]>
Commit | Line | Data |
---|---|---|
f6bed0ef SL |
1 | /* |
2 | * Copyright (C) 2015 Shaohua Li <shli@fb.com> | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify it | |
5 | * under the terms and conditions of the GNU General Public License, | |
6 | * version 2, as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope it will be useful, but WITHOUT | |
9 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
10 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
11 | * more details. | |
12 | * | |
13 | */ | |
14 | #include <linux/kernel.h> | |
15 | #include <linux/wait.h> | |
16 | #include <linux/blkdev.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/raid/md_p.h> | |
5cb2fbd6 | 19 | #include <linux/crc32c.h> |
f6bed0ef SL |
20 | #include <linux/random.h> |
21 | #include "md.h" | |
22 | #include "raid5.h" | |
23 | ||
24 | /* | |
25 | * metadata/data stored in disk with 4k size unit (a block) regardless | |
26 | * underneath hardware sector size. only works with PAGE_SIZE == 4096 | |
27 | */ | |
28 | #define BLOCK_SECTORS (8) | |
29 | ||
0576b1c6 SL |
30 | /* |
31 | * reclaim runs every 1/4 disk size or 10G reclaimable space. This can prevent | |
32 | * recovery scans a very long log | |
33 | */ | |
34 | #define RECLAIM_MAX_FREE_SPACE (10 * 1024 * 1024 * 2) /* sector */ | |
35 | #define RECLAIM_MAX_FREE_SPACE_SHIFT (2) | |
36 | ||
f6bed0ef SL |
37 | struct r5l_log { |
38 | struct md_rdev *rdev; | |
39 | ||
40 | u32 uuid_checksum; | |
41 | ||
42 | sector_t device_size; /* log device size, round to | |
43 | * BLOCK_SECTORS */ | |
0576b1c6 SL |
44 | sector_t max_free_space; /* reclaim run if free space is at |
45 | * this size */ | |
f6bed0ef SL |
46 | |
47 | sector_t last_checkpoint; /* log tail. where recovery scan | |
48 | * starts from */ | |
49 | u64 last_cp_seq; /* log tail sequence */ | |
50 | ||
51 | sector_t log_start; /* log head. where new data appends */ | |
52 | u64 seq; /* log head sequence */ | |
53 | ||
54 | struct mutex io_mutex; | |
55 | struct r5l_io_unit *current_io; /* current io_unit accepting new data */ | |
56 | ||
57 | spinlock_t io_list_lock; | |
58 | struct list_head running_ios; /* io_units which are still running, | |
59 | * and have not yet been completely | |
60 | * written to the log */ | |
61 | struct list_head io_end_ios; /* io_units which have been completely | |
62 | * written to the log but not yet written | |
63 | * to the RAID */ | |
a8c34f91 SL |
64 | struct list_head flushing_ios; /* io_units which are waiting for log |
65 | * cache flush */ | |
66 | struct list_head flushed_ios; /* io_units which settle down in log disk */ | |
67 | struct bio flush_bio; | |
0576b1c6 SL |
68 | struct list_head stripe_end_ios;/* io_units which have been completely |
69 | * written to the RAID but have not yet | |
70 | * been considered for updating super */ | |
f6bed0ef SL |
71 | |
72 | struct kmem_cache *io_kc; | |
73 | ||
0576b1c6 SL |
74 | struct md_thread *reclaim_thread; |
75 | unsigned long reclaim_target; /* number of space that need to be | |
76 | * reclaimed. if it's 0, reclaim spaces | |
77 | * used by io_units which are in | |
78 | * IO_UNIT_STRIPE_END state (eg, reclaim | |
79 | * dones't wait for specific io_unit | |
80 | * switching to IO_UNIT_STRIPE_END | |
81 | * state) */ | |
0fd22b45 | 82 | wait_queue_head_t iounit_wait; |
0576b1c6 | 83 | |
f6bed0ef SL |
84 | struct list_head no_space_stripes; /* pending stripes, log has no space */ |
85 | spinlock_t no_space_stripes_lock; | |
86 | }; | |
87 | ||
88 | /* | |
89 | * an IO range starts from a meta data block and end at the next meta data | |
90 | * block. The io unit's the meta data block tracks data/parity followed it. io | |
91 | * unit is written to log disk with normal write, as we always flush log disk | |
92 | * first and then start move data to raid disks, there is no requirement to | |
93 | * write io unit with FLUSH/FUA | |
94 | */ | |
95 | struct r5l_io_unit { | |
96 | struct r5l_log *log; | |
97 | ||
98 | struct page *meta_page; /* store meta block */ | |
99 | int meta_offset; /* current offset in meta_page */ | |
100 | ||
101 | struct bio_list bios; | |
102 | atomic_t pending_io; /* pending bios not written to log yet */ | |
103 | struct bio *current_bio;/* current_bio accepting new data */ | |
104 | ||
105 | atomic_t pending_stripe;/* how many stripes not flushed to raid */ | |
106 | u64 seq; /* seq number of the metablock */ | |
107 | sector_t log_start; /* where the io_unit starts */ | |
108 | sector_t log_end; /* where the io_unit ends */ | |
109 | struct list_head log_sibling; /* log->running_ios */ | |
110 | struct list_head stripe_list; /* stripes added to the io_unit */ | |
111 | ||
112 | int state; | |
f6bed0ef SL |
113 | }; |
114 | ||
115 | /* r5l_io_unit state */ | |
116 | enum r5l_io_unit_state { | |
117 | IO_UNIT_RUNNING = 0, /* accepting new IO */ | |
118 | IO_UNIT_IO_START = 1, /* io_unit bio start writing to log, | |
119 | * don't accepting new bio */ | |
120 | IO_UNIT_IO_END = 2, /* io_unit bio finish writing to log */ | |
a8c34f91 | 121 | IO_UNIT_STRIPE_END = 3, /* stripes data finished writing to raid */ |
f6bed0ef SL |
122 | }; |
