Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid5.c : Multiple Devices driver for Linux | |
3 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
4 | * Copyright (C) 1999, 2000 Ingo Molnar | |
16a53ecc | 5 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 6 | * |
16a53ecc N |
7 | * RAID-4/5/6 management functions. |
8 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
9 | * by donating a test server! | |
1da177e4 LT |
10 | * |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
ae3c20cc N |
21 | /* |
22 | * BITMAP UNPLUGGING: | |
23 | * | |
24 | * The sequencing for updating the bitmap reliably is a little | |
25 | * subtle (and I got it wrong the first time) so it deserves some | |
26 | * explanation. | |
27 | * | |
28 | * We group bitmap updates into batches. Each batch has a number. | |
29 | * We may write out several batches at once, but that isn't very important. | |
30 | * conf->bm_write is the number of the last batch successfully written. | |
31 | * conf->bm_flush is the number of the last batch that was closed to | |
32 | * new additions. | |
33 | * When we discover that we will need to write to any block in a stripe | |
34 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
35 | * the number of the batch it will be in. This is bm_flush+1. | |
36 | * When we are ready to do a write, if that batch hasn't been written yet, | |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
1da177e4 LT |
46 | #include <linux/module.h> |
47 | #include <linux/slab.h> | |
1da177e4 LT |
48 | #include <linux/highmem.h> |
49 | #include <linux/bitops.h> | |
f6705578 | 50 | #include <linux/kthread.h> |
1da177e4 | 51 | #include <asm/atomic.h> |
16a53ecc | 52 | #include "raid6.h" |
1da177e4 | 53 | |
72626685 | 54 | #include <linux/raid/bitmap.h> |
91c00924 | 55 | #include <linux/async_tx.h> |
72626685 | 56 | |
1da177e4 LT |
57 | /* |
58 | * Stripe cache | |
59 | */ | |
60 | ||
61 | #define NR_STRIPES 256 | |
62 | #define STRIPE_SIZE PAGE_SIZE | |
63 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
64 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
65 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 66 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 67 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
68 | #define HASH_MASK (NR_HASH - 1) |
69 | ||
fccddba0 | 70 | #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK])) |
1da177e4 LT |
71 | |
72 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
73 | * order without overlap. There may be several bio's per stripe+device, and | |
74 | * a bio could span several devices. | |
75 | * When walking this list for a particular stripe+device, we must never proceed | |
76 | * beyond a bio that extends past this device, as the next bio might no longer | |
77 | * be valid. | |
78 | * This macro is used to determine the 'next' bio in the list, given the sector | |
79 | * of the current stripe+device | |
80 | */ | |
81 | #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) | |
82 | /* | |
83 | * The following can be used to debug the driver | |
84 | */ | |
1da177e4 LT |
85 | #define RAID5_PARANOIA 1 |
86 | #if RAID5_PARANOIA && defined(CONFIG_SMP) | |
87 | # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) | |
88 | #else | |
89 | # define CHECK_DEVLOCK() | |
90 | #endif | |
91 | ||
45b4233c | 92 | #ifdef DEBUG |
1da177e4 LT |
93 | #define inline |
94 | #define __inline__ | |
95 | #endif | |
96 | ||
6be9d494 BS |
97 | #define printk_rl(args...) ((void) (printk_ratelimit() && printk(args))) |
98 | ||
16a53ecc N |
99 | #if !RAID6_USE_EMPTY_ZERO_PAGE |
100 | /* In .bss so it's zeroed */ | |
101 | const char raid6_empty_zero_page[PAGE_SIZE] __attribute__((aligned(256))); | |
102 | #endif | |
103 | ||
104 | static inline int raid6_next_disk(int disk, int raid_disks) | |
105 | { | |
106 | disk++; | |
107 | return (disk < raid_disks) ? disk : 0; | |
108 | } | |
a4456856 DW |
109 | |
110 | static void return_io(struct bio *return_bi) | |
111 | { | |
112 | struct bio *bi = return_bi; | |
113 | while (bi) { | |
a4456856 DW |
114 | |
115 | return_bi = bi->bi_next; | |
116 | bi->bi_next = NULL; | |
117 | bi->bi_size = 0; | |
6712ecf8 | 118 | bi->bi_end_io(bi, |
a4456856 DW |
119 | test_bit(BIO_UPTODATE, &bi->bi_flags) |
120 | ? 0 : -EIO); | |
121 | bi = return_bi; | |
122 | } | |
123 | } | |
124 | ||
1da177e4 LT |
125 | static void print_raid5_conf (raid5_conf_t *conf); |
126 | ||
858119e1 | 127 | static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 LT |
128 | { |
129 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
130 | BUG_ON(!list_empty(&sh->lru)); |
131 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 132 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
7c785b7a | 133 | if (test_bit(STRIPE_DELAYED, &sh->state)) { |
1da177e4 | 134 | list_add_tail(&sh->lru, &conf->delayed_list); |
7c785b7a N |
135 | blk_plug_device(conf->mddev->queue); |
136 | } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && | |
ae3c20cc | 137 | sh->bm_seq - conf->seq_write > 0) { |
72626685 | 138 | list_add_tail(&sh->lru, &conf->bitmap_list); |
7c785b7a N |
139 | blk_plug_device(conf->mddev->queue); |
140 | } else { | |
72626685 | 141 | clear_bit(STRIPE_BIT_DELAY, &sh->state); |
1da177e4 | 142 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 143 | } |
1da177e4 LT |
144 | md_wakeup_thread(conf->mddev->thread); |
145 | } else { | |
d84e0f10 | 146 | BUG_ON(sh->ops.pending); |
1da177e4 LT |
147 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { |
148 | atomic_dec(&conf->preread_active_stripes); | |
149 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
150 | md_wakeup_thread(conf->mddev->thread); | |
151 | } | |
1da177e4 | 152 | atomic_dec(&conf->active_stripes); |
ccfcc3c1 N |
153 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
154 | list_add_tail(&sh->lru, &conf->inactive_list); | |
1da177e4 | 155 | wake_up(&conf->wait_for_stripe); |
46031f9a RBJ |
156 | if (conf->retry_read_aligned) |
157 | md_wakeup_thread(conf->mddev->thread); | |
ccfcc3c1 | 158 | } |
1da177e4 LT |
159 | } |
160 | } | |
161 | } | |
162 | static void release_stripe(struct stripe_head *sh) | |
163 | { | |
164 | raid5_conf_t *conf = sh->raid_conf; | |
165 | unsigned long flags; | |
16a53ecc | 166 | |
1da177e4 LT |
167 | spin_lock_irqsave(&conf->device_lock, flags); |
168 | __release_stripe(conf, sh); | |
169 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
170 | } | |
171 | ||
fccddba0 | 172 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 173 | { |
45b4233c DW |
174 | pr_debug("remove_hash(), stripe %llu\n", |
175 | (unsigned long long)sh->sector); | |
1da177e4 | 176 | |
fccddba0 | 177 | hlist_del_init(&sh->hash); |
1da177e4 LT |
178 | } |
179 | ||
16a53ecc | 180 | static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 | 181 | { |
fccddba0 | 182 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 183 | |
45b4233c DW |
184 | pr_debug("insert_hash(), stripe %llu\n", |
185 | (unsigned long long)sh->sector); | |
1da177e4 LT |
186 | |
187 | CHECK_DEVLOCK(); | |
fccddba0 | 188 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
189 | } |
190 | ||
191 | ||
192 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
193 | static struct stripe_head *get_free_stripe(raid5_conf_t *conf) | |
194 | { | |
195 | struct stripe_head *sh = NULL; | |
196 | struct list_head *first; | |
197 | ||
198 | CHECK_DEVLOCK(); | |
199 | if (list_empty(&conf->inactive_list)) | |
200 | goto out; | |
201 | first = conf->inactive_list.next; | |
202 | sh = list_entry(first, struct stripe_head, lru); | |
203 | list_del_init(first); | |
204 | remove_hash(sh); | |
205 | atomic_inc(&conf->active_stripes); | |
206 | out: | |
207 | return sh; | |
208 | } | |
209 | ||
210 | static void shrink_buffers(struct stripe_head *sh, int num) | |
211 | { | |
212 | struct page *p; | |
213 | int i; | |
214 | ||
215 | for (i=0; i<num ; i++) { | |
216 | p = sh->dev[i].page; | |
217 | if (!p) | |
218 | continue; | |
219 | sh->dev[i].page = NULL; | |
2d1f3b5d | 220 | put_page(p); |
1da177e4 LT |
221 | } |
222 | } | |
223 | ||
224 | static int grow_buffers(struct stripe_head *sh, int num) | |
225 | { | |
226 | int i; | |
227 | ||
228 | for (i=0; i<num; i++) { | |
229 | struct page *page; | |
230 | ||
231 | if (!(page = alloc_page(GFP_KERNEL))) { | |
232 | return 1; | |
233 | } | |
234 | sh->dev[i].page = page; | |
235 | } | |
236 | return 0; | |
237 | } | |
238 | ||
239 | static void raid5_build_block (struct stripe_head *sh, int i); | |
240 | ||
7ecaa1e6 | 241 | static void init_stripe(struct stripe_head *sh, sector_t sector, int pd_idx, int disks) |
1da177e4 LT |
242 | { |
243 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 244 | int i; |
1da177e4 | 245 | |
78bafebd ES |
246 | BUG_ON(atomic_read(&sh->count) != 0); |
247 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
d84e0f10 DW |
248 | BUG_ON(sh->ops.pending || sh->ops.ack || sh->ops.complete); |
249 | ||
1da177e4 | 250 | CHECK_DEVLOCK(); |
45b4233c | 251 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
252 | (unsigned long long)sh->sector); |
253 | ||
254 | remove_hash(sh); | |
16a53ecc | 255 | |
1da177e4 LT |
256 | sh->sector = sector; |
257 | sh->pd_idx = pd_idx; | |
258 | sh->state = 0; | |
259 | ||
7ecaa1e6 N |
260 | sh->disks = disks; |
261 | ||
262 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
263 | struct r5dev *dev = &sh->dev[i]; |
264 | ||
d84e0f10 | 265 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 266 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 267 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 268 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 269 | dev->read, dev->towrite, dev->written, |
1da177e4 LT |
270 | test_bit(R5_LOCKED, &dev->flags)); |
271 | BUG(); | |
272 | } | |
273 | dev->flags = 0; | |
274 | raid5_build_block(sh, i); | |
275 | } | |
276 | insert_hash(conf, sh); | |
277 | } | |
278 | ||
7ecaa1e6 | 279 | static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, int disks) |
1da177e4 LT |
280 | { |
281 | struct stripe_head *sh; | |
fccddba0 | 282 | struct hlist_node *hn; |
1da177e4 LT |
283 | |
284 | CHECK_DEVLOCK(); | |
45b4233c | 285 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 286 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
7ecaa1e6 | 287 | if (sh->sector == sector && sh->disks == disks) |
1da177e4 | 288 | return sh; |
45b4233c | 289 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
290 | return NULL; |
291 | } | |
292 | ||
293 | static void unplug_slaves(mddev_t *mddev); | |
165125e1 | 294 | static void raid5_unplug_device(struct request_queue *q); |
1da177e4 | 295 | |
7ecaa1e6 N |
296 | static struct stripe_head *get_active_stripe(raid5_conf_t *conf, sector_t sector, int disks, |
297 | int pd_idx, int noblock) | |
1da177e4 LT |
298 | { |
299 | struct stripe_head *sh; | |
300 | ||
45b4233c | 301 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
302 | |
303 | spin_lock_irq(&conf->device_lock); | |
304 | ||
305 | do { | |
72626685 N |
306 | wait_event_lock_irq(conf->wait_for_stripe, |
307 | conf->quiesce == 0, | |
308 | conf->device_lock, /* nothing */); | |
7ecaa1e6 | 309 | sh = __find_stripe(conf, sector, disks); |
1da177e4 LT |
310 | if (!sh) { |
311 | if (!conf->inactive_blocked) | |
312 | sh = get_free_stripe(conf); | |
313 | if (noblock && sh == NULL) | |
314 | break; | |
315 | if (!sh) { | |
316 | conf->inactive_blocked = 1; | |
317 | wait_event_lock_irq(conf->wait_for_stripe, | |
318 | !list_empty(&conf->inactive_list) && | |
5036805b N |
319 | (atomic_read(&conf->active_stripes) |
320 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
321 | || !conf->inactive_blocked), |
322 | conf->device_lock, | |
f4370781 | 323 | raid5_unplug_device(conf->mddev->queue) |
1da177e4 LT |
324 | ); |
325 | conf->inactive_blocked = 0; | |
326 | } else | |
7ecaa1e6 | 327 | init_stripe(sh, sector, pd_idx, disks); |
1da177e4 LT |
328 | } else { |
329 | if (atomic_read(&sh->count)) { | |
78bafebd | 330 | BUG_ON(!list_empty(&sh->lru)); |
1da177e4 LT |
331 | } else { |
332 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
333 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
334 | if (list_empty(&sh->lru) && |
335 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
336 | BUG(); |
337 | list_del_init(&sh->lru); | |
1da177e4 LT |
338 | } |
339 | } | |
340 | } while (sh == NULL); | |
341 | ||
342 | if (sh) | |
343 | atomic_inc(&sh->count); | |
344 | ||
345 | spin_unlock_irq(&conf->device_lock); | |
346 | return sh; | |
347 | } | |
348 | ||
d84e0f10 DW |
349 | /* test_and_ack_op() ensures that we only dequeue an operation once */ |
350 | #define test_and_ack_op(op, pend) \ | |
351 | do { \ | |
352 | if (test_bit(op, &sh->ops.pending) && \ | |
353 | !test_bit(op, &sh->ops.complete)) { \ | |
354 | if (test_and_set_bit(op, &sh->ops.ack)) \ | |
355 | clear_bit(op, &pend); \ | |
356 | else \ | |
357 | ack++; \ | |
358 | } else \ | |
359 | clear_bit(op, &pend); \ | |
360 | } while (0) | |
361 | ||
362 | /* find new work to run, do not resubmit work that is already | |
363 | * in flight | |
364 | */ | |
365 | static unsigned long get_stripe_work(struct stripe_head *sh) | |
366 | { | |
367 | unsigned long pending; | |
368 | int ack = 0; | |
369 | ||
370 | pending = sh->ops.pending; | |
371 | ||
372 | test_and_ack_op(STRIPE_OP_BIOFILL, pending); | |
373 | test_and_ack_op(STRIPE_OP_COMPUTE_BLK, pending); | |
374 | test_and_ack_op(STRIPE_OP_PREXOR, pending); | |
375 | test_and_ack_op(STRIPE_OP_BIODRAIN, pending); | |
376 | test_and_ack_op(STRIPE_OP_POSTXOR, pending); | |
377 | test_and_ack_op(STRIPE_OP_CHECK, pending); | |
378 | if (test_and_clear_bit(STRIPE_OP_IO, &sh->ops.pending)) | |
379 | ack++; | |
380 | ||
381 | sh->ops.count -= ack; | |
4ae3f847 DW |
382 | if (unlikely(sh->ops.count < 0)) { |
383 | printk(KERN_ERR "pending: %#lx ops.pending: %#lx ops.ack: %#lx " | |
384 | "ops.complete: %#lx\n", pending, sh->ops.pending, | |
385 | sh->ops.ack, sh->ops.complete); | |
386 | BUG(); | |
387 | } | |
d84e0f10 DW |
388 | |
389 | return pending; | |
390 | } | |
391 | ||
6712ecf8 N |
392 | static void |
393 | raid5_end_read_request(struct bio *bi, int error); | |
394 | static void | |
395 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 DW |
396 | |
397 | static void ops_run_io(struct stripe_head *sh) | |
398 | { | |
399 | raid5_conf_t *conf = sh->raid_conf; | |
400 | int i, disks = sh->disks; | |
401 | ||
402 | might_sleep(); | |
403 | ||
8b3e6cdc | 404 | set_bit(STRIPE_IO_STARTED, &sh->state); |
91c00924 DW |
405 | for (i = disks; i--; ) { |
406 | int rw; | |
407 | struct bio *bi; | |
408 | mdk_rdev_t *rdev; | |
409 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
410 | rw = WRITE; | |
411 | else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
412 | rw = READ; | |
413 | else | |
414 | continue; | |
415 | ||
416 | bi = &sh->dev[i].req; | |
417 | ||
418 | bi->bi_rw = rw; | |
419 | if (rw == WRITE) | |
420 | bi->bi_end_io = raid5_end_write_request; | |
421 | else | |
422 | bi->bi_end_io = raid5_end_read_request; | |
423 | ||
424 | rcu_read_lock(); | |
425 | rdev = rcu_dereference(conf->disks[i].rdev); | |
426 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
427 | rdev = NULL; | |
428 | if (rdev) | |
429 | atomic_inc(&rdev->nr_pending); | |
430 | rcu_read_unlock(); | |
431 | ||
432 | if (rdev) { | |
433 | if (test_bit(STRIPE_SYNCING, &sh->state) || | |
434 | test_bit(STRIPE_EXPAND_SOURCE, &sh->state) || | |
435 | test_bit(STRIPE_EXPAND_READY, &sh->state)) | |
436 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); | |
437 | ||
438 | bi->bi_bdev = rdev->bdev; | |
439 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 440 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
441 | bi->bi_rw, i); |
442 | atomic_inc(&sh->count); | |
443 | bi->bi_sector = sh->sector + rdev->data_offset; | |
444 | bi->bi_flags = 1 << BIO_UPTODATE; | |
445 | bi->bi_vcnt = 1; | |
446 | bi->bi_max_vecs = 1; | |
447 | bi->bi_idx = 0; | |
448 | bi->bi_io_vec = &sh->dev[i].vec; | |
449 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
450 | bi->bi_io_vec[0].bv_offset = 0; | |
451 | bi->bi_size = STRIPE_SIZE; | |
452 | bi->bi_next = NULL; | |
453 | if (rw == WRITE && | |
454 | test_bit(R5_ReWrite, &sh->dev[i].flags)) | |
455 | atomic_add(STRIPE_SECTORS, | |
456 | &rdev->corrected_errors); | |
457 | generic_make_request(bi); | |
458 | } else { | |
459 | if (rw == WRITE) | |
460 | set_bit(STRIPE_DEGRADED, &sh->state); | |
461 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
462 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
463 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
464 | set_bit(STRIPE_HANDLE, &sh->state); | |
465 | } | |
466 | } | |
467 | } | |
468 | ||
469 | static struct dma_async_tx_descriptor * | |
470 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
471 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
472 | { | |
473 | struct bio_vec *bvl; | |
474 | struct page *bio_page; | |
475 | int i; | |
476 | int page_offset; | |
477 | ||
478 | if (bio->bi_sector >= sector) | |
479 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
480 | else | |
481 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
482 | bio_for_each_segment(bvl, bio, i) { | |
483 | int len = bio_iovec_idx(bio, i)->bv_len; | |
484 | int clen; | |
485 | int b_offset = 0; | |
486 | ||
487 | if (page_offset < 0) { | |
488 | b_offset = -page_offset; | |
489 | page_offset += b_offset; | |
490 | len -= b_offset; | |
491 | } | |
492 | ||
493 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
494 | clen = STRIPE_SIZE - page_offset; | |
495 | else | |
496 | clen = len; | |
497 | ||
498 | if (clen > 0) { | |
499 | b_offset += bio_iovec_idx(bio, i)->bv_offset; | |
500 | bio_page = bio_iovec_idx(bio, i)->bv_page; | |
501 | if (frombio) | |
502 | tx = async_memcpy(page, bio_page, page_offset, | |
503 | b_offset, clen, | |
eb0645a8 | 504 | ASYNC_TX_DEP_ACK, |
91c00924 DW |
505 | tx, NULL, NULL); |
506 | else | |
507 | tx = async_memcpy(bio_page, page, b_offset, | |
508 | page_offset, clen, | |
eb0645a8 | 509 | ASYNC_TX_DEP_ACK, |
91c00924 DW |
510 | tx, NULL, NULL); |
511 | } | |
512 | if (clen < len) /* hit end of page */ | |
513 | break; | |
514 | page_offset += len; | |
515 | } | |
516 | ||
517 | return tx; | |
518 | } | |
519 | ||
520 | static void ops_complete_biofill(void *stripe_head_ref) | |
521 | { | |
522 | struct stripe_head *sh = stripe_head_ref; | |
523 | struct bio *return_bi = NULL; | |
524 | raid5_conf_t *conf = sh->raid_conf; | |
e4d84909 | 525 | int i; |
91c00924 | 526 | |
e46b272b | 527 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
528 | (unsigned long long)sh->sector); |
529 | ||
530 | /* clear completed biofills */ | |
531 | for (i = sh->disks; i--; ) { | |
532 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
533 | |
534 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
535 | /* and check if we need to reply to a read request, |
536 | * new R5_Wantfill requests are held off until | |
537 | * !test_bit(STRIPE_OP_BIOFILL, &sh->ops.pending) | |
538 | */ | |
539 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 540 | struct bio *rbi, *rbi2; |
91c00924 DW |
541 | |
542 | /* The access to dev->read is outside of the | |
543 | * spin_lock_irq(&conf->device_lock), but is protected | |
544 | * by the STRIPE_OP_BIOFILL pending bit | |
545 | */ | |
546 | BUG_ON(!dev->read); | |
547 | rbi = dev->read; | |
548 | dev->read = NULL; | |
549 | while (rbi && rbi->bi_sector < | |
550 | dev->sector + STRIPE_SECTORS) { | |
551 | rbi2 = r5_next_bio(rbi, dev->sector); | |
552 | spin_lock_irq(&conf->device_lock); | |
553 | if (--rbi->bi_phys_segments == 0) { | |
554 | rbi->bi_next = return_bi; | |
555 | return_bi = rbi; | |
556 | } | |
557 | spin_unlock_irq(&conf->device_lock); | |
558 | rbi = rbi2; | |
559 | } | |
560 | } | |
561 | } | |
4ae3f847 | 562 | set_bit(STRIPE_OP_BIOFILL, &sh->ops.complete); |
91c00924 DW |
563 | |
564 | return_io(return_bi); | |
565 | ||
e4d84909 | 566 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
567 | release_stripe(sh); |
568 | } | |
569 | ||
570 | static void ops_run_biofill(struct stripe_head *sh) | |
571 | { | |
572 | struct dma_async_tx_descriptor *tx = NULL; | |
573 | raid5_conf_t *conf = sh->raid_conf; | |
574 | int i; | |
575 | ||
e46b272b | 576 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
577 | (unsigned long long)sh->sector); |
578 | ||
579 | for (i = sh->disks; i--; ) { | |
580 | struct r5dev *dev = &sh->dev[i]; | |
581 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
582 | struct bio *rbi; | |
583 | spin_lock_irq(&conf->device_lock); | |
584 | dev->read = rbi = dev->toread; | |
585 | dev->toread = NULL; | |
586 | spin_unlock_irq(&conf->device_lock); | |
587 | while (rbi && rbi->bi_sector < | |
588 | dev->sector + STRIPE_SECTORS) { | |
589 | tx = async_copy_data(0, rbi, dev->page, | |
590 | dev->sector, tx); | |
591 | rbi = r5_next_bio(rbi, dev->sector); | |
592 | } | |
593 | } | |
594 | } | |
595 | ||
596 | atomic_inc(&sh->count); | |
597 | async_trigger_callback(ASYNC_TX_DEP_ACK | ASYNC_TX_ACK, tx, | |
598 | ops_complete_biofill, sh); | |
599 | } | |
600 | ||
601 | static void ops_complete_compute5(void *stripe_head_ref) | |
602 | { | |
603 | struct stripe_head *sh = stripe_head_ref; | |
604 | int target = sh->ops.target; | |
605 | struct r5dev *tgt = &sh->dev[target]; | |
606 | ||
e46b272b | 607 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
608 | (unsigned long long)sh->sector); |
609 | ||
610 | set_bit(R5_UPTODATE, &tgt->flags); | |
611 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
612 | clear_bit(R5_Wantcompute, &tgt->flags); | |
613 | set_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.complete); | |
614 | set_bit(STRIPE_HANDLE, &sh->state); | |
615 | release_stripe(sh); | |
616 | } | |
617 | ||
618 | static struct dma_async_tx_descriptor * | |
619 | ops_run_compute5(struct stripe_head *sh, unsigned long pending) | |
620 | { | |
621 | /* kernel stack size limits the total number of disks */ | |
622 | int disks = sh->disks; | |
623 | struct page *xor_srcs[disks]; | |
624 | int target = sh->ops.target; | |
625 | struct r5dev *tgt = &sh->dev[target]; | |
626 | struct page *xor_dest = tgt->page; | |
627 | int count = 0; | |
628 | struct dma_async_tx_descriptor *tx; | |
629 | int i; | |
630 | ||
631 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 632 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
633 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
634 | ||
635 | for (i = disks; i--; ) | |
636 | if (i != target) | |
637 | xor_srcs[count++] = sh->dev[i].page; | |
638 | ||
639 | atomic_inc(&sh->count); | |
640 | ||
641 | if (unlikely(count == 1)) | |
642 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, | |
643 | 0, NULL, ops_complete_compute5, sh); | |
644 | else | |
645 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
646 | ASYNC_TX_XOR_ZERO_DST, NULL, | |
647 | ops_complete_compute5, sh); | |
648 | ||
649 | /* ack now if postxor is not set to be run */ | |
650 | if (tx && !test_bit(STRIPE_OP_POSTXOR, &pending)) | |
651 | async_tx_ack(tx); | |
652 | ||
653 | return tx; | |
654 | } | |
655 | ||
656 | static void ops_complete_prexor(void *stripe_head_ref) | |
657 | { | |
658 | struct stripe_head *sh = stripe_head_ref; | |
659 | ||
e46b272b | 660 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
661 | (unsigned long long)sh->sector); |
662 | ||
663 | set_bit(STRIPE_OP_PREXOR, &sh->ops.complete); | |
664 | } | |
665 | ||
666 | static struct dma_async_tx_descriptor * | |
667 | ops_run_prexor(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) | |
668 | { | |
669 | /* kernel stack size limits the total number of disks */ | |
670 | int disks = sh->disks; | |
671 | struct page *xor_srcs[disks]; | |
672 | int count = 0, pd_idx = sh->pd_idx, i; | |
673 | ||
674 | /* existing parity data subtracted */ | |
675 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
676 | ||
e46b272b | 677 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
