_attr(bitmap, page); | |
806 | clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); | |
807 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
808 | if ((attr & BITMAP_PAGE_DIRTY)) | |
809 | wait = 1; | |
810 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
811 | ||
8a5e9cf1 N |
812 | if (attr & (BITMAP_PAGE_DIRTY | BITMAP_PAGE_NEEDWRITE)) { |
813 | err = write_page(bitmap, page, 0); | |
814 | if (err == -EAGAIN) { | |
815 | if (attr & BITMAP_PAGE_DIRTY) | |
816 | err = write_page(bitmap, page, 1); | |
817 | else | |
818 | err = 0; | |
819 | } | |
820 | if (err) | |
bfb39fba | 821 | return 1; |
8a5e9cf1 | 822 | } |
32a7627c N |
823 | } |
824 | if (wait) { /* if any writes were performed, we need to wait on them */ | |
a654b9d8 N |
825 | if (bitmap->file) { |
826 | spin_lock_irq(&bitmap->write_lock); | |
827 | wait_event_lock_irq(bitmap->write_wait, | |
828 | list_empty(&bitmap->complete_pages), bitmap->write_lock, | |
829 | wake_up_process(bitmap->writeback_daemon->tsk)); | |
830 | spin_unlock_irq(&bitmap->write_lock); | |
831 | } else | |
a9701a30 | 832 | md_super_wait(bitmap->mddev); |
32a7627c N |
833 | } |
834 | return 0; | |
835 | } | |
836 | ||
6a07997f | 837 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
838 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
839 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
840 | * memory mapping of the bitmap file | |
841 | * Special cases: | |
842 | * if there's no bitmap file, or if the bitmap file had been | |
843 | * previously kicked from the array, we mark all the bits as | |
844 | * 1's in order to cause a full resync. | |
6a07997f N |
845 | * |
846 | * We ignore all bits for sectors that end earlier than 'start'. | |
847 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 848 | */ |
6a07997f | 849 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
850 | { |
851 | unsigned long i, chunks, index, oldindex, bit; | |
852 | struct page *page = NULL, *oldpage = NULL; | |
853 | unsigned long num_pages, bit_cnt = 0; | |
854 | struct file *file; | |
855 | unsigned long bytes, offset, dummy; | |
856 | int outofdate; | |
857 | int ret = -ENOSPC; | |
ea03aff9 | 858 | void *paddr; |
32a7627c N |
859 | |
860 | chunks = bitmap->chunks; | |
861 | file = bitmap->file; | |
862 | ||
a654b9d8 | 863 | BUG_ON(!file && !bitmap->offset); |
32a7627c | 864 | |
44456d37 | 865 | #ifdef INJECT_FAULTS_3 |
32a7627c N |
866 | outofdate = 1; |
867 | #else | |
868 | outofdate = bitmap->flags & BITMAP_STALE; | |
869 | #endif | |
870 | if (outofdate) | |
871 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
872 | "recovery\n", bmname(bitmap)); | |
873 | ||
874 | bytes = (chunks + 7) / 8; | |
bc7f77de | 875 | |
cdbb4cc2 | 876 | num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE; |
bc7f77de | 877 | |
a654b9d8 | 878 | if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) { |
32a7627c N |
879 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
880 | bmname(bitmap), | |
881 | (unsigned long) i_size_read(file->f_mapping->host), | |
882 | bytes + sizeof(bitmap_super_t)); | |
883 | goto out; | |
884 | } | |
bc7f77de N |
885 | |
886 | ret = -ENOMEM; | |
887 | ||
32a7627c | 888 | bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL); |
bc7f77de | 889 | if (!bitmap->filemap) |
32a7627c | 890 | goto out; |
32a7627c | 891 | |
9ffae0cf | 892 | bitmap->filemap_attr = kzalloc(sizeof(long) * num_pages, GFP_KERNEL); |
bc7f77de | 893 | if (!bitmap->filemap_attr) |
32a7627c | 894 | goto out; |
32a7627c | 895 | |
32a7627c N |
896 | oldindex = ~0L; |
897 | ||
898 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 899 | int b; |
32a7627c N |
900 | index = file_page_index(i); |
901 | bit = file_page_offset(i); | |
902 | if (index != oldindex) { /* this is a new page, read it in */ | |
903 | /* unmap the old page, we're done with it */ | |