123 | ||
124 | static sector_t r5l_ring_add(struct r5l_log *log, sector_t start, sector_t inc) | |
125 | { | |
126 | start += inc; | |
127 | if (start >= log->device_size) | |
128 | start = start - log->device_size; | |
129 | return start; | |
130 | } | |
131 | ||
132 | static sector_t r5l_ring_distance(struct r5l_log *log, sector_t start, | |
133 | sector_t end) | |
134 | { | |
135 | if (end >= start) | |
136 | return end - start; | |
137 | else | |
138 | return end + log->device_size - start; | |
139 | } | |
140 | ||
141 | static bool r5l_has_free_space(struct r5l_log *log, sector_t size) | |
142 | { | |
143 | sector_t used_size; | |
144 | ||
145 | used_size = r5l_ring_distance(log, log->last_checkpoint, | |
146 | log->log_start); | |
147 | ||
148 | return log->device_size > used_size + size; | |
149 | } | |
150 | ||
151 | static struct r5l_io_unit *r5l_alloc_io_unit(struct r5l_log *log) | |
152 | { | |
153 | struct r5l_io_unit *io; | |
154 | /* We can't handle memory allocate failure so far */ | |
155 | gfp_t gfp = GFP_NOIO | __GFP_NOFAIL; | |
156 | ||
157 | io = kmem_cache_zalloc(log->io_kc, gfp); | |
158 | io->log = log; | |
159 | io->meta_page = alloc_page(gfp | __GFP_ZERO); | |
160 | ||
161 | bio_list_init(&io->bios); | |
162 | INIT_LIST_HEAD(&io->log_sibling); | |
163 | INIT_LIST_HEAD(&io->stripe_list); | |
164 | io->state = IO_UNIT_RUNNING; | |
f6bed0ef SL |
165 | return io; |
166 | } | |
167 | ||
168 | static void r5l_free_io_unit(struct r5l_log *log, struct r5l_io_unit *io) | |
169 | { | |
170 | __free_page(io->meta_page); | |
171 | kmem_cache_free(log->io_kc, io); | |
172 | } | |
173 | ||
174 | static void r5l_move_io_unit_list(struct list_head *from, struct list_head *to, | |
175 | enum r5l_io_unit_state state) | |
176 | { | |
177 | struct r5l_io_unit *io; | |
178 | ||
179 | while (!list_empty(from)) { | |
180 | io = list_first_entry(from, struct r5l_io_unit, log_sibling); | |
181 | /* don't change list order */ | |
182 | if (io->state >= state) | |
183 | list_move_tail(&io->log_sibling, to); | |
184 | else | |
185 | break; | |
186 | } | |
187 | } | |
188 | ||
0576b1c6 SL |
189 | /* |
190 | * We don't want too many io_units reside in stripe_end_ios list, which will | |
191 | * waste a lot of memory. So we try to remove some. But we must keep at least 2 | |
192 | * io_units. The superblock must point to a valid meta, if it's the last meta, | |
193 | * recovery can scan less | |
194 | */ | |
195 | static void r5l_compress_stripe_end_list(struct r5l_log *log) | |
196 | { | |
197 | struct r5l_io_unit *first, *last, *io; | |
198 | ||
199 | first = list_first_entry(&log->stripe_end_ios, | |
200 | struct r5l_io_unit, log_sibling); | |
201 | last = list_last_entry(&log->stripe_end_ios, | |
202 | struct r5l_io_unit, log_sibling); | |
203 | if (first == last) | |
204 | return; | |
205 | list_del(&first->log_sibling); | |
206 | list_del(&last->log_sibling); | |
207 | while (!list_empty(&log->stripe_end_ios)) { | |
208 | io = list_first_entry(&log->stripe_end_ios, | |
209 | struct r5l_io_unit, log_sibling); | |
210 | list_del(&io->log_sibling); | |
211 | first->log_end = io->log_end; | |
212 | r5l_free_io_unit(log, io); | |
213 | } | |
214 | list_add_tail(&first->log_sibling, &log->stripe_end_ios); | |
215 | list_add_tail(&last->log_sibling, &log->stripe_end_ios); | |
216 | } | |
217 | ||
f6bed0ef SL |
218 | static void r5l_wake_reclaim(struct r5l_log *log, sector_t space); |
219 | static void __r5l_set_io_unit_state(struct r5l_io_unit *io, | |
220 | enum r5l_io_unit_state state) | |
221 | { | |
222 | struct r5l_log *log = io->log; | |
223 | ||
224 | if (WARN_ON(io->state >= state)) | |
225 | return; | |
226 | io->state = state; | |
227 | if (state == IO_UNIT_IO_END) | |
228 | r5l_move_io_unit_list(&log->running_ios, &log->io_end_ios, | |
229 | IO_UNIT_IO_END); | |
0576b1c6 SL |
230 | if (state == IO_UNIT_STRIPE_END) { |
231 | struct r5l_io_unit *last; | |
232 | sector_t reclaimable_space; | |
233 | ||
a8c34f91 | 234 | r5l_move_io_unit_list(&log->flushed_ios, &log->stripe_end_ios, |
0576b1c6 SL |
235 | IO_UNIT_STRIPE_END); |
236 | ||
237 | last = list_last_entry(&log->stripe_end_ios, | |
238 | struct r5l_io_unit, log_sibling); | |
239 | reclaimable_space = r5l_ring_distance(log, log->last_checkpoint, | |
240 | last->log_end); | |
241 | if (reclaimable_space >= log->max_free_space) | |
242 | r5l_wake_reclaim(log, 0); | |
243 | ||
244 | r5l_compress_stripe_end_list(log); | |
0fd22b45 | 245 | wake_up(&log->iounit_wait); |
0576b1c6 | 246 | } |
f6bed0ef SL |
247 | } |
248 | ||
249 | static void r5l_set_io_unit_state(struct r5l_io_unit *io, | |
250 | enum r5l_io_unit_state state) | |
251 | { | |
252 | struct r5l_log *log = io->log; | |
253 | unsigned long flags; | |
254 | ||
255 | spin_lock_irqsave(&log->io_list_lock, flags); | |
256 | __r5l_set_io_unit_state(io, state); | |
257 | spin_unlock_irqrestore(&log->io_list_lock, flags); | |
258 | } | |
259 | ||
260 | /* XXX: totally ignores I/O errors */ | |
261 | static void r5l_log_endio(struct bio *bio) | |
262 | { | |
263 | struct r5l_io_unit *io = bio->bi_private; | |
264 | struct r5l_log *log = io->log; | |
265 | ||