678 | (unsigned long long)sh->sector); |
679 | ||
680 | for (i = disks; i--; ) { | |
681 | struct r5dev *dev = &sh->dev[i]; | |
682 | /* Only process blocks that are known to be uptodate */ | |
683 | if (dev->towrite && test_bit(R5_Wantprexor, &dev->flags)) | |
684 | xor_srcs[count++] = dev->page; | |
685 | } | |
686 | ||
687 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
688 | ASYNC_TX_DEP_ACK | ASYNC_TX_XOR_DROP_DST, tx, | |
689 | ops_complete_prexor, sh); | |
690 | ||
691 | return tx; | |
692 | } | |
693 | ||
694 | static struct dma_async_tx_descriptor * | |
6c55be8b DW |
695 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx, |
696 | unsigned long pending) | |
91c00924 DW |
697 | { |
698 | int disks = sh->disks; | |
699 | int pd_idx = sh->pd_idx, i; | |
700 | ||
701 | /* check if prexor is active which means only process blocks | |
702 | * that are part of a read-modify-write (Wantprexor) | |
703 | */ | |
6c55be8b | 704 | int prexor = test_bit(STRIPE_OP_PREXOR, &pending); |
91c00924 | 705 | |
e46b272b | 706 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
707 | (unsigned long long)sh->sector); |
708 | ||
709 | for (i = disks; i--; ) { | |
710 | struct r5dev *dev = &sh->dev[i]; | |
711 | struct bio *chosen; | |
712 | int towrite; | |
713 | ||
714 | towrite = 0; | |
715 | if (prexor) { /* rmw */ | |
716 | if (dev->towrite && | |
717 | test_bit(R5_Wantprexor, &dev->flags)) | |
718 | towrite = 1; | |
719 | } else { /* rcw */ | |
720 | if (i != pd_idx && dev->towrite && | |
721 | test_bit(R5_LOCKED, &dev->flags)) | |
722 | towrite = 1; | |
723 | } | |
724 | ||
725 | if (towrite) { | |
726 | struct bio *wbi; | |
727 | ||
728 | spin_lock(&sh->lock); | |
729 | chosen = dev->towrite; | |
730 | dev->towrite = NULL; | |
731 | BUG_ON(dev->written); | |
732 | wbi = dev->written = chosen; | |
733 | spin_unlock(&sh->lock); | |
734 | ||
735 | while (wbi && wbi->bi_sector < | |
736 | dev->sector + STRIPE_SECTORS) { | |
737 | tx = async_copy_data(1, wbi, dev->page, | |
738 | dev->sector, tx); | |
739 | wbi = r5_next_bio(wbi, dev->sector); | |
740 | } | |
741 | } | |
742 | } | |
743 | ||
744 | return tx; | |
745 | } | |
746 | ||
747 | static void ops_complete_postxor(void *stripe_head_ref) | |
748 | { | |
749 | struct stripe_head *sh = stripe_head_ref; | |
750 | ||
e46b272b | 751 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
752 | (unsigned long long)sh->sector); |
753 | ||
754 | set_bit(STRIPE_OP_POSTXOR, &sh->ops.complete); | |
755 | set_bit(STRIPE_HANDLE, &sh->state); | |
756 | release_stripe(sh); | |
757 | } | |
758 | ||
759 | static void ops_complete_write(void *stripe_head_ref) | |
760 | { | |
761 | struct stripe_head *sh = stripe_head_ref; | |
762 | int disks = sh->disks, i, pd_idx = sh->pd_idx; | |
763 | ||
e46b272b | 764 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
765 | (unsigned long long)sh->sector); |
766 | ||
767 | for (i = disks; i--; ) { | |
768 | struct r5dev *dev = &sh->dev[i]; | |
769 | if (dev->written || i == pd_idx) | |
770 | set_bit(R5_UPTODATE, &dev->flags); | |
771 | } | |
772 | ||
773 | set_bit(STRIPE_OP_BIODRAIN, &sh->ops.complete); | |
774 | set_bit(STRIPE_OP_POSTXOR, &sh->ops.complete); | |
775 | ||
776 | set_bit(STRIPE_HANDLE, &sh->state); | |
777 | release_stripe(sh); | |
778 | } | |
779 | ||
780 | static void | |
6c55be8b DW |
781 | ops_run_postxor(struct stripe_head *sh, struct dma_async_tx_descriptor *tx, |
782 | unsigned long pending) | |
91c00924 DW |
783 | { |
784 | /* kernel stack size limits the total number of disks */ | |
785 | int disks = sh->disks; | |
786 | struct page *xor_srcs[disks]; | |
787 | ||
788 | int count = 0, pd_idx = sh->pd_idx, i; | |
789 | struct page *xor_dest; | |
6c55be8b | 790 | int prexor = test_bit(STRIPE_OP_PREXOR, &pending); |
91c00924 DW |
791 | unsigned long flags; |
792 | dma_async_tx_callback callback; | |
793 | ||
e46b272b | 794 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
795 | (unsigned long long)sh->sector); |
796 | ||
797 | /* check if prexor is active which means only process blocks | |
798 | * that are part of a read-modify-write (written) | |
799 | */ | |
800 | if (prexor) { | |
801 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
802 | for (i = disks; i--; ) { | |
803 | struct r5dev *dev = &sh->dev[i]; | |
804 | if (dev->written) | |
805 | xor_srcs[count++] = dev->page; | |
806 | } | |
807 | } else { | |
808 | xor_dest = sh->dev[pd_idx].page; | |
809 | for (i = disks; i--; ) { | |
810 | struct r5dev *dev = &sh->dev[i]; | |
811 | if (i != pd_idx) | |
812 | xor_srcs[count++] = dev->page; | |
813 | } | |
814 | } | |
815 | ||
816 | /* check whether this postxor is part of a write */ | |
6c55be8b | 817 | callback = test_bit(STRIPE_OP_BIODRAIN, &pending) ? |
91c00924 DW |
818 | ops_complete_write : ops_complete_postxor; |
819 | ||
820 | /* 1/ if we prexor'd then the dest is reused as a source | |
821 | * 2/ if we did not prexor then we are redoing the parity | |
822 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
823 | * for the synchronous xor case | |
824 | */ | |
825 | flags = ASYNC_TX_DEP_ACK | ASYNC_TX_ACK | | |
826 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
827 | ||
828 | atomic_inc(&sh->count); | |
829 | ||
830 | if (unlikely(count == 1)) { | |
831 | flags &= ~(ASYNC_TX_XOR_DROP_DST | ASYNC_TX_XOR_ZERO_DST); | |
832 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, | |
833 | flags, tx, callback, sh); | |
834 | } else | |
835 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
836 | flags, tx, callback, sh); | |
837 | } | |
838 | ||
839 | static void ops_complete_check(void *stripe_head_ref) | |
840 | { | |
841 | struct stripe_head *sh = stripe_head_ref; | |
842 | int pd_idx = sh->pd_idx; | |
843 | ||
e46b272b | 844 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
845 | (unsigned long long)sh->sector); |
846 | ||
847 | if (test_and_clear_bit(STRIPE_OP_MOD_DMA_CHECK, &sh->ops.pending) && | |
848 | sh->ops.zero_sum_result == 0) | |
849 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
850 | ||
851 | set_bit(STRIPE_OP_CHECK, &sh->ops.complete); | |
852 | set_bit(STRIPE_HANDLE, &sh->state); | |
853 | release_stripe(sh); | |
854 | } | |
855 | ||
856 | static void ops_run_check(struct stripe_head *sh) | |
857 | { | |
858 | /* kernel stack size limits the total number of disks */ | |
859 | int disks = sh->disks; | |
860 | struct page *xor_srcs[disks]; | |
861 | struct dma_async_tx_descriptor *tx; | |
862 | ||
863 | int count = 0, pd_idx = sh->pd_idx, i; | |
864 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
865 | ||
e46b272b | 866 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
867 | (unsigned long long)sh->sector); |
868 | ||
869 | for (i = disks; i--; ) { | |
870 | struct r5dev *dev = &sh->dev[i]; | |
871 | if (i != pd_idx) | |
872 | xor_srcs[count++] = dev->page; | |
873 | } | |
874 | ||
875 | tx = async_xor_zero_sum(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
876 | &sh->ops.zero_sum_result, 0, NULL, NULL, NULL); | |
877 | ||
878 | if (tx) | |
879 | set_bit(STRIPE_OP_MOD_DMA_CHECK, &sh->ops.pending); | |
880 | else | |
881 | clear_bit(STRIPE_OP_MOD_DMA_CHECK, &sh->ops.pending); | |
882 | ||
883 | atomic_inc(&sh->count); | |
884 | tx = async_trigger_callback(ASYNC_TX_DEP_ACK | ASYNC_TX_ACK, tx, | |
885 | ops_complete_check, sh); | |
886 | } | |
887 | ||
888 | static void raid5_run_ops(struct stripe_head *sh, unsigned long pending) | |
889 | { | |
890 | int overlap_clear = 0, i, disks = sh->disks; | |
891 | struct dma_async_tx_descriptor *tx = NULL; | |
892 | ||
893 | if (test_bit(STRIPE_OP_BIOFILL, &pending)) { | |
894 | ops_run_biofill(sh); | |
895 | overlap_clear++; | |
896 | } | |
897 | ||
898 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &pending)) | |
899 | tx = ops_run_compute5(sh, pending); | |
900 | ||
901 | if (test_bit(STRIPE_OP_PREXOR, &pending)) | |
902 | tx = ops_run_prexor(sh, tx); | |
903 | ||
904 | if (test_bit(STRIPE_OP_BIODRAIN, &pending)) { | |
6c55be8b | 905 | tx = ops_run_biodrain(sh, tx, pending); |
91c00924 DW |
906 | overlap_clear++; |
907 | } | |
908 | ||
909 | if (test_bit(STRIPE_OP_POSTXOR, &pending)) | |
6c55be8b | 910 | ops_run_postxor(sh, tx, pending); |
91c00924 DW |
911 | |
912 | if (test_bit(STRIPE_OP_CHECK, &pending)) | |
913 | ops_run_check(sh); | |
914 | ||
915 | if (test_bit(STRIPE_OP_IO, &pending)) | |
916 | ops_run_io(sh); | |
917 | ||
918 | if (overlap_clear) | |
919 | for (i = disks; i--; ) { | |
920 | struct r5dev *dev = &sh->dev[i]; | |
921 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
922 | wake_up(&sh->raid_conf->wait_for_overlap); | |
923 | } | |
924 | } | |
925 | ||
3f294f4f | 926 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
927 | { |
928 | struct stripe_head *sh; | |
3f294f4f N |
929 | sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL); |
930 | if (!sh) | |
931 | return 0; | |
932 | memset(sh, 0, sizeof(*sh) + (conf->raid_disks-1)*sizeof(struct r5dev)); | |
933 | sh->raid_conf = conf; | |
934 | spin_lock_init(&sh->lock); | |
935 | ||
936 | if (grow_buffers(sh, conf->raid_disks)) { | |
937 | shrink_buffers(sh, conf->raid_disks); | |
938 | kmem_cache_free(conf->slab_cache, sh); | |
939 | return 0; | |
940 | } | |
7ecaa1e6 | 941 | sh->disks = conf->raid_disks; |
3f294f4f N |
942 | /* we just created an active stripe so... */ |
943 | atomic_set(&sh->count, 1); | |
944 | atomic_inc(&conf->active_stripes); | |
945 | INIT_LIST_HEAD(&sh->lru); | |
946 | release_stripe(sh); | |
947 | return 1; | |
948 | } | |
949 | ||
950 | static int grow_stripes(raid5_conf_t *conf, int num) | |
951 | { | |
e18b890b | 952 | struct kmem_cache *sc; |
1da177e4 LT |
953 | int devs = conf->raid_disks; |
954 | ||
42b9bebe N |
955 | sprintf(conf->cache_name[0], "raid5-%s", mdname(conf->mddev)); |
956 | sprintf(conf->cache_name[1], "raid5-%s-alt", mdname(conf->mddev)); | |
ad01c9e3 N |
957 | conf->active_name = 0; |
958 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 959 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 960 | 0, 0, NULL); |
1da177e4 LT |
961 | if (!sc) |
962 | return 1; | |
963 | conf->slab_cache = sc; | |
ad01c9e3 | 964 | conf->pool_size = devs; |
16a53ecc | 965 | while (num--) |
3f294f4f | 966 | if (!grow_one_stripe(conf)) |
1da177e4 | 967 | return 1; |
1da177e4 LT |
968 | return 0; |
969 | } | |
29269553 N |
970 | |
971 | #ifdef CONFIG_MD_RAID5_RESHAPE | |
ad01c9e3 N |
972 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
973 | { | |
974 | /* Make all the stripes able to hold 'newsize' devices. | |
975 | * New slots in each stripe get 'page' set to a new page. | |
976 | * | |
977 | * This happens in stages: | |
978 | * 1/ create a new kmem_cache and allocate the required number of | |
979 | * stripe_heads. | |
980 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
981 | * to the new stripe_heads. This will have the side effect of | |
982 | * freezing the array as once all stripe_heads have been collected, | |
983 | * no IO will be possible. Old stripe heads are freed once their | |
984 | * pages have been transferred over, and the old kmem_cache is | |
985 | * freed when all stripes are done. | |
986 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
987 | * we simple return a failre status - no need to clean anything up. | |
988 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
989 | * If this fails, we don't bother trying the shrink the | |
990 | * stripe_heads down again, we just leave them as they are. | |
991 | * As each stripe_head is processed the new one is released into | |
992 | * active service. | |
993 | * | |
994 | * Once step2 is started, we cannot afford to wait for a write, | |
995 | * so we use GFP_NOIO allocations. | |
996 | */ | |
997 | struct stripe_head *osh, *nsh; | |
998 | LIST_HEAD(newstripes); | |
999 | struct disk_info *ndisks; | |
1000 | int err = 0; | |
e18b890b | 1001 | struct kmem_cache *sc; |
ad01c9e3 N |
1002 | int i; |
1003 | ||
1004 | if (newsize <= conf->pool_size) | |
1005 | return 0; /* never bother to shrink */ | |
1006 | ||
2a2275d6 N |
1007 | md_allow_write(conf->mddev); |
1008 | ||
ad01c9e3 N |
1009 | /* Step 1 */ |
1010 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1011 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1012 | 0, 0, NULL); |
ad01c9e3 N |
1013 | if (!sc) |
1014 | return -ENOMEM; | |
1015 | ||
1016 | for (i = conf->max_nr_stripes; i; i--) { | |
1017 | nsh = kmem_cache_alloc(sc, GFP_KERNEL); | |
1018 | if (!nsh) | |
1019 | break; | |
1020 | ||
1021 | memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev)); | |
1022 | ||
1023 | nsh->raid_conf = conf; | |
1024 | spin_lock_init(&nsh->lock); | |
1025 | ||
1026 | list_add(&nsh->lru, &newstripes); | |
1027 | } | |
1028 | if (i) { | |
1029 | /* didn't get enough, give up */ | |
1030 | while (!list_empty(&newstripes)) { | |
1031 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1032 | list_del(&nsh->lru); | |
1033 | kmem_cache_free(sc, nsh); | |
1034 | } | |
1035 | kmem_cache_destroy(sc); | |
1036 | return -ENOMEM; | |
1037 | } | |
1038 | /* Step 2 - Must use GFP_NOIO now. | |
1039 | * OK, we have enough stripes, start collecting inactive | |
1040 | * stripes and copying them over | |
1041 | */ | |
1042 | list_for_each_entry(nsh, &newstripes, lru) { | |
1043 | spin_lock_irq(&conf->device_lock); | |
1044 | wait_event_lock_irq(conf->wait_for_stripe, | |
1045 | !list_empty(&conf->inactive_list), | |
1046 | conf->device_lock, | |
b3b46be3 | 1047 | unplug_slaves(conf->mddev) |
ad01c9e3 N |
1048 | ); |
1049 | osh = get_free_stripe(conf); | |
1050 | spin_unlock_irq(&conf->device_lock); | |
1051 | atomic_set(&nsh->count, 1); | |
1052 | for(i=0; i<conf->pool_size; i++) | |
1053 | nsh->dev[i].page = osh->dev[i].page; | |
1054 | for( ; i<newsize; i++) | |
1055 | nsh->dev[i].page = NULL; | |
1056 | kmem_cache_free(conf->slab_cache, osh); | |
1057 | } | |
1058 | kmem_cache_destroy(conf->slab_cache); | |
1059 | ||
1060 | /* Step 3. | |
1061 | * At this point, we are holding all the stripes so the array | |
1062 | * is completely stalled, so now is a good time to resize | |
1063 | * conf->disks. | |
1064 | */ | |
1065 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1066 | if (ndisks) { | |
1067 | for (i=0; i<conf->raid_disks; i++) | |
1068 | ndisks[i] = conf->disks[i]; | |
1069 | kfree(conf->disks); | |
1070 | conf->disks = ndisks; | |
1071 | } else | |
1072 | err = -ENOMEM; | |
1073 | ||
1074 | /* Step 4, return new stripes to service */ | |
1075 | while(!list_empty(&newstripes)) { | |
1076 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1077 | list_del_init(&nsh->lru); | |
1078 | for (i=conf->raid_disks; i < newsize; i++) | |
1079 | if (nsh->dev[i].page == NULL) { | |
1080 | struct page *p = alloc_page(GFP_NOIO); | |
1081 | nsh->dev[i].page = p; | |
1082 | if (!p) | |
1083 | err = -ENOMEM; | |
1084 | } | |
1085 | release_stripe(nsh); | |
1086 | } | |
1087 | /* critical section pass, GFP_NOIO no longer needed */ | |
1088 | ||
1089 | conf->slab_cache = sc; | |
1090 | conf->active_name = 1-conf->active_name; | |
1091 | conf->pool_size = newsize; | |
1092 | return err; | |
1093 | } | |
29269553 | 1094 | #endif |
1da177e4 | 1095 | |
3f294f4f | 1096 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1097 | { |
1098 | struct stripe_head *sh; | |
1099 | ||
3f294f4f N |
1100 | spin_lock_irq(&conf->device_lock); |
1101 | sh = get_free_stripe(conf); | |
1102 | spin_unlock_irq(&conf->device_lock); | |
1103 | if (!sh) | |
1104 | return 0; | |
78bafebd | 1105 | BUG_ON(atomic_read(&sh->count)); |
ad01c9e3 | 1106 | shrink_buffers(sh, conf->pool_size); |
3f294f4f N |
1107 | kmem_cache_free(conf->slab_cache, sh); |
1108 | atomic_dec(&conf->active_stripes); | |
1109 | return 1; | |
1110 | } | |
1111 | ||
1112 | static void shrink_stripes(raid5_conf_t *conf) | |
1113 | { | |
1114 | while (drop_one_stripe(conf)) | |
1115 | ; | |
1116 | ||
29fc7e3e N |
1117 | if (conf->slab_cache) |
1118 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1119 | conf->slab_cache = NULL; |
1120 | } | |
1121 | ||
6712ecf8 | 1122 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 LT |
1123 | { |
1124 | struct stripe_head *sh = bi->bi_private; | |
1125 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 1126 | int disks = sh->disks, i; |
1da177e4 | 1127 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
1128 | char b[BDEVNAME_SIZE]; |
1129 | mdk_rdev_t *rdev; | |
1da177e4 | 1130 | |
1da177e4 LT |
1131 | |
1132 | for (i=0 ; i<disks; i++) | |
1133 | if (bi == &sh->dev[i].req) | |
1134 | break; | |
1135 | ||
45b4233c DW |
1136 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1137 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1138 | uptodate); |
1139 | if (i == disks) { | |
1140 | BUG(); | |
6712ecf8 | 1141 | return; |
1da177e4 LT |
1142 | } |
1143 | ||
1144 | if (uptodate) { | |
1da177e4 | 1145 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1146 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1147 | rdev = conf->disks[i].rdev; |
6be9d494 BS |
1148 | printk_rl(KERN_INFO "raid5:%s: read error corrected" |
1149 | " (%lu sectors at %llu on %s)\n", | |
1150 | mdname(conf->mddev), STRIPE_SECTORS, | |
1151 | (unsigned long long)(sh->sector | |
1152 | + rdev->data_offset), | |
1153 | bdevname(rdev->bdev, b)); | |
4e5314b5 N |
1154 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1155 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1156 | } | |
ba22dcbf N |
1157 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1158 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1159 | } else { |
d6950432 | 1160 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1161 | int retry = 0; |
d6950432 N |
1162 | rdev = conf->disks[i].rdev; |
1163 | ||
1da177e4 | 1164 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1165 | atomic_inc(&rdev->read_errors); |
ba22dcbf | 1166 | if (conf->mddev->degraded) |
6be9d494 BS |
1167 | printk_rl(KERN_WARNING |
1168 | "raid5:%s: read error not correctable " | |
1169 | "(sector %llu on %s).\n", | |
1170 | mdname(conf->mddev), | |
1171 | (unsigned long long)(sh->sector | |
1172 | + rdev->data_offset), | |
1173 | bdn); | |
ba22dcbf | 1174 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1175 | /* Oh, no!!! */ |
6be9d494 BS |
1176 | printk_rl(KERN_WARNING |
1177 | "raid5:%s: read error NOT corrected!! " | |
1178 | "(sector %llu on %s).\n", | |
1179 | mdname(conf->mddev), | |
1180 | (unsigned long long)(sh->sector | |
1181 | + rdev->data_offset), | |
1182 | bdn); | |
d6950432 | 1183 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1184 | > conf->max_nr_stripes) |
14f8d26b | 1185 | printk(KERN_WARNING |
d6950432 N |
1186 | "raid5:%s: Too many read errors, failing device %s.\n", |
1187 | mdname(conf->mddev), bdn); | |
ba22dcbf N |
1188 | else |
1189 | retry = 1; | |
1190 | if (retry) | |
1191 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1192 | else { | |
4e5314b5 N |
1193 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1194 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1195 | md_error(conf->mddev, rdev); |
ba22dcbf | 1196 | } |
1da177e4 LT |
1197 | } |
1198 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1199 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1200 | set_bit(STRIPE_HANDLE, &sh->state); | |
1201 | release_stripe(sh); | |
1da177e4 LT |
1202 | } |
1203 | ||
6712ecf8 | 1204 | static void raid5_end_write_request (struct bio *bi, int error) |
1da177e4 LT |
1205 | { |
1206 | struct stripe_head *sh = bi->bi_private; | |
1207 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 1208 | int disks = sh->disks, i; |
1da177e4 LT |
1209 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
1210 | ||
1da177e4 LT |
1211 | for (i=0 ; i<disks; i++) |
1212 | if (bi == &sh->dev[i].req) | |
1213 | break; | |
1214 | ||
45b4233c | 1215 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1216 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1217 | uptodate); | |
1218 | if (i == disks) { | |
1219 | BUG(); | |
6712ecf8 | 1220 | return; |
1da177e4 LT |
1221 | } |
1222 | ||
1da177e4 LT |
1223 | if (!uptodate) |
1224 | md_error(conf->mddev, conf->disks[i].rdev); | |
1225 | ||
1226 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1227 | ||
1228 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1229 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1230 | release_stripe(sh); |
1da177e4 LT |
1231 | } |
1232 | ||
1233 | ||
1234 | static sector_t compute_blocknr(struct stripe_head *sh, int i); | |
1235 | ||
1236 | static void raid5_build_block (struct stripe_head *sh, int i) | |
1237 | { | |
1238 | struct r5dev *dev = &sh->dev[i]; | |
1239 | ||
1240 | bio_init(&dev->req); | |
1241 | dev->req.bi_io_vec = &dev->vec; | |
1242 | dev->req.bi_vcnt++; | |
1243 | dev->req.bi_max_vecs++; | |
1244 | dev->vec.bv_page = dev->page; | |
1245 | dev->vec.bv_len = STRIPE_SIZE; | |
1246 | dev->vec.bv_offset = 0; | |
1247 | ||
1248 | dev->req.bi_sector = sh->sector; | |
1249 | dev->req.bi_private = sh; | |
1250 | ||
1251 | dev->flags = 0; | |
16a53ecc | 1252 | dev->sector = compute_blocknr(sh, i); |
1da177e4 LT |
1253 | } |
1254 | ||
1255 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1256 | { | |
1257 | char b[BDEVNAME_SIZE]; | |
1258 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
45b4233c | 1259 | pr_debug("raid5: error called\n"); |
1da177e4 | 1260 | |
b2d444d7 | 1261 | if (!test_bit(Faulty, &rdev->flags)) { |
850b2b42 | 1262 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
c04be0aa N |
1263 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1264 | unsigned long flags; | |
1265 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1266 | mddev->degraded++; |
c04be0aa | 1267 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1268 | /* |
1269 | * if recovery was running, make sure it aborts. | |
1270 | */ | |
dfc70645 | 1271 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1272 | } |
b2d444d7 | 1273 | set_bit(Faulty, &rdev->flags); |
1da177e4 | 1274 | printk (KERN_ALERT |
d7a420c9 NA |
1275 | "raid5: Disk failure on %s, disabling device.\n" |
1276 | "raid5: Operation continuing on %d devices.\n", | |
02c2de8c | 1277 | bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded); |
1da177e4 | 1278 | } |
16a53ecc | 1279 | } |
1da177e4 LT |
1280 | |
1281 | /* | |
1282 | * Input: a 'big' sector number, | |
1283 | * Output: index of the data and parity disk, and the sector # in them. | |
1284 | */ | |
1285 | static sector_t raid5_compute_sector(sector_t r_sector, unsigned int raid_disks, | |
1286 | unsigned int data_disks, unsigned int * dd_idx, | |
1287 | unsigned int * pd_idx, raid5_conf_t *conf) | |
1288 | { | |
1289 | long stripe; | |
1290 | unsigned long chunk_number; | |
1291 | unsigned int chunk_offset; | |
1292 | sector_t new_sector; | |
1293 | int sectors_per_chunk = conf->chunk_size >> 9; | |
1294 | ||
1295 | /* First compute the information on this sector */ | |
1296 | ||
1297 | /* | |
1298 | * Compute the chunk number and the sector offset inside the chunk | |
1299 | */ | |
1300 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1301 | chunk_number = r_sector; | |
1302 | BUG_ON(r_sector != chunk_number); | |
1303 | ||
1304 | /* | |
1305 | * Compute the stripe number | |
1306 | */ | |
1307 | stripe = chunk_number / data_disks; | |
1308 | ||
1309 | /* | |
1310 | * Compute the data disk and parity disk indexes inside the stripe | |
1311 | */ | |
1312 | *dd_idx = chunk_number % data_disks; | |
1313 | ||
1314 | /* | |
1315 | * Select the parity disk based on the user selected algorithm. | |
1316 | */ | |
16a53ecc N |
1317 | switch(conf->level) { |
1318 | case 4: | |
1da177e4 | 1319 | *pd_idx = data_disks; |
16a53ecc N |
1320 | break; |
1321 | case 5: | |
1322 | switch (conf->algorithm) { | |
1da177e4 LT |
1323 | case ALGORITHM_LEFT_ASYMMETRIC: |
1324 | *pd_idx = data_disks - stripe % raid_disks; | |
1325 | if (*dd_idx >= *pd_idx) | |
1326 | (*dd_idx)++; | |
1327 | break; | |
1328 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1329 | *pd_idx = stripe % raid_disks; | |