32a7627c N |
904 | if (index == 0) { |
905 | /* | |
906 | * if we're here then the superblock page | |
907 | * contains some bits (PAGE_SIZE != sizeof sb) | |
908 | * we've already read it in, so just use it | |
909 | */ | |
910 | page = bitmap->sb_page; | |
911 | offset = sizeof(bitmap_super_t); | |
a654b9d8 | 912 | } else if (file) { |
32a7627c | 913 | page = read_page(file, index, &dummy); |
a654b9d8 N |
914 | offset = 0; |
915 | } else { | |
916 | page = read_sb_page(bitmap->mddev, bitmap->offset, index); | |
32a7627c N |
917 | offset = 0; |
918 | } | |
a654b9d8 N |
919 | if (IS_ERR(page)) { /* read error */ |
920 | ret = PTR_ERR(page); | |
921 | goto out; | |
922 | } | |
923 | ||
32a7627c N |
924 | oldindex = index; |
925 | oldpage = page; | |
32a7627c N |
926 | |
927 | if (outofdate) { | |
928 | /* | |
929 | * if bitmap is out of date, dirty the | |
930 | * whole page and write it out | |
931 | */ | |
ea03aff9 N |
932 | paddr = kmap_atomic(page, KM_USER0); |
933 | memset(paddr + offset, 0xff, | |
6a07997f | 934 | PAGE_SIZE - offset); |
ea03aff9 | 935 | kunmap_atomic(paddr, KM_USER0); |
77ad4bc7 | 936 | ret = write_page(bitmap, page, 1); |
32a7627c | 937 | if (ret) { |
32a7627c | 938 | /* release, page not in filemap yet */ |
2d1f3b5d | 939 | put_page(page); |
32a7627c N |
940 | goto out; |
941 | } | |
942 | } | |
943 | ||
944 | bitmap->filemap[bitmap->file_pages++] = page; | |
945 | } | |
ea03aff9 | 946 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 947 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 948 | b = test_bit(bit, paddr); |
bd926c63 | 949 | else |
ea03aff9 N |
950 | b = ext2_test_bit(bit, paddr); |
951 | kunmap_atomic(paddr, KM_USER0); | |
bd926c63 | 952 | if (b) { |
32a7627c | 953 | /* if the disk bit is set, set the memory bit */ |
6a07997f N |
954 | bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap), |
955 | ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start) | |
956 | ); | |
32a7627c | 957 | bit_cnt++; |
6a07997f | 958 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); |
32a7627c | 959 | } |
32a7627c N |
960 | } |
961 | ||
962 | /* everything went OK */ | |
963 | ret = 0; | |
964 | bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET); | |
965 | ||
32a7627c N |
966 | if (bit_cnt) { /* Kick recovery if any bits were set */ |
967 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
968 | md_wakeup_thread(bitmap->mddev->thread); | |
969 | } | |
970 | ||
971 | out: | |
972 | printk(KERN_INFO "%s: bitmap initialized from disk: " | |
973 | "read %lu/%lu pages, set %lu bits, status: %d\n", | |
974 | bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret); | |
975 | ||
976 | return ret; | |
977 | } | |
978 | ||
a654b9d8 N |
979 | void bitmap_write_all(struct bitmap *bitmap) |
980 | { | |
981 | /* We don't actually write all bitmap blocks here, | |
982 | * just flag them as needing to be written | |
983 | */ | |
984 | ||
985 | unsigned long chunks = bitmap->chunks; | |
986 | unsigned long bytes = (chunks+7)/8 + sizeof(bitmap_super_t); | |
987 | unsigned long num_pages = (bytes + PAGE_SIZE-1) / PAGE_SIZE; | |
988 | while (num_pages--) | |
989 | bitmap->filemap_attr[num_pages] |= BITMAP_PAGE_NEEDWRITE; | |
990 | } | |
991 | ||
32a7627c N |
992 | |
993 | static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc) | |
994 | { | |
995 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
996 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
997 | bitmap->bp[page].count += inc; | |
998 | /* | |
999 | if (page == 0) printk("count page 0, offset %llu: %d gives %d\n", | |
1000 | (unsigned long long)offset, inc, bitmap->bp[page].count); | |
1001 | */ | |
1002 | bitmap_checkfree(bitmap, page); | |
1003 | } | |
1004 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, | |
1005 | sector_t offset, int *blocks, | |
1006 | int create); | |
1007 | ||