266 | bio_put(bio); | |
267 | ||
268 | if (!atomic_dec_and_test(&io->pending_io)) | |
269 | return; | |
270 | ||
271 | r5l_set_io_unit_state(io, IO_UNIT_IO_END); | |
272 | md_wakeup_thread(log->rdev->mddev->thread); | |
273 | } | |
274 | ||
275 | static void r5l_submit_current_io(struct r5l_log *log) | |
276 | { | |
277 | struct r5l_io_unit *io = log->current_io; | |
278 | struct r5l_meta_block *block; | |
279 | struct bio *bio; | |
280 | u32 crc; | |
281 | ||
282 | if (!io) | |
283 | return; | |
284 | ||
285 | block = page_address(io->meta_page); | |
286 | block->meta_size = cpu_to_le32(io->meta_offset); | |
5cb2fbd6 | 287 | crc = crc32c_le(log->uuid_checksum, block, PAGE_SIZE); |
f6bed0ef SL |
288 | block->checksum = cpu_to_le32(crc); |
289 | ||
290 | log->current_io = NULL; | |
291 | r5l_set_io_unit_state(io, IO_UNIT_IO_START); | |
292 | ||
293 | while ((bio = bio_list_pop(&io->bios))) { | |
294 | /* all IO must start from rdev->data_offset */ | |
295 | bio->bi_iter.bi_sector += log->rdev->data_offset; | |
296 | submit_bio(WRITE, bio); | |
297 | } | |
298 | } | |
299 | ||
300 | static struct r5l_io_unit *r5l_new_meta(struct r5l_log *log) | |
301 | { | |
302 | struct r5l_io_unit *io; | |
303 | struct r5l_meta_block *block; | |
304 | struct bio *bio; | |
305 | ||
306 | io = r5l_alloc_io_unit(log); | |
307 | ||
308 | block = page_address(io->meta_page); | |
309 | block->magic = cpu_to_le32(R5LOG_MAGIC); | |
310 | block->version = R5LOG_VERSION; | |
311 | block->seq = cpu_to_le64(log->seq); | |
312 | block->position = cpu_to_le64(log->log_start); | |
313 | ||
314 | io->log_start = log->log_start; | |
315 | io->meta_offset = sizeof(struct r5l_meta_block); | |
316 | io->seq = log->seq; | |
317 | ||
318 | bio = bio_kmalloc(GFP_NOIO | __GFP_NOFAIL, BIO_MAX_PAGES); | |
319 | io->current_bio = bio; | |
320 | bio->bi_rw = WRITE; | |
321 | bio->bi_bdev = log->rdev->bdev; | |
322 | bio->bi_iter.bi_sector = log->log_start; | |
323 | bio_add_page(bio, io->meta_page, PAGE_SIZE, 0); | |
324 | bio->bi_end_io = r5l_log_endio; | |
325 | bio->bi_private = io; | |
326 | ||
327 | bio_list_add(&io->bios, bio); | |
328 | atomic_inc(&io->pending_io); | |
329 | ||
330 | log->seq++; | |
331 | log->log_start = r5l_ring_add(log, log->log_start, BLOCK_SECTORS); | |
332 | io->log_end = log->log_start; | |
333 | /* current bio hit disk end */ | |
334 | if (log->log_start == 0) | |
335 | io->current_bio = NULL; | |
336 | ||
337 | spin_lock_irq(&log->io_list_lock); | |
338 | list_add_tail(&io->log_sibling, &log->running_ios); | |
339 | spin_unlock_irq(&log->io_list_lock); | |
340 | ||
341 | return io; | |
342 | } | |
343 | ||
344 | static int r5l_get_meta(struct r5l_log *log, unsigned int payload_size) | |
345 | { | |
346 | struct r5l_io_unit *io; | |
347 | ||
348 | io = log->current_io; | |
349 | if (io && io->meta_offset + payload_size > PAGE_SIZE) | |
350 | r5l_submit_current_io(log); | |
351 | io = log->current_io; | |
352 | if (io) | |
353 | return 0; | |
354 | ||
355 | log->current_io = r5l_new_meta(log); | |
356 | return 0; | |
357 | } | |
358 | ||
359 | static void r5l_append_payload_meta(struct r5l_log *log, u16 type, | |
360 | sector_t location, | |
361 | u32 checksum1, u32 checksum2, | |
362 | bool checksum2_valid) | |
363 | { | |
364 | struct r5l_io_unit *io = log->current_io; | |
365 | struct r5l_payload_data_parity *payload; | |
366 | ||
367 | payload = page_address(io->meta_page) + io->meta_offset; | |
368 | payload->header.type = cpu_to_le16(type); | |
369 | payload->header.flags = cpu_to_le16(0); | |
370 | payload->size = cpu_to_le32((1 + !!checksum2_valid) << | |
371 | (PAGE_SHIFT - 9)); | |
372 | payload->location = cpu_to_le64(location); | |
373 | payload->checksum[0] = cpu_to_le32(checksum1); | |
374 | if (checksum2_valid) | |
375 | payload->checksum[1] = cpu_to_le32(checksum2); | |
376 | ||
377 | io->meta_offset += sizeof(struct r5l_payload_data_parity) + | |
378 | sizeof(__le32) * (1 + !!checksum2_valid); | |
379 | } | |
380 | ||
381 | static void r5l_append_payload_page(struct r5l_log *log, struct page *page) | |
382 | { | |
383 | struct r5l_io_unit *io = log->current_io; | |
384 | ||
385 | alloc_bio: | |
386 | if (!io->current_bio) { | |
387 | struct bio *bio; | |
388 | ||
389 | bio = bio_kmalloc(GFP_NOIO | __GFP_NOFAIL, BIO_MAX_PAGES); | |
390 | bio->bi_rw = WRITE; | |
391 | bio->bi_bdev = log->rdev->bdev; | |
392 | bio->bi_iter.bi_sector = log->log_start; | |
393 | bio->bi_end_io = r5l_log_endio; | |
394 | bio->bi_private = io; | |
395 | bio_list_add(&io->bios, bio); | |
396 | atomic_inc(&io->pending_io); | |
397 | io->current_bio = bio; | |
398 | } | |
399 | if (!bio_add_page(io->current_bio, page, PAGE_SIZE, 0)) { | |
400 | io->current_bio = NULL; | |
401 | goto alloc_bio; | |
402 | } | |
403 | log->log_start = r5l_ring_add(log, log->log_start, | |
404 | BLOCK_SECTORS); | |
405 | /* current bio hit disk end */ | |
406 | if (log->log_start == 0) | |
407 | io->current_bio = NULL; | |
408 | ||
409 | io->log_end = log->log_start; | |
410 | } | |
411 | ||
412 | static void r5l_log_stripe(struct r5l_log *log, struct stripe_head *sh, | |
413 | int data_pages, int parity_pages) | |