1330 | if (*dd_idx >= *pd_idx) | |
1331 | (*dd_idx)++; | |
1332 | break; | |
1333 | case ALGORITHM_LEFT_SYMMETRIC: | |
1334 | *pd_idx = data_disks - stripe % raid_disks; | |
1335 | *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks; | |
1336 | break; | |
1337 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1338 | *pd_idx = stripe % raid_disks; | |
1339 | *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks; | |
1340 | break; | |
1341 | default: | |
14f8d26b | 1342 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
1da177e4 | 1343 | conf->algorithm); |
16a53ecc N |
1344 | } |
1345 | break; | |
1346 | case 6: | |
1347 | ||
1348 | /**** FIX THIS ****/ | |
1349 | switch (conf->algorithm) { | |
1350 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1351 | *pd_idx = raid_disks - 1 - (stripe % raid_disks); | |
1352 | if (*pd_idx == raid_disks-1) | |
1353 | (*dd_idx)++; /* Q D D D P */ | |
1354 | else if (*dd_idx >= *pd_idx) | |
1355 | (*dd_idx) += 2; /* D D P Q D */ | |
1356 | break; | |
1357 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1358 | *pd_idx = stripe % raid_disks; | |
1359 | if (*pd_idx == raid_disks-1) | |
1360 | (*dd_idx)++; /* Q D D D P */ | |
1361 | else if (*dd_idx >= *pd_idx) | |
1362 | (*dd_idx) += 2; /* D D P Q D */ | |
1363 | break; | |
1364 | case ALGORITHM_LEFT_SYMMETRIC: | |
1365 | *pd_idx = raid_disks - 1 - (stripe % raid_disks); | |
1366 | *dd_idx = (*pd_idx + 2 + *dd_idx) % raid_disks; | |
1367 | break; | |
1368 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1369 | *pd_idx = stripe % raid_disks; | |
1370 | *dd_idx = (*pd_idx + 2 + *dd_idx) % raid_disks; | |
1371 | break; | |
1372 | default: | |
1373 | printk (KERN_CRIT "raid6: unsupported algorithm %d\n", | |
1374 | conf->algorithm); | |
1375 | } | |
1376 | break; | |
1da177e4 LT |
1377 | } |
1378 | ||
1379 | /* | |
1380 | * Finally, compute the new sector number | |
1381 | */ | |
1382 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1383 | return new_sector; | |
1384 | } | |
1385 | ||
1386 | ||
1387 | static sector_t compute_blocknr(struct stripe_head *sh, int i) | |
1388 | { | |
1389 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
1390 | int raid_disks = sh->disks; |
1391 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1392 | sector_t new_sector = sh->sector, check; |
1393 | int sectors_per_chunk = conf->chunk_size >> 9; | |
1394 | sector_t stripe; | |
1395 | int chunk_offset; | |
1396 | int chunk_number, dummy1, dummy2, dd_idx = i; | |
1397 | sector_t r_sector; | |
1398 | ||
16a53ecc | 1399 | |
1da177e4 LT |
1400 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1401 | stripe = new_sector; | |
1402 | BUG_ON(new_sector != stripe); | |
1403 | ||
16a53ecc N |
1404 | if (i == sh->pd_idx) |
1405 | return 0; | |
1406 | switch(conf->level) { | |
1407 | case 4: break; | |
1408 | case 5: | |
1409 | switch (conf->algorithm) { | |
1da177e4 LT |
1410 | case ALGORITHM_LEFT_ASYMMETRIC: |
1411 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1412 | if (i > sh->pd_idx) | |
1413 | i--; | |
1414 | break; | |
1415 | case ALGORITHM_LEFT_SYMMETRIC: | |
1416 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1417 | if (i < sh->pd_idx) | |
1418 | i += raid_disks; | |
1419 | i -= (sh->pd_idx + 1); | |
1420 | break; | |
1421 | default: | |
14f8d26b | 1422 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
16a53ecc N |
1423 | conf->algorithm); |
1424 | } | |
1425 | break; | |
1426 | case 6: | |
16a53ecc N |
1427 | if (i == raid6_next_disk(sh->pd_idx, raid_disks)) |
1428 | return 0; /* It is the Q disk */ | |
1429 | switch (conf->algorithm) { | |
1430 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1431 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1432 | if (sh->pd_idx == raid_disks-1) | |
1433 | i--; /* Q D D D P */ | |
1434 | else if (i > sh->pd_idx) | |
1435 | i -= 2; /* D D P Q D */ | |
1436 | break; | |
1437 | case ALGORITHM_LEFT_SYMMETRIC: | |
1438 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1439 | if (sh->pd_idx == raid_disks-1) | |
1440 | i--; /* Q D D D P */ | |
1441 | else { | |
1442 | /* D D P Q D */ | |
1443 | if (i < sh->pd_idx) | |
1444 | i += raid_disks; | |
1445 | i -= (sh->pd_idx + 2); | |
1446 | } | |
1447 | break; | |
1448 | default: | |
1449 | printk (KERN_CRIT "raid6: unsupported algorithm %d\n", | |
1da177e4 | 1450 | conf->algorithm); |
16a53ecc N |
1451 | } |
1452 | break; | |
1da177e4 LT |
1453 | } |
1454 | ||
1455 | chunk_number = stripe * data_disks + i; | |
1456 | r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset; | |
1457 | ||
1458 | check = raid5_compute_sector (r_sector, raid_disks, data_disks, &dummy1, &dummy2, conf); | |
1459 | if (check != sh->sector || dummy1 != dd_idx || dummy2 != sh->pd_idx) { | |
14f8d26b | 1460 | printk(KERN_ERR "compute_blocknr: map not correct\n"); |
1da177e4 LT |
1461 | return 0; |
1462 | } | |
1463 | return r_sector; | |
1464 | } | |
1465 | ||
1466 | ||
1467 | ||
1468 | /* | |
16a53ecc N |
1469 | * Copy data between a page in the stripe cache, and one or more bion |
1470 | * The page could align with the middle of the bio, or there could be | |
1471 | * several bion, each with several bio_vecs, which cover part of the page | |
1472 | * Multiple bion are linked together on bi_next. There may be extras | |
1473 | * at the end of this list. We ignore them. | |
1da177e4 LT |
1474 | */ |
1475 | static void copy_data(int frombio, struct bio *bio, | |
1476 | struct page *page, | |
1477 | sector_t sector) | |
1478 | { | |
1479 | char *pa = page_address(page); | |
1480 | struct bio_vec *bvl; | |
1481 | int i; | |
1482 | int page_offset; | |
1483 | ||
1484 | if (bio->bi_sector >= sector) | |
1485 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
1486 | else | |
1487 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
1488 | bio_for_each_segment(bvl, bio, i) { | |
1489 | int len = bio_iovec_idx(bio,i)->bv_len; | |
1490 | int clen; | |
1491 | int b_offset = 0; | |
1492 | ||
1493 | if (page_offset < 0) { | |
1494 | b_offset = -page_offset; | |
1495 | page_offset += b_offset; | |
1496 | len -= b_offset; | |
1497 | } | |
1498 | ||
1499 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1500 | clen = STRIPE_SIZE - page_offset; | |
1501 | else clen = len; | |
16a53ecc | 1502 | |
1da177e4 LT |
1503 | if (clen > 0) { |
1504 | char *ba = __bio_kmap_atomic(bio, i, KM_USER0); | |
1505 | if (frombio) | |
1506 | memcpy(pa+page_offset, ba+b_offset, clen); | |
1507 | else | |
1508 | memcpy(ba+b_offset, pa+page_offset, clen); | |
1509 | __bio_kunmap_atomic(ba, KM_USER0); | |
1510 | } | |
1511 | if (clen < len) /* hit end of page */ | |
1512 | break; | |
1513 | page_offset += len; | |
1514 | } | |
1515 | } | |
1516 | ||
9bc89cd8 DW |
1517 | #define check_xor() do { \ |
1518 | if (count == MAX_XOR_BLOCKS) { \ | |
1519 | xor_blocks(count, STRIPE_SIZE, dest, ptr);\ | |
1520 | count = 0; \ | |
1521 | } \ | |
1da177e4 LT |
1522 | } while(0) |
1523 | ||
16a53ecc N |
1524 | static void compute_parity6(struct stripe_head *sh, int method) |
1525 | { | |
1526 | raid6_conf_t *conf = sh->raid_conf; | |
f416885e | 1527 | int i, pd_idx = sh->pd_idx, qd_idx, d0_idx, disks = sh->disks, count; |
16a53ecc N |
1528 | struct bio *chosen; |
1529 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
1530 | void *ptrs[disks]; | |
1531 | ||
1532 | qd_idx = raid6_next_disk(pd_idx, disks); | |
1533 | d0_idx = raid6_next_disk(qd_idx, disks); | |
1534 | ||
45b4233c | 1535 | pr_debug("compute_parity, stripe %llu, method %d\n", |
16a53ecc N |
1536 | (unsigned long long)sh->sector, method); |
1537 | ||
1538 | switch(method) { | |
1539 | case READ_MODIFY_WRITE: | |
1540 | BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */ | |
1541 | case RECONSTRUCT_WRITE: | |
1542 | for (i= disks; i-- ;) | |
1543 | if ( i != pd_idx && i != qd_idx && sh->dev[i].towrite ) { | |
1544 | chosen = sh->dev[i].towrite; | |
1545 | sh->dev[i].towrite = NULL; | |
1546 | ||
1547 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1548 | wake_up(&conf->wait_for_overlap); | |
1549 | ||
52e5f9d1 | 1550 | BUG_ON(sh->dev[i].written); |
16a53ecc N |
1551 | sh->dev[i].written = chosen; |
1552 | } | |
1553 | break; | |
1554 | case CHECK_PARITY: | |
1555 | BUG(); /* Not implemented yet */ | |
1556 | } | |
1557 | ||
1558 | for (i = disks; i--;) | |
1559 | if (sh->dev[i].written) { | |
1560 | sector_t sector = sh->dev[i].sector; | |
1561 | struct bio *wbi = sh->dev[i].written; | |
1562 | while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) { | |
1563 | copy_data(1, wbi, sh->dev[i].page, sector); | |
1564 | wbi = r5_next_bio(wbi, sector); | |
1565 | } | |
1566 | ||
1567 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
1568 | set_bit(R5_UPTODATE, &sh->dev[i].flags); | |
1569 | } | |
1570 | ||
1571 | // switch(method) { | |
1572 | // case RECONSTRUCT_WRITE: | |
1573 | // case CHECK_PARITY: | |
1574 | // case UPDATE_PARITY: | |
1575 | /* Note that unlike RAID-5, the ordering of the disks matters greatly. */ | |
1576 | /* FIX: Is this ordering of drives even remotely optimal? */ | |
1577 | count = 0; | |
1578 | i = d0_idx; | |
1579 | do { | |
1580 | ptrs[count++] = page_address(sh->dev[i].page); | |
1581 | if (count <= disks-2 && !test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
1582 | printk("block %d/%d not uptodate on parity calc\n", i,count); | |
1583 | i = raid6_next_disk(i, disks); | |
1584 | } while ( i != d0_idx ); | |
1585 | // break; | |
1586 | // } | |
1587 | ||
1588 | raid6_call.gen_syndrome(disks, STRIPE_SIZE, ptrs); | |
1589 | ||
1590 | switch(method) { | |
1591 | case RECONSTRUCT_WRITE: | |
1592 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1593 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
1594 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
1595 | set_bit(R5_LOCKED, &sh->dev[qd_idx].flags); | |
1596 | break; | |
1597 | case UPDATE_PARITY: | |
1598 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1599 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
1600 | break; | |
1601 | } | |
1602 | } | |
1603 | ||
1604 | ||
1605 | /* Compute one missing block */ | |
1606 | static void compute_block_1(struct stripe_head *sh, int dd_idx, int nozero) | |
1607 | { | |
f416885e | 1608 | int i, count, disks = sh->disks; |
9bc89cd8 | 1609 | void *ptr[MAX_XOR_BLOCKS], *dest, *p; |
16a53ecc N |
1610 | int pd_idx = sh->pd_idx; |
1611 | int qd_idx = raid6_next_disk(pd_idx, disks); | |
1612 | ||
45b4233c | 1613 | pr_debug("compute_block_1, stripe %llu, idx %d\n", |
16a53ecc N |
1614 | (unsigned long long)sh->sector, dd_idx); |
1615 | ||
1616 | if ( dd_idx == qd_idx ) { | |
1617 | /* We're actually computing the Q drive */ | |
1618 | compute_parity6(sh, UPDATE_PARITY); | |
1619 | } else { | |
9bc89cd8 DW |
1620 | dest = page_address(sh->dev[dd_idx].page); |
1621 | if (!nozero) memset(dest, 0, STRIPE_SIZE); | |
1622 | count = 0; | |
16a53ecc N |
1623 | for (i = disks ; i--; ) { |
1624 | if (i == dd_idx || i == qd_idx) | |
1625 | continue; | |
1626 | p = page_address(sh->dev[i].page); | |
1627 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
1628 | ptr[count++] = p; | |
1629 | else | |
1630 | printk("compute_block() %d, stripe %llu, %d" | |
1631 | " not present\n", dd_idx, | |
1632 | (unsigned long long)sh->sector, i); | |
1633 | ||
1634 | check_xor(); | |
1635 | } | |
9bc89cd8 DW |
1636 | if (count) |
1637 | xor_blocks(count, STRIPE_SIZE, dest, ptr); | |
16a53ecc N |
1638 | if (!nozero) set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); |
1639 | else clear_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); | |
1640 | } | |
1641 | } | |
1642 | ||
1643 | /* Compute two missing blocks */ | |
1644 | static void compute_block_2(struct stripe_head *sh, int dd_idx1, int dd_idx2) | |
1645 | { | |
f416885e | 1646 | int i, count, disks = sh->disks; |
16a53ecc N |
1647 | int pd_idx = sh->pd_idx; |
1648 | int qd_idx = raid6_next_disk(pd_idx, disks); | |
1649 | int d0_idx = raid6_next_disk(qd_idx, disks); | |
1650 | int faila, failb; | |
1651 | ||
1652 | /* faila and failb are disk numbers relative to d0_idx */ | |
1653 | /* pd_idx become disks-2 and qd_idx become disks-1 */ | |
1654 | faila = (dd_idx1 < d0_idx) ? dd_idx1+(disks-d0_idx) : dd_idx1-d0_idx; | |
1655 | failb = (dd_idx2 < d0_idx) ? dd_idx2+(disks-d0_idx) : dd_idx2-d0_idx; | |
1656 | ||
1657 | BUG_ON(faila == failb); | |
1658 | if ( failb < faila ) { int tmp = faila; faila = failb; failb = tmp; } | |
1659 | ||
45b4233c | 1660 | pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n", |
16a53ecc N |
1661 | (unsigned long long)sh->sector, dd_idx1, dd_idx2, faila, failb); |
1662 | ||
1663 | if ( failb == disks-1 ) { | |
1664 | /* Q disk is one of the missing disks */ | |
1665 | if ( faila == disks-2 ) { | |
1666 | /* Missing P+Q, just recompute */ | |
1667 | compute_parity6(sh, UPDATE_PARITY); | |
1668 | return; | |
1669 | } else { | |
1670 | /* We're missing D+Q; recompute D from P */ | |
1671 | compute_block_1(sh, (dd_idx1 == qd_idx) ? dd_idx2 : dd_idx1, 0); | |
1672 | compute_parity6(sh, UPDATE_PARITY); /* Is this necessary? */ | |
1673 | return; | |
1674 | } | |
1675 | } | |
1676 | ||
1677 | /* We're missing D+P or D+D; build pointer table */ | |
1678 | { | |
1679 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
1680 | void *ptrs[disks]; | |
1681 | ||
1682 | count = 0; | |
1683 | i = d0_idx; | |
1684 | do { | |
1685 | ptrs[count++] = page_address(sh->dev[i].page); | |
1686 | i = raid6_next_disk(i, disks); | |
1687 | if (i != dd_idx1 && i != dd_idx2 && | |
1688 | !test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
1689 | printk("compute_2 with missing block %d/%d\n", count, i); | |
1690 | } while ( i != d0_idx ); | |
1691 | ||
1692 | if ( failb == disks-2 ) { | |
1693 | /* We're missing D+P. */ | |
1694 | raid6_datap_recov(disks, STRIPE_SIZE, faila, ptrs); | |
1695 | } else { | |
1696 | /* We're missing D+D. */ | |
1697 | raid6_2data_recov(disks, STRIPE_SIZE, faila, failb, ptrs); | |
1698 | } | |
1699 | ||
1700 | /* Both the above update both missing blocks */ | |
1701 | set_bit(R5_UPTODATE, &sh->dev[dd_idx1].flags); | |
1702 | set_bit(R5_UPTODATE, &sh->dev[dd_idx2].flags); | |
1703 | } | |
1704 | } | |
1705 | ||
e33129d8 DW |
1706 | static int |
1707 | handle_write_operations5(struct stripe_head *sh, int rcw, int expand) | |
1708 | { | |
1709 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
1710 | int locked = 0; | |
1711 | ||
1712 | if (rcw) { | |
1713 | /* if we are not expanding this is a proper write request, and | |
1714 | * there will be bios with new data to be drained into the | |
1715 | * stripe cache | |
1716 | */ | |
1717 | if (!expand) { | |
1718 | set_bit(STRIPE_OP_BIODRAIN, &sh->ops.pending); | |
1719 | sh->ops.count++; | |
1720 | } | |
16a53ecc | 1721 | |
e33129d8 DW |
1722 | set_bit(STRIPE_OP_POSTXOR, &sh->ops.pending); |
1723 | sh->ops.count++; | |
1724 | ||
1725 | for (i = disks; i--; ) { | |
1726 | struct r5dev *dev = &sh->dev[i]; | |
1727 | ||
1728 | if (dev->towrite) { | |
1729 | set_bit(R5_LOCKED, &dev->flags); | |
1730 | if (!expand) | |
1731 | clear_bit(R5_UPTODATE, &dev->flags); | |
1732 | locked++; | |
1733 | } | |
1734 | } | |
8b3e6cdc DW |
1735 | if (locked + 1 == disks) |
1736 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) | |
1737 | atomic_inc(&sh->raid_conf->pending_full_writes); | |
e33129d8 DW |
1738 | } else { |
1739 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
1740 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
1741 | ||
1742 | set_bit(STRIPE_OP_PREXOR, &sh->ops.pending); | |
1743 | set_bit(STRIPE_OP_BIODRAIN, &sh->ops.pending); | |
1744 | set_bit(STRIPE_OP_POSTXOR, &sh->ops.pending); | |
1745 | ||
1746 | sh->ops.count += 3; | |
1747 | ||
1748 | for (i = disks; i--; ) { | |
1749 | struct r5dev *dev = &sh->dev[i]; | |
1750 | if (i == pd_idx) | |
1751 | continue; | |
1752 | ||
1753 | /* For a read-modify write there may be blocks that are | |
1754 | * locked for reading while others are ready to be | |
1755 | * written so we distinguish these blocks by the | |
1756 | * R5_Wantprexor bit | |
1757 | */ | |
1758 | if (dev->towrite && | |
1759 | (test_bit(R5_UPTODATE, &dev->flags) || | |
1760 | test_bit(R5_Wantcompute, &dev->flags))) { | |
1761 | set_bit(R5_Wantprexor, &dev->flags); | |
1762 | set_bit(R5_LOCKED, &dev->flags); | |
1763 | clear_bit(R5_UPTODATE, &dev->flags); | |
1764 | locked++; | |
1765 | } | |
1766 | } | |
1767 | } | |
1768 | ||
1769 | /* keep the parity disk locked while asynchronous operations | |
1770 | * are in flight | |
1771 | */ | |
1772 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
1773 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1774 | locked++; | |
1775 | ||
1776 | pr_debug("%s: stripe %llu locked: %d pending: %lx\n", | |
e46b272b | 1777 | __func__, (unsigned long long)sh->sector, |
e33129d8 DW |
1778 | locked, sh->ops.pending); |
1779 | ||
1780 | return locked; | |
1781 | } | |
16a53ecc | 1782 | |
1da177e4 LT |
1783 | /* |
1784 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 1785 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
1786 | * The bi_next chain must be in order. |
1787 | */ | |
1788 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
1789 | { | |
1790 | struct bio **bip; | |
1791 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 1792 | int firstwrite=0; |
1da177e4 | 1793 | |
45b4233c | 1794 | pr_debug("adding bh b#%llu to stripe s#%llu\n", |
1da177e4 LT |
1795 | (unsigned long long)bi->bi_sector, |
1796 | (unsigned long long)sh->sector); | |
1797 | ||
1798 | ||
1799 | spin_lock(&sh->lock); | |
1800 | spin_lock_irq(&conf->device_lock); | |
72626685 | 1801 | if (forwrite) { |
1da177e4 | 1802 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
1803 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
1804 | firstwrite = 1; | |
1805 | } else | |
1da177e4 LT |
1806 | bip = &sh->dev[dd_idx].toread; |
1807 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
1808 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
1809 | goto overlap; | |
1810 | bip = & (*bip)->bi_next; | |
1811 | } | |
1812 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
1813 | goto overlap; | |
1814 | ||
78bafebd | 1815 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
1816 | if (*bip) |
1817 | bi->bi_next = *bip; | |
1818 | *bip = bi; | |
1819 | bi->bi_phys_segments ++; | |
1820 | spin_unlock_irq(&conf->device_lock); | |
1821 | spin_unlock(&sh->lock); | |
1822 | ||
45b4233c | 1823 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", |
1da177e4 LT |
1824 | (unsigned long long)bi->bi_sector, |
1825 | (unsigned long long)sh->sector, dd_idx); | |
1826 | ||
72626685 | 1827 | if (conf->mddev->bitmap && firstwrite) { |
72626685 N |
1828 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
1829 | STRIPE_SECTORS, 0); | |
ae3c20cc | 1830 | sh->bm_seq = conf->seq_flush+1; |
72626685 N |
1831 | set_bit(STRIPE_BIT_DELAY, &sh->state); |
1832 | } | |
1833 | ||
1da177e4 LT |
1834 | if (forwrite) { |
1835 | /* check if page is covered */ | |
1836 | sector_t sector = sh->dev[dd_idx].sector; | |
1837 | for (bi=sh->dev[dd_idx].towrite; | |
1838 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
1839 | bi && bi->bi_sector <= sector; | |
1840 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
1841 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
1842 | sector = bi->bi_sector + (bi->bi_size>>9); | |
1843 | } | |
1844 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
1845 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
1846 | } | |
1847 | return 1; | |
1848 | ||
1849 | overlap: | |
1850 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
1851 | spin_unlock_irq(&conf->device_lock); | |
1852 | spin_unlock(&sh->lock); | |
1853 | return 0; | |
1854 | } | |
1855 | ||
29269553 N |
1856 | static void end_reshape(raid5_conf_t *conf); |
1857 | ||
16a53ecc N |
1858 | static int page_is_zero(struct page *p) |
1859 | { | |
1860 | char *a = page_address(p); | |
1861 | return ((*(u32*)a) == 0 && | |
1862 | memcmp(a, a+4, STRIPE_SIZE-4)==0); | |
1863 | } | |
1864 | ||
ccfcc3c1 N |
1865 | static int stripe_to_pdidx(sector_t stripe, raid5_conf_t *conf, int disks) |
1866 | { | |
1867 | int sectors_per_chunk = conf->chunk_size >> 9; | |
ccfcc3c1 | 1868 | int pd_idx, dd_idx; |
2d2063ce CQH |
1869 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
1870 | ||
b875e531 N |
1871 | raid5_compute_sector(stripe * (disks - conf->max_degraded) |
1872 | *sectors_per_chunk + chunk_offset, | |
1873 | disks, disks - conf->max_degraded, | |
1874 | &dd_idx, &pd_idx, conf); | |
ccfcc3c1 N |
1875 | return pd_idx; |
1876 | } | |
1877 | ||
a4456856 DW |
1878 | static void |
1879 | handle_requests_to_failed_array(raid5_conf_t *conf, struct stripe_head *sh, | |
1880 | struct stripe_head_state *s, int disks, | |
1881 | struct bio **return_bi) | |
1882 | { | |
1883 | int i; | |
1884 | for (i = disks; i--; ) { | |
1885 | struct bio *bi; | |
1886 | int bitmap_end = 0; | |
1887 | ||
1888 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
1889 | mdk_rdev_t *rdev; | |
1890 | rcu_read_lock(); | |
1891 | rdev = rcu_dereference(conf->disks[i].rdev); | |
1892 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
1893 | /* multiple read failures in one stripe */ | |
1894 | md_error(conf->mddev, rdev); | |
1895 | rcu_read_unlock(); | |
1896 | } | |
1897 | spin_lock_irq(&conf->device_lock); | |
1898 | /* fail all writes first */ | |
1899 | bi = sh->dev[i].towrite; | |
1900 | sh->dev[i].towrite = NULL; | |
1901 | if (bi) { | |
1902 | s->to_write--; | |
1903 | bitmap_end = 1; | |
1904 | } | |
1905 | ||
1906 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1907 | wake_up(&conf->wait_for_overlap); | |
1908 | ||
1909 | while (bi && bi->bi_sector < | |
1910 | sh->dev[i].sector + STRIPE_SECTORS) { | |
1911 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
1912 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1913 | if (--bi->bi_phys_segments == 0) { | |
1914 | md_write_end(conf->mddev); | |
1915 | bi->bi_next = *return_bi; | |
1916 | *return_bi = bi; | |
1917 | } | |
1918 | bi = nextbi; | |
1919 | } | |
1920 | /* and fail all 'written' */ | |
1921 | bi = sh->dev[i].written; | |
1922 | sh->dev[i].written = NULL; | |
1923 | if (bi) bitmap_end = 1; | |
1924 | while (bi && bi->bi_sector < | |
1925 | sh->dev[i].sector + STRIPE_SECTORS) { | |
1926 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
1927 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1928 | if (--bi->bi_phys_segments == 0) { | |
1929 | md_write_end(conf->mddev); | |
1930 | bi->bi_next = *return_bi; | |
1931 | *return_bi = bi; | |
1932 | } | |
1933 | bi = bi2; | |
1934 | } | |
1935 | ||
b5e98d65 DW |
1936 | /* fail any reads if this device is non-operational and |
1937 | * the data has not reached the cache yet. | |
1938 | */ | |
1939 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
1940 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
1941 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
1942 | bi = sh->dev[i].toread; |
1943 | sh->dev[i].toread = NULL; | |
1944 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1945 | wake_up(&conf->wait_for_overlap); | |
1946 | if (bi) s->to_read--; | |
1947 | while (bi && bi->bi_sector < | |
1948 | sh->dev[i].sector + STRIPE_SECTORS) { | |
1949 | struct bio *nextbi = | |
1950 | r5_next_bio(bi, sh->dev[i].sector); | |
1951 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1952 | if (--bi->bi_phys_segments == 0) { | |
1953 | bi->bi_next = *return_bi; | |
1954 | *return_bi = bi; | |
1955 | } | |
1956 | bi = nextbi; | |
1957 | } | |
1958 | } | |
1959 | spin_unlock_irq(&conf->device_lock); | |
1960 | if (bitmap_end) | |
1961 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
1962 | STRIPE_SECTORS, 0, 0); | |
1963 | } | |
1964 | ||
8b3e6cdc DW |
1965 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
1966 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
1967 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
1968 | } |
1969 | ||
f38e1219 DW |