1008 | /* | |
1009 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1010 | * out to disk | |
1011 | */ | |
1012 | ||
1013 | int bitmap_daemon_work(struct bitmap *bitmap) | |
1014 | { | |
aa3163f8 | 1015 | unsigned long j; |
32a7627c N |
1016 | unsigned long flags; |
1017 | struct page *page = NULL, *lastpage = NULL; | |
1018 | int err = 0; | |
1019 | int blocks; | |
1020 | int attr; | |
ea03aff9 | 1021 | void *paddr; |
32a7627c N |
1022 | |
1023 | if (bitmap == NULL) | |
1024 | return 0; | |
1025 | if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ)) | |
1026 | return 0; | |
1027 | bitmap->daemon_lastrun = jiffies; | |
1028 | ||
1029 | for (j = 0; j < bitmap->chunks; j++) { | |
1030 | bitmap_counter_t *bmc; | |
1031 | spin_lock_irqsave(&bitmap->lock, flags); | |
a654b9d8 | 1032 | if (!bitmap->filemap) { |
32a7627c N |
1033 | /* error or shutdown */ |
1034 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1035 | break; | |
1036 | } | |
1037 | ||
1038 | page = filemap_get_page(bitmap, j); | |
32a7627c N |
1039 | |
1040 | if (page != lastpage) { | |
aa3163f8 N |
1041 | /* skip this page unless it's marked as needing cleaning */ |
1042 | if (!((attr=get_page_attr(bitmap, page)) & BITMAP_PAGE_CLEAN)) { | |
1043 | if (attr & BITMAP_PAGE_NEEDWRITE) { | |
2d1f3b5d | 1044 | get_page(page); |
aa3163f8 N |
1045 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); |
1046 | } | |
1047 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1048 | if (attr & BITMAP_PAGE_NEEDWRITE) { | |
8a5e9cf1 N |
1049 | switch (write_page(bitmap, page, 0)) { |
1050 | case -EAGAIN: | |
1051 | set_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
1052 | break; | |
1053 | case 0: | |
1054 | break; | |
1055 | default: | |
aa3163f8 | 1056 | bitmap_file_kick(bitmap); |
8a5e9cf1 | 1057 | } |
2d1f3b5d | 1058 | put_page(page); |
aa3163f8 N |
1059 | } |
1060 | continue; | |
1061 | } | |
1062 | ||
32a7627c | 1063 | /* grab the new page, sync and release the old */ |
2d1f3b5d | 1064 | get_page(page); |
32a7627c N |
1065 | if (lastpage != NULL) { |
1066 | if (get_page_attr(bitmap, lastpage) & BITMAP_PAGE_NEEDWRITE) { | |
1067 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1068 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
77ad4bc7 | 1069 | err = write_page(bitmap, lastpage, 0); |
8a5e9cf1 N |
1070 | if (err == -EAGAIN) { |
1071 | err = 0; | |
1072 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1073 | } | |
32a7627c N |
1074 | } else { |
1075 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1076 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1077 | } | |
2d1f3b5d | 1078 | put_page(lastpage); |
32a7627c N |
1079 | if (err) |
1080 | bitmap_file_kick(bitmap); | |
1081 | } else | |
1082 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1083 | lastpage = page; | |
32a7627c N |
1084 | /* |
1085 | printk("bitmap clean at page %lu\n", j); | |
1086 | */ | |
1087 | spin_lock_irqsave(&bitmap->lock, flags); | |
1088 | clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1089 | } | |
1090 | bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap), | |
1091 | &blocks, 0); | |
1092 | if (bmc) { | |
1093 | /* | |
1094 | if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc); | |
1095 | */ | |
1096 | if (*bmc == 2) { | |
1097 | *bmc=1; /* maybe clear the bit next time */ | |
1098 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1099 | } else if (*bmc == 1) { | |
1100 | /* we can clear the bit */ | |
1101 | *bmc = 0; | |
1102 | bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap), | |
1103 | -1); | |
1104 | ||
1105 | /* clear the bit */ | |
ea03aff9 | 1106 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 1107 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1108 | clear_bit(file_page_offset(j), paddr); |
bd926c63 | 1109 | else |
ea03aff9 N |
1110 | ext2_clear_bit(file_page_offset(j), paddr); |