414 | { | |
415 | int i; | |
416 | int meta_size; | |
417 | struct r5l_io_unit *io; | |
418 | ||
419 | meta_size = | |
420 | ((sizeof(struct r5l_payload_data_parity) + sizeof(__le32)) | |
421 | * data_pages) + | |
422 | sizeof(struct r5l_payload_data_parity) + | |
423 | sizeof(__le32) * parity_pages; | |
424 | ||
425 | r5l_get_meta(log, meta_size); | |
426 | io = log->current_io; | |
427 | ||
428 | for (i = 0; i < sh->disks; i++) { | |
429 | if (!test_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
430 | continue; | |
431 | if (i == sh->pd_idx || i == sh->qd_idx) | |
432 | continue; | |
433 | r5l_append_payload_meta(log, R5LOG_PAYLOAD_DATA, | |
434 | raid5_compute_blocknr(sh, i, 0), | |
435 | sh->dev[i].log_checksum, 0, false); | |
436 | r5l_append_payload_page(log, sh->dev[i].page); | |
437 | } | |
438 | ||
439 | if (sh->qd_idx >= 0) { | |
440 | r5l_append_payload_meta(log, R5LOG_PAYLOAD_PARITY, | |
441 | sh->sector, sh->dev[sh->pd_idx].log_checksum, | |
442 | sh->dev[sh->qd_idx].log_checksum, true); | |
443 | r5l_append_payload_page(log, sh->dev[sh->pd_idx].page); | |
444 | r5l_append_payload_page(log, sh->dev[sh->qd_idx].page); | |
445 | } else { | |
446 | r5l_append_payload_meta(log, R5LOG_PAYLOAD_PARITY, | |
447 | sh->sector, sh->dev[sh->pd_idx].log_checksum, | |
448 | 0, false); | |
449 | r5l_append_payload_page(log, sh->dev[sh->pd_idx].page); | |
450 | } | |
451 | ||
452 | list_add_tail(&sh->log_list, &io->stripe_list); | |
453 | atomic_inc(&io->pending_stripe); | |
454 | sh->log_io = io; | |
455 | } | |
456 | ||
457 | /* | |
458 | * running in raid5d, where reclaim could wait for raid5d too (when it flushes | |
459 | * data from log to raid disks), so we shouldn't wait for reclaim here | |
460 | */ | |
461 | int r5l_write_stripe(struct r5l_log *log, struct stripe_head *sh) | |
462 | { | |
463 | int write_disks = 0; | |
464 | int data_pages, parity_pages; | |
465 | int meta_size; | |
466 | int reserve; | |
467 | int i; | |
468 | ||
469 | if (!log) | |
470 | return -EAGAIN; | |
471 | /* Don't support stripe batch */ | |
472 | if (sh->log_io || !test_bit(R5_Wantwrite, &sh->dev[sh->pd_idx].flags) || | |
473 | test_bit(STRIPE_SYNCING, &sh->state)) { | |
474 | /* the stripe is written to log, we start writing it to raid */ | |
475 | clear_bit(STRIPE_LOG_TRAPPED, &sh->state); | |
476 | return -EAGAIN; | |
477 | } | |
478 | ||
479 | for (i = 0; i < sh->disks; i++) { | |
480 | void *addr; | |
481 | ||
482 | if (!test_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
483 | continue; | |
484 | write_disks++; | |
485 | /* checksum is already calculated in last run */ | |
486 | if (test_bit(STRIPE_LOG_TRAPPED, &sh->state)) | |
487 | continue; | |
488 | addr = kmap_atomic(sh->dev[i].page); | |
5cb2fbd6 SL |
489 | sh->dev[i].log_checksum = crc32c_le(log->uuid_checksum, |
490 | addr, PAGE_SIZE); | |
f6bed0ef SL |
491 | kunmap_atomic(addr); |
492 | } | |
493 | parity_pages = 1 + !!(sh->qd_idx >= 0); | |
494 | data_pages = write_disks - parity_pages; | |
495 | ||
496 | meta_size = | |
497 | ((sizeof(struct r5l_payload_data_parity) + sizeof(__le32)) | |
498 | * data_pages) + | |
499 | sizeof(struct r5l_payload_data_parity) + | |
500 | sizeof(__le32) * parity_pages; | |
501 | /* Doesn't work with very big raid array */ | |
502 | if (meta_size + sizeof(struct r5l_meta_block) > PAGE_SIZE) | |
503 | return -EINVAL; | |
504 | ||
505 | set_bit(STRIPE_LOG_TRAPPED, &sh->state); | |
506 | atomic_inc(&sh->count); | |
507 | ||
508 | mutex_lock(&log->io_mutex); | |
509 | /* meta + data */ | |
510 | reserve = (1 + write_disks) << (PAGE_SHIFT - 9); | |
511 | if (r5l_has_free_space(log, reserve)) | |
512 | r5l_log_stripe(log, sh, data_pages, parity_pages); | |
513 | else { | |
514 | spin_lock(&log->no_space_stripes_lock); | |
515 | list_add_tail(&sh->log_list, &log->no_space_stripes); | |
516 | spin_unlock(&log->no_space_stripes_lock); | |
517 | ||
518 | r5l_wake_reclaim(log, reserve); | |
519 | } | |
520 | mutex_unlock(&log->io_mutex); | |
521 | ||
522 | return 0; | |
523 | } | |
524 | ||
525 | void r5l_write_stripe_run(struct r5l_log *log) | |
526 | { | |
527 | if (!log) | |
528 | return; | |
529 | mutex_lock(&log->io_mutex); | |
530 | r5l_submit_current_io(log); | |
531 | mutex_unlock(&log->io_mutex); | |
532 | } | |
533 | ||
534 | /* This will run after log space is reclaimed */ | |
535 | static void r5l_run_no_space_stripes(struct r5l_log *log) | |
536 | { | |
537 | struct stripe_head *sh; | |
538 | ||
539 | spin_lock(&log->no_space_stripes_lock); | |
540 | while (!list_empty(&log->no_space_stripes)) { | |
541 | sh = list_first_entry(&log->no_space_stripes, | |
542 | struct stripe_head, log_list); | |
543 | list_del_init(&sh->log_list); | |
544 | set_bit(STRIPE_HANDLE, &sh->state); | |
545 | raid5_release_stripe(sh); | |
546 | } | |
547 | spin_unlock(&log->no_space_stripes_lock); | |
548 | } | |
549 | ||
0576b1c6 SL |
550 | void r5l_stripe_write_finished(struct stripe_head *sh) |
551 | { | |
552 | struct r5l_io_unit *io; | |
553 | ||
554 | /* Don't support stripe batch */ | |
555 | io = sh->log_io; | |
556 | if (!io) | |
557 | return; | |
558 | sh->log_io = NULL; | |