1970 | /* __handle_issuing_new_read_requests5 - returns 0 if there are no more disks |
1971 | * to process | |
1972 | */ | |
1973 | static int __handle_issuing_new_read_requests5(struct stripe_head *sh, | |
1974 | struct stripe_head_state *s, int disk_idx, int disks) | |
1975 | { | |
1976 | struct r5dev *dev = &sh->dev[disk_idx]; | |
1977 | struct r5dev *failed_dev = &sh->dev[s->failed_num]; | |
1978 | ||
1979 | /* don't schedule compute operations or reads on the parity block while | |
1980 | * a check is in flight | |
1981 | */ | |
1982 | if ((disk_idx == sh->pd_idx) && | |
1983 | test_bit(STRIPE_OP_CHECK, &sh->ops.pending)) | |
1984 | return ~0; | |
1985 | ||
1986 | /* is the data in this block needed, and can we get it? */ | |
1987 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1988 | !test_bit(R5_UPTODATE, &dev->flags) && (dev->toread || | |
1989 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
1990 | s->syncing || s->expanding || (s->failed && | |
1991 | (failed_dev->toread || (failed_dev->towrite && | |
1992 | !test_bit(R5_OVERWRITE, &failed_dev->flags) | |
1993 | ))))) { | |
1994 | /* 1/ We would like to get this block, possibly by computing it, | |
1995 | * but we might not be able to. | |
1996 | * | |
1997 | * 2/ Since parity check operations potentially make the parity | |
1998 | * block !uptodate it will need to be refreshed before any | |
1999 | * compute operations on data disks are scheduled. | |
2000 | * | |
2001 | * 3/ We hold off parity block re-reads until check operations | |
2002 | * have quiesced. | |
2003 | */ | |
2004 | if ((s->uptodate == disks - 1) && | |
c337869d | 2005 | (s->failed && disk_idx == s->failed_num) && |
f38e1219 DW |
2006 | !test_bit(STRIPE_OP_CHECK, &sh->ops.pending)) { |
2007 | set_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending); | |
2008 | set_bit(R5_Wantcompute, &dev->flags); | |
2009 | sh->ops.target = disk_idx; | |
2010 | s->req_compute = 1; | |
2011 | sh->ops.count++; | |
2012 | /* Careful: from this point on 'uptodate' is in the eye | |
2013 | * of raid5_run_ops which services 'compute' operations | |
2014 | * before writes. R5_Wantcompute flags a block that will | |
2015 | * be R5_UPTODATE by the time it is needed for a | |
2016 | * subsequent operation. | |
2017 | */ | |
2018 | s->uptodate++; | |
2019 | return 0; /* uptodate + compute == disks */ | |
2020 | } else if ((s->uptodate < disks - 1) && | |
2021 | test_bit(R5_Insync, &dev->flags)) { | |
2022 | /* Note: we hold off compute operations while checks are | |
2023 | * in flight, but we still prefer 'compute' over 'read' | |
2024 | * hence we only read if (uptodate < * disks-1) | |
2025 | */ | |
2026 | set_bit(R5_LOCKED, &dev->flags); | |
2027 | set_bit(R5_Wantread, &dev->flags); | |
2028 | if (!test_and_set_bit(STRIPE_OP_IO, &sh->ops.pending)) | |
2029 | sh->ops.count++; | |
2030 | s->locked++; | |
2031 | pr_debug("Reading block %d (sync=%d)\n", disk_idx, | |
2032 | s->syncing); | |
2033 | } | |
2034 | } | |
2035 | ||
2036 | return ~0; | |
2037 | } | |
2038 | ||
a4456856 DW |
2039 | static void handle_issuing_new_read_requests5(struct stripe_head *sh, |
2040 | struct stripe_head_state *s, int disks) | |
2041 | { | |
2042 | int i; | |
f38e1219 DW |
2043 | |
2044 | /* Clear completed compute operations. Parity recovery | |
2045 | * (STRIPE_OP_MOD_REPAIR_PD) implies a write-back which is handled | |
2046 | * later on in this routine | |
2047 | */ | |
2048 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.complete) && | |
2049 | !test_bit(STRIPE_OP_MOD_REPAIR_PD, &sh->ops.pending)) { | |
2050 | clear_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.complete); | |
2051 | clear_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.ack); | |
2052 | clear_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending); | |
2053 | } | |
2054 | ||
2055 | /* look for blocks to read/compute, skip this if a compute | |
2056 | * is already in flight, or if the stripe contents are in the | |
2057 | * midst of changing due to a write | |
2058 | */ | |
2059 | if (!test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending) && | |
2060 | !test_bit(STRIPE_OP_PREXOR, &sh->ops.pending) && | |
2061 | !test_bit(STRIPE_OP_POSTXOR, &sh->ops.pending)) { | |
2062 | for (i = disks; i--; ) | |
2063 | if (__handle_issuing_new_read_requests5( | |
2064 | sh, s, i, disks) == 0) | |
2065 | break; | |
a4456856 DW |
2066 | } |
2067 | set_bit(STRIPE_HANDLE, &sh->state); | |
2068 | } | |
2069 | ||
2070 | static void handle_issuing_new_read_requests6(struct stripe_head *sh, | |
2071 | struct stripe_head_state *s, struct r6_state *r6s, | |
2072 | int disks) | |
2073 | { | |
2074 | int i; | |
2075 | for (i = disks; i--; ) { | |
2076 | struct r5dev *dev = &sh->dev[i]; | |
2077 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2078 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2079 | (dev->toread || (dev->towrite && | |
2080 | !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2081 | s->syncing || s->expanding || | |
2082 | (s->failed >= 1 && | |
2083 | (sh->dev[r6s->failed_num[0]].toread || | |
2084 | s->to_write)) || | |
2085 | (s->failed >= 2 && | |
2086 | (sh->dev[r6s->failed_num[1]].toread || | |
2087 | s->to_write)))) { | |
2088 | /* we would like to get this block, possibly | |
2089 | * by computing it, but we might not be able to | |
2090 | */ | |
c337869d DW |
2091 | if ((s->uptodate == disks - 1) && |
2092 | (s->failed && (i == r6s->failed_num[0] || | |
2093 | i == r6s->failed_num[1]))) { | |
45b4233c | 2094 | pr_debug("Computing stripe %llu block %d\n", |
a4456856 DW |
2095 | (unsigned long long)sh->sector, i); |
2096 | compute_block_1(sh, i, 0); | |
2097 | s->uptodate++; | |
2098 | } else if ( s->uptodate == disks-2 && s->failed >= 2 ) { | |
2099 | /* Computing 2-failure is *very* expensive; only | |
2100 | * do it if failed >= 2 | |
2101 | */ | |
2102 | int other; | |
2103 | for (other = disks; other--; ) { | |
2104 | if (other == i) | |
2105 | continue; | |
2106 | if (!test_bit(R5_UPTODATE, | |
2107 | &sh->dev[other].flags)) | |
2108 | break; | |
2109 | } | |
2110 | BUG_ON(other < 0); | |
45b4233c | 2111 | pr_debug("Computing stripe %llu blocks %d,%d\n", |
a4456856 DW |
2112 | (unsigned long long)sh->sector, |
2113 | i, other); | |
2114 | compute_block_2(sh, i, other); | |
2115 | s->uptodate += 2; | |
2116 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2117 | set_bit(R5_LOCKED, &dev->flags); | |
2118 | set_bit(R5_Wantread, &dev->flags); | |
2119 | s->locked++; | |
45b4233c | 2120 | pr_debug("Reading block %d (sync=%d)\n", |
a4456856 DW |
2121 | i, s->syncing); |
2122 | } | |
2123 | } | |
2124 | } | |
2125 | set_bit(STRIPE_HANDLE, &sh->state); | |
2126 | } | |
2127 | ||
2128 | ||
2129 | /* handle_completed_write_requests | |
2130 | * any written block on an uptodate or failed drive can be returned. | |
2131 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2132 | * never LOCKED, so we don't need to test 'failed' directly. | |
2133 | */ | |
2134 | static void handle_completed_write_requests(raid5_conf_t *conf, | |
2135 | struct stripe_head *sh, int disks, struct bio **return_bi) | |
2136 | { | |
2137 | int i; | |
2138 | struct r5dev *dev; | |
2139 | ||
2140 | for (i = disks; i--; ) | |
2141 | if (sh->dev[i].written) { | |
2142 | dev = &sh->dev[i]; | |
2143 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2144 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2145 | /* We can return any write requests */ | |
2146 | struct bio *wbi, *wbi2; | |
2147 | int bitmap_end = 0; | |
45b4233c | 2148 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2149 | spin_lock_irq(&conf->device_lock); |
2150 | wbi = dev->written; | |
2151 | dev->written = NULL; | |
2152 | while (wbi && wbi->bi_sector < | |
2153 | dev->sector + STRIPE_SECTORS) { | |
2154 | wbi2 = r5_next_bio(wbi, dev->sector); | |
2155 | if (--wbi->bi_phys_segments == 0) { | |
2156 | md_write_end(conf->mddev); | |
2157 | wbi->bi_next = *return_bi; | |
2158 | *return_bi = wbi; | |
2159 | } | |
2160 | wbi = wbi2; | |
2161 | } | |
2162 | if (dev->towrite == NULL) | |
2163 | bitmap_end = 1; | |
2164 | spin_unlock_irq(&conf->device_lock); | |
2165 | if (bitmap_end) | |
2166 | bitmap_endwrite(conf->mddev->bitmap, | |
2167 | sh->sector, | |
2168 | STRIPE_SECTORS, | |
2169 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2170 | 0); | |
2171 | } | |
2172 | } | |
8b3e6cdc DW |
2173 | |
2174 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2175 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2176 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2177 | } |
2178 | ||
2179 | static void handle_issuing_new_write_requests5(raid5_conf_t *conf, | |
2180 | struct stripe_head *sh, struct stripe_head_state *s, int disks) | |
2181 | { | |
2182 | int rmw = 0, rcw = 0, i; | |
2183 | for (i = disks; i--; ) { | |
2184 | /* would I have to read this buffer for read_modify_write */ | |
2185 | struct r5dev *dev = &sh->dev[i]; | |
2186 | if ((dev->towrite || i == sh->pd_idx) && | |
2187 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2188 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2189 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2190 | if (test_bit(R5_Insync, &dev->flags)) |
2191 | rmw++; | |
2192 | else | |
2193 | rmw += 2*disks; /* cannot read it */ | |
2194 | } | |
2195 | /* Would I have to read this buffer for reconstruct_write */ | |
2196 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2197 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2198 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2199 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2200 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2201 | else |
2202 | rcw += 2*disks; | |
2203 | } | |
2204 | } | |
45b4233c | 2205 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2206 | (unsigned long long)sh->sector, rmw, rcw); |
2207 | set_bit(STRIPE_HANDLE, &sh->state); | |
2208 | if (rmw < rcw && rmw > 0) | |
2209 | /* prefer read-modify-write, but need to get some data */ | |
2210 | for (i = disks; i--; ) { | |
2211 | struct r5dev *dev = &sh->dev[i]; | |
2212 | if ((dev->towrite || i == sh->pd_idx) && | |
2213 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2214 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2215 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2216 | test_bit(R5_Insync, &dev->flags)) { |
2217 | if ( | |
2218 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2219 | pr_debug("Read_old block " |
a4456856 DW |
2220 | "%d for r-m-w\n", i); |
2221 | set_bit(R5_LOCKED, &dev->flags); | |
2222 | set_bit(R5_Wantread, &dev->flags); | |
830ea016 DW |
2223 | if (!test_and_set_bit( |
2224 | STRIPE_OP_IO, &sh->ops.pending)) | |
2225 | sh->ops.count++; | |
a4456856 DW |
2226 | s->locked++; |
2227 | } else { | |
2228 | set_bit(STRIPE_DELAYED, &sh->state); | |
2229 | set_bit(STRIPE_HANDLE, &sh->state); | |
2230 | } | |
2231 | } | |
2232 | } | |
2233 | if (rcw <= rmw && rcw > 0) | |
2234 | /* want reconstruct write, but need to get some data */ | |
2235 | for (i = disks; i--; ) { | |
2236 | struct r5dev *dev = &sh->dev[i]; | |
2237 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2238 | i != sh->pd_idx && | |
2239 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2240 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2241 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2242 | test_bit(R5_Insync, &dev->flags)) { |
2243 | if ( | |
2244 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2245 | pr_debug("Read_old block " |
a4456856 DW |
2246 | "%d for Reconstruct\n", i); |
2247 | set_bit(R5_LOCKED, &dev->flags); | |
2248 | set_bit(R5_Wantread, &dev->flags); | |
830ea016 DW |
2249 | if (!test_and_set_bit( |
2250 | STRIPE_OP_IO, &sh->ops.pending)) | |
2251 | sh->ops.count++; | |
a4456856 DW |
2252 | s->locked++; |
2253 | } else { | |
2254 | set_bit(STRIPE_DELAYED, &sh->state); | |
2255 | set_bit(STRIPE_HANDLE, &sh->state); | |
2256 | } | |
2257 | } | |
2258 | } | |
2259 | /* now if nothing is locked, and if we have enough data, | |
2260 | * we can start a write request | |
2261 | */ | |
f38e1219 DW |
2262 | /* since handle_stripe can be called at any time we need to handle the |
2263 | * case where a compute block operation has been submitted and then a | |
2264 | * subsequent call wants to start a write request. raid5_run_ops only | |
2265 | * handles the case where compute block and postxor are requested | |
2266 | * simultaneously. If this is not the case then new writes need to be | |
2267 | * held off until the compute completes. | |
2268 | */ | |
2269 | if ((s->req_compute || | |
2270 | !test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending)) && | |
2271 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2272 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
e33129d8 | 2273 | s->locked += handle_write_operations5(sh, rcw == 0, 0); |
a4456856 DW |
2274 | } |
2275 | ||
2276 | static void handle_issuing_new_write_requests6(raid5_conf_t *conf, | |
2277 | struct stripe_head *sh, struct stripe_head_state *s, | |
2278 | struct r6_state *r6s, int disks) | |
2279 | { | |
2280 | int rcw = 0, must_compute = 0, pd_idx = sh->pd_idx, i; | |
2281 | int qd_idx = r6s->qd_idx; | |
2282 | for (i = disks; i--; ) { | |
2283 | struct r5dev *dev = &sh->dev[i]; | |
2284 | /* Would I have to read this buffer for reconstruct_write */ | |
2285 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
2286 | && i != pd_idx && i != qd_idx | |
2287 | && (!test_bit(R5_LOCKED, &dev->flags) | |
2288 | ) && | |
2289 | !test_bit(R5_UPTODATE, &dev->flags)) { | |
2290 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
2291 | else { | |
45b4233c | 2292 | pr_debug("raid6: must_compute: " |
a4456856 DW |
2293 | "disk %d flags=%#lx\n", i, dev->flags); |
2294 | must_compute++; | |
2295 | } | |
2296 | } | |
2297 | } | |
45b4233c | 2298 | pr_debug("for sector %llu, rcw=%d, must_compute=%d\n", |
a4456856 DW |
2299 | (unsigned long long)sh->sector, rcw, must_compute); |
2300 | set_bit(STRIPE_HANDLE, &sh->state); | |
2301 | ||
2302 | if (rcw > 0) | |
2303 | /* want reconstruct write, but need to get some data */ | |
2304 | for (i = disks; i--; ) { | |
2305 | struct r5dev *dev = &sh->dev[i]; | |
2306 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
2307 | && !(s->failed == 0 && (i == pd_idx || i == qd_idx)) | |
2308 | && !test_bit(R5_LOCKED, &dev->flags) && | |
2309 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2310 | test_bit(R5_Insync, &dev->flags)) { | |
2311 | if ( | |
2312 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2313 | pr_debug("Read_old stripe %llu " |
a4456856 DW |
2314 | "block %d for Reconstruct\n", |
2315 | (unsigned long long)sh->sector, i); | |
2316 | set_bit(R5_LOCKED, &dev->flags); | |
2317 | set_bit(R5_Wantread, &dev->flags); | |
2318 | s->locked++; | |
2319 | } else { | |
45b4233c | 2320 | pr_debug("Request delayed stripe %llu " |
a4456856 DW |
2321 | "block %d for Reconstruct\n", |
2322 | (unsigned long long)sh->sector, i); | |
2323 | set_bit(STRIPE_DELAYED, &sh->state); | |
2324 | set_bit(STRIPE_HANDLE, &sh->state); | |
2325 | } | |
2326 | } | |
2327 | } | |
2328 | /* now if nothing is locked, and if we have enough data, we can start a | |
2329 | * write request | |
2330 | */ | |
2331 | if (s->locked == 0 && rcw == 0 && | |
2332 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { | |
2333 | if (must_compute > 0) { | |
2334 | /* We have failed blocks and need to compute them */ | |
2335 | switch (s->failed) { | |
2336 | case 0: | |
2337 | BUG(); | |
2338 | case 1: | |
2339 | compute_block_1(sh, r6s->failed_num[0], 0); | |
2340 | break; | |
2341 | case 2: | |
2342 | compute_block_2(sh, r6s->failed_num[0], | |
2343 | r6s->failed_num[1]); | |
2344 | break; | |
2345 | default: /* This request should have been failed? */ | |
2346 | BUG(); | |
2347 | } | |
2348 | } | |
2349 | ||
45b4233c | 2350 | pr_debug("Computing parity for stripe %llu\n", |
a4456856 DW |
2351 | (unsigned long long)sh->sector); |
2352 | compute_parity6(sh, RECONSTRUCT_WRITE); | |
2353 | /* now every locked buffer is ready to be written */ | |
2354 | for (i = disks; i--; ) | |
2355 | if (test_bit(R5_LOCKED, &sh->dev[i].flags)) { | |
45b4233c | 2356 | pr_debug("Writing stripe %llu block %d\n", |
a4456856 DW |
2357 | (unsigned long long)sh->sector, i); |
2358 | s->locked++; | |
2359 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
2360 | } | |
8b3e6cdc DW |
2361 | if (s->locked == disks) |
2362 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2363 | atomic_inc(&conf->pending_full_writes); | |
a4456856 DW |
2364 | /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */ |
2365 | set_bit(STRIPE_INSYNC, &sh->state); | |
2366 | ||
2367 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2368 | atomic_dec(&conf->preread_active_stripes); | |
2369 | if (atomic_read(&conf->preread_active_stripes) < | |
2370 | IO_THRESHOLD) | |
2371 | md_wakeup_thread(conf->mddev->thread); | |
2372 | } | |
2373 | } | |
2374 | } | |
2375 | ||
2376 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, | |
2377 | struct stripe_head_state *s, int disks) | |
2378 | { | |
bd2ab670 DW |
2379 | int canceled_check = 0; |
2380 | ||
a4456856 | 2381 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2382 | |
bd2ab670 DW |
2383 | /* complete a check operation */ |
2384 | if (test_and_clear_bit(STRIPE_OP_CHECK, &sh->ops.complete)) { | |
c8894419 DW |
2385 | clear_bit(STRIPE_OP_CHECK, &sh->ops.ack); |
2386 | clear_bit(STRIPE_OP_CHECK, &sh->ops.pending); | |
bd2ab670 | 2387 | if (s->failed == 0) { |
e89f8962 DW |
2388 | if (sh->ops.zero_sum_result == 0) |
2389 | /* parity is correct (on disc, | |
2390 | * not in buffer any more) | |
2391 | */ | |
a4456856 DW |
2392 | set_bit(STRIPE_INSYNC, &sh->state); |
2393 | else { | |
e89f8962 DW |
2394 | conf->mddev->resync_mismatches += |
2395 | STRIPE_SECTORS; | |
2396 | if (test_bit( | |
2397 | MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2398 | /* don't try to repair!! */ | |
2399 | set_bit(STRIPE_INSYNC, &sh->state); | |
2400 | else { | |
2401 | set_bit(STRIPE_OP_COMPUTE_BLK, | |
2402 | &sh->ops.pending); | |
2403 | set_bit(STRIPE_OP_MOD_REPAIR_PD, | |
2404 | &sh->ops.pending); | |
2405 | set_bit(R5_Wantcompute, | |
2406 | &sh->dev[sh->pd_idx].flags); | |
2407 | sh->ops.target = sh->pd_idx; | |
2408 | sh->ops.count++; | |
2409 | s->uptodate++; | |
2410 | } | |
a4456856 | 2411 | } |
bd2ab670 DW |
2412 | } else |
2413 | canceled_check = 1; /* STRIPE_INSYNC is not set */ | |
a4456856 | 2414 | } |
e89f8962 | 2415 | |
bd2ab670 DW |
2416 | /* start a new check operation if there are no failures, the stripe is |
2417 | * not insync, and a repair is not in flight | |
2418 | */ | |
2419 | if (s->failed == 0 && | |
2420 | !test_bit(STRIPE_INSYNC, &sh->state) && | |
2421 | !test_bit(STRIPE_OP_MOD_REPAIR_PD, &sh->ops.pending)) { | |
2422 | if (!test_and_set_bit(STRIPE_OP_CHECK, &sh->ops.pending)) { | |
2423 | BUG_ON(s->uptodate != disks); | |
2424 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
2425 | sh->ops.count++; | |
2426 | s->uptodate--; | |
2427 | } | |
2428 | } | |
2429 | ||
c8894419 DW |
2430 | /* check if we can clear a parity disk reconstruct */ |
2431 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.complete) && | |
2432 | test_bit(STRIPE_OP_MOD_REPAIR_PD, &sh->ops.pending)) { | |
2433 | ||
2434 | clear_bit(STRIPE_OP_MOD_REPAIR_PD, &sh->ops.pending); | |
2435 | clear_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.complete); | |
2436 | clear_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.ack); | |
2437 | clear_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending); | |
2438 | } | |
2439 | ||
2440 | ||
e89f8962 | 2441 | /* Wait for check parity and compute block operations to complete |
bd2ab670 DW |
2442 | * before write-back. If a failure occurred while the check operation |
2443 | * was in flight we need to cycle this stripe through handle_stripe | |
2444 | * since the parity block may not be uptodate | |
e89f8962 | 2445 | */ |
bd2ab670 DW |
2446 | if (!canceled_check && !test_bit(STRIPE_INSYNC, &sh->state) && |
2447 | !test_bit(STRIPE_OP_CHECK, &sh->ops.pending) && | |
2448 | !test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending)) { | |
a4456856 DW |
2449 | struct r5dev *dev; |
2450 | /* either failed parity check, or recovery is happening */ | |
2451 | if (s->failed == 0) | |
2452 | s->failed_num = sh->pd_idx; | |
2453 | dev = &sh->dev[s->failed_num]; | |
2454 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); | |
2455 | BUG_ON(s->uptodate != disks); | |
2456 | ||
2457 | set_bit(R5_LOCKED, &dev->flags); | |
2458 | set_bit(R5_Wantwrite, &dev->flags); | |
830ea016 DW |
2459 | if (!test_and_set_bit(STRIPE_OP_IO, &sh->ops.pending)) |
2460 | sh->ops.count++; | |
2461 | ||
a4456856 DW |
2462 | clear_bit(STRIPE_DEGRADED, &sh->state); |
2463 | s->locked++; | |
2464 | set_bit(STRIPE_INSYNC, &sh->state); | |
2465 | } | |
2466 | } | |
2467 | ||
2468 | ||
2469 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
2470 | struct stripe_head_state *s, | |
2471 | struct r6_state *r6s, struct page *tmp_page, | |
2472 | int disks) | |
2473 | { | |
2474 | int update_p = 0, update_q = 0; | |
2475 | struct r5dev *dev; | |
2476 | int pd_idx = sh->pd_idx; | |
2477 | int qd_idx = r6s->qd_idx; | |
2478 | ||
2479 | set_bit(STRIPE_HANDLE, &sh->state); | |
2480 | ||
2481 | BUG_ON(s->failed > 2); | |
2482 | BUG_ON(s->uptodate < disks); | |
2483 | /* Want to check and possibly repair P and Q. | |
2484 | * However there could be one 'failed' device, in which | |
2485 | * case we can only check one of them, possibly using the | |
2486 | * other to generate missing data | |
2487 | */ | |
2488 | ||
2489 | /* If !tmp_page, we cannot do the calculations, | |
2490 | * but as we have set STRIPE_HANDLE, we will soon be called | |
2491 | * by stripe_handle with a tmp_page - just wait until then. | |
2492 | */ | |
2493 | if (tmp_page) { | |
2494 | if (s->failed == r6s->q_failed) { | |
2495 | /* The only possible failed device holds 'Q', so it | |
2496 | * makes sense to check P (If anything else were failed, | |
2497 | * we would have used P to recreate it). | |
2498 | */ | |
2499 | compute_block_1(sh, pd_idx, 1); | |
2500 | if (!page_is_zero(sh->dev[pd_idx].page)) { | |
2501 | compute_block_1(sh, pd_idx, 0); | |
2502 | update_p = 1; | |
2503 | } | |
2504 | } | |
2505 | if (!r6s->q_failed && s->failed < 2) { | |
2506 | /* q is not failed, and we didn't use it to generate | |
2507 | * anything, so it makes sense to check it | |
2508 | */ | |
2509 | memcpy(page_address(tmp_page), | |
2510 | page_address(sh->dev[qd_idx].page), | |
2511 | STRIPE_SIZE); | |
2512 | compute_parity6(sh, UPDATE_PARITY); | |
2513 | if (memcmp(page_address(tmp_page), | |
2514 | page_address(sh->dev[qd_idx].page), | |
2515 | STRIPE_SIZE) != 0) { | |
2516 | clear_bit(STRIPE_INSYNC, &sh->state); | |
2517 | update_q = 1; | |
2518 | } | |
2519 | } | |
2520 | if (update_p || update_q) { | |
2521 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2522 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2523 | /* don't try to repair!! */ | |
2524 | update_p = update_q = 0; | |
2525 | } | |
2526 | ||
2527 | /* now write out any block on a failed drive, | |
2528 | * or P or Q if they need it | |
2529 | */ | |
2530 | ||
2531 | if (s->failed == 2) { | |
2532 | dev = &sh->dev[r6s->failed_num[1]]; | |
2533 | s->locked++; | |
2534 | set_bit(R5_LOCKED, &dev->flags); | |
2535 | set_bit(R5_Wantwrite, &dev->flags); | |
2536 | } | |
2537 | if (s->failed >= 1) { | |
2538 | dev = &sh->dev[r6s->failed_num[0]]; | |
2539 | s->locked++; | |
2540 | set_bit(R5_LOCKED, &dev->flags); | |
2541 | set_bit(R5_Wantwrite, &dev->flags); | |
2542 | } | |
2543 | ||
2544 | if (update_p) { | |
2545 | dev = &sh->dev[pd_idx]; | |
2546 | s->locked++; | |
2547 | set_bit(R5_LOCKED, &dev->flags); | |
2548 | set_bit(R5_Wantwrite, &dev->flags); | |
2549 | } | |
2550 | if (update_q) { | |
2551 | dev = &sh->dev[qd_idx]; | |
2552 | s->locked++; | |
2553 | set_bit(R5_LOCKED, &dev->flags); | |