1111 | kunmap_atomic(paddr, KM_USER0); | |
32a7627c N |
1112 | } |
1113 | } | |
1114 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1115 | } | |
1116 | ||
1117 | /* now sync the final page */ | |
1118 | if (lastpage != NULL) { | |
32a7627c N |
1119 | spin_lock_irqsave(&bitmap->lock, flags); |
1120 | if (get_page_attr(bitmap, lastpage) &BITMAP_PAGE_NEEDWRITE) { | |
1121 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1122 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
77ad4bc7 | 1123 | err = write_page(bitmap, lastpage, 0); |
8a5e9cf1 N |
1124 | if (err == -EAGAIN) { |
1125 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1126 | err = 0; | |
1127 | } | |
32a7627c N |
1128 | } else { |
1129 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1130 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1131 | } | |
1132 | ||
2d1f3b5d | 1133 | put_page(lastpage); |
32a7627c N |
1134 | } |
1135 | ||
1136 | return err; | |
1137 | } | |
1138 | ||
1139 | static void daemon_exit(struct bitmap *bitmap, mdk_thread_t **daemon) | |
1140 | { | |
1141 | mdk_thread_t *dmn; | |
1142 | unsigned long flags; | |
1143 | ||
1144 | /* if no one is waiting on us, we'll free the md thread struct | |
1145 | * and exit, otherwise we let the waiter clean things up */ | |
1146 | spin_lock_irqsave(&bitmap->lock, flags); | |
1147 | if ((dmn = *daemon)) { /* no one is waiting, cleanup and exit */ | |
1148 | *daemon = NULL; | |
1149 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1150 | kfree(dmn); | |
1151 | complete_and_exit(NULL, 0); /* do_exit not exported */ | |
1152 | } | |
1153 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1154 | } | |
1155 | ||
1156 | static void bitmap_writeback_daemon(mddev_t *mddev) | |
1157 | { | |
1158 | struct bitmap *bitmap = mddev->bitmap; | |
1159 | struct page *page; | |
1160 | struct page_list *item; | |
1161 | int err = 0; | |
1162 | ||
77ad4bc7 N |
1163 | if (signal_pending(current)) { |
1164 | printk(KERN_INFO | |
1165 | "%s: bitmap writeback daemon got signal, exiting...\n", | |
1166 | bmname(bitmap)); | |
1167 | err = -EINTR; | |
1168 | goto out; | |
1169 | } | |
9ba00538 N |
1170 | if (bitmap == NULL) |
1171 | /* about to be stopped. */ | |
1172 | return; | |
32a7627c | 1173 | |
77ad4bc7 N |
1174 | PRINTK("%s: bitmap writeback daemon woke up...\n", bmname(bitmap)); |
1175 | /* wait on bitmap page writebacks */ | |
1176 | while ((item = dequeue_page(bitmap))) { | |
1177 | page = item->page; | |
1178 | mempool_free(item, bitmap->write_pool); | |
1179 | PRINTK("wait on page writeback: %p\n", page); | |
1180 | wait_on_page_writeback(page); | |
1181 | PRINTK("finished page writeback: %p\n", page); | |
1182 | ||
1183 | err = PageError(page); | |
2d1f3b5d | 1184 | put_page(page); |
77ad4bc7 N |
1185 | if (err) { |
1186 | printk(KERN_WARNING "%s: bitmap file writeback " | |
1187 | "failed (page %lu): %d\n", | |
1188 | bmname(bitmap), page->index, err); | |
1189 | bitmap_file_kick(bitmap); | |
1190 | goto out; | |
32a7627c N |
1191 | } |
1192 | } | |
77ad4bc7 N |
1193 | out: |
1194 | wake_up(&bitmap->write_wait); | |
32a7627c N |
1195 | if (err) { |
1196 | printk(KERN_INFO "%s: bitmap writeback daemon exiting (%d)\n", | |
77ad4bc7 | 1197 | bmname(bitmap), err); |
32a7627c N |
1198 | daemon_exit(bitmap, &bitmap->writeback_daemon); |
1199 | } | |
32a7627c N |
1200 | } |
1201 | ||
500af87a | 1202 | static mdk_thread_t *bitmap_start_daemon(struct bitmap *bitmap, |
32a7627c N |
1203 | void (*func)(mddev_t *), char *name) |
1204 | { | |
1205 | mdk_thread_t *daemon; | |
32a7627c N |
1206 | char namebuf[32]; |
1207 | ||
44456d37 | 1208 | #ifdef INJECT_FATAL_FAULT_2 |
32a7627c N |
1209 | daemon = NULL; |
1210 | #else | |
1211 | sprintf(namebuf, "%%s_%s", name); | |
1212 | daemon = md_register_thread(func, bitmap->mddev, namebuf); | |
1213 | #endif | |
