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
5 * Copyright (C) 2002, 2003 H. Peter Anvin
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!
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)
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.
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
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
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
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
46 #include <linux/blkdev.h>
47 #include <linux/kthread.h>
48 #include <linux/raid/pq.h>
49 #include <linux/async_tx.h>
50 #include <linux/seq_file.h>
59 #define NR_STRIPES 256
60 #define STRIPE_SIZE PAGE_SIZE
61 #define STRIPE_SHIFT (PAGE_SHIFT - 9)
62 #define STRIPE_SECTORS (STRIPE_SIZE>>9)
63 #define IO_THRESHOLD 1
64 #define BYPASS_THRESHOLD 1
65 #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
66 #define HASH_MASK (NR_HASH - 1)
68 #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
70 /* bio's attached to a stripe+device for I/O are linked together in bi_sector
71 * order without overlap. There may be several bio's per stripe+device, and
72 * a bio could span several devices.
73 * When walking this list for a particular stripe+device, we must never proceed
74 * beyond a bio that extends past this device, as the next bio might no longer
76 * This macro is used to determine the 'next' bio in the list, given the sector
77 * of the current stripe+device
79 #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
81 * The following can be used to debug the driver
83 #define RAID5_PARANOIA 1
84 #if RAID5_PARANOIA && defined(CONFIG_SMP)
85 # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
87 # define CHECK_DEVLOCK()
95 #define printk_rl(args...) ((void) (printk_ratelimit() && printk(args)))
98 * We maintain a biased count of active stripes in the bottom 16 bits of
99 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
101 static inline int raid5_bi_phys_segments(struct bio *bio)
103 return bio->bi_phys_segments & 0xffff;
106 static inline int raid5_bi_hw_segments(struct bio *bio)
108 return (bio->bi_phys_segments >> 16) & 0xffff;
111 static inline int raid5_dec_bi_phys_segments(struct bio *bio)
113 --bio->bi_phys_segments;
114 return raid5_bi_phys_segments(bio);
117 static inline int raid5_dec_bi_hw_segments(struct bio *bio)
119 unsigned short val = raid5_bi_hw_segments(bio);
122 bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio);
126 static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt)
128 bio->bi_phys_segments = raid5_bi_phys_segments(bio) || (cnt << 16);
131 /* Find first data disk in a raid6 stripe */
132 static inline int raid6_d0(struct stripe_head *sh)
135 /* ddf always start from first device */
137 /* md starts just after Q block */
138 if (sh->qd_idx == sh->disks - 1)
141 return sh->qd_idx + 1;
143 static inline int raid6_next_disk(int disk, int raid_disks)
146 return (disk < raid_disks) ? disk : 0;
149 /* When walking through the disks in a raid5, starting at raid6_d0,
150 * We need to map each disk to a 'slot', where the data disks are slot
151 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
152 * is raid_disks-1. This help does that mapping.
154 static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
155 int *count, int syndrome_disks)
159 if (idx == sh->pd_idx)
160 return syndrome_disks;
161 if (idx == sh->qd_idx)
162 return syndrome_disks + 1;
167 static void return_io(struct bio *return_bi)
169 struct bio *bi = return_bi;
172 return_bi = bi->bi_next;
180 static void print_raid5_conf (raid5_conf_t *conf);
182 static int stripe_operations_active(struct stripe_head *sh)
184 return sh->check_state || sh->reconstruct_state ||
185 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
186 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
189 static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh)
191 if (atomic_dec_and_test(&sh->count)) {
192 BUG_ON(!list_empty(&sh->lru));
193 BUG_ON(atomic_read(&conf->active_stripes)==0);
194 if (test_bit(STRIPE_HANDLE, &sh->state)) {
195 if (test_bit(STRIPE_DELAYED, &sh->state)) {
196 list_add_tail(&sh->lru, &conf->delayed_list);
197 blk_plug_device(conf->mddev->queue);
198 } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
199 sh->bm_seq - conf->seq_write > 0) {
200 list_add_tail(&sh->lru, &conf->bitmap_list);
201 blk_plug_device(conf->mddev->queue);
203 clear_bit(STRIPE_BIT_DELAY, &sh->state);
204 list_add_tail(&sh->lru, &conf->handle_list);
206 md_wakeup_thread(conf->mddev->thread);
208 BUG_ON(stripe_operations_active(sh));
209 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
210 atomic_dec(&conf->preread_active_stripes);