2554 | set_bit(R5_Wantwrite, &dev->flags); | |
2555 | } | |
2556 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2557 | ||
2558 | set_bit(STRIPE_INSYNC, &sh->state); | |
2559 | } | |
2560 | } | |
2561 | ||
2562 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh, | |
2563 | struct r6_state *r6s) | |
2564 | { | |
2565 | int i; | |
2566 | ||
2567 | /* We have read all the blocks in this stripe and now we need to | |
2568 | * copy some of them into a target stripe for expand. | |
2569 | */ | |
f0a50d37 | 2570 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2571 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2572 | for (i = 0; i < sh->disks; i++) | |
a2e08551 | 2573 | if (i != sh->pd_idx && (!r6s || i != r6s->qd_idx)) { |
a4456856 DW |
2574 | int dd_idx, pd_idx, j; |
2575 | struct stripe_head *sh2; | |
2576 | ||
2577 | sector_t bn = compute_blocknr(sh, i); | |
2578 | sector_t s = raid5_compute_sector(bn, conf->raid_disks, | |
2579 | conf->raid_disks - | |
2580 | conf->max_degraded, &dd_idx, | |
2581 | &pd_idx, conf); | |
2582 | sh2 = get_active_stripe(conf, s, conf->raid_disks, | |
2583 | pd_idx, 1); | |
2584 | if (sh2 == NULL) | |
2585 | /* so far only the early blocks of this stripe | |
2586 | * have been requested. When later blocks | |
2587 | * get requested, we will try again | |
2588 | */ | |
2589 | continue; | |
2590 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
2591 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
2592 | /* must have already done this block */ | |
2593 | release_stripe(sh2); | |
2594 | continue; | |
2595 | } | |
f0a50d37 DW |
2596 | |
2597 | /* place all the copies on one channel */ | |
2598 | tx = async_memcpy(sh2->dev[dd_idx].page, | |
2599 | sh->dev[i].page, 0, 0, STRIPE_SIZE, | |
2600 | ASYNC_TX_DEP_ACK, tx, NULL, NULL); | |
2601 | ||
a4456856 DW |
2602 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
2603 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
2604 | for (j = 0; j < conf->raid_disks; j++) | |
2605 | if (j != sh2->pd_idx && | |
a2e08551 N |
2606 | (!r6s || j != raid6_next_disk(sh2->pd_idx, |
2607 | sh2->disks)) && | |
a4456856 DW |
2608 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
2609 | break; | |
2610 | if (j == conf->raid_disks) { | |
2611 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
2612 | set_bit(STRIPE_HANDLE, &sh2->state); | |
2613 | } | |
2614 | release_stripe(sh2); | |
f0a50d37 | 2615 | |
a4456856 | 2616 | } |
a2e08551 N |
2617 | /* done submitting copies, wait for them to complete */ |
2618 | if (tx) { | |
2619 | async_tx_ack(tx); | |
2620 | dma_wait_for_async_tx(tx); | |
2621 | } | |
a4456856 | 2622 | } |
1da177e4 | 2623 | |
6bfe0b49 | 2624 | |
1da177e4 LT |
2625 | /* |
2626 | * handle_stripe - do things to a stripe. | |
2627 | * | |
2628 | * We lock the stripe and then examine the state of various bits | |
2629 | * to see what needs to be done. | |
2630 | * Possible results: | |
2631 | * return some read request which now have data | |
2632 | * return some write requests which are safely on disc | |
2633 | * schedule a read on some buffers | |
2634 | * schedule a write of some buffers | |
2635 | * return confirmation of parity correctness | |
2636 | * | |
1da177e4 LT |
2637 | * buffers are taken off read_list or write_list, and bh_cache buffers |
2638 | * get BH_Lock set before the stripe lock is released. | |
2639 | * | |
2640 | */ | |
a4456856 | 2641 | |
16a53ecc | 2642 | static void handle_stripe5(struct stripe_head *sh) |
1da177e4 LT |
2643 | { |
2644 | raid5_conf_t *conf = sh->raid_conf; | |
a4456856 DW |
2645 | int disks = sh->disks, i; |
2646 | struct bio *return_bi = NULL; | |
2647 | struct stripe_head_state s; | |
1da177e4 | 2648 | struct r5dev *dev; |
d84e0f10 | 2649 | unsigned long pending = 0; |
6bfe0b49 | 2650 | mdk_rdev_t *blocked_rdev = NULL; |
e0a115e5 | 2651 | int prexor; |
1da177e4 | 2652 | |
a4456856 | 2653 | memset(&s, 0, sizeof(s)); |
d84e0f10 DW |
2654 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d " |
2655 | "ops=%lx:%lx:%lx\n", (unsigned long long)sh->sector, sh->state, | |
2656 | atomic_read(&sh->count), sh->pd_idx, | |
2657 | sh->ops.pending, sh->ops.ack, sh->ops.complete); | |
1da177e4 LT |
2658 | |
2659 | spin_lock(&sh->lock); | |
2660 | clear_bit(STRIPE_HANDLE, &sh->state); | |
2661 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2662 | ||
a4456856 DW |
2663 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
2664 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2665 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
1da177e4 LT |
2666 | /* Now to look around and see what can be done */ |
2667 | ||
def6ae26 NB |
2668 | /* clean-up completed biofill operations */ |
2669 | if (test_bit(STRIPE_OP_BIOFILL, &sh->ops.complete)) { | |
2670 | clear_bit(STRIPE_OP_BIOFILL, &sh->ops.pending); | |
2671 | clear_bit(STRIPE_OP_BIOFILL, &sh->ops.ack); | |
2672 | clear_bit(STRIPE_OP_BIOFILL, &sh->ops.complete); | |
2673 | } | |
2674 | ||
9910f16a | 2675 | rcu_read_lock(); |
1da177e4 LT |
2676 | for (i=disks; i--; ) { |
2677 | mdk_rdev_t *rdev; | |
a4456856 | 2678 | struct r5dev *dev = &sh->dev[i]; |
1da177e4 | 2679 | clear_bit(R5_Insync, &dev->flags); |
1da177e4 | 2680 | |
b5e98d65 DW |
2681 | pr_debug("check %d: state 0x%lx toread %p read %p write %p " |
2682 | "written %p\n", i, dev->flags, dev->toread, dev->read, | |
2683 | dev->towrite, dev->written); | |
2684 | ||
2685 | /* maybe we can request a biofill operation | |
2686 | * | |
2687 | * new wantfill requests are only permitted while | |
2688 | * STRIPE_OP_BIOFILL is clear | |
2689 | */ | |
2690 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
2691 | !test_bit(STRIPE_OP_BIOFILL, &sh->ops.pending)) | |
2692 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 LT |
2693 | |
2694 | /* now count some things */ | |
a4456856 DW |
2695 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
2696 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
f38e1219 | 2697 | if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++; |
1da177e4 | 2698 | |
b5e98d65 DW |
2699 | if (test_bit(R5_Wantfill, &dev->flags)) |
2700 | s.to_fill++; | |
2701 | else if (dev->toread) | |
a4456856 | 2702 | s.to_read++; |
1da177e4 | 2703 | if (dev->towrite) { |
a4456856 | 2704 | s.to_write++; |
1da177e4 | 2705 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 2706 | s.non_overwrite++; |
1da177e4 | 2707 | } |
a4456856 DW |
2708 | if (dev->written) |
2709 | s.written++; | |
9910f16a | 2710 | rdev = rcu_dereference(conf->disks[i].rdev); |
6bfe0b49 DW |
2711 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
2712 | blocked_rdev = rdev; | |
2713 | atomic_inc(&rdev->nr_pending); | |
2714 | break; | |
2715 | } | |
b2d444d7 | 2716 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
14f8d26b | 2717 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
2718 | clear_bit(R5_ReadError, &dev->flags); |
2719 | clear_bit(R5_ReWrite, &dev->flags); | |
2720 | } | |
b2d444d7 | 2721 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
4e5314b5 | 2722 | || test_bit(R5_ReadError, &dev->flags)) { |
a4456856 DW |
2723 | s.failed++; |
2724 | s.failed_num = i; | |
1da177e4 LT |
2725 | } else |
2726 | set_bit(R5_Insync, &dev->flags); | |
2727 | } | |
9910f16a | 2728 | rcu_read_unlock(); |
b5e98d65 | 2729 | |
6bfe0b49 DW |
2730 | if (unlikely(blocked_rdev)) { |
2731 | set_bit(STRIPE_HANDLE, &sh->state); | |
2732 | goto unlock; | |
2733 | } | |
2734 | ||
b5e98d65 DW |
2735 | if (s.to_fill && !test_and_set_bit(STRIPE_OP_BIOFILL, &sh->ops.pending)) |
2736 | sh->ops.count++; | |
2737 | ||
45b4233c | 2738 | pr_debug("locked=%d uptodate=%d to_read=%d" |
1da177e4 | 2739 | " to_write=%d failed=%d failed_num=%d\n", |
a4456856 DW |
2740 | s.locked, s.uptodate, s.to_read, s.to_write, |
2741 | s.failed, s.failed_num); | |
1da177e4 LT |
2742 | /* check if the array has lost two devices and, if so, some requests might |
2743 | * need to be failed | |
2744 | */ | |
a4456856 DW |
2745 | if (s.failed > 1 && s.to_read+s.to_write+s.written) |
2746 | handle_requests_to_failed_array(conf, sh, &s, disks, | |
2747 | &return_bi); | |
2748 | if (s.failed > 1 && s.syncing) { | |
1da177e4 LT |
2749 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
2750 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 2751 | s.syncing = 0; |
1da177e4 LT |
2752 | } |
2753 | ||
2754 | /* might be able to return some write requests if the parity block | |
2755 | * is safe, or on a failed drive | |
2756 | */ | |
2757 | dev = &sh->dev[sh->pd_idx]; | |
a4456856 DW |
2758 | if ( s.written && |
2759 | ((test_bit(R5_Insync, &dev->flags) && | |
2760 | !test_bit(R5_LOCKED, &dev->flags) && | |
2761 | test_bit(R5_UPTODATE, &dev->flags)) || | |
2762 | (s.failed == 1 && s.failed_num == sh->pd_idx))) | |
2763 | handle_completed_write_requests(conf, sh, disks, &return_bi); | |
1da177e4 LT |
2764 | |
2765 | /* Now we might consider reading some blocks, either to check/generate | |
2766 | * parity, or to satisfy requests | |
2767 | * or to load a block that is being partially written. | |
2768 | */ | |
a4456856 | 2769 | if (s.to_read || s.non_overwrite || |
f38e1219 DW |
2770 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding || |
2771 | test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending)) | |
a4456856 | 2772 | handle_issuing_new_read_requests5(sh, &s, disks); |
1da177e4 | 2773 | |
e33129d8 DW |
2774 | /* Now we check to see if any write operations have recently |
2775 | * completed | |
2776 | */ | |
2777 | ||
2778 | /* leave prexor set until postxor is done, allows us to distinguish | |
2779 | * a rmw from a rcw during biodrain | |
2780 | */ | |
e0a115e5 | 2781 | prexor = 0; |
e33129d8 DW |
2782 | if (test_bit(STRIPE_OP_PREXOR, &sh->ops.complete) && |
2783 | test_bit(STRIPE_OP_POSTXOR, &sh->ops.complete)) { | |
2784 | ||
e0a115e5 | 2785 | prexor = 1; |
e33129d8 DW |
2786 | clear_bit(STRIPE_OP_PREXOR, &sh->ops.complete); |
2787 | clear_bit(STRIPE_OP_PREXOR, &sh->ops.ack); | |
2788 | clear_bit(STRIPE_OP_PREXOR, &sh->ops.pending); | |
2789 | ||
2790 | for (i = disks; i--; ) | |
2791 | clear_bit(R5_Wantprexor, &sh->dev[i].flags); | |
2792 | } | |
2793 | ||
2794 | /* if only POSTXOR is set then this is an 'expand' postxor */ | |
2795 | if (test_bit(STRIPE_OP_BIODRAIN, &sh->ops.complete) && | |
2796 | test_bit(STRIPE_OP_POSTXOR, &sh->ops.complete)) { | |
2797 | ||
2798 | clear_bit(STRIPE_OP_BIODRAIN, &sh->ops.complete); | |
2799 | clear_bit(STRIPE_OP_BIODRAIN, &sh->ops.ack); | |
2800 | clear_bit(STRIPE_OP_BIODRAIN, &sh->ops.pending); | |
2801 | ||
2802 | clear_bit(STRIPE_OP_POSTXOR, &sh->ops.complete); | |
2803 | clear_bit(STRIPE_OP_POSTXOR, &sh->ops.ack); | |
2804 | clear_bit(STRIPE_OP_POSTXOR, &sh->ops.pending); | |
2805 | ||
2806 | /* All the 'written' buffers and the parity block are ready to | |
2807 | * be written back to disk | |
2808 | */ | |
2809 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
2810 | for (i = disks; i--; ) { | |
2811 | dev = &sh->dev[i]; | |
2812 | if (test_bit(R5_LOCKED, &dev->flags) && | |
2813 | (i == sh->pd_idx || dev->written)) { | |
2814 | pr_debug("Writing block %d\n", i); | |
2815 | set_bit(R5_Wantwrite, &dev->flags); | |
2816 | if (!test_and_set_bit( | |
2817 | STRIPE_OP_IO, &sh->ops.pending)) | |
2818 | sh->ops.count++; | |
e0a115e5 DW |
2819 | if (prexor) |
2820 | continue; | |
e33129d8 DW |
2821 | if (!test_bit(R5_Insync, &dev->flags) || |
2822 | (i == sh->pd_idx && s.failed == 0)) | |
2823 | set_bit(STRIPE_INSYNC, &sh->state); | |
2824 | } | |
2825 | } | |
2826 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2827 | atomic_dec(&conf->preread_active_stripes); | |
2828 | if (atomic_read(&conf->preread_active_stripes) < | |
2829 | IO_THRESHOLD) | |
2830 | md_wakeup_thread(conf->mddev->thread); | |
2831 | } | |
2832 | } | |
2833 | ||
2834 | /* Now to consider new write requests and what else, if anything | |
2835 | * should be read. We do not handle new writes when: | |
2836 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
2837 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
2838 | * block. | |
2839 | */ | |
2840 | if (s.to_write && !test_bit(STRIPE_OP_POSTXOR, &sh->ops.pending) && | |
2841 | !test_bit(STRIPE_OP_CHECK, &sh->ops.pending)) | |
a4456856 | 2842 | handle_issuing_new_write_requests5(conf, sh, &s, disks); |
1da177e4 LT |
2843 | |
2844 | /* maybe we need to check and possibly fix the parity for this stripe | |
e89f8962 DW |
2845 | * Any reads will already have been scheduled, so we just see if enough |
2846 | * data is available. The parity check is held off while parity | |
2847 | * dependent operations are in flight. | |
1da177e4 | 2848 | */ |
e89f8962 DW |
2849 | if ((s.syncing && s.locked == 0 && |
2850 | !test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending) && | |
2851 | !test_bit(STRIPE_INSYNC, &sh->state)) || | |
2852 | test_bit(STRIPE_OP_CHECK, &sh->ops.pending) || | |
2853 | test_bit(STRIPE_OP_MOD_REPAIR_PD, &sh->ops.pending)) | |
a4456856 | 2854 | handle_parity_checks5(conf, sh, &s, disks); |
e89f8962 | 2855 | |
a4456856 | 2856 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
1da177e4 LT |
2857 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
2858 | clear_bit(STRIPE_SYNCING, &sh->state); | |
2859 | } | |
4e5314b5 N |
2860 | |
2861 | /* If the failed drive is just a ReadError, then we might need to progress | |
2862 | * the repair/check process | |
2863 | */ | |
a4456856 DW |
2864 | if (s.failed == 1 && !conf->mddev->ro && |
2865 | test_bit(R5_ReadError, &sh->dev[s.failed_num].flags) | |
2866 | && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags) | |
2867 | && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags) | |
4e5314b5 | 2868 | ) { |
a4456856 | 2869 | dev = &sh->dev[s.failed_num]; |
4e5314b5 N |
2870 | if (!test_bit(R5_ReWrite, &dev->flags)) { |
2871 | set_bit(R5_Wantwrite, &dev->flags); | |
830ea016 DW |
2872 | if (!test_and_set_bit(STRIPE_OP_IO, &sh->ops.pending)) |
2873 | sh->ops.count++; | |
4e5314b5 N |
2874 | set_bit(R5_ReWrite, &dev->flags); |
2875 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 2876 | s.locked++; |
4e5314b5 N |
2877 | } else { |
2878 | /* let's read it back */ | |
2879 | set_bit(R5_Wantread, &dev->flags); | |
830ea016 DW |
2880 | if (!test_and_set_bit(STRIPE_OP_IO, &sh->ops.pending)) |
2881 | sh->ops.count++; | |
4e5314b5 | 2882 | set_bit(R5_LOCKED, &dev->flags); |
a4456856 | 2883 | s.locked++; |
4e5314b5 N |
2884 | } |
2885 | } | |
2886 | ||
f0a50d37 DW |
2887 | /* Finish postxor operations initiated by the expansion |
2888 | * process | |
2889 | */ | |
2890 | if (test_bit(STRIPE_OP_POSTXOR, &sh->ops.complete) && | |
2891 | !test_bit(STRIPE_OP_BIODRAIN, &sh->ops.pending)) { | |
2892 | ||
2893 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
2894 | ||
2895 | clear_bit(STRIPE_OP_POSTXOR, &sh->ops.pending); | |
2896 | clear_bit(STRIPE_OP_POSTXOR, &sh->ops.ack); | |
2897 | clear_bit(STRIPE_OP_POSTXOR, &sh->ops.complete); | |
2898 | ||
a4456856 | 2899 | for (i = conf->raid_disks; i--; ) { |
ccfcc3c1 | 2900 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
efe31143 NB |
2901 | set_bit(R5_LOCKED, &dev->flags); |
2902 | s.locked++; | |
f0a50d37 DW |
2903 | if (!test_and_set_bit(STRIPE_OP_IO, &sh->ops.pending)) |
2904 | sh->ops.count++; | |
ccfcc3c1 | 2905 | } |
f0a50d37 DW |
2906 | } |
2907 | ||
2908 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
2909 | !test_bit(STRIPE_OP_POSTXOR, &sh->ops.pending)) { | |
2910 | /* Need to write out all blocks after computing parity */ | |
2911 | sh->disks = conf->raid_disks; | |
2912 | sh->pd_idx = stripe_to_pdidx(sh->sector, conf, | |
2913 | conf->raid_disks); | |
a2e08551 | 2914 | s.locked += handle_write_operations5(sh, 1, 1); |
f0a50d37 | 2915 | } else if (s.expanded && |
efe31143 | 2916 | s.locked == 0 && |
f0a50d37 | 2917 | !test_bit(STRIPE_OP_POSTXOR, &sh->ops.pending)) { |
ccfcc3c1 | 2918 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
f6705578 | 2919 | atomic_dec(&conf->reshape_stripes); |
ccfcc3c1 N |
2920 | wake_up(&conf->wait_for_overlap); |
2921 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
2922 | } | |
2923 | ||
0f94e87c DW |
2924 | if (s.expanding && s.locked == 0 && |
2925 | !test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending)) | |
a4456856 | 2926 | handle_stripe_expansion(conf, sh, NULL); |
ccfcc3c1 | 2927 | |
d84e0f10 DW |
2928 | if (sh->ops.count) |
2929 | pending = get_stripe_work(sh); | |
2930 | ||
6bfe0b49 | 2931 | unlock: |
1da177e4 LT |
2932 | spin_unlock(&sh->lock); |
2933 | ||
6bfe0b49 DW |
2934 | /* wait for this device to become unblocked */ |
2935 | if (unlikely(blocked_rdev)) | |
2936 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
2937 | ||
d84e0f10 DW |
2938 | if (pending) |
2939 | raid5_run_ops(sh, pending); | |
2940 | ||
a4456856 | 2941 | return_io(return_bi); |
1da177e4 | 2942 | |
1da177e4 LT |
2943 | } |
2944 | ||
16a53ecc | 2945 | static void handle_stripe6(struct stripe_head *sh, struct page *tmp_page) |
1da177e4 | 2946 | { |
16a53ecc | 2947 | raid6_conf_t *conf = sh->raid_conf; |
f416885e | 2948 | int disks = sh->disks; |
a4456856 DW |
2949 | struct bio *return_bi = NULL; |
2950 | int i, pd_idx = sh->pd_idx; | |
2951 | struct stripe_head_state s; | |
2952 | struct r6_state r6s; | |
16a53ecc | 2953 | struct r5dev *dev, *pdev, *qdev; |
6bfe0b49 | 2954 | mdk_rdev_t *blocked_rdev = NULL; |
1da177e4 | 2955 | |
a4456856 | 2956 | r6s.qd_idx = raid6_next_disk(pd_idx, disks); |
45b4233c | 2957 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " |
a4456856 DW |
2958 | "pd_idx=%d, qd_idx=%d\n", |
2959 | (unsigned long long)sh->sector, sh->state, | |
2960 | atomic_read(&sh->count), pd_idx, r6s.qd_idx); | |
2961 | memset(&s, 0, sizeof(s)); | |
72626685 | 2962 | |
16a53ecc N |
2963 | spin_lock(&sh->lock); |
2964 | clear_bit(STRIPE_HANDLE, &sh->state); | |
2965 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2966 | ||
a4456856 DW |
2967 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
2968 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2969 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
16a53ecc | 2970 | /* Now to look around and see what can be done */ |
1da177e4 LT |
2971 | |
2972 | rcu_read_lock(); | |
16a53ecc N |
2973 | for (i=disks; i--; ) { |
2974 | mdk_rdev_t *rdev; | |
2975 | dev = &sh->dev[i]; | |
2976 | clear_bit(R5_Insync, &dev->flags); | |
1da177e4 | 2977 | |
45b4233c | 2978 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc N |
2979 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
2980 | /* maybe we can reply to a read */ | |
2981 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) { | |
2982 | struct bio *rbi, *rbi2; | |
45b4233c | 2983 | pr_debug("Return read for disc %d\n", i); |
16a53ecc N |
2984 | spin_lock_irq(&conf->device_lock); |
2985 | rbi = dev->toread; | |
2986 | dev->toread = NULL; | |
2987 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
2988 | wake_up(&conf->wait_for_overlap); | |
2989 | spin_unlock_irq(&conf->device_lock); | |
2990 | while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) { | |
2991 | copy_data(0, rbi, dev->page, dev->sector); | |
2992 | rbi2 = r5_next_bio(rbi, dev->sector); | |
2993 | spin_lock_irq(&conf->device_lock); | |
2994 | if (--rbi->bi_phys_segments == 0) { | |
2995 | rbi->bi_next = return_bi; | |
2996 | return_bi = rbi; | |
2997 | } | |
2998 | spin_unlock_irq(&conf->device_lock); | |
2999 | rbi = rbi2; | |
3000 | } | |
3001 | } | |
1da177e4 | 3002 | |
16a53ecc | 3003 | /* now count some things */ |
a4456856 DW |
3004 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
3005 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
1da177e4 | 3006 | |
16a53ecc | 3007 | |
a4456856 DW |
3008 | if (dev->toread) |
3009 | s.to_read++; | |
16a53ecc | 3010 | if (dev->towrite) { |
a4456856 | 3011 | s.to_write++; |
16a53ecc | 3012 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 3013 | s.non_overwrite++; |
16a53ecc | 3014 | } |
a4456856 DW |
3015 | if (dev->written) |
3016 | s.written++; | |
16a53ecc | 3017 | rdev = rcu_dereference(conf->disks[i].rdev); |
6bfe0b49 DW |
3018 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
3019 | blocked_rdev = rdev; | |
3020 | atomic_inc(&rdev->nr_pending); | |
3021 | break; | |
3022 | } | |
16a53ecc N |
3023 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
3024 | /* The ReadError flag will just be confusing now */ | |
3025 | clear_bit(R5_ReadError, &dev->flags); | |
3026 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3027 | } |
16a53ecc N |
3028 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
3029 | || test_bit(R5_ReadError, &dev->flags)) { | |
a4456856 DW |
3030 | if (s.failed < 2) |
3031 | r6s.failed_num[s.failed] = i; | |
3032 | s.failed++; | |
16a53ecc N |
3033 | } else |
3034 | set_bit(R5_Insync, &dev->flags); | |
1da177e4 LT |
3035 | } |
3036 | rcu_read_unlock(); | |
6bfe0b49 DW |
3037 | |
3038 | if (unlikely(blocked_rdev)) { | |
3039 | set_bit(STRIPE_HANDLE, &sh->state); | |
3040 | goto unlock; | |
3041 | } | |
45b4233c | 3042 | pr_debug("locked=%d uptodate=%d to_read=%d" |
16a53ecc | 3043 | " to_write=%d failed=%d failed_num=%d,%d\n", |
a4456856 DW |
3044 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, |
3045 | r6s.failed_num[0], r6s.failed_num[1]); | |
3046 | /* check if the array has lost >2 devices and, if so, some requests | |
3047 | * might need to be failed | |
16a53ecc | 3048 | */ |
a4456856 DW |
3049 | if (s.failed > 2 && s.to_read+s.to_write+s.written) |
3050 | handle_requests_to_failed_array(conf, sh, &s, disks, | |
3051 | &return_bi); | |
3052 | if (s.failed > 2 && s.syncing) { | |
16a53ecc N |
3053 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3054 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3055 | s.syncing = 0; |
16a53ecc N |
3056 | } |
3057 | ||
3058 | /* | |
3059 | * might be able to return some write requests if the parity blocks | |
3060 | * are safe, or on a failed drive | |
3061 | */ | |
3062 | pdev = &sh->dev[pd_idx]; | |
a4456856 DW |
3063 | r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx) |
3064 | || (s.failed >= 2 && r6s.failed_num[1] == pd_idx); | |
3065 | qdev = &sh->dev[r6s.qd_idx]; | |
3066 | r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == r6s.qd_idx) | |
3067 | || (s.failed >= 2 && r6s.failed_num[1] == r6s.qd_idx); | |
3068 | ||
3069 | if ( s.written && | |
3070 | ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
16a53ecc | 3071 | && !test_bit(R5_LOCKED, &pdev->flags) |
a4456856 DW |
3072 | && test_bit(R5_UPTODATE, &pdev->flags)))) && |
3073 | ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
16a53ecc | 3074 | && !test_bit(R5_LOCKED, &qdev->flags) |
a4456856 DW |
3075 | && test_bit(R5_UPTODATE, &qdev->flags))))) |
3076 | handle_completed_write_requests(conf, sh, disks, &return_bi); | |
16a53ecc N |
3077 | |
3078 | /* Now we might consider reading some blocks, either to check/generate | |
3079 | * parity, or to satisfy requests | |
3080 | * or to load a block that is being partially written. | |
3081 | */ | |
a4456856 DW |
3082 | if (s.to_read || s.non_overwrite || (s.to_write && s.failed) || |
3083 | (s.syncing && (s.uptodate < disks)) || s.expanding) | |
3084 | handle_issuing_new_read_requests6(sh, &s, &r6s, disks); | |
16a53ecc N |
3085 | |
3086 | /* now to consider writing and what else, if anything should be read */ | |
a4456856 DW |
3087 | if (s.to_write) |
3088 | handle_issuing_new_write_requests6(conf, sh, &s, &r6s, disks); | |
16a53ecc N |
3089 | |
3090 | /* maybe we need to check and possibly fix the parity for this stripe | |
a4456856 DW |
3091 | * Any reads will already have been scheduled, so we just see if enough |
3092 | * data is available | |
16a53ecc | 3093 | */ |
a4456856 DW |
3094 | if (s.syncing && s.locked == 0 && !test_bit(STRIPE_INSYNC, &sh->state)) |
3095 | handle_parity_checks6(conf, sh, &s, &r6s, tmp_page, disks); | |
16a53ecc | 3096 | |
a4456856 | 3097 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
16a53ecc N |