1214 | if (!daemon) { | |
1215 | printk(KERN_ERR "%s: failed to start bitmap daemon\n", | |
1216 | bmname(bitmap)); | |
500af87a | 1217 | return ERR_PTR(-ECHILD); |
32a7627c N |
1218 | } |
1219 | ||
32a7627c N |
1220 | md_wakeup_thread(daemon); /* start it running */ |
1221 | ||
1222 | PRINTK("%s: %s daemon (pid %d) started...\n", | |
d80a138c | 1223 | bmname(bitmap), name, daemon->tsk->pid); |
32a7627c | 1224 | |
500af87a | 1225 | return daemon; |
32a7627c N |
1226 | } |
1227 | ||
500af87a | 1228 | static void bitmap_stop_daemon(struct bitmap *bitmap) |
32a7627c | 1229 | { |
500af87a N |
1230 | /* the daemon can't stop itself... it'll just exit instead... */ |
1231 | if (bitmap->writeback_daemon && ! IS_ERR(bitmap->writeback_daemon) && | |
1232 | current->pid != bitmap->writeback_daemon->tsk->pid) { | |
1233 | mdk_thread_t *daemon; | |
1234 | unsigned long flags; | |
32a7627c | 1235 | |
500af87a N |
1236 | spin_lock_irqsave(&bitmap->lock, flags); |
1237 | daemon = bitmap->writeback_daemon; | |
1238 | bitmap->writeback_daemon = NULL; | |
1239 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1240 | if (daemon && ! IS_ERR(daemon)) | |
1241 | md_unregister_thread(daemon); /* destroy the thread */ | |
1242 | } | |
32a7627c N |
1243 | } |
1244 | ||
1245 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, | |
1246 | sector_t offset, int *blocks, | |
1247 | int create) | |
1248 | { | |
1249 | /* If 'create', we might release the lock and reclaim it. | |
1250 | * The lock must have been taken with interrupts enabled. | |
1251 | * If !create, we don't release the lock. | |
1252 | */ | |
1253 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1254 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1255 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1256 | sector_t csize; | |
1257 | ||
1258 | if (bitmap_checkpage(bitmap, page, create) < 0) { | |
1259 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); | |
1260 | *blocks = csize - (offset & (csize- 1)); | |
1261 | return NULL; | |
1262 | } | |
1263 | /* now locked ... */ | |
1264 | ||
1265 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1266 | /* should we use the first or second counter field | |
1267 | * of the hijacked pointer? */ | |
1268 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
1269 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) + | |
1270 | PAGE_COUNTER_SHIFT - 1); | |
1271 | *blocks = csize - (offset & (csize- 1)); | |
1272 | return &((bitmap_counter_t *) | |
1273 | &bitmap->bp[page].map)[hi]; | |
1274 | } else { /* page is allocated */ | |
1275 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); | |
1276 | *blocks = csize - (offset & (csize- 1)); | |
1277 | return (bitmap_counter_t *) | |
1278 | &(bitmap->bp[page].map[pageoff]); | |
1279 | } | |
1280 | } | |
1281 | ||
4b6d287f | 1282 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c N |
1283 | { |
1284 | if (!bitmap) return 0; | |
4b6d287f N |
1285 | |
1286 | if (behind) { | |
1287 | atomic_inc(&bitmap->behind_writes); | |
1288 | PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n", | |
1289 | atomic_read(&bitmap->behind_writes), bitmap->max_write_behind); | |
1290 | } | |
1291 | ||
32a7627c N |
1292 | while (sectors) { |
1293 | int blocks; | |
1294 | bitmap_counter_t *bmc; | |
1295 | ||
1296 | spin_lock_irq(&bitmap->lock); | |
1297 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 1); | |
1298 | if (!bmc) { | |
1299 | spin_unlock_irq(&bitmap->lock); | |
1300 | return 0; | |
1301 | } | |
1302 | ||
1303 | switch(*bmc) { | |
1304 | case 0: | |
1305 | bitmap_file_set_bit(bitmap, offset); | |
1306 | bitmap_count_page(bitmap,offset, 1); | |
1307 | blk_plug_device(bitmap->mddev->queue); | |
1308 | /* fall through */ | |
1309 | case 1: | |
1310 | *bmc = 2; | |
1311 | } | |
1312 | if ((*bmc & COUNTER_MAX) == COUNTER_MAX) BUG(); | |