211 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD)
212 md_wakeup_thread(conf->mddev->thread);
214 atomic_dec(&conf->active_stripes);
215 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
216 list_add_tail(&sh->lru, &conf->inactive_list);
217 wake_up(&conf->wait_for_stripe);
218 if (conf->retry_read_aligned)
219 md_wakeup_thread(conf->mddev->thread);
225 static void release_stripe(struct stripe_head *sh)
227 raid5_conf_t *conf = sh->raid_conf;
230 spin_lock_irqsave(&conf->device_lock, flags);
231 __release_stripe(conf, sh);
232 spin_unlock_irqrestore(&conf->device_lock, flags);
235 static inline void remove_hash(struct stripe_head *sh)
237 pr_debug("remove_hash(), stripe %llu\n",
238 (unsigned long long)sh->sector);
240 hlist_del_init(&sh->hash);
243 static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh)
245 struct hlist_head *hp = stripe_hash(conf, sh->sector);
247 pr_debug("insert_hash(), stripe %llu\n",
248 (unsigned long long)sh->sector);
251 hlist_add_head(&sh->hash, hp);
255 /* find an idle stripe, make sure it is unhashed, and return it. */
256 static struct stripe_head *get_free_stripe(raid5_conf_t *conf)
258 struct stripe_head *sh = NULL;
259 struct list_head *first;
262 if (list_empty(&conf->inactive_list))
264 first = conf->inactive_list.next;
265 sh = list_entry(first, struct stripe_head, lru);
266 list_del_init(first);
268 atomic_inc(&conf->active_stripes);
273 static void shrink_buffers(struct stripe_head *sh, int num)
278 for (i=0; i<num ; i++) {
282 sh->dev[i].page = NULL;
287 static int grow_buffers(struct stripe_head *sh, int num)
291 for (i=0; i<num; i++) {
294 if (!(page = alloc_page(GFP_KERNEL))) {
297 sh->dev[i].page = page;
302 static void raid5_build_block(struct stripe_head *sh, int i);
303 static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
304 struct stripe_head *sh);
306 static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
308 raid5_conf_t *conf = sh->raid_conf;
311 BUG_ON(atomic_read(&sh->count) != 0);
312 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
313 BUG_ON(stripe_operations_active(sh));
316 pr_debug("init_stripe called, stripe %llu\n",
317 (unsigned long long)sh->sector);
321 sh->generation = conf->generation - previous;
322 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
324 stripe_set_idx(sector, conf, previous, sh);
328 for (i = sh->disks; i--; ) {
329 struct r5dev *dev = &sh->dev[i];
331 if (dev->toread || dev->read || dev->towrite || dev->written ||
332 test_bit(R5_LOCKED, &dev->flags)) {
333 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
334 (unsigned long long)sh->sector, i, dev->toread,
335 dev->read, dev->towrite, dev->written,
336 test_bit(R5_LOCKED, &dev->flags));
340 raid5_build_block(sh, i);
342 insert_hash(conf, sh);
345 static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector,
348 struct stripe_head *sh;
349 struct hlist_node *hn;
352 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
353 hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash)
354 if (sh->sector == sector && sh->generation == generation)
356 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
360 static void unplug_slaves(mddev_t *mddev);
361 static void raid5_unplug_device(struct request_queue *q);
363 static struct stripe_head *
364 get_active_stripe(raid5_conf_t *conf, sector_t sector,
365 int previous, int noblock)
367 struct stripe_head *sh;
369 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
371 spin_lock_irq(&conf->device_lock);
374 wait_event_lock_irq(conf->wait_for_stripe,
376 conf->device_lock, /* nothing */);
377 sh = __find_stripe(conf, sector, conf->generation - previous);
379 if (!conf->inactive_blocked)
380 sh = get_free_stripe(conf);
381 if (noblock && sh == NULL)
384 conf->inactive_blocked = 1;
385 wait_event_lock_irq(conf->wait_for_stripe,
386 !list_empty(&conf->inactive_list) &&
387 (atomic_read(&conf->active_stripes)
388 < (conf->max_nr_stripes *3/4)
389 || !conf->inactive_blocked),
391 raid5_unplug_device(conf->mddev->queue)
393 conf->inactive_blocked = 0;
395 init_stripe(sh, sector, previous);
397 if (atomic_read(&sh->count)) {
398 BUG_ON(!list_empty(&sh->lru));
400 if (!test_bit(STRIPE_HANDLE, &sh->state))
401 atomic_inc(&conf->active_stripes);
402 if (list_empty(&sh->lru) &&
403 !test_bit(STRIPE_EXPANDING, &sh->state))
405 list_del_init(&sh->lru);
408 } while (sh == NULL);
411 atomic_inc(&sh->count);
413 spin_unlock_irq(&conf->device_lock);
418 raid5_end_read_request(struct bio *bi, int error);
420 raid5_end_write_request(struct bio *bi, int error);
422 static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
424 raid5_conf_t *conf = sh->raid_conf;
425 int i, disks = sh->disks;
429 for (i = disks; i--; ) {