3098 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3099 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3100 | } | |
3101 | ||
3102 | /* If the failed drives are just a ReadError, then we might need | |
3103 | * to progress the repair/check process | |
3104 | */ | |
a4456856 DW |
3105 | if (s.failed <= 2 && !conf->mddev->ro) |
3106 | for (i = 0; i < s.failed; i++) { | |
3107 | dev = &sh->dev[r6s.failed_num[i]]; | |
16a53ecc N |
3108 | if (test_bit(R5_ReadError, &dev->flags) |
3109 | && !test_bit(R5_LOCKED, &dev->flags) | |
3110 | && test_bit(R5_UPTODATE, &dev->flags) | |
3111 | ) { | |
3112 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3113 | set_bit(R5_Wantwrite, &dev->flags); | |
3114 | set_bit(R5_ReWrite, &dev->flags); | |
3115 | set_bit(R5_LOCKED, &dev->flags); | |
3116 | } else { | |
3117 | /* let's read it back */ | |
3118 | set_bit(R5_Wantread, &dev->flags); | |
3119 | set_bit(R5_LOCKED, &dev->flags); | |
3120 | } | |
3121 | } | |
3122 | } | |
f416885e | 3123 | |
a4456856 | 3124 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state)) { |
f416885e N |
3125 | /* Need to write out all blocks after computing P&Q */ |
3126 | sh->disks = conf->raid_disks; | |
3127 | sh->pd_idx = stripe_to_pdidx(sh->sector, conf, | |
3128 | conf->raid_disks); | |
3129 | compute_parity6(sh, RECONSTRUCT_WRITE); | |
3130 | for (i = conf->raid_disks ; i-- ; ) { | |
3131 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3132 | s.locked++; |
f416885e N |
3133 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
3134 | } | |
3135 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
a4456856 | 3136 | } else if (s.expanded) { |
f416885e N |
3137 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
3138 | atomic_dec(&conf->reshape_stripes); | |
3139 | wake_up(&conf->wait_for_overlap); | |
3140 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3141 | } | |
3142 | ||
0f94e87c DW |
3143 | if (s.expanding && s.locked == 0 && |
3144 | !test_bit(STRIPE_OP_COMPUTE_BLK, &sh->ops.pending)) | |
a4456856 | 3145 | handle_stripe_expansion(conf, sh, &r6s); |
f416885e | 3146 | |
6bfe0b49 | 3147 | unlock: |
16a53ecc N |
3148 | spin_unlock(&sh->lock); |
3149 | ||
6bfe0b49 DW |
3150 | /* wait for this device to become unblocked */ |
3151 | if (unlikely(blocked_rdev)) | |
3152 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3153 | ||
a4456856 | 3154 | return_io(return_bi); |
16a53ecc | 3155 | |
16a53ecc N |
3156 | for (i=disks; i-- ;) { |
3157 | int rw; | |
3158 | struct bio *bi; | |
3159 | mdk_rdev_t *rdev; | |
3160 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
802ba064 | 3161 | rw = WRITE; |
16a53ecc | 3162 | else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
802ba064 | 3163 | rw = READ; |
16a53ecc N |
3164 | else |
3165 | continue; | |
3166 | ||
8b3e6cdc DW |
3167 | set_bit(STRIPE_IO_STARTED, &sh->state); |
3168 | ||
16a53ecc N |
3169 | bi = &sh->dev[i].req; |
3170 | ||
3171 | bi->bi_rw = rw; | |
802ba064 | 3172 | if (rw == WRITE) |
16a53ecc N |
3173 | bi->bi_end_io = raid5_end_write_request; |
3174 | else | |
3175 | bi->bi_end_io = raid5_end_read_request; | |
3176 | ||
3177 | rcu_read_lock(); | |
3178 | rdev = rcu_dereference(conf->disks[i].rdev); | |
3179 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
3180 | rdev = NULL; | |
3181 | if (rdev) | |
3182 | atomic_inc(&rdev->nr_pending); | |
3183 | rcu_read_unlock(); | |
3184 | ||
3185 | if (rdev) { | |
a4456856 | 3186 | if (s.syncing || s.expanding || s.expanded) |
16a53ecc N |
3187 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
3188 | ||
3189 | bi->bi_bdev = rdev->bdev; | |
45b4233c | 3190 | pr_debug("for %llu schedule op %ld on disc %d\n", |
16a53ecc N |
3191 | (unsigned long long)sh->sector, bi->bi_rw, i); |
3192 | atomic_inc(&sh->count); | |
3193 | bi->bi_sector = sh->sector + rdev->data_offset; | |
3194 | bi->bi_flags = 1 << BIO_UPTODATE; | |
3195 | bi->bi_vcnt = 1; | |
3196 | bi->bi_max_vecs = 1; | |
3197 | bi->bi_idx = 0; | |
3198 | bi->bi_io_vec = &sh->dev[i].vec; | |
3199 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
3200 | bi->bi_io_vec[0].bv_offset = 0; | |
3201 | bi->bi_size = STRIPE_SIZE; | |
3202 | bi->bi_next = NULL; | |
3203 | if (rw == WRITE && | |
3204 | test_bit(R5_ReWrite, &sh->dev[i].flags)) | |
3205 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); | |
3206 | generic_make_request(bi); | |
3207 | } else { | |
802ba064 | 3208 | if (rw == WRITE) |
16a53ecc | 3209 | set_bit(STRIPE_DEGRADED, &sh->state); |
45b4233c | 3210 | pr_debug("skip op %ld on disc %d for sector %llu\n", |
16a53ecc N |
3211 | bi->bi_rw, i, (unsigned long long)sh->sector); |
3212 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
3213 | set_bit(STRIPE_HANDLE, &sh->state); | |
3214 | } | |
3215 | } | |
3216 | } | |
3217 | ||
3218 | static void handle_stripe(struct stripe_head *sh, struct page *tmp_page) | |
3219 | { | |
3220 | if (sh->raid_conf->level == 6) | |
3221 | handle_stripe6(sh, tmp_page); | |
3222 | else | |
3223 | handle_stripe5(sh); | |
3224 | } | |
3225 | ||
3226 | ||
3227 | ||
3228 | static void raid5_activate_delayed(raid5_conf_t *conf) | |
3229 | { | |
3230 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3231 | while (!list_empty(&conf->delayed_list)) { | |
3232 | struct list_head *l = conf->delayed_list.next; | |
3233 | struct stripe_head *sh; | |
3234 | sh = list_entry(l, struct stripe_head, lru); | |
3235 | list_del_init(l); | |
3236 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3237 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3238 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3239 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3240 | } |
6ed3003c N |
3241 | } else |
3242 | blk_plug_device(conf->mddev->queue); | |
16a53ecc N |
3243 | } |
3244 | ||
3245 | static void activate_bit_delay(raid5_conf_t *conf) | |
3246 | { | |
3247 | /* device_lock is held */ | |
3248 | struct list_head head; | |
3249 | list_add(&head, &conf->bitmap_list); | |
3250 | list_del_init(&conf->bitmap_list); | |
3251 | while (!list_empty(&head)) { | |
3252 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3253 | list_del_init(&sh->lru); | |
3254 | atomic_inc(&sh->count); | |
3255 | __release_stripe(conf, sh); | |
3256 | } | |
3257 | } | |
3258 | ||
3259 | static void unplug_slaves(mddev_t *mddev) | |
3260 | { | |
3261 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3262 | int i; | |
3263 | ||
3264 | rcu_read_lock(); | |
3265 | for (i=0; i<mddev->raid_disks; i++) { | |
3266 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); | |
3267 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { | |
165125e1 | 3268 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); |
16a53ecc N |
3269 | |
3270 | atomic_inc(&rdev->nr_pending); | |
3271 | rcu_read_unlock(); | |
3272 | ||
2ad8b1ef | 3273 | blk_unplug(r_queue); |
16a53ecc N |
3274 | |
3275 | rdev_dec_pending(rdev, mddev); | |
3276 | rcu_read_lock(); | |
3277 | } | |
3278 | } | |
3279 | rcu_read_unlock(); | |
3280 | } | |
3281 | ||
165125e1 | 3282 | static void raid5_unplug_device(struct request_queue *q) |
16a53ecc N |
3283 | { |
3284 | mddev_t *mddev = q->queuedata; | |
3285 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3286 | unsigned long flags; | |
3287 | ||
3288 | spin_lock_irqsave(&conf->device_lock, flags); | |
3289 | ||
3290 | if (blk_remove_plug(q)) { | |
3291 | conf->seq_flush++; | |
3292 | raid5_activate_delayed(conf); | |
72626685 | 3293 | } |
1da177e4 LT |
3294 | md_wakeup_thread(mddev->thread); |
3295 | ||
3296 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3297 | ||
3298 | unplug_slaves(mddev); | |
3299 | } | |
3300 | ||
f022b2fd N |
3301 | static int raid5_congested(void *data, int bits) |
3302 | { | |
3303 | mddev_t *mddev = data; | |
3304 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3305 | ||
3306 | /* No difference between reads and writes. Just check | |
3307 | * how busy the stripe_cache is | |
3308 | */ | |
3309 | if (conf->inactive_blocked) | |
3310 | return 1; | |
3311 | if (conf->quiesce) | |
3312 | return 1; | |
3313 | if (list_empty_careful(&conf->inactive_list)) | |
3314 | return 1; | |
3315 | ||
3316 | return 0; | |
3317 | } | |
3318 | ||
23032a0e RBJ |
3319 | /* We want read requests to align with chunks where possible, |
3320 | * but write requests don't need to. | |
3321 | */ | |
165125e1 | 3322 | static int raid5_mergeable_bvec(struct request_queue *q, struct bio *bio, struct bio_vec *biovec) |
23032a0e RBJ |
3323 | { |
3324 | mddev_t *mddev = q->queuedata; | |
3325 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
3326 | int max; | |
3327 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
3328 | unsigned int bio_sectors = bio->bi_size >> 9; | |
3329 | ||
802ba064 | 3330 | if (bio_data_dir(bio) == WRITE) |
23032a0e RBJ |
3331 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3332 | ||
3333 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; | |
3334 | if (max < 0) max = 0; | |
3335 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3336 | return biovec->bv_len; | |
3337 | else | |
3338 | return max; | |
3339 | } | |
3340 | ||
f679623f RBJ |
3341 | |
3342 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
3343 | { | |
3344 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
3345 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
3346 | unsigned int bio_sectors = bio->bi_size >> 9; | |
3347 | ||
3348 | return chunk_sectors >= | |
3349 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3350 | } | |
3351 | ||
46031f9a RBJ |
3352 | /* |
3353 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3354 | * later sampled by raid5d. | |
3355 | */ | |
3356 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
3357 | { | |
3358 | unsigned long flags; | |
3359 | ||
3360 | spin_lock_irqsave(&conf->device_lock, flags); | |
3361 | ||
3362 | bi->bi_next = conf->retry_read_aligned_list; | |
3363 | conf->retry_read_aligned_list = bi; | |
3364 | ||
3365 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3366 | md_wakeup_thread(conf->mddev->thread); | |
3367 | } | |
3368 | ||
3369 | ||
3370 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
3371 | { | |
3372 | struct bio *bi; | |
3373 | ||
3374 | bi = conf->retry_read_aligned; | |
3375 | if (bi) { | |
3376 | conf->retry_read_aligned = NULL; | |
3377 | return bi; | |
3378 | } | |
3379 | bi = conf->retry_read_aligned_list; | |
3380 | if(bi) { | |
387bb173 | 3381 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a RBJ |
3382 | bi->bi_next = NULL; |
3383 | bi->bi_phys_segments = 1; /* biased count of active stripes */ | |
3384 | bi->bi_hw_segments = 0; /* count of processed stripes */ | |
3385 | } | |
3386 | ||
3387 | return bi; | |
3388 | } | |
3389 | ||
3390 | ||
f679623f RBJ |
3391 | /* |
3392 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3393 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3394 | * first). | |
3395 | * If the read failed.. | |
3396 | */ | |
6712ecf8 | 3397 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3398 | { |
3399 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
3400 | mddev_t *mddev; |
3401 | raid5_conf_t *conf; | |
3402 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
3403 | mdk_rdev_t *rdev; | |
3404 | ||
f679623f | 3405 | bio_put(bi); |
46031f9a RBJ |
3406 | |
3407 | mddev = raid_bi->bi_bdev->bd_disk->queue->queuedata; | |
3408 | conf = mddev_to_conf(mddev); | |
3409 | rdev = (void*)raid_bi->bi_next; | |
3410 | raid_bi->bi_next = NULL; | |
3411 | ||
3412 | rdev_dec_pending(rdev, conf->mddev); | |
3413 | ||
3414 | if (!error && uptodate) { | |
6712ecf8 | 3415 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3416 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3417 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3418 | return; |
46031f9a RBJ |
3419 | } |
3420 | ||
3421 | ||
45b4233c | 3422 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3423 | |
3424 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3425 | } |
3426 | ||
387bb173 NB |
3427 | static int bio_fits_rdev(struct bio *bi) |
3428 | { | |
165125e1 | 3429 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 NB |
3430 | |
3431 | if ((bi->bi_size>>9) > q->max_sectors) | |
3432 | return 0; | |
3433 | blk_recount_segments(q, bi); | |
3434 | if (bi->bi_phys_segments > q->max_phys_segments || | |
3435 | bi->bi_hw_segments > q->max_hw_segments) | |
3436 | return 0; | |
3437 | ||
3438 | if (q->merge_bvec_fn) | |
3439 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3440 | * just just give up | |
3441 | */ | |
3442 | return 0; | |
3443 | ||
3444 | return 1; | |
3445 | } | |
3446 | ||
3447 | ||
165125e1 | 3448 | static int chunk_aligned_read(struct request_queue *q, struct bio * raid_bio) |
f679623f RBJ |
3449 | { |
3450 | mddev_t *mddev = q->queuedata; | |
3451 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3452 | const unsigned int raid_disks = conf->raid_disks; | |
46031f9a | 3453 | const unsigned int data_disks = raid_disks - conf->max_degraded; |
f679623f RBJ |
3454 | unsigned int dd_idx, pd_idx; |
3455 | struct bio* align_bi; | |
3456 | mdk_rdev_t *rdev; | |
3457 | ||
3458 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3459 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3460 | return 0; |
3461 | } | |
3462 | /* | |
3463 | * use bio_clone to make a copy of the bio | |
3464 | */ | |
3465 | align_bi = bio_clone(raid_bio, GFP_NOIO); | |
3466 | if (!align_bi) | |
3467 | return 0; | |
3468 | /* | |
3469 | * set bi_end_io to a new function, and set bi_private to the | |
3470 | * original bio. | |
3471 | */ | |
3472 | align_bi->bi_end_io = raid5_align_endio; | |
3473 | align_bi->bi_private = raid_bio; | |
3474 | /* | |
3475 | * compute position | |
3476 | */ | |
3477 | align_bi->bi_sector = raid5_compute_sector(raid_bio->bi_sector, | |
3478 | raid_disks, | |
3479 | data_disks, | |
3480 | &dd_idx, | |
3481 | &pd_idx, | |
3482 | conf); | |
3483 | ||
3484 | rcu_read_lock(); | |
3485 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3486 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
3487 | atomic_inc(&rdev->nr_pending); |
3488 | rcu_read_unlock(); | |
46031f9a RBJ |
3489 | raid_bio->bi_next = (void*)rdev; |
3490 | align_bi->bi_bdev = rdev->bdev; | |
3491 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3492 | align_bi->bi_sector += rdev->data_offset; | |
3493 | ||
387bb173 NB |
3494 | if (!bio_fits_rdev(align_bi)) { |
3495 | /* too big in some way */ | |
3496 | bio_put(align_bi); | |
3497 | rdev_dec_pending(rdev, mddev); | |
3498 | return 0; | |
3499 | } | |
3500 | ||
46031f9a RBJ |
3501 | spin_lock_irq(&conf->device_lock); |
3502 | wait_event_lock_irq(conf->wait_for_stripe, | |
3503 | conf->quiesce == 0, | |
3504 | conf->device_lock, /* nothing */); | |
3505 | atomic_inc(&conf->active_aligned_reads); | |
3506 | spin_unlock_irq(&conf->device_lock); | |
3507 | ||
f679623f RBJ |
3508 | generic_make_request(align_bi); |
3509 | return 1; | |
3510 | } else { | |
3511 | rcu_read_unlock(); | |
46031f9a | 3512 | bio_put(align_bi); |
f679623f RBJ |
3513 | return 0; |
3514 | } | |
3515 | } | |
3516 | ||
8b3e6cdc DW |
3517 | /* __get_priority_stripe - get the next stripe to process |
3518 | * | |
3519 | * Full stripe writes are allowed to pass preread active stripes up until | |
3520 | * the bypass_threshold is exceeded. In general the bypass_count | |
3521 | * increments when the handle_list is handled before the hold_list; however, it | |
3522 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3523 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3524 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3525 | * handle_list. | |
3526 | */ | |
3527 | static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf) | |
3528 | { | |
3529 | struct stripe_head *sh; | |
3530 | ||
3531 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3532 | __func__, | |
3533 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3534 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3535 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3536 | ||
3537 | if (!list_empty(&conf->handle_list)) { | |
3538 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3539 | ||
3540 | if (list_empty(&conf->hold_list)) | |
3541 | conf->bypass_count = 0; | |
3542 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3543 | if (conf->hold_list.next == conf->last_hold) | |
3544 | conf->bypass_count++; | |
3545 | else { | |
3546 | conf->last_hold = conf->hold_list.next; | |
3547 | conf->bypass_count -= conf->bypass_threshold; | |
3548 | if (conf->bypass_count < 0) | |
3549 | conf->bypass_count = 0; | |
3550 | } | |
3551 | } | |
3552 | } else if (!list_empty(&conf->hold_list) && | |
3553 | ((conf->bypass_threshold && | |
3554 | conf->bypass_count > conf->bypass_threshold) || | |
3555 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3556 | sh = list_entry(conf->hold_list.next, | |
3557 | typeof(*sh), lru); | |
3558 | conf->bypass_count -= conf->bypass_threshold; | |
3559 | if (conf->bypass_count < 0) | |
3560 | conf->bypass_count = 0; | |
3561 | } else | |
3562 | return NULL; | |
3563 | ||
3564 | list_del_init(&sh->lru); | |
3565 | atomic_inc(&sh->count); | |
3566 | BUG_ON(atomic_read(&sh->count) != 1); | |
3567 | return sh; | |
3568 | } | |
f679623f | 3569 | |
165125e1 | 3570 | static int make_request(struct request_queue *q, struct bio * bi) |
1da177e4 LT |
3571 | { |
3572 | mddev_t *mddev = q->queuedata; | |
3573 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
1da177e4 LT |
3574 | unsigned int dd_idx, pd_idx; |
3575 | sector_t new_sector; | |
3576 | sector_t logical_sector, last_sector; | |
3577 | struct stripe_head *sh; | |
a362357b | 3578 | const int rw = bio_data_dir(bi); |
f6344757 | 3579 | int remaining; |
1da177e4 | 3580 | |
e5dcdd80 | 3581 | if (unlikely(bio_barrier(bi))) { |
6712ecf8 | 3582 | bio_endio(bi, -EOPNOTSUPP); |
e5dcdd80 N |
3583 | return 0; |
3584 | } | |
3585 | ||
3d310eb7 | 3586 | md_write_start(mddev, bi); |
06d91a5f | 3587 | |
a362357b JA |
3588 | disk_stat_inc(mddev->gendisk, ios[rw]); |
3589 | disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bi)); | |
1da177e4 | 3590 | |
802ba064 | 3591 | if (rw == READ && |
52488615 RBJ |
3592 | mddev->reshape_position == MaxSector && |
3593 | chunk_aligned_read(q,bi)) | |
3594 | return 0; | |
3595 | ||
1da177e4 LT |
3596 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3597 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3598 | bi->bi_next = NULL; | |
3599 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3600 | |
1da177e4 LT |
3601 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3602 | DEFINE_WAIT(w); | |
16a53ecc | 3603 | int disks, data_disks; |
b578d55f | 3604 | |
7ecaa1e6 | 3605 | retry: |
b578d55f | 3606 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
7ecaa1e6 N |
3607 | if (likely(conf->expand_progress == MaxSector)) |
3608 | disks = conf->raid_disks; | |
3609 | else { | |
df8e7f76 N |
3610 | /* spinlock is needed as expand_progress may be |
3611 | * 64bit on a 32bit platform, and so it might be | |
3612 | * possible to see a half-updated value | |
3613 | * Ofcourse expand_progress could change after | |
3614 | * the lock is dropped, so once we get a reference | |
3615 | * to the stripe that we think it is, we will have | |
3616 | * to check again. | |
3617 | */ | |
7ecaa1e6 N |
3618 | spin_lock_irq(&conf->device_lock); |
3619 | disks = conf->raid_disks; | |
3620 | if (logical_sector >= conf->expand_progress) | |
3621 | disks = conf->previous_raid_disks; | |
b578d55f N |
3622 | else { |
3623 | if (logical_sector >= conf->expand_lo) { | |
3624 | spin_unlock_irq(&conf->device_lock); | |
3625 | schedule(); | |
3626 | goto retry; | |
3627 | } | |
3628 | } | |
7ecaa1e6 N |
3629 | spin_unlock_irq(&conf->device_lock); |
3630 | } | |
16a53ecc N |
3631 | data_disks = disks - conf->max_degraded; |
3632 | ||
3633 | new_sector = raid5_compute_sector(logical_sector, disks, data_disks, | |
7ecaa1e6 | 3634 | &dd_idx, &pd_idx, conf); |
45b4233c | 3635 | pr_debug("raid5: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3636 | (unsigned long long)new_sector, |
3637 | (unsigned long long)logical_sector); | |
3638 | ||
7ecaa1e6 | 3639 | sh = get_active_stripe(conf, new_sector, disks, pd_idx, (bi->bi_rw&RWA_MASK)); |
1da177e4 | 3640 | if (sh) { |
7ecaa1e6 N |
3641 | if (unlikely(conf->expand_progress != MaxSector)) { |
3642 | /* expansion might have moved on while waiting for a | |
df8e7f76 N |
3643 | * stripe, so we must do the range check again. |
3644 | * Expansion could still move past after this | |
3645 | * test, but as we are holding a reference to | |
3646 | * 'sh', we know that if that happens, | |
3647 | * STRIPE_EXPANDING will get set and the expansion | |
3648 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
3649 | */ |
3650 | int must_retry = 0; | |
3651 | spin_lock_irq(&conf->device_lock); | |
3652 | if (logical_sector < conf->expand_progress && | |
3653 | disks == conf->previous_raid_disks) | |
3654 | /* mismatch, need to try again */ | |
3655 | must_retry = 1; | |
3656 | spin_unlock_irq(&conf->device_lock); | |
3657 | if (must_retry) { | |
3658 | release_stripe(sh); | |
3659 | goto retry; | |
3660 | } | |
3661 | } | |
e464eafd N |
3662 | /* FIXME what if we get a false positive because these |
3663 | * are being updated. | |
3664 | */ | |
3665 | if (logical_sector >= mddev->suspend_lo && | |
3666 | logical_sector < mddev->suspend_hi) { | |
3667 | release_stripe(sh); | |
3668 | schedule(); | |
3669 | goto retry; | |
3670 | } | |
7ecaa1e6 N |
3671 | |
3672 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
3673 | !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { | |
3674 | /* Stripe is busy expanding or | |
3675 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
3676 | * and wait a while |
3677 | */ | |
3678 | raid5_unplug_device(mddev->queue); | |
3679 | release_stripe(sh); | |
3680 | schedule(); | |
3681 | goto retry; | |
3682 | } | |
3683 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
3684 | set_bit(STRIPE_HANDLE, &sh->state); |
3685 | clear_bit(STRIPE_DELAYED, &sh->state); | |
1da177e4 | 3686 | release_stripe(sh); |
1da177e4 LT |
3687 | } else { |
3688 | /* cannot get stripe for read-ahead, just give-up */ | |
3689 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
3690 | finish_wait(&conf->wait_for_overlap, &w); | |
3691 | break; | |
3692 | } | |
3693 | ||
3694 | } | |
3695 | spin_lock_irq(&conf->device_lock); | |
f6344757 N |
3696 | remaining = --bi->bi_phys_segments; |
3697 | spin_unlock_irq(&conf->device_lock); | |
3698 | if (remaining == 0) { | |
1da177e4 | 3699 | |
16a53ecc | 3700 | if ( rw == WRITE ) |
1da177e4 | 3701 | md_write_end(mddev); |
6712ecf8 N |
3702 | |
3703 | bi->bi_end_io(bi, | |
c2b00852 N |
3704 | test_bit(BIO_UPTODATE, &bi->bi_flags) |
3705 | ? 0 : -EIO); | |
1da177e4 | 3706 | } |
1da177e4 LT |
3707 | return 0; |
3708 | } | |
3709 | ||
52c03291 | 3710 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 3711 | { |
52c03291 N |
3712 | /* reshaping is quite different to recovery/resync so it is |
3713 | * handled quite separately ... here. | |
3714 | * | |
3715 | * On each call to sync_request, we gather one chunk worth of | |
3716 | * destination stripes and flag them as expanding. | |
3717 | * Then we find all the source stripes and request reads. | |
3718 | * As the reads complete, handle_stripe will copy the data | |
3719 | * into the destination stripe and release that stripe. | |
3720 | */ | |
1da177e4 LT |
3721 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; |
3722 | struct stripe_head *sh; | |
ccfcc3c1 N |
3723 | int pd_idx; |
3724 | sector_t first_sector, last_sector; | |
f416885e N |
3725 | int raid_disks = conf->previous_raid_disks; |
3726 | int data_disks = raid_disks - conf->max_degraded; | |
3727 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
3728 | int i; |
3729 | int dd_idx; | |
3730 | sector_t writepos, safepos, gap; | |
3731 | ||
3732 | if (sector_nr == 0 && | |
3733 | conf->expand_progress != 0) { | |
3734 | /* restarting in the middle, skip the initial sectors */ | |