1313 | (*bmc)++; | |
1314 | ||
1315 | spin_unlock_irq(&bitmap->lock); | |
1316 | ||
1317 | offset += blocks; | |
1318 | if (sectors > blocks) | |
1319 | sectors -= blocks; | |
1320 | else sectors = 0; | |
1321 | } | |
1322 | return 0; | |
1323 | } | |
1324 | ||
1325 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1326 | int success, int behind) |
32a7627c N |
1327 | { |
1328 | if (!bitmap) return; | |
4b6d287f N |
1329 | if (behind) { |
1330 | atomic_dec(&bitmap->behind_writes); | |
1331 | PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n", | |
1332 | atomic_read(&bitmap->behind_writes), bitmap->max_write_behind); | |
1333 | } | |
1334 | ||
32a7627c N |
1335 | while (sectors) { |
1336 | int blocks; | |
1337 | unsigned long flags; | |
1338 | bitmap_counter_t *bmc; | |
1339 | ||
1340 | spin_lock_irqsave(&bitmap->lock, flags); | |
1341 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 0); | |
1342 | if (!bmc) { | |
1343 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1344 | return; | |
1345 | } | |
1346 | ||
1347 | if (!success && ! (*bmc & NEEDED_MASK)) | |
1348 | *bmc |= NEEDED_MASK; | |
1349 | ||
1350 | (*bmc)--; | |
1351 | if (*bmc <= 2) { | |
1352 | set_page_attr(bitmap, | |
1353 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
1354 | BITMAP_PAGE_CLEAN); | |
1355 | } | |
1356 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1357 | offset += blocks; | |
1358 | if (sectors > blocks) | |
1359 | sectors -= blocks; | |
1360 | else sectors = 0; | |
1361 | } | |
1362 | } | |
1363 | ||
6a806c51 N |
1364 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, |
1365 | int degraded) | |
32a7627c N |
1366 | { |
1367 | bitmap_counter_t *bmc; | |
1368 | int rv; | |
1369 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1370 | *blocks = 1024; | |
1371 | return 1; /* always resync if no bitmap */ | |
1372 | } | |
1373 | spin_lock_irq(&bitmap->lock); | |
1374 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1375 | rv = 0; | |
1376 | if (bmc) { | |
1377 | /* locked */ | |
1378 | if (RESYNC(*bmc)) | |
1379 | rv = 1; | |
1380 | else if (NEEDED(*bmc)) { | |
1381 | rv = 1; | |
6a806c51 N |
1382 | if (!degraded) { /* don't set/clear bits if degraded */ |
1383 | *bmc |= RESYNC_MASK; | |
1384 | *bmc &= ~NEEDED_MASK; | |
1385 | } | |
32a7627c N |
1386 | } |
1387 | } | |
1388 | spin_unlock_irq(&bitmap->lock); | |
1389 | return rv; | |
1390 | } | |
1391 | ||
1392 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted) | |
1393 | { | |
1394 | bitmap_counter_t *bmc; | |
1395 | unsigned long flags; | |
1396 | /* | |
1397 | if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted); | |
1398 | */ if (bitmap == NULL) { | |
1399 | *blocks = 1024; | |
1400 | return; | |
1401 | } | |
1402 | spin_lock_irqsave(&bitmap->lock, flags); | |
1403 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1404 | if (bmc == NULL) | |
1405 | goto unlock; | |
1406 | /* locked */ | |
1407 | /* | |
1408 | if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks); | |
1409 | */ | |
1410 | if (RESYNC(*bmc)) { | |
1411 | *bmc &= ~RESYNC_MASK; | |
1412 | ||
1413 | if (!NEEDED(*bmc) && aborted) | |
1414 | *bmc |= NEEDED_MASK; | |
1415 | else { | |
1416 | if (*bmc <= 2) { | |
1417 | set_page_attr(bitmap, | |
1418 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
1419 | BITMAP_PAGE_CLEAN); | |
1420 | } | |
1421 | } | |
1422 | } | |
1423 | unlock: | |
1424 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1425 | } | |
1426 | ||
1427 | void bitmap_close_sync(struct bitmap *bitmap) | |
1428 | { | |
1429 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1430 | * properly have been aborted. It remains to us to clear the | |
1431 | * RESYNC bit wherever it is still on | |
1432 | */ | |
1433 | sector_t sector = 0; | |
1434 | int blocks; | |
1435 | if (!bitmap) return; | |
1436 | while (sector < bitmap->mddev->resync_max_sectors) { | |