559 | ||
560 | if (atomic_dec_and_test(&io->pending_stripe)) | |
561 | r5l_set_io_unit_state(io, IO_UNIT_STRIPE_END); | |
562 | } | |
563 | ||
a8c34f91 SL |
564 | static void r5l_log_flush_endio(struct bio *bio) |
565 | { | |
566 | struct r5l_log *log = container_of(bio, struct r5l_log, | |
567 | flush_bio); | |
568 | unsigned long flags; | |
569 | struct r5l_io_unit *io; | |
570 | struct stripe_head *sh; | |
571 | ||
572 | spin_lock_irqsave(&log->io_list_lock, flags); | |
573 | list_for_each_entry(io, &log->flushing_ios, log_sibling) { | |
574 | while (!list_empty(&io->stripe_list)) { | |
575 | sh = list_first_entry(&io->stripe_list, | |
576 | struct stripe_head, log_list); | |
577 | list_del_init(&sh->log_list); | |
578 | set_bit(STRIPE_HANDLE, &sh->state); | |
579 | raid5_release_stripe(sh); | |
580 | } | |
581 | } | |
582 | list_splice_tail_init(&log->flushing_ios, &log->flushed_ios); | |
583 | spin_unlock_irqrestore(&log->io_list_lock, flags); | |
584 | } | |
585 | ||
0576b1c6 SL |
586 | /* |
587 | * Starting dispatch IO to raid. | |
588 | * io_unit(meta) consists of a log. There is one situation we want to avoid. A | |
589 | * broken meta in the middle of a log causes recovery can't find meta at the | |
590 | * head of log. If operations require meta at the head persistent in log, we | |
591 | * must make sure meta before it persistent in log too. A case is: | |
592 | * | |
593 | * stripe data/parity is in log, we start write stripe to raid disks. stripe | |
594 | * data/parity must be persistent in log before we do the write to raid disks. | |
595 | * | |
596 | * The solution is we restrictly maintain io_unit list order. In this case, we | |
597 | * only write stripes of an io_unit to raid disks till the io_unit is the first | |
598 | * one whose data/parity is in log. | |
599 | */ | |
600 | void r5l_flush_stripe_to_raid(struct r5l_log *log) | |
601 | { | |
a8c34f91 | 602 | bool do_flush; |
0576b1c6 SL |
603 | if (!log) |
604 | return; | |
0576b1c6 SL |
605 | |
606 | spin_lock_irq(&log->io_list_lock); | |
a8c34f91 SL |
607 | /* flush bio is running */ |
608 | if (!list_empty(&log->flushing_ios)) { | |
609 | spin_unlock_irq(&log->io_list_lock); | |
610 | return; | |
0576b1c6 | 611 | } |
a8c34f91 SL |
612 | list_splice_tail_init(&log->io_end_ios, &log->flushing_ios); |
613 | do_flush = !list_empty(&log->flushing_ios); | |
0576b1c6 | 614 | spin_unlock_irq(&log->io_list_lock); |
a8c34f91 SL |
615 | |
616 | if (!do_flush) | |
617 | return; | |
618 | bio_reset(&log->flush_bio); | |
619 | log->flush_bio.bi_bdev = log->rdev->bdev; | |
620 | log->flush_bio.bi_end_io = r5l_log_flush_endio; | |
621 | submit_bio(WRITE_FLUSH, &log->flush_bio); | |
0576b1c6 SL |
622 | } |
623 | ||
0fd22b45 | 624 | static void r5l_kick_io_unit(struct r5l_log *log) |
0576b1c6 | 625 | { |
a8c34f91 | 626 | md_wakeup_thread(log->rdev->mddev->thread); |
0fd22b45 SL |
627 | wait_event_lock_irq(log->iounit_wait, !list_empty(&log->stripe_end_ios), |
628 | log->io_list_lock); | |
0576b1c6 SL |
629 | } |
630 | ||
631 | static void r5l_write_super(struct r5l_log *log, sector_t cp); | |
632 | static void r5l_do_reclaim(struct r5l_log *log) | |
633 | { | |
634 | struct r5l_io_unit *io, *last; | |
635 | LIST_HEAD(list); | |
636 | sector_t free = 0; | |
637 | sector_t reclaim_target = xchg(&log->reclaim_target, 0); | |
638 | ||
639 | spin_lock_irq(&log->io_list_lock); | |
640 | /* | |
641 | * move proper io_unit to reclaim list. We should not change the order. | |
642 | * reclaimable/unreclaimable io_unit can be mixed in the list, we | |
643 | * shouldn't reuse space of an unreclaimable io_unit | |
644 | */ | |
645 | while (1) { | |
a8c34f91 SL |
646 | struct list_head *target_list = NULL; |
647 | ||
0576b1c6 SL |
648 | while (!list_empty(&log->stripe_end_ios)) { |
649 | io = list_first_entry(&log->stripe_end_ios, | |
650 | struct r5l_io_unit, log_sibling); | |
651 | list_move_tail(&io->log_sibling, &list); | |
652 | free += r5l_ring_distance(log, io->log_start, | |
653 | io->log_end); | |
654 | } | |
655 | ||
656 | if (free >= reclaim_target || | |
657 | (list_empty(&log->running_ios) && | |
658 | list_empty(&log->io_end_ios) && | |
a8c34f91 SL |
659 | list_empty(&log->flushing_ios) && |
660 | list_empty(&log->flushed_ios))) | |
0576b1c6 SL |
661 | break; |
662 | ||
663 | /* Below waiting mostly happens when we shutdown the raid */ | |
a8c34f91 SL |
664 | if (!list_empty(&log->flushed_ios)) |
665 | target_list = &log->flushed_ios; | |
666 | else if (!list_empty(&log->flushing_ios)) | |
667 | target_list = &log->flushing_ios; | |
668 | else if (!list_empty(&log->io_end_ios)) | |
669 | target_list = &log->io_end_ios; | |
670 | else if (!list_empty(&log->running_ios)) | |
671 | target_list = &log->running_ios; | |
672 | ||
0fd22b45 | 673 | r5l_kick_io_unit(log); |
0576b1c6 SL |
674 | } |
675 | spin_unlock_irq(&log->io_list_lock); | |
676 | ||
677 | if (list_empty(&list)) | |
678 | return; | |
679 | ||
680 | /* super always point to last valid meta */ | |
681 | last = list_last_entry(&list, struct r5l_io_unit, log_sibling); | |
682 | /* | |
683 | * write_super will flush cache of each raid disk. We must write super | |