3735 | sector_nr = conf->expand_progress; | |
f416885e | 3736 | sector_div(sector_nr, new_data_disks); |
52c03291 N |
3737 | *skipped = 1; |
3738 | return sector_nr; | |
3739 | } | |
3740 | ||
3741 | /* we update the metadata when there is more than 3Meg | |
3742 | * in the block range (that is rather arbitrary, should | |
3743 | * probably be time based) or when the data about to be | |
3744 | * copied would over-write the source of the data at | |
3745 | * the front of the range. | |
3746 | * i.e. one new_stripe forward from expand_progress new_maps | |
3747 | * to after where expand_lo old_maps to | |
3748 | */ | |
3749 | writepos = conf->expand_progress + | |
f416885e N |
3750 | conf->chunk_size/512*(new_data_disks); |
3751 | sector_div(writepos, new_data_disks); | |
52c03291 | 3752 | safepos = conf->expand_lo; |
f416885e | 3753 | sector_div(safepos, data_disks); |
52c03291 N |
3754 | gap = conf->expand_progress - conf->expand_lo; |
3755 | ||
3756 | if (writepos >= safepos || | |
f416885e | 3757 | gap > (new_data_disks)*3000*2 /*3Meg*/) { |
52c03291 N |
3758 | /* Cannot proceed until we've updated the superblock... */ |
3759 | wait_event(conf->wait_for_overlap, | |
3760 | atomic_read(&conf->reshape_stripes)==0); | |
3761 | mddev->reshape_position = conf->expand_progress; | |
850b2b42 | 3762 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 3763 | md_wakeup_thread(mddev->thread); |
850b2b42 | 3764 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
3765 | kthread_should_stop()); |
3766 | spin_lock_irq(&conf->device_lock); | |
3767 | conf->expand_lo = mddev->reshape_position; | |
3768 | spin_unlock_irq(&conf->device_lock); | |
3769 | wake_up(&conf->wait_for_overlap); | |
3770 | } | |
3771 | ||
3772 | for (i=0; i < conf->chunk_size/512; i+= STRIPE_SECTORS) { | |
3773 | int j; | |
3774 | int skipped = 0; | |
3775 | pd_idx = stripe_to_pdidx(sector_nr+i, conf, conf->raid_disks); | |
3776 | sh = get_active_stripe(conf, sector_nr+i, | |
3777 | conf->raid_disks, pd_idx, 0); | |
3778 | set_bit(STRIPE_EXPANDING, &sh->state); | |
3779 | atomic_inc(&conf->reshape_stripes); | |
3780 | /* If any of this stripe is beyond the end of the old | |
3781 | * array, then we need to zero those blocks | |
3782 | */ | |
3783 | for (j=sh->disks; j--;) { | |
3784 | sector_t s; | |
3785 | if (j == sh->pd_idx) | |
3786 | continue; | |
f416885e N |
3787 | if (conf->level == 6 && |
3788 | j == raid6_next_disk(sh->pd_idx, sh->disks)) | |
3789 | continue; | |
52c03291 N |
3790 | s = compute_blocknr(sh, j); |
3791 | if (s < (mddev->array_size<<1)) { | |
3792 | skipped = 1; | |
3793 | continue; | |
3794 | } | |
3795 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
3796 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
3797 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
3798 | } | |
3799 | if (!skipped) { | |
3800 | set_bit(STRIPE_EXPAND_READY, &sh->state); | |
3801 | set_bit(STRIPE_HANDLE, &sh->state); | |
3802 | } | |
3803 | release_stripe(sh); | |
3804 | } | |
3805 | spin_lock_irq(&conf->device_lock); | |
6d3baf2e | 3806 | conf->expand_progress = (sector_nr + i) * new_data_disks; |
52c03291 N |
3807 | spin_unlock_irq(&conf->device_lock); |
3808 | /* Ok, those stripe are ready. We can start scheduling | |
3809 | * reads on the source stripes. | |
3810 | * The source stripes are determined by mapping the first and last | |
3811 | * block on the destination stripes. | |
3812 | */ | |
52c03291 | 3813 | first_sector = |
f416885e | 3814 | raid5_compute_sector(sector_nr*(new_data_disks), |
52c03291 N |
3815 | raid_disks, data_disks, |
3816 | &dd_idx, &pd_idx, conf); | |
3817 | last_sector = | |
3818 | raid5_compute_sector((sector_nr+conf->chunk_size/512) | |
f416885e | 3819 | *(new_data_disks) -1, |
52c03291 N |
3820 | raid_disks, data_disks, |
3821 | &dd_idx, &pd_idx, conf); | |
3822 | if (last_sector >= (mddev->size<<1)) | |
3823 | last_sector = (mddev->size<<1)-1; | |
3824 | while (first_sector <= last_sector) { | |
f416885e N |
3825 | pd_idx = stripe_to_pdidx(first_sector, conf, |
3826 | conf->previous_raid_disks); | |
52c03291 N |
3827 | sh = get_active_stripe(conf, first_sector, |
3828 | conf->previous_raid_disks, pd_idx, 0); | |
3829 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
3830 | set_bit(STRIPE_HANDLE, &sh->state); | |
3831 | release_stripe(sh); | |
3832 | first_sector += STRIPE_SECTORS; | |
3833 | } | |
c6207277 N |
3834 | /* If this takes us to the resync_max point where we have to pause, |
3835 | * then we need to write out the superblock. | |
3836 | */ | |
3837 | sector_nr += conf->chunk_size>>9; | |
3838 | if (sector_nr >= mddev->resync_max) { | |
3839 | /* Cannot proceed until we've updated the superblock... */ | |
3840 | wait_event(conf->wait_for_overlap, | |
3841 | atomic_read(&conf->reshape_stripes) == 0); | |
3842 | mddev->reshape_position = conf->expand_progress; | |
3843 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
3844 | md_wakeup_thread(mddev->thread); | |
3845 | wait_event(mddev->sb_wait, | |
3846 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
3847 | || kthread_should_stop()); | |
3848 | spin_lock_irq(&conf->device_lock); | |
3849 | conf->expand_lo = mddev->reshape_position; | |
3850 | spin_unlock_irq(&conf->device_lock); | |
3851 | wake_up(&conf->wait_for_overlap); | |
3852 | } | |
52c03291 N |
3853 | return conf->chunk_size>>9; |
3854 | } | |
3855 | ||
3856 | /* FIXME go_faster isn't used */ | |
3857 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
3858 | { | |
3859 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
3860 | struct stripe_head *sh; | |
3861 | int pd_idx; | |
1da177e4 | 3862 | int raid_disks = conf->raid_disks; |
72626685 N |
3863 | sector_t max_sector = mddev->size << 1; |
3864 | int sync_blocks; | |
16a53ecc N |
3865 | int still_degraded = 0; |
3866 | int i; | |
1da177e4 | 3867 | |
72626685 | 3868 | if (sector_nr >= max_sector) { |
1da177e4 LT |
3869 | /* just being told to finish up .. nothing much to do */ |
3870 | unplug_slaves(mddev); | |
29269553 N |
3871 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
3872 | end_reshape(conf); | |
3873 | return 0; | |
3874 | } | |
72626685 N |
3875 | |
3876 | if (mddev->curr_resync < max_sector) /* aborted */ | |
3877 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
3878 | &sync_blocks, 1); | |
16a53ecc | 3879 | else /* completed sync */ |
72626685 N |
3880 | conf->fullsync = 0; |
3881 | bitmap_close_sync(mddev->bitmap); | |
3882 | ||
1da177e4 LT |
3883 | return 0; |
3884 | } | |
ccfcc3c1 | 3885 | |
52c03291 N |
3886 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3887 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 3888 | |
c6207277 N |
3889 | /* No need to check resync_max as we never do more than one |
3890 | * stripe, and as resync_max will always be on a chunk boundary, | |
3891 | * if the check in md_do_sync didn't fire, there is no chance | |
3892 | * of overstepping resync_max here | |
3893 | */ | |
3894 | ||
16a53ecc | 3895 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
3896 | * to resync, then assert that we are finished, because there is |
3897 | * nothing we can do. | |
3898 | */ | |
3285edf1 | 3899 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 3900 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
57afd89f N |
3901 | sector_t rv = (mddev->size << 1) - sector_nr; |
3902 | *skipped = 1; | |
1da177e4 LT |
3903 | return rv; |
3904 | } | |
72626685 | 3905 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 3906 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
3907 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
3908 | /* we can skip this block, and probably more */ | |
3909 | sync_blocks /= STRIPE_SECTORS; | |
3910 | *skipped = 1; | |
3911 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
3912 | } | |
1da177e4 | 3913 | |
b47490c9 N |
3914 | |
3915 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
3916 | ||
ccfcc3c1 | 3917 | pd_idx = stripe_to_pdidx(sector_nr, conf, raid_disks); |
7ecaa1e6 | 3918 | sh = get_active_stripe(conf, sector_nr, raid_disks, pd_idx, 1); |
1da177e4 | 3919 | if (sh == NULL) { |
7ecaa1e6 | 3920 | sh = get_active_stripe(conf, sector_nr, raid_disks, pd_idx, 0); |
1da177e4 | 3921 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 3922 | * is trying to get access |
1da177e4 | 3923 | */ |
66c006a5 | 3924 | schedule_timeout_uninterruptible(1); |
1da177e4 | 3925 | } |
16a53ecc N |
3926 | /* Need to check if array will still be degraded after recovery/resync |
3927 | * We don't need to check the 'failed' flag as when that gets set, | |
3928 | * recovery aborts. | |
3929 | */ | |
3930 | for (i=0; i<mddev->raid_disks; i++) | |
3931 | if (conf->disks[i].rdev == NULL) | |
3932 | still_degraded = 1; | |
3933 | ||
3934 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
3935 | ||
3936 | spin_lock(&sh->lock); | |
1da177e4 LT |
3937 | set_bit(STRIPE_SYNCING, &sh->state); |
3938 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3939 | spin_unlock(&sh->lock); | |
3940 | ||
16a53ecc | 3941 | handle_stripe(sh, NULL); |
1da177e4 LT |
3942 | release_stripe(sh); |
3943 | ||
3944 | return STRIPE_SECTORS; | |
3945 | } | |
3946 | ||
46031f9a RBJ |
3947 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
3948 | { | |
3949 | /* We may not be able to submit a whole bio at once as there | |
3950 | * may not be enough stripe_heads available. | |
3951 | * We cannot pre-allocate enough stripe_heads as we may need | |
3952 | * more than exist in the cache (if we allow ever large chunks). | |
3953 | * So we do one stripe head at a time and record in | |
3954 | * ->bi_hw_segments how many have been done. | |
3955 | * | |
3956 | * We *know* that this entire raid_bio is in one chunk, so | |
3957 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
3958 | */ | |
3959 | struct stripe_head *sh; | |
3960 | int dd_idx, pd_idx; | |
3961 | sector_t sector, logical_sector, last_sector; | |
3962 | int scnt = 0; | |
3963 | int remaining; | |
3964 | int handled = 0; | |
3965 | ||
3966 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
3967 | sector = raid5_compute_sector( logical_sector, | |
3968 | conf->raid_disks, | |
3969 | conf->raid_disks - conf->max_degraded, | |
3970 | &dd_idx, | |
3971 | &pd_idx, | |
3972 | conf); | |
3973 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); | |
3974 | ||
3975 | for (; logical_sector < last_sector; | |
387bb173 NB |
3976 | logical_sector += STRIPE_SECTORS, |
3977 | sector += STRIPE_SECTORS, | |
3978 | scnt++) { | |
46031f9a RBJ |
3979 | |
3980 | if (scnt < raid_bio->bi_hw_segments) | |
3981 | /* already done this stripe */ | |
3982 | continue; | |
3983 | ||
3984 | sh = get_active_stripe(conf, sector, conf->raid_disks, pd_idx, 1); | |
3985 | ||
3986 | if (!sh) { | |
3987 | /* failed to get a stripe - must wait */ | |
3988 | raid_bio->bi_hw_segments = scnt; | |
3989 | conf->retry_read_aligned = raid_bio; | |
3990 | return handled; | |
3991 | } | |
3992 | ||
3993 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
3994 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
3995 | release_stripe(sh); | |
3996 | raid_bio->bi_hw_segments = scnt; | |
3997 | conf->retry_read_aligned = raid_bio; | |
3998 | return handled; | |
3999 | } | |
4000 | ||
46031f9a RBJ |
4001 | handle_stripe(sh, NULL); |
4002 | release_stripe(sh); | |
4003 | handled++; | |
4004 | } | |
4005 | spin_lock_irq(&conf->device_lock); | |
4006 | remaining = --raid_bio->bi_phys_segments; | |
4007 | spin_unlock_irq(&conf->device_lock); | |
4008 | if (remaining == 0) { | |
46031f9a | 4009 | |
6712ecf8 | 4010 | raid_bio->bi_end_io(raid_bio, |
c2b00852 N |
4011 | test_bit(BIO_UPTODATE, &raid_bio->bi_flags) |
4012 | ? 0 : -EIO); | |
46031f9a RBJ |
4013 | } |
4014 | if (atomic_dec_and_test(&conf->active_aligned_reads)) | |
4015 | wake_up(&conf->wait_for_stripe); | |
4016 | return handled; | |
4017 | } | |
4018 | ||
4019 | ||
4020 | ||
1da177e4 LT |
4021 | /* |
4022 | * This is our raid5 kernel thread. | |
4023 | * | |
4024 | * We scan the hash table for stripes which can be handled now. | |
4025 | * During the scan, completed stripes are saved for us by the interrupt | |
4026 | * handler, so that they will not have to wait for our next wakeup. | |
4027 | */ | |
6ed3003c | 4028 | static void raid5d(mddev_t *mddev) |
1da177e4 LT |
4029 | { |
4030 | struct stripe_head *sh; | |
4031 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4032 | int handled; | |
4033 | ||
45b4233c | 4034 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4035 | |
4036 | md_check_recovery(mddev); | |
1da177e4 LT |
4037 | |
4038 | handled = 0; | |
4039 | spin_lock_irq(&conf->device_lock); | |
4040 | while (1) { | |
46031f9a | 4041 | struct bio *bio; |
1da177e4 | 4042 | |
ae3c20cc | 4043 | if (conf->seq_flush != conf->seq_write) { |
72626685 | 4044 | int seq = conf->seq_flush; |
700e432d | 4045 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4046 | bitmap_unplug(mddev->bitmap); |
700e432d | 4047 | spin_lock_irq(&conf->device_lock); |
72626685 N |
4048 | conf->seq_write = seq; |
4049 | activate_bit_delay(conf); | |
4050 | } | |
4051 | ||
46031f9a RBJ |
4052 | while ((bio = remove_bio_from_retry(conf))) { |
4053 | int ok; | |
4054 | spin_unlock_irq(&conf->device_lock); | |
4055 | ok = retry_aligned_read(conf, bio); | |
4056 | spin_lock_irq(&conf->device_lock); | |
4057 | if (!ok) | |
4058 | break; | |
4059 | handled++; | |
4060 | } | |
4061 | ||
8b3e6cdc DW |
4062 | sh = __get_priority_stripe(conf); |
4063 | ||
4064 | if (!sh) { | |
d84e0f10 | 4065 | async_tx_issue_pending_all(); |
1da177e4 | 4066 | break; |
d84e0f10 | 4067 | } |
1da177e4 LT |
4068 | spin_unlock_irq(&conf->device_lock); |
4069 | ||
4070 | handled++; | |
16a53ecc | 4071 | handle_stripe(sh, conf->spare_page); |
1da177e4 LT |
4072 | release_stripe(sh); |
4073 | ||
4074 | spin_lock_irq(&conf->device_lock); | |
4075 | } | |
45b4233c | 4076 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4077 | |
4078 | spin_unlock_irq(&conf->device_lock); | |
4079 | ||
4080 | unplug_slaves(mddev); | |
4081 | ||
45b4233c | 4082 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4083 | } |
4084 | ||
3f294f4f | 4085 | static ssize_t |
007583c9 | 4086 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 4087 | { |
007583c9 | 4088 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
4089 | if (conf) |
4090 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4091 | else | |
4092 | return 0; | |
3f294f4f N |
4093 | } |
4094 | ||
4095 | static ssize_t | |
007583c9 | 4096 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) |
3f294f4f | 4097 | { |
007583c9 | 4098 | raid5_conf_t *conf = mddev_to_conf(mddev); |
4ef197d8 | 4099 | unsigned long new; |
3f294f4f N |
4100 | if (len >= PAGE_SIZE) |
4101 | return -EINVAL; | |
96de1e66 N |
4102 | if (!conf) |
4103 | return -ENODEV; | |
3f294f4f | 4104 | |
4ef197d8 | 4105 | if (strict_strtoul(page, 10, &new)) |
3f294f4f N |
4106 | return -EINVAL; |
4107 | if (new <= 16 || new > 32768) | |
4108 | return -EINVAL; | |
4109 | while (new < conf->max_nr_stripes) { | |
4110 | if (drop_one_stripe(conf)) | |
4111 | conf->max_nr_stripes--; | |
4112 | else | |
4113 | break; | |
4114 | } | |
2a2275d6 | 4115 | md_allow_write(mddev); |
3f294f4f N |
4116 | while (new > conf->max_nr_stripes) { |
4117 | if (grow_one_stripe(conf)) | |
4118 | conf->max_nr_stripes++; | |
4119 | else break; | |
4120 | } | |
4121 | return len; | |
4122 | } | |
007583c9 | 4123 | |
96de1e66 N |
4124 | static struct md_sysfs_entry |
4125 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4126 | raid5_show_stripe_cache_size, | |
4127 | raid5_store_stripe_cache_size); | |
3f294f4f | 4128 | |
8b3e6cdc DW |
4129 | static ssize_t |
4130 | raid5_show_preread_threshold(mddev_t *mddev, char *page) | |
4131 | { | |
4132 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4133 | if (conf) | |
4134 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4135 | else | |
4136 | return 0; | |
4137 | } | |
4138 | ||
4139 | static ssize_t | |
4140 | raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len) | |
4141 | { | |
4142 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4ef197d8 | 4143 | unsigned long new; |
8b3e6cdc DW |
4144 | if (len >= PAGE_SIZE) |
4145 | return -EINVAL; | |
4146 | if (!conf) | |
4147 | return -ENODEV; | |
4148 | ||
4ef197d8 | 4149 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4150 | return -EINVAL; |
4ef197d8 | 4151 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4152 | return -EINVAL; |
4153 | conf->bypass_threshold = new; | |
4154 | return len; | |
4155 | } | |
4156 | ||
4157 | static struct md_sysfs_entry | |
4158 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4159 | S_IRUGO | S_IWUSR, | |
4160 | raid5_show_preread_threshold, | |
4161 | raid5_store_preread_threshold); | |
4162 | ||
3f294f4f | 4163 | static ssize_t |
96de1e66 | 4164 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 4165 | { |
007583c9 | 4166 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
4167 | if (conf) |
4168 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4169 | else | |
4170 | return 0; | |
3f294f4f N |
4171 | } |
4172 | ||
96de1e66 N |
4173 | static struct md_sysfs_entry |
4174 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4175 | |
007583c9 | 4176 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4177 | &raid5_stripecache_size.attr, |
4178 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4179 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4180 | NULL, |
4181 | }; | |
007583c9 N |
4182 | static struct attribute_group raid5_attrs_group = { |
4183 | .name = NULL, | |
4184 | .attrs = raid5_attrs, | |
3f294f4f N |
4185 | }; |
4186 | ||
72626685 | 4187 | static int run(mddev_t *mddev) |
1da177e4 LT |
4188 | { |
4189 | raid5_conf_t *conf; | |
4190 | int raid_disk, memory; | |
4191 | mdk_rdev_t *rdev; | |
4192 | struct disk_info *disk; | |
4193 | struct list_head *tmp; | |
02c2de8c | 4194 | int working_disks = 0; |
1da177e4 | 4195 | |
16a53ecc N |
4196 | if (mddev->level != 5 && mddev->level != 4 && mddev->level != 6) { |
4197 | printk(KERN_ERR "raid5: %s: raid level not set to 4/5/6 (%d)\n", | |
14f8d26b | 4198 | mdname(mddev), mddev->level); |
1da177e4 LT |
4199 | return -EIO; |
4200 | } | |
4201 | ||
f6705578 N |
4202 | if (mddev->reshape_position != MaxSector) { |
4203 | /* Check that we can continue the reshape. | |
4204 | * Currently only disks can change, it must | |
4205 | * increase, and we must be past the point where | |
4206 | * a stripe over-writes itself | |
4207 | */ | |
4208 | sector_t here_new, here_old; | |
4209 | int old_disks; | |
f416885e | 4210 | int max_degraded = (mddev->level == 5 ? 1 : 2); |
f6705578 N |
4211 | |
4212 | if (mddev->new_level != mddev->level || | |
4213 | mddev->new_layout != mddev->layout || | |
4214 | mddev->new_chunk != mddev->chunk_size) { | |
f416885e N |
4215 | printk(KERN_ERR "raid5: %s: unsupported reshape " |
4216 | "required - aborting.\n", | |
f6705578 N |
4217 | mdname(mddev)); |
4218 | return -EINVAL; | |
4219 | } | |
4220 | if (mddev->delta_disks <= 0) { | |
f416885e N |
4221 | printk(KERN_ERR "raid5: %s: unsupported reshape " |
4222 | "(reduce disks) required - aborting.\n", | |
f6705578 N |
4223 | mdname(mddev)); |
4224 | return -EINVAL; | |
4225 | } | |
4226 | old_disks = mddev->raid_disks - mddev->delta_disks; | |
4227 | /* reshape_position must be on a new-stripe boundary, and one | |
f416885e N |
4228 | * further up in new geometry must map after here in old |
4229 | * geometry. | |
f6705578 N |
4230 | */ |
4231 | here_new = mddev->reshape_position; | |
f416885e N |
4232 | if (sector_div(here_new, (mddev->chunk_size>>9)* |
4233 | (mddev->raid_disks - max_degraded))) { | |
4234 | printk(KERN_ERR "raid5: reshape_position not " | |
4235 | "on a stripe boundary\n"); | |
f6705578 N |
4236 | return -EINVAL; |
4237 | } | |
4238 | /* here_new is the stripe we will write to */ | |
4239 | here_old = mddev->reshape_position; | |
f416885e N |
4240 | sector_div(here_old, (mddev->chunk_size>>9)* |
4241 | (old_disks-max_degraded)); | |
4242 | /* here_old is the first stripe that we might need to read | |
4243 | * from */ | |
f6705578 N |
4244 | if (here_new >= here_old) { |
4245 | /* Reading from the same stripe as writing to - bad */ | |
f416885e N |
4246 | printk(KERN_ERR "raid5: reshape_position too early for " |
4247 | "auto-recovery - aborting.\n"); | |
f6705578 N |
4248 | return -EINVAL; |
4249 | } | |
4250 | printk(KERN_INFO "raid5: reshape will continue\n"); | |
4251 | /* OK, we should be able to continue; */ | |
4252 | } | |
4253 | ||
4254 | ||
b55e6bfc | 4255 | mddev->private = kzalloc(sizeof (raid5_conf_t), GFP_KERNEL); |
1da177e4 LT |
4256 | if ((conf = mddev->private) == NULL) |
4257 | goto abort; | |
f6705578 N |
4258 | if (mddev->reshape_position == MaxSector) { |
4259 | conf->previous_raid_disks = conf->raid_disks = mddev->raid_disks; | |
4260 | } else { | |
4261 | conf->raid_disks = mddev->raid_disks; | |
4262 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; | |
4263 | } | |
4264 | ||
4265 | conf->disks = kzalloc(conf->raid_disks * sizeof(struct disk_info), | |
b55e6bfc N |
4266 | GFP_KERNEL); |
4267 | if (!conf->disks) | |
4268 | goto abort; | |
9ffae0cf | 4269 | |
1da177e4 LT |
4270 | conf->mddev = mddev; |
4271 | ||
fccddba0 | 4272 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4273 | goto abort; |
1da177e4 | 4274 | |
16a53ecc N |
4275 | if (mddev->level == 6) { |
4276 | conf->spare_page = alloc_page(GFP_KERNEL); | |
4277 | if (!conf->spare_page) | |
4278 | goto abort; | |
4279 | } | |
1da177e4 | 4280 | spin_lock_init(&conf->device_lock); |
e7e72bf6 | 4281 | mddev->queue->queue_lock = &conf->device_lock; |
1da177e4 LT |
4282 | init_waitqueue_head(&conf->wait_for_stripe); |
4283 | init_waitqueue_head(&conf->wait_for_overlap); | |
4284 | INIT_LIST_HEAD(&conf->handle_list); | |
8b3e6cdc | 4285 | INIT_LIST_HEAD(&conf->hold_list); |
1da177e4 | 4286 | INIT_LIST_HEAD(&conf->delayed_list); |
72626685 | 4287 | INIT_LIST_HEAD(&conf->bitmap_list); |
1da177e4 LT |
4288 | INIT_LIST_HEAD(&conf->inactive_list); |
4289 | atomic_set(&conf->active_stripes, 0); | |
4290 | atomic_set(&conf->preread_active_stripes, 0); | |
46031f9a | 4291 | atomic_set(&conf->active_aligned_reads, 0); |
8b3e6cdc | 4292 | conf->bypass_threshold = BYPASS_THRESHOLD; |
1da177e4 | 4293 | |
45b4233c | 4294 | pr_debug("raid5: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4295 | |
d089c6af | 4296 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 | 4297 | raid_disk = rdev->raid_disk; |
f6705578 | 4298 | if (raid_disk >= conf->raid_disks |
1da177e4 LT |
4299 | || raid_disk < 0) |
4300 | continue; | |
4301 | disk = conf->disks + raid_disk; | |
4302 | ||
4303 | disk->rdev = rdev; | |
4304 | ||
b2d444d7 | 4305 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4306 | char b[BDEVNAME_SIZE]; |
4307 | printk(KERN_INFO "raid5: device %s operational as raid" | |
4308 | " disk %d\n", bdevname(rdev->bdev,b), | |
4309 | raid_disk); | |
02c2de8c | 4310 | working_disks++; |
8c2e870a NB |
4311 | } else |
4312 | /* Cannot rely on bitmap to complete recovery */ | |
4313 | conf->fullsync = 1; | |
1da177e4 LT |
4314 | } |
4315 | ||
1da177e4 | 4316 | /* |
16a53ecc | 4317 | * 0 for a fully functional array, 1 or 2 for a degraded array. |
1da177e4 | 4318 | */ |
02c2de8c | 4319 | mddev->degraded = conf->raid_disks - working_disks; |
1da177e4 LT |
4320 | conf->mddev = mddev; |
4321 | conf->chunk_size = mddev->chunk_size; | |
4322 | conf->level = mddev->level; | |
16a53ecc N |
4323 | if (conf->level == 6) |
4324 | conf->max_degraded = 2; | |
4325 | else | |
4326 | conf->max_degraded = 1; | |
1da177e4 LT |
4327 | conf->algorithm = mddev->layout; |
4328 | conf->max_nr_stripes = NR_STRIPES; | |
f6705578 | 4329 | conf->expand_progress = mddev->reshape_position; |
1da177e4 LT |
4330 | |
4331 | /* device size must be a multiple of chunk size */ | |
4332 | mddev->size &= ~(mddev->chunk_size/1024 -1); | |
b1581566 | 4333 | mddev->resync_max_sectors = mddev->size << 1; |
1da177e4 | 4334 | |