1437 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
1438 | /* | |
1439 | if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n", | |
1440 | (unsigned long long)sector, blocks); | |
1441 | */ sector += blocks; | |
1442 | } | |
1443 | } | |
1444 | ||
6a07997f | 1445 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1446 | { |
1447 | /* For each chunk covered by any of these sectors, set the | |
193f1c93 | 1448 | * counter to 1 and set resync_needed. They should all |
32a7627c N |
1449 | * be 0 at this point |
1450 | */ | |
193f1c93 N |
1451 | |
1452 | int secs; | |
1453 | bitmap_counter_t *bmc; | |
1454 | spin_lock_irq(&bitmap->lock); | |
1455 | bmc = bitmap_get_counter(bitmap, offset, &secs, 1); | |
1456 | if (!bmc) { | |
32a7627c | 1457 | spin_unlock_irq(&bitmap->lock); |
193f1c93 | 1458 | return; |
32a7627c | 1459 | } |
193f1c93 N |
1460 | if (! *bmc) { |
1461 | struct page *page; | |
6a07997f | 1462 | *bmc = 1 | (needed?NEEDED_MASK:0); |
193f1c93 N |
1463 | bitmap_count_page(bitmap, offset, 1); |
1464 | page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)); | |
1465 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1466 | } | |
1467 | spin_unlock_irq(&bitmap->lock); | |
1468 | ||
32a7627c N |
1469 | } |
1470 | ||
6b8b3e8a N |
1471 | /* |
1472 | * flush out any pending updates | |
1473 | */ | |
1474 | void bitmap_flush(mddev_t *mddev) | |
1475 | { | |
1476 | struct bitmap *bitmap = mddev->bitmap; | |
1477 | int sleep; | |
1478 | ||
1479 | if (!bitmap) /* there was no bitmap */ | |
1480 | return; | |
1481 | ||
1482 | /* run the daemon_work three time to ensure everything is flushed | |
1483 | * that can be | |
1484 | */ | |
1485 | sleep = bitmap->daemon_sleep; | |
1486 | bitmap->daemon_sleep = 0; | |
1487 | bitmap_daemon_work(bitmap); | |
1488 | bitmap_daemon_work(bitmap); | |
1489 | bitmap_daemon_work(bitmap); | |
1490 | bitmap->daemon_sleep = sleep; | |
1491 | bitmap_update_sb(bitmap); | |
1492 | } | |
1493 | ||
32a7627c N |
1494 | /* |
1495 | * free memory that was allocated | |
1496 | */ | |
3178b0db | 1497 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1498 | { |
1499 | unsigned long k, pages; | |
1500 | struct bitmap_page *bp; | |
32a7627c N |
1501 | |
1502 | if (!bitmap) /* there was no bitmap */ | |
1503 | return; | |
1504 | ||
32a7627c N |
1505 | /* release the bitmap file and kill the daemon */ |
1506 | bitmap_file_put(bitmap); | |
1507 | ||
1508 | bp = bitmap->bp; | |
1509 | pages = bitmap->pages; | |
1510 | ||
1511 | /* free all allocated memory */ | |
1512 | ||
1513 | mempool_destroy(bitmap->write_pool); | |
1514 | ||
1515 | if (bp) /* deallocate the page memory */ | |
1516 | for (k = 0; k < pages; k++) | |
1517 | if (bp[k].map && !bp[k].hijacked) | |
1518 | kfree(bp[k].map); | |
1519 | kfree(bp); | |
1520 | kfree(bitmap); | |
1521 | } | |
3178b0db N |
1522 | void bitmap_destroy(mddev_t *mddev) |
1523 | { | |
1524 | struct bitmap *bitmap = mddev->bitmap; | |
1525 | ||
1526 | if (!bitmap) /* there was no bitmap */ | |
1527 | return; | |
1528 | ||
1529 | mddev->bitmap = NULL; /* disconnect from the md device */ | |
b15c2e57 N |
1530 | if (mddev->thread) |
1531 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db N |
1532 | |
1533 | bitmap_free(bitmap); | |
1534 | } | |
32a7627c N |
1535 | |
1536 | /* | |
1537 | * initialize the bitmap structure | |
1538 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1539 | */ | |
1540 | int bitmap_create(mddev_t *mddev) | |
1541 | { | |
1542 | struct bitmap *bitmap; | |
1543 | unsigned long blocks = mddev->resync_max_sectors; | |
1544 | unsigned long chunks; | |
1545 | unsigned long pages; | |
1546 | struct file *file = mddev->bitmap_file; | |
1547 | int err; | |
6a07997f | 1548 | sector_t start; |
32a7627c N |
1549 | |
1550 | BUG_ON(sizeof(bitmap_super_t) != 256); | |
1551 | ||