684 | * here, because the log area might be reused soon and we don't want to | |
685 | * confuse recovery | |
686 | */ | |
687 | r5l_write_super(log, last->log_start); | |
688 | ||
689 | mutex_lock(&log->io_mutex); | |
690 | log->last_checkpoint = last->log_start; | |
691 | log->last_cp_seq = last->seq; | |
692 | mutex_unlock(&log->io_mutex); | |
693 | r5l_run_no_space_stripes(log); | |
694 | ||
695 | while (!list_empty(&list)) { | |
696 | io = list_first_entry(&list, struct r5l_io_unit, log_sibling); | |
697 | list_del(&io->log_sibling); | |
698 | r5l_free_io_unit(log, io); | |
699 | } | |
700 | } | |
701 | ||
702 | static void r5l_reclaim_thread(struct md_thread *thread) | |
703 | { | |
704 | struct mddev *mddev = thread->mddev; | |
705 | struct r5conf *conf = mddev->private; | |
706 | struct r5l_log *log = conf->log; | |
707 | ||
708 | if (!log) | |
709 | return; | |
710 | r5l_do_reclaim(log); | |
711 | } | |
712 | ||
f6bed0ef SL |
713 | static void r5l_wake_reclaim(struct r5l_log *log, sector_t space) |
714 | { | |
0576b1c6 SL |
715 | unsigned long target; |
716 | unsigned long new = (unsigned long)space; /* overflow in theory */ | |
717 | ||
718 | do { | |
719 | target = log->reclaim_target; | |
720 | if (new < target) | |
721 | return; | |
722 | } while (cmpxchg(&log->reclaim_target, target, new) != target); | |
723 | md_wakeup_thread(log->reclaim_thread); | |
f6bed0ef SL |
724 | } |
725 | ||
355810d1 SL |
726 | struct r5l_recovery_ctx { |
727 | struct page *meta_page; /* current meta */ | |
728 | sector_t meta_total_blocks; /* total size of current meta and data */ | |
729 | sector_t pos; /* recovery position */ | |
730 | u64 seq; /* recovery position seq */ | |
731 | }; | |
732 | ||
733 | static int r5l_read_meta_block(struct r5l_log *log, | |
734 | struct r5l_recovery_ctx *ctx) | |
735 | { | |
736 | struct page *page = ctx->meta_page; | |
737 | struct r5l_meta_block *mb; | |
738 | u32 crc, stored_crc; | |
739 | ||
740 | if (!sync_page_io(log->rdev, ctx->pos, PAGE_SIZE, page, READ, false)) | |
741 | return -EIO; | |
742 | ||
743 | mb = page_address(page); | |
744 | stored_crc = le32_to_cpu(mb->checksum); | |
745 | mb->checksum = 0; | |
746 | ||
747 | if (le32_to_cpu(mb->magic) != R5LOG_MAGIC || | |
748 | le64_to_cpu(mb->seq) != ctx->seq || | |
749 | mb->version != R5LOG_VERSION || | |
750 | le64_to_cpu(mb->position) != ctx->pos) | |
751 | return -EINVAL; | |
752 | ||
5cb2fbd6 | 753 | crc = crc32c_le(log->uuid_checksum, mb, PAGE_SIZE); |
355810d1 SL |
754 | if (stored_crc != crc) |
755 | return -EINVAL; | |
756 | ||
757 | if (le32_to_cpu(mb->meta_size) > PAGE_SIZE) | |
758 | return -EINVAL; | |
759 | ||
760 | ctx->meta_total_blocks = BLOCK_SECTORS; | |
761 | ||
762 | return 0; | |
763 | } | |
764 | ||
765 | static int r5l_recovery_flush_one_stripe(struct r5l_log *log, | |
766 | struct r5l_recovery_ctx *ctx, | |
767 | sector_t stripe_sect, | |
768 | int *offset, sector_t *log_offset) | |
769 | { | |
770 | struct r5conf *conf = log->rdev->mddev->private; | |
771 | struct stripe_head *sh; | |
772 | struct r5l_payload_data_parity *payload; | |
773 | int disk_index; | |
774 | ||
775 | sh = raid5_get_active_stripe(conf, stripe_sect, 0, 0, 0); | |
776 | while (1) { | |
777 | payload = page_address(ctx->meta_page) + *offset; | |
778 | ||
779 | if (le16_to_cpu(payload->header.type) == R5LOG_PAYLOAD_DATA) { | |
780 | raid5_compute_sector(conf, | |
781 | le64_to_cpu(payload->location), 0, | |
782 | &disk_index, sh); | |
783 | ||
784 | sync_page_io(log->rdev, *log_offset, PAGE_SIZE, | |
785 | sh->dev[disk_index].page, READ, false); | |
786 | sh->dev[disk_index].log_checksum = | |
787 | le32_to_cpu(payload->checksum[0]); | |
788 | set_bit(R5_Wantwrite, &sh->dev[disk_index].flags); | |
789 | ctx->meta_total_blocks += BLOCK_SECTORS; | |
790 | } else { | |
791 | disk_index = sh->pd_idx; | |
792 | sync_page_io(log->rdev, *log_offset, PAGE_SIZE, | |
793 | sh->dev[disk_index].page, READ, false); | |
794 | sh->dev[disk_index].log_checksum = | |
795 | le32_to_cpu(payload->checksum[0]); | |
796 | set_bit(R5_Wantwrite, &sh->dev[disk_index].flags); | |
797 | ||
798 | if (sh->qd_idx >= 0) { | |
799 | disk_index = sh->qd_idx; | |
800 | sync_page_io(log->rdev, | |
801 | r5l_ring_add(log, *log_offset, BLOCK_SECTORS), | |
802 | PAGE_SIZE, sh->dev[disk_index].page, | |
803 | READ, false); | |
804 | sh->dev[disk_index].log_checksum = | |
805 | le32_to_cpu(payload->checksum[1]); | |
806 | set_bit(R5_Wantwrite, | |
807 | &sh->dev[disk_index].flags); | |
808 | } | |
809 | ctx->meta_total_blocks += BLOCK_SECTORS * conf->max_degraded; | |
810 | } | |
811 | ||
812 | *log_offset = r5l_ring_add(log, *log_offset, | |
813 | le32_to_cpu(payload->size)); | |
814 | *offset += sizeof(struct r5l_payload_data_parity) + | |
815 | sizeof(__le32) * | |
816 | (le32_to_cpu(payload->size) >> (PAGE_SHIFT - 9)); | |
817 | if (le16_to_cpu(payload->header.type) == R5LOG_PAYLOAD_PARITY) | |
818 | break; | |
819 | } | |
820 | ||
821 | for (disk_index = 0; disk_index < sh->disks; disk_index++) { | |
822 | void *addr; | |
823 | u32 checksum; | |
824 | ||