16a53ecc N |
4335 | if (conf->level == 6 && conf->raid_disks < 4) { |
4336 | printk(KERN_ERR "raid6: not enough configured devices for %s (%d, minimum 4)\n", | |
4337 | mdname(mddev), conf->raid_disks); | |
4338 | goto abort; | |
4339 | } | |
1da177e4 LT |
4340 | if (!conf->chunk_size || conf->chunk_size % 4) { |
4341 | printk(KERN_ERR "raid5: invalid chunk size %d for %s\n", | |
4342 | conf->chunk_size, mdname(mddev)); | |
4343 | goto abort; | |
4344 | } | |
4345 | if (conf->algorithm > ALGORITHM_RIGHT_SYMMETRIC) { | |
4346 | printk(KERN_ERR | |
4347 | "raid5: unsupported parity algorithm %d for %s\n", | |
4348 | conf->algorithm, mdname(mddev)); | |
4349 | goto abort; | |
4350 | } | |
16a53ecc | 4351 | if (mddev->degraded > conf->max_degraded) { |
1da177e4 LT |
4352 | printk(KERN_ERR "raid5: not enough operational devices for %s" |
4353 | " (%d/%d failed)\n", | |
02c2de8c | 4354 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
4355 | goto abort; |
4356 | } | |
4357 | ||
16a53ecc | 4358 | if (mddev->degraded > 0 && |
1da177e4 | 4359 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
4360 | if (mddev->ok_start_degraded) |
4361 | printk(KERN_WARNING | |
4362 | "raid5: starting dirty degraded array: %s" | |
4363 | "- data corruption possible.\n", | |
4364 | mdname(mddev)); | |
4365 | else { | |
4366 | printk(KERN_ERR | |
4367 | "raid5: cannot start dirty degraded array for %s\n", | |
4368 | mdname(mddev)); | |
4369 | goto abort; | |
4370 | } | |
1da177e4 LT |
4371 | } |
4372 | ||
4373 | { | |
4374 | mddev->thread = md_register_thread(raid5d, mddev, "%s_raid5"); | |
4375 | if (!mddev->thread) { | |
4376 | printk(KERN_ERR | |
4377 | "raid5: couldn't allocate thread for %s\n", | |
4378 | mdname(mddev)); | |
4379 | goto abort; | |
4380 | } | |
4381 | } | |
5036805b | 4382 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
1da177e4 LT |
4383 | conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4384 | if (grow_stripes(conf, conf->max_nr_stripes)) { | |
4385 | printk(KERN_ERR | |
4386 | "raid5: couldn't allocate %dkB for buffers\n", memory); | |
4387 | shrink_stripes(conf); | |
4388 | md_unregister_thread(mddev->thread); | |
4389 | goto abort; | |
4390 | } else | |
4391 | printk(KERN_INFO "raid5: allocated %dkB for %s\n", | |
4392 | memory, mdname(mddev)); | |
4393 | ||
4394 | if (mddev->degraded == 0) | |
4395 | printk("raid5: raid level %d set %s active with %d out of %d" | |
4396 | " devices, algorithm %d\n", conf->level, mdname(mddev), | |
4397 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
4398 | conf->algorithm); | |
4399 | else | |
4400 | printk(KERN_ALERT "raid5: raid level %d set %s active with %d" | |
4401 | " out of %d devices, algorithm %d\n", conf->level, | |
4402 | mdname(mddev), mddev->raid_disks - mddev->degraded, | |
4403 | mddev->raid_disks, conf->algorithm); | |
4404 | ||
4405 | print_raid5_conf(conf); | |
4406 | ||
f6705578 N |
4407 | if (conf->expand_progress != MaxSector) { |
4408 | printk("...ok start reshape thread\n"); | |
b578d55f | 4409 | conf->expand_lo = conf->expand_progress; |
f6705578 N |
4410 | atomic_set(&conf->reshape_stripes, 0); |
4411 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4412 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4413 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4414 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4415 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4416 | "%s_reshape"); | |
f6705578 N |
4417 | } |
4418 | ||
1da177e4 | 4419 | /* read-ahead size must cover two whole stripes, which is |
16a53ecc | 4420 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices |
1da177e4 LT |
4421 | */ |
4422 | { | |
16a53ecc N |
4423 | int data_disks = conf->previous_raid_disks - conf->max_degraded; |
4424 | int stripe = data_disks * | |
8932c2e0 | 4425 | (mddev->chunk_size / PAGE_SIZE); |
1da177e4 LT |
4426 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
4427 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4428 | } | |
4429 | ||
4430 | /* Ok, everything is just fine now */ | |
5e55e2f5 N |
4431 | if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) |
4432 | printk(KERN_WARNING | |
4433 | "raid5: failed to create sysfs attributes for %s\n", | |
4434 | mdname(mddev)); | |
7a5febe9 N |
4435 | |
4436 | mddev->queue->unplug_fn = raid5_unplug_device; | |
f022b2fd | 4437 | mddev->queue->backing_dev_info.congested_data = mddev; |
041ae52e | 4438 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; |
f022b2fd | 4439 | |
16a53ecc N |
4440 | mddev->array_size = mddev->size * (conf->previous_raid_disks - |
4441 | conf->max_degraded); | |
7a5febe9 | 4442 | |
23032a0e RBJ |
4443 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
4444 | ||
1da177e4 LT |
4445 | return 0; |
4446 | abort: | |
4447 | if (conf) { | |
4448 | print_raid5_conf(conf); | |
16a53ecc | 4449 | safe_put_page(conf->spare_page); |
b55e6bfc | 4450 | kfree(conf->disks); |
fccddba0 | 4451 | kfree(conf->stripe_hashtbl); |
1da177e4 LT |
4452 | kfree(conf); |
4453 | } | |
4454 | mddev->private = NULL; | |
4455 | printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev)); | |
4456 | return -EIO; | |
4457 | } | |
4458 | ||
4459 | ||
4460 | ||
3f294f4f | 4461 | static int stop(mddev_t *mddev) |
1da177e4 LT |
4462 | { |
4463 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4464 | ||
4465 | md_unregister_thread(mddev->thread); | |
4466 | mddev->thread = NULL; | |
4467 | shrink_stripes(conf); | |
fccddba0 | 4468 | kfree(conf->stripe_hashtbl); |
041ae52e | 4469 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 4470 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
007583c9 | 4471 | sysfs_remove_group(&mddev->kobj, &raid5_attrs_group); |
b55e6bfc | 4472 | kfree(conf->disks); |
96de1e66 | 4473 | kfree(conf); |
1da177e4 LT |
4474 | mddev->private = NULL; |
4475 | return 0; | |
4476 | } | |
4477 | ||
45b4233c | 4478 | #ifdef DEBUG |
16a53ecc | 4479 | static void print_sh (struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
4480 | { |
4481 | int i; | |
4482 | ||
16a53ecc N |
4483 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
4484 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
4485 | seq_printf(seq, "sh %llu, count %d.\n", | |
4486 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
4487 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 4488 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
4489 | seq_printf(seq, "(cache%d: %p %ld) ", |
4490 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 4491 | } |
16a53ecc | 4492 | seq_printf(seq, "\n"); |
1da177e4 LT |
4493 | } |
4494 | ||
16a53ecc | 4495 | static void printall (struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
4496 | { |
4497 | struct stripe_head *sh; | |
fccddba0 | 4498 | struct hlist_node *hn; |
1da177e4 LT |
4499 | int i; |
4500 | ||
4501 | spin_lock_irq(&conf->device_lock); | |
4502 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 4503 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
4504 | if (sh->raid_conf != conf) |
4505 | continue; | |
16a53ecc | 4506 | print_sh(seq, sh); |
1da177e4 LT |
4507 | } |
4508 | } | |
4509 | spin_unlock_irq(&conf->device_lock); | |
4510 | } | |
4511 | #endif | |
4512 | ||
4513 | static void status (struct seq_file *seq, mddev_t *mddev) | |
4514 | { | |
4515 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4516 | int i; | |
4517 | ||
4518 | seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout); | |
02c2de8c | 4519 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
4520 | for (i = 0; i < conf->raid_disks; i++) |
4521 | seq_printf (seq, "%s", | |
4522 | conf->disks[i].rdev && | |
b2d444d7 | 4523 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 4524 | seq_printf (seq, "]"); |
45b4233c | 4525 | #ifdef DEBUG |
16a53ecc N |
4526 | seq_printf (seq, "\n"); |
4527 | printall(seq, conf); | |
1da177e4 LT |
4528 | #endif |
4529 | } | |
4530 | ||
4531 | static void print_raid5_conf (raid5_conf_t *conf) | |
4532 | { | |
4533 | int i; | |
4534 | struct disk_info *tmp; | |
4535 | ||
4536 | printk("RAID5 conf printout:\n"); | |
4537 | if (!conf) { | |
4538 | printk("(conf==NULL)\n"); | |
4539 | return; | |
4540 | } | |
02c2de8c N |
4541 | printk(" --- rd:%d wd:%d\n", conf->raid_disks, |
4542 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
4543 | |
4544 | for (i = 0; i < conf->raid_disks; i++) { | |
4545 | char b[BDEVNAME_SIZE]; | |
4546 | tmp = conf->disks + i; | |
4547 | if (tmp->rdev) | |
4548 | printk(" disk %d, o:%d, dev:%s\n", | |
b2d444d7 | 4549 | i, !test_bit(Faulty, &tmp->rdev->flags), |
1da177e4 LT |
4550 | bdevname(tmp->rdev->bdev,b)); |
4551 | } | |
4552 | } | |
4553 | ||
4554 | static int raid5_spare_active(mddev_t *mddev) | |
4555 | { | |
4556 | int i; | |
4557 | raid5_conf_t *conf = mddev->private; | |
4558 | struct disk_info *tmp; | |
4559 | ||
4560 | for (i = 0; i < conf->raid_disks; i++) { | |
4561 | tmp = conf->disks + i; | |
4562 | if (tmp->rdev | |
b2d444d7 | 4563 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa N |
4564 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
4565 | unsigned long flags; | |
4566 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 4567 | mddev->degraded--; |
c04be0aa | 4568 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
4569 | } |
4570 | } | |
4571 | print_raid5_conf(conf); | |
4572 | return 0; | |
4573 | } | |
4574 | ||
4575 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
4576 | { | |
4577 | raid5_conf_t *conf = mddev->private; | |
4578 | int err = 0; | |
4579 | mdk_rdev_t *rdev; | |
4580 | struct disk_info *p = conf->disks + number; | |
4581 | ||
4582 | print_raid5_conf(conf); | |
4583 | rdev = p->rdev; | |
4584 | if (rdev) { | |
b2d444d7 | 4585 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
4586 | atomic_read(&rdev->nr_pending)) { |
4587 | err = -EBUSY; | |
4588 | goto abort; | |
4589 | } | |
dfc70645 N |
4590 | /* Only remove non-faulty devices if recovery |
4591 | * isn't possible. | |
4592 | */ | |
4593 | if (!test_bit(Faulty, &rdev->flags) && | |
4594 | mddev->degraded <= conf->max_degraded) { | |
4595 | err = -EBUSY; | |
4596 | goto abort; | |
4597 | } | |
1da177e4 | 4598 | p->rdev = NULL; |
fbd568a3 | 4599 | synchronize_rcu(); |
1da177e4 LT |
4600 | if (atomic_read(&rdev->nr_pending)) { |
4601 | /* lost the race, try later */ | |
4602 | err = -EBUSY; | |
4603 | p->rdev = rdev; | |
4604 | } | |
4605 | } | |
4606 | abort: | |
4607 | ||
4608 | print_raid5_conf(conf); | |
4609 | return err; | |
4610 | } | |
4611 | ||
4612 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
4613 | { | |
4614 | raid5_conf_t *conf = mddev->private; | |
4615 | int found = 0; | |
4616 | int disk; | |
4617 | struct disk_info *p; | |
4618 | ||
16a53ecc | 4619 | if (mddev->degraded > conf->max_degraded) |
1da177e4 LT |
4620 | /* no point adding a device */ |
4621 | return 0; | |
4622 | ||
4623 | /* | |
16a53ecc N |
4624 | * find the disk ... but prefer rdev->saved_raid_disk |
4625 | * if possible. | |
1da177e4 | 4626 | */ |
16a53ecc N |
4627 | if (rdev->saved_raid_disk >= 0 && |
4628 | conf->disks[rdev->saved_raid_disk].rdev == NULL) | |
4629 | disk = rdev->saved_raid_disk; | |
4630 | else | |
4631 | disk = 0; | |
4632 | for ( ; disk < conf->raid_disks; disk++) | |
1da177e4 | 4633 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 4634 | clear_bit(In_sync, &rdev->flags); |
1da177e4 LT |
4635 | rdev->raid_disk = disk; |
4636 | found = 1; | |
72626685 N |
4637 | if (rdev->saved_raid_disk != disk) |
4638 | conf->fullsync = 1; | |
d6065f7b | 4639 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
4640 | break; |
4641 | } | |
4642 | print_raid5_conf(conf); | |
4643 | return found; | |
4644 | } | |
4645 | ||
4646 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
4647 | { | |
4648 | /* no resync is happening, and there is enough space | |
4649 | * on all devices, so we can resize. | |
4650 | * We need to make sure resync covers any new space. | |
4651 | * If the array is shrinking we should possibly wait until | |
4652 | * any io in the removed space completes, but it hardly seems | |
4653 | * worth it. | |
4654 | */ | |
16a53ecc N |
4655 | raid5_conf_t *conf = mddev_to_conf(mddev); |
4656 | ||
1da177e4 | 4657 | sectors &= ~((sector_t)mddev->chunk_size/512 - 1); |
16a53ecc | 4658 | mddev->array_size = (sectors * (mddev->raid_disks-conf->max_degraded))>>1; |
1da177e4 | 4659 | set_capacity(mddev->gendisk, mddev->array_size << 1); |
44ce6294 | 4660 | mddev->changed = 1; |
1da177e4 LT |
4661 | if (sectors/2 > mddev->size && mddev->recovery_cp == MaxSector) { |
4662 | mddev->recovery_cp = mddev->size << 1; | |
4663 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4664 | } | |
4665 | mddev->size = sectors /2; | |
4b5c7ae8 | 4666 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
4667 | return 0; |
4668 | } | |
4669 | ||
29269553 | 4670 | #ifdef CONFIG_MD_RAID5_RESHAPE |
63c70c4f | 4671 | static int raid5_check_reshape(mddev_t *mddev) |
29269553 N |
4672 | { |
4673 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4674 | int err; | |
29269553 | 4675 | |
63c70c4f N |
4676 | if (mddev->delta_disks < 0 || |
4677 | mddev->new_level != mddev->level) | |
4678 | return -EINVAL; /* Cannot shrink array or change level yet */ | |
4679 | if (mddev->delta_disks == 0) | |
29269553 N |
4680 | return 0; /* nothing to do */ |
4681 | ||
4682 | /* Can only proceed if there are plenty of stripe_heads. | |
4683 | * We need a minimum of one full stripe,, and for sensible progress | |
4684 | * it is best to have about 4 times that. | |
4685 | * If we require 4 times, then the default 256 4K stripe_heads will | |
4686 | * allow for chunk sizes up to 256K, which is probably OK. | |
4687 | * If the chunk size is greater, user-space should request more | |
4688 | * stripe_heads first. | |
4689 | */ | |
63c70c4f N |
4690 | if ((mddev->chunk_size / STRIPE_SIZE) * 4 > conf->max_nr_stripes || |
4691 | (mddev->new_chunk / STRIPE_SIZE) * 4 > conf->max_nr_stripes) { | |
29269553 N |
4692 | printk(KERN_WARNING "raid5: reshape: not enough stripes. Needed %lu\n", |
4693 | (mddev->chunk_size / STRIPE_SIZE)*4); | |
4694 | return -ENOSPC; | |
4695 | } | |
4696 | ||
63c70c4f N |
4697 | err = resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
4698 | if (err) | |
4699 | return err; | |
4700 | ||
b4c4c7b8 N |
4701 | if (mddev->degraded > conf->max_degraded) |
4702 | return -EINVAL; | |
63c70c4f N |
4703 | /* looks like we might be able to manage this */ |
4704 | return 0; | |
4705 | } | |
4706 | ||
4707 | static int raid5_start_reshape(mddev_t *mddev) | |
4708 | { | |
4709 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4710 | mdk_rdev_t *rdev; | |
4711 | struct list_head *rtmp; | |
4712 | int spares = 0; | |
4713 | int added_devices = 0; | |
c04be0aa | 4714 | unsigned long flags; |
63c70c4f | 4715 | |
f416885e | 4716 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
4717 | return -EBUSY; |
4718 | ||
d089c6af | 4719 | rdev_for_each(rdev, rtmp, mddev) |
29269553 N |
4720 | if (rdev->raid_disk < 0 && |
4721 | !test_bit(Faulty, &rdev->flags)) | |
4722 | spares++; | |
63c70c4f | 4723 | |
f416885e | 4724 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
4725 | /* Not enough devices even to make a degraded array |
4726 | * of that size | |
4727 | */ | |
4728 | return -EINVAL; | |
4729 | ||
f6705578 | 4730 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
4731 | spin_lock_irq(&conf->device_lock); |
4732 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 4733 | conf->raid_disks += mddev->delta_disks; |
29269553 | 4734 | conf->expand_progress = 0; |
b578d55f | 4735 | conf->expand_lo = 0; |
29269553 N |
4736 | spin_unlock_irq(&conf->device_lock); |
4737 | ||
4738 | /* Add some new drives, as many as will fit. | |
4739 | * We know there are enough to make the newly sized array work. | |
4740 | */ | |
d089c6af | 4741 | rdev_for_each(rdev, rtmp, mddev) |
29269553 N |
4742 | if (rdev->raid_disk < 0 && |
4743 | !test_bit(Faulty, &rdev->flags)) { | |
4744 | if (raid5_add_disk(mddev, rdev)) { | |
4745 | char nm[20]; | |
4746 | set_bit(In_sync, &rdev->flags); | |
29269553 | 4747 | added_devices++; |
5fd6c1dc | 4748 | rdev->recovery_offset = 0; |
29269553 | 4749 | sprintf(nm, "rd%d", rdev->raid_disk); |
5e55e2f5 N |
4750 | if (sysfs_create_link(&mddev->kobj, |
4751 | &rdev->kobj, nm)) | |
4752 | printk(KERN_WARNING | |
4753 | "raid5: failed to create " | |
4754 | " link %s for %s\n", | |
4755 | nm, mdname(mddev)); | |
29269553 N |
4756 | } else |
4757 | break; | |
4758 | } | |
4759 | ||
c04be0aa | 4760 | spin_lock_irqsave(&conf->device_lock, flags); |
63c70c4f | 4761 | mddev->degraded = (conf->raid_disks - conf->previous_raid_disks) - added_devices; |
c04be0aa | 4762 | spin_unlock_irqrestore(&conf->device_lock, flags); |
63c70c4f | 4763 | mddev->raid_disks = conf->raid_disks; |
f6705578 | 4764 | mddev->reshape_position = 0; |
850b2b42 | 4765 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 4766 | |
29269553 N |
4767 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
4768 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4769 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4770 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4771 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4772 | "%s_reshape"); | |
4773 | if (!mddev->sync_thread) { | |
4774 | mddev->recovery = 0; | |
4775 | spin_lock_irq(&conf->device_lock); | |
4776 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
4777 | conf->expand_progress = MaxSector; | |
4778 | spin_unlock_irq(&conf->device_lock); | |
4779 | return -EAGAIN; | |
4780 | } | |
4781 | md_wakeup_thread(mddev->sync_thread); | |
4782 | md_new_event(mddev); | |
4783 | return 0; | |
4784 | } | |
4785 | #endif | |
4786 | ||
4787 | static void end_reshape(raid5_conf_t *conf) | |
4788 | { | |
4789 | struct block_device *bdev; | |
4790 | ||
f6705578 | 4791 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f416885e N |
4792 | conf->mddev->array_size = conf->mddev->size * |
4793 | (conf->raid_disks - conf->max_degraded); | |
f6705578 | 4794 | set_capacity(conf->mddev->gendisk, conf->mddev->array_size << 1); |
44ce6294 | 4795 | conf->mddev->changed = 1; |
f6705578 N |
4796 | |
4797 | bdev = bdget_disk(conf->mddev->gendisk, 0); | |
4798 | if (bdev) { | |
4799 | mutex_lock(&bdev->bd_inode->i_mutex); | |
0692c6b1 | 4800 | i_size_write(bdev->bd_inode, (loff_t)conf->mddev->array_size << 10); |
f6705578 N |
4801 | mutex_unlock(&bdev->bd_inode->i_mutex); |
4802 | bdput(bdev); | |
4803 | } | |
4804 | spin_lock_irq(&conf->device_lock); | |
4805 | conf->expand_progress = MaxSector; | |
4806 | spin_unlock_irq(&conf->device_lock); | |
4807 | conf->mddev->reshape_position = MaxSector; | |
16a53ecc N |
4808 | |
4809 | /* read-ahead size must cover two whole stripes, which is | |
4810 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
4811 | */ | |
4812 | { | |
4813 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
4814 | int stripe = data_disks * | |
4815 | (conf->mddev->chunk_size / PAGE_SIZE); | |
4816 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
4817 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4818 | } | |
29269553 | 4819 | } |
29269553 N |
4820 | } |
4821 | ||
72626685 N |
4822 | static void raid5_quiesce(mddev_t *mddev, int state) |
4823 | { | |
4824 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4825 | ||
4826 | switch(state) { | |
e464eafd N |
4827 | case 2: /* resume for a suspend */ |
4828 | wake_up(&conf->wait_for_overlap); | |
4829 | break; | |
4830 | ||
72626685 N |
4831 | case 1: /* stop all writes */ |
4832 | spin_lock_irq(&conf->device_lock); | |
4833 | conf->quiesce = 1; | |
4834 | wait_event_lock_irq(conf->wait_for_stripe, | |
46031f9a RBJ |
4835 | atomic_read(&conf->active_stripes) == 0 && |
4836 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 N |
4837 | conf->device_lock, /* nothing */); |
4838 | spin_unlock_irq(&conf->device_lock); | |
4839 | break; | |
4840 | ||
4841 | case 0: /* re-enable writes */ | |
4842 | spin_lock_irq(&conf->device_lock); | |
4843 | conf->quiesce = 0; | |
4844 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 4845 | wake_up(&conf->wait_for_overlap); |
72626685 N |
4846 | spin_unlock_irq(&conf->device_lock); |
4847 | break; | |
4848 | } | |
72626685 | 4849 | } |
b15c2e57 | 4850 | |
16a53ecc N |
4851 | static struct mdk_personality raid6_personality = |
4852 | { | |
4853 | .name = "raid6", | |
4854 | .level = 6, | |
4855 | .owner = THIS_MODULE, | |
4856 | .make_request = make_request, | |
4857 | .run = run, | |
4858 | .stop = stop, | |
4859 | .status = status, | |
4860 | .error_handler = error, | |
4861 | .hot_add_disk = raid5_add_disk, | |
4862 | .hot_remove_disk= raid5_remove_disk, | |
4863 | .spare_active = raid5_spare_active, | |
4864 | .sync_request = sync_request, | |
4865 | .resize = raid5_resize, | |
f416885e N |
4866 | #ifdef CONFIG_MD_RAID5_RESHAPE |
4867 | .check_reshape = raid5_check_reshape, | |
4868 | .start_reshape = raid5_start_reshape, | |
4869 | #endif | |
16a53ecc N |
4870 | .quiesce = raid5_quiesce, |
4871 | }; | |
2604b703 | 4872 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
4873 | { |
4874 | .name = "raid5", | |
2604b703 | 4875 | .level = 5, |
1da177e4 LT |
4876 | .owner = THIS_MODULE, |
4877 | .make_request = make_request, | |
4878 | .run = run, | |
4879 | .stop = stop, | |
4880 | .status = status, | |
4881 | .error_handler = error, | |
4882 | .hot_add_disk = raid5_add_disk, | |
4883 | .hot_remove_disk= raid5_remove_disk, | |
4884 | .spare_active = raid5_spare_active, | |
4885 | .sync_request = sync_request, | |
4886 | .resize = raid5_resize, | |
29269553 | 4887 | #ifdef CONFIG_MD_RAID5_RESHAPE |
63c70c4f N |
4888 | .check_reshape = raid5_check_reshape, |
4889 | .start_reshape = raid5_start_reshape, | |
29269553 | 4890 | #endif |
72626685 | 4891 | .quiesce = raid5_quiesce, |
1da177e4 LT |
4892 | }; |
4893 | ||
2604b703 | 4894 | static struct mdk_personality raid4_personality = |
1da177e4 | 4895 | { |
2604b703 N |
4896 | .name = "raid4", |
4897 | .level = 4, | |
4898 | .owner = THIS_MODULE, | |
4899 | .make_request = make_request, | |
4900 | .run = run, | |
4901 | .stop = stop, | |
4902 | .status = status, | |
4903 | .error_handler = error, | |
4904 | .hot_add_disk = raid5_add_disk, | |
4905 | .hot_remove_disk= raid5_remove_disk, | |
4906 | .spare_active = raid5_spare_active, | |
4907 | .sync_request = sync_request, | |
4908 | .resize = raid5_resize, | |
3d37890b N |
4909 | #ifdef CONFIG_MD_RAID5_RESHAPE |
4910 | .check_reshape = raid5_check_reshape, | |
4911 | .start_reshape = raid5_start_reshape, | |
4912 | #endif | |
2604b703 N |
4913 | .quiesce = raid5_quiesce, |
4914 | }; | |
4915 | ||
4916 | static int __init raid5_init(void) | |
4917 | { | |
16a53ecc N |
4918 | int e; |
4919 | ||
4920 | e = raid6_select_algo(); | |
4921 | if ( e ) | |
4922 | return e; | |
4923 | register_md_personality(&raid6_personality); | |
2604b703 N |
4924 | register_md_personality(&raid5_personality); |
4925 | register_md_personality(&raid4_personality); | |
4926 | return 0; | |
1da177e4 LT |
4927 | } |
4928 | ||
2604b703 | 4929 | static void raid5_exit(void) |
1da177e4 | 4930 | { |
16a53ecc | 4931 | unregister_md_personality(&raid6_personality); |
2604b703 N |
4932 | unregister_md_personality(&raid5_personality); |
4933 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
4934 | } |
4935 | ||
4936 | module_init(raid5_init); | |
4937 | module_exit(raid5_exit); | |
4938 | MODULE_LICENSE("GPL"); | |
4939 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ | |
d9d166c2 N |
4940 | MODULE_ALIAS("md-raid5"); |
4941 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
4942 | MODULE_ALIAS("md-level-5"); |
4943 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
4944 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
4945 | MODULE_ALIAS("md-raid6"); | |
4946 | MODULE_ALIAS("md-level-6"); | |
4947 | ||
4948 | /* This used to be two separate modules, they were: */ | |
4949 | MODULE_ALIAS("raid5"); | |
4950 | MODULE_ALIAS("raid6"); |