a654b9d8 | 1552 | if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */ |
32a7627c N |
1553 | return 0; |
1554 | ||
a654b9d8 N |
1555 | BUG_ON(file && mddev->bitmap_offset); |
1556 | ||
9ffae0cf | 1557 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1558 | if (!bitmap) |
1559 | return -ENOMEM; | |
1560 | ||
32a7627c N |
1561 | spin_lock_init(&bitmap->lock); |
1562 | bitmap->mddev = mddev; | |
32a7627c N |
1563 | |
1564 | spin_lock_init(&bitmap->write_lock); | |
32a7627c N |
1565 | INIT_LIST_HEAD(&bitmap->complete_pages); |
1566 | init_waitqueue_head(&bitmap->write_wait); | |
1567 | bitmap->write_pool = mempool_create(WRITE_POOL_SIZE, write_pool_alloc, | |
1568 | write_pool_free, NULL); | |
3178b0db | 1569 | err = -ENOMEM; |
32a7627c | 1570 | if (!bitmap->write_pool) |
3178b0db | 1571 | goto error; |
32a7627c N |
1572 | |
1573 | bitmap->file = file; | |
a654b9d8 N |
1574 | bitmap->offset = mddev->bitmap_offset; |
1575 | if (file) get_file(file); | |
32a7627c N |
1576 | /* read superblock from bitmap file (this sets bitmap->chunksize) */ |
1577 | err = bitmap_read_sb(bitmap); | |
1578 | if (err) | |
3178b0db | 1579 | goto error; |
32a7627c N |
1580 | |
1581 | bitmap->chunkshift = find_first_bit(&bitmap->chunksize, | |
1582 | sizeof(bitmap->chunksize)); | |
1583 | ||
1584 | /* now that chunksize and chunkshift are set, we can use these macros */ | |
1585 | chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) / | |
1586 | CHUNK_BLOCK_RATIO(bitmap); | |
1587 | pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO; | |
1588 | ||
1589 | BUG_ON(!pages); | |
1590 | ||
1591 | bitmap->chunks = chunks; | |
1592 | bitmap->pages = pages; | |
1593 | bitmap->missing_pages = pages; | |
1594 | bitmap->counter_bits = COUNTER_BITS; | |
1595 | ||
1596 | bitmap->syncchunk = ~0UL; | |
1597 | ||
44456d37 | 1598 | #ifdef INJECT_FATAL_FAULT_1 |
32a7627c N |
1599 | bitmap->bp = NULL; |
1600 | #else | |
9ffae0cf | 1601 | bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL); |
32a7627c | 1602 | #endif |
3178b0db | 1603 | err = -ENOMEM; |
32a7627c | 1604 | if (!bitmap->bp) |
3178b0db | 1605 | goto error; |
32a7627c N |
1606 | |
1607 | bitmap->flags |= BITMAP_ACTIVE; | |
1608 | ||
1609 | /* now that we have some pages available, initialize the in-memory | |
1610 | * bitmap from the on-disk bitmap */ | |
6a07997f N |
1611 | start = 0; |
1612 | if (mddev->degraded == 0 | |
1613 | || bitmap->events_cleared == mddev->events) | |
1614 | /* no need to keep dirty bits to optimise a re-add of a missing device */ | |
1615 | start = mddev->recovery_cp; | |
1616 | err = bitmap_init_from_disk(bitmap, start); | |
193f1c93 | 1617 | |
32a7627c | 1618 | if (err) |
3178b0db | 1619 | goto error; |
32a7627c N |
1620 | |
1621 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", | |
1622 | pages, bmname(bitmap)); | |
1623 | ||
3178b0db N |
1624 | mddev->bitmap = bitmap; |
1625 | ||
500af87a N |
1626 | if (file) |
1627 | /* kick off the bitmap writeback daemon */ | |
1628 | bitmap->writeback_daemon = | |
1629 | bitmap_start_daemon(bitmap, | |
1630 | bitmap_writeback_daemon, | |
1631 | "bitmap_wb"); | |
1632 | ||
1633 | if (IS_ERR(bitmap->writeback_daemon)) | |
1634 | return PTR_ERR(bitmap->writeback_daemon); | |
b15c2e57 N |
1635 | mddev->thread->timeout = bitmap->daemon_sleep * HZ; |
1636 | ||
32a7627c | 1637 | return bitmap_update_sb(bitmap); |
3178b0db N |
1638 | |
1639 | error: | |
1640 | bitmap_free(bitmap); | |
1641 | return err; | |
32a7627c N |
1642 | } |
1643 | ||
1644 | /* the bitmap API -- for raid personalities */ | |
1645 | EXPORT_SYMBOL(bitmap_startwrite); | |
1646 | EXPORT_SYMBOL(bitmap_endwrite); | |
1647 | EXPORT_SYMBOL(bitmap_start_sync); | |
1648 | EXPORT_SYMBOL(bitmap_end_sync); | |
1649 | EXPORT_SYMBOL(bitmap_unplug); | |
1650 | EXPORT_SYMBOL(bitmap_close_sync); | |
1651 | EXPORT_SYMBOL(bitmap_daemon_work); |