825 | if (!test_bit(R5_Wantwrite, &sh->dev[disk_index].flags)) | |
826 | continue; | |
827 | addr = kmap_atomic(sh->dev[disk_index].page); | |
5cb2fbd6 | 828 | checksum = crc32c_le(log->uuid_checksum, addr, PAGE_SIZE); |
355810d1 SL |
829 | kunmap_atomic(addr); |
830 | if (checksum != sh->dev[disk_index].log_checksum) | |
831 | goto error; | |
832 | } | |
833 | ||
834 | for (disk_index = 0; disk_index < sh->disks; disk_index++) { | |
835 | struct md_rdev *rdev, *rrdev; | |
836 | ||
837 | if (!test_and_clear_bit(R5_Wantwrite, | |
838 | &sh->dev[disk_index].flags)) | |
839 | continue; | |
840 | ||
841 | /* in case device is broken */ | |
842 | rdev = rcu_dereference(conf->disks[disk_index].rdev); | |
843 | if (rdev) | |
844 | sync_page_io(rdev, stripe_sect, PAGE_SIZE, | |
845 | sh->dev[disk_index].page, WRITE, false); | |
846 | rrdev = rcu_dereference(conf->disks[disk_index].replacement); | |
847 | if (rrdev) | |
848 | sync_page_io(rrdev, stripe_sect, PAGE_SIZE, | |
849 | sh->dev[disk_index].page, WRITE, false); | |
850 | } | |
851 | raid5_release_stripe(sh); | |
852 | return 0; | |
853 | ||
854 | error: | |
855 | for (disk_index = 0; disk_index < sh->disks; disk_index++) | |
856 | sh->dev[disk_index].flags = 0; | |
857 | raid5_release_stripe(sh); | |
858 | return -EINVAL; | |
859 | } | |
860 | ||
861 | static int r5l_recovery_flush_one_meta(struct r5l_log *log, | |
862 | struct r5l_recovery_ctx *ctx) | |
863 | { | |
864 | struct r5conf *conf = log->rdev->mddev->private; | |
865 | struct r5l_payload_data_parity *payload; | |
866 | struct r5l_meta_block *mb; | |
867 | int offset; | |
868 | sector_t log_offset; | |
869 | sector_t stripe_sector; | |
870 | ||
871 | mb = page_address(ctx->meta_page); | |
872 | offset = sizeof(struct r5l_meta_block); | |
873 | log_offset = r5l_ring_add(log, ctx->pos, BLOCK_SECTORS); | |
874 | ||
875 | while (offset < le32_to_cpu(mb->meta_size)) { | |
876 | int dd; | |
877 | ||
878 | payload = (void *)mb + offset; | |
879 | stripe_sector = raid5_compute_sector(conf, | |
880 | le64_to_cpu(payload->location), 0, &dd, NULL); | |
881 | if (r5l_recovery_flush_one_stripe(log, ctx, stripe_sector, | |
882 | &offset, &log_offset)) | |
883 | return -EINVAL; | |
884 | } | |
885 | return 0; | |
886 | } | |
887 | ||
888 | /* copy data/parity from log to raid disks */ | |
889 | static void r5l_recovery_flush_log(struct r5l_log *log, | |
890 | struct r5l_recovery_ctx *ctx) | |
891 | { | |
892 | while (1) { | |
893 | if (r5l_read_meta_block(log, ctx)) | |
894 | return; | |
895 | if (r5l_recovery_flush_one_meta(log, ctx)) | |
896 | return; | |
897 | ctx->seq++; | |
898 | ctx->pos = r5l_ring_add(log, ctx->pos, ctx->meta_total_blocks); | |
899 | } | |
900 | } | |
901 | ||
902 | static int r5l_log_write_empty_meta_block(struct r5l_log *log, sector_t pos, | |
903 | u64 seq) | |
904 | { | |
905 | struct page *page; | |
906 | struct r5l_meta_block *mb; | |
907 | u32 crc; | |
908 | ||
909 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
910 | if (!page) | |
911 | return -ENOMEM; | |
912 | mb = page_address(page); | |
913 | mb->magic = cpu_to_le32(R5LOG_MAGIC); | |
914 | mb->version = R5LOG_VERSION; | |
915 | mb->meta_size = cpu_to_le32(sizeof(struct r5l_meta_block)); | |
916 | mb->seq = cpu_to_le64(seq); | |
917 | mb->position = cpu_to_le64(pos); | |
5cb2fbd6 | 918 | crc = crc32c_le(log->uuid_checksum, mb, PAGE_SIZE); |
355810d1 SL |
919 | mb->checksum = cpu_to_le32(crc); |
920 | ||
921 | if (!sync_page_io(log->rdev, pos, PAGE_SIZE, page, WRITE_FUA, false)) { | |
922 | __free_page(page); | |
923 | return -EIO; | |
924 | } | |
925 | __free_page(page); | |
926 | return 0; | |
927 | } | |
928 | ||
f6bed0ef SL |
929 | static int r5l_recovery_log(struct r5l_log *log) |
930 | { | |
355810d1 SL |
931 | struct r5l_recovery_ctx ctx; |
932 | ||
933 | ctx.pos = log->last_checkpoint; | |
934 | ctx.seq = log->last_cp_seq; | |
935 | ctx.meta_page = alloc_page(GFP_KERNEL); | |
936 | if (!ctx.meta_page) | |
937 | return -ENOMEM; | |
938 | ||
939 | r5l_recovery_flush_log(log, &ctx); | |
940 | __free_page(ctx.meta_page); | |
941 | ||
942 | /* | |
943 | * we did a recovery. Now ctx.pos points to an invalid meta block. New | |
944 | * log will start here. but we can't let superblock point to last valid | |
945 | * meta block. The log might looks like: | |
946 | * | meta 1| meta 2| meta 3| | |
947 | * meta 1 is valid, meta 2 is invalid. meta 3 could be valid. If | |
948 | * superblock points to meta 1, we write a new valid meta 2n. if crash | |
949 | * happens again, new recovery will start from meta 1. Since meta 2n is | |
950 | * valid now, recovery will think meta 3 is valid, which is wrong. | |
951 | * The solution is we create a new meta in meta2 with its seq == meta | |
952 | * 1's seq + 10 and let superblock points to meta2. The same recovery will | |
953 | * not think meta 3 is a valid meta, because its seq doesn't match | |
954 | */ | |
955 | if (ctx.seq > log->last_cp_seq + 1) { | |
956 | int ret; | |
957 | ||
958 | ret = r5l_log_write_empty_meta_block(log, ctx.pos, ctx.seq + 10); | |
959 | if (ret) | |
960 | return ret; | |
961 | log->seq = ctx.seq + 11; | |
962 | log->log_start = r5l_ring_add(log, ctx.pos, BLOCK_SECTORS); | |
963 | r5l_write_super(log, ctx.pos); | |
964 | } else { | |
965 | log->log_start = ctx.pos; | |
966 | log->seq = ctx.seq; | |
967 | } | |
f6bed0ef SL |
968 | return 0; |
969 | } | |
970 | ||
971 | static void r5l_write_super(struct r5l_log *log, sector_t cp) | |
972 | { | |
973 | struct mddev *mddev = log->rdev->mddev; | |
974 | ||
975 | log->rdev->journal_tail = cp; | |
976 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
977 | } | |
978 | ||
979 | static int r5l_load_log(struct r5l_log *log) | |
980 | { | |
981 | struct md_rdev *rdev = log->rdev; | |
982 | struct page *page; | |
983 | struct r5l_meta_block *mb; | |
984 | sector_t cp = log->rdev->journal_tail; | |
985 | u32 stored_crc, expected_crc; | |
986 | bool create_super = false; | |
987 | int ret; | |
988 | ||
989 | /* Make sure it's valid */ | |
990 | if (cp >= rdev->sectors || round_down(cp, BLOCK_SECTORS) != cp) | |
991 | cp = 0; | |
992 | page = alloc_page(GFP_KERNEL); | |
993 | if (!page) | |
994 | return -ENOMEM; | |
995 | ||
996 | if (!sync_page_io(rdev, cp, PAGE_SIZE, page, READ, false)) { | |
997 | ret = -EIO; | |
998 | goto ioerr; | |
999 | } | |
1000 | mb = page_address(page); | |
1001 | ||
1002 | if (le32_to_cpu(mb->magic) != R5LOG_MAGIC || | |
1003 | mb->version != R5LOG_VERSION) { | |
1004 | create_super = true; | |
1005 | goto create; | |
1006 | } | |
1007 | stored_crc = le32_to_cpu(mb->checksum); | |
1008 | mb->checksum = 0; | |
5cb2fbd6 | 1009 | expected_crc = crc32c_le(log->uuid_checksum, mb, PAGE_SIZE); |
f6bed0ef SL |
1010 | if (stored_crc != expected_crc) { |
1011 | create_super = true; | |
1012 | goto create; | |
1013 | } | |
1014 | if (le64_to_cpu(mb->position) != cp) { | |
1015 | create_super = true; | |
1016 | goto create; | |
1017 | } | |
1018 | create: | |
1019 | if (create_super) { | |
1020 | log->last_cp_seq = prandom_u32(); | |
1021 | cp = 0; | |
1022 | /* | |
1023 | * Make sure super points to correct address. Log might have | |
1024 | * data very soon. If super hasn't correct log tail address, | |
1025 | * recovery can't find the log | |
1026 | */ | |
1027 | r5l_write_super(log, cp); | |
1028 | } else | |
1029 | log->last_cp_seq = le64_to_cpu(mb->seq); | |
1030 | ||
1031 | log->device_size = round_down(rdev->sectors, BLOCK_SECTORS); | |
0576b1c6 SL |
1032 | log->max_free_space = log->device_size >> RECLAIM_MAX_FREE_SPACE_SHIFT; |
1033 | if (log->max_free_space > RECLAIM_MAX_FREE_SPACE) | |
1034 | log->max_free_space = RECLAIM_MAX_FREE_SPACE; | |
f6bed0ef SL |
1035 | log->last_checkpoint = cp; |
1036 | ||
1037 | __free_page(page); | |
1038 | ||
1039 | return r5l_recovery_log(log); | |
1040 | ioerr: | |
1041 | __free_page(page); | |
1042 | return ret; | |
1043 | } | |
1044 | ||
1045 | int r5l_init_log(struct r5conf *conf, struct md_rdev *rdev) | |
1046 | { | |
1047 | struct r5l_log *log; | |
1048 | ||
1049 | if (PAGE_SIZE != 4096) | |
1050 | return -EINVAL; | |
1051 | log = kzalloc(sizeof(*log), GFP_KERNEL); | |
1052 | if (!log) | |
1053 | return -ENOMEM; | |
1054 | log->rdev = rdev; | |
1055 | ||
5cb2fbd6 SL |
1056 | log->uuid_checksum = crc32c_le(~0, rdev->mddev->uuid, |
1057 | sizeof(rdev->mddev->uuid)); | |
f6bed0ef SL |
1058 | |
1059 | mutex_init(&log->io_mutex); | |
1060 | ||
1061 | spin_lock_init(&log->io_list_lock); | |
1062 | INIT_LIST_HEAD(&log->running_ios); | |
0576b1c6 SL |
1063 | INIT_LIST_HEAD(&log->io_end_ios); |
1064 | INIT_LIST_HEAD(&log->stripe_end_ios); | |
a8c34f91 SL |
1065 | INIT_LIST_HEAD(&log->flushing_ios); |
1066 | INIT_LIST_HEAD(&log->flushed_ios); | |
1067 | bio_init(&log->flush_bio); | |
f6bed0ef SL |
1068 | |
1069 | log->io_kc = KMEM_CACHE(r5l_io_unit, 0); | |
1070 | if (!log->io_kc) | |
1071 | goto io_kc; | |
1072 | ||
0576b1c6 SL |
1073 | log->reclaim_thread = md_register_thread(r5l_reclaim_thread, |
1074 | log->rdev->mddev, "reclaim"); | |
1075 | if (!log->reclaim_thread) | |
1076 | goto reclaim_thread; | |
0fd22b45 | 1077 | init_waitqueue_head(&log->iounit_wait); |
0576b1c6 | 1078 | |
f6bed0ef SL |
1079 | INIT_LIST_HEAD(&log->no_space_stripes); |
1080 | spin_lock_init(&log->no_space_stripes_lock); | |
1081 | ||
1082 | if (r5l_load_log(log)) | |
1083 | goto error; | |
1084 | ||
1085 | conf->log = log; | |
1086 | return 0; | |
1087 | error: | |
0576b1c6 SL |
1088 | md_unregister_thread(&log->reclaim_thread); |
1089 | reclaim_thread: | |
f6bed0ef SL |
1090 | kmem_cache_destroy(log->io_kc); |
1091 | io_kc: | |
1092 | kfree(log); | |
1093 | return -EINVAL; | |
1094 | } | |
1095 | ||
1096 | void r5l_exit_log(struct r5l_log *log) | |
1097 | { | |
0576b1c6 SL |
1098 | /* |
1099 | * at this point all stripes are finished, so io_unit is at least in | |
1100 | * STRIPE_END state | |
1101 | */ | |
1102 | r5l_wake_reclaim(log, -1L); | |
1103 | md_unregister_thread(&log->reclaim_thread); | |
1104 | r5l_do_reclaim(log); | |
1105 | /* | |
1106 | * force a super update, r5l_do_reclaim might updated the super. | |
1107 | * mddev->thread is already stopped | |
1108 | */ | |
1109 | md_update_sb(log->rdev->mddev, 1); | |
1110 | ||
f6bed0ef SL |
1111 | kmem_cache_destroy(log->io_kc); |
1112 | kfree(log); | |
1113 | } |