433 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags))
435 else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
440 bi = &sh->dev[i].req;
444 bi->bi_end_io = raid5_end_write_request;
446 bi->bi_end_io = raid5_end_read_request;
449 rdev = rcu_dereference(conf->disks[i].rdev);
450 if (rdev && test_bit(Faulty, &rdev->flags))
453 atomic_inc(&rdev->nr_pending);
457 if (s->syncing || s->expanding || s->expanded)
458 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
460 set_bit(STRIPE_IO_STARTED, &sh->state);
462 bi->bi_bdev = rdev->bdev;
463 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
464 __func__, (unsigned long long)sh->sector,
466 atomic_inc(&sh->count);
467 bi->bi_sector = sh->sector + rdev->data_offset;
468 bi->bi_flags = 1 << BIO_UPTODATE;
472 bi->bi_io_vec = &sh->dev[i].vec;
473 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
474 bi->bi_io_vec[0].bv_offset = 0;
475 bi->bi_size = STRIPE_SIZE;
478 test_bit(R5_ReWrite, &sh->dev[i].flags))
479 atomic_add(STRIPE_SECTORS,
480 &rdev->corrected_errors);
481 generic_make_request(bi);
484 set_bit(STRIPE_DEGRADED, &sh->state);
485 pr_debug("skip op %ld on disc %d for sector %llu\n",
486 bi->bi_rw, i, (unsigned long long)sh->sector);
487 clear_bit(R5_LOCKED, &sh->dev[i].flags);
488 set_bit(STRIPE_HANDLE, &sh->state);
493 static struct dma_async_tx_descriptor *
494 async_copy_data(int frombio, struct bio *bio, struct page *page,
495 sector_t sector, struct dma_async_tx_descriptor *tx)
498 struct page *bio_page;
502 if (bio->bi_sector >= sector)
503 page_offset = (signed)(bio->bi_sector - sector) * 512;
505 page_offset = (signed)(sector - bio->bi_sector) * -512;
506 bio_for_each_segment(bvl, bio, i) {
507 int len = bio_iovec_idx(bio, i)->bv_len;
511 if (page_offset < 0) {
512 b_offset = -page_offset;
513 page_offset += b_offset;
517 if (len > 0 && page_offset + len > STRIPE_SIZE)
518 clen = STRIPE_SIZE - page_offset;
523 b_offset += bio_iovec_idx(bio, i)->bv_offset;
524 bio_page = bio_iovec_idx(bio, i)->bv_page;
526 tx = async_memcpy(page, bio_page, page_offset,
531 tx = async_memcpy(bio_page, page, b_offset,
536 if (clen < len) /* hit end of page */
544 static void ops_complete_biofill(void *stripe_head_ref)
546 struct stripe_head *sh = stripe_head_ref;
547 struct bio *return_bi = NULL;
548 raid5_conf_t *conf = sh->raid_conf;
551 pr_debug("%s: stripe %llu\n", __func__,
552 (unsigned long long)sh->sector);
554 /* clear completed biofills */
555 spin_lock_irq(&conf->device_lock);
556 for (i = sh->disks; i--; ) {
557 struct r5dev *dev = &sh->dev[i];
559 /* acknowledge completion of a biofill operation */
560 /* and check if we need to reply to a read request,
561 * new R5_Wantfill requests are held off until
562 * !STRIPE_BIOFILL_RUN
564 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
565 struct bio *rbi, *rbi2;
570 while (rbi && rbi->bi_sector <
571 dev->sector + STRIPE_SECTORS) {
572 rbi2 = r5_next_bio(rbi, dev->sector);
573 if (!raid5_dec_bi_phys_segments(rbi)) {
574 rbi->bi_next = return_bi;
581 spin_unlock_irq(&conf->device_lock);
582 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
584 return_io(return_bi);
586 set_bit(STRIPE_HANDLE, &sh->state);
590 static void ops_run_biofill(struct stripe_head *sh)
592 struct dma_async_tx_descriptor *tx = NULL;
593 raid5_conf_t *conf = sh->raid_conf;
596 pr_debug("%s: stripe %llu\n", __func__,
597 (unsigned long long)sh->sector);
599 for (i = sh->disks; i--; ) {
600 struct r5dev *dev = &sh->dev[i];
601 if (test_bit(R5_Wantfill, &dev->flags)) {
603 spin_lock_irq(&conf->device_lock);
604 dev->read = rbi = dev->toread;
606 spin_unlock_irq(&conf->device_lock);
607 while (rbi && rbi->bi_sector <
608 dev->sector + STRIPE_SECTORS) {
609 tx = async_copy_data(0, rbi, dev->page,
611 rbi = r5_next_bio(rbi, dev->sector);
616 atomic_inc(&sh->count);
617 async_trigger_callback(ASYNC_TX_DEP_ACK | ASYNC_TX_ACK, tx,
618 ops_complete_biofill, sh);
621 static void ops_complete_compute5(void *stripe_head_ref)
623 struct stripe_head *sh = stripe_head_ref;
624 int target = sh->ops.target;
625 struct r5dev *tgt = &sh->dev[target];
627 pr_debug("%s: stripe %llu\n", __func__,
628 (unsigned long long)sh->sector);
630 set_bit(R5_UPTODATE, &tgt->flags);
631 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
632 clear_bit(R5_Wantcompute, &tgt->flags);
633 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
634 if (sh->check_state == check_state_compute_run)
635 sh->check_state = check_state_compute_result;
636 set_bit(STRIPE_HANDLE, &sh->state);
640 static struct dma_async_tx_descriptor *ops_run_compute5(struct stripe_head *sh)
642 /* kernel stack size limits the total number of disks */
643 int disks = sh->disks;
644 struct page *xor_srcs[disks];
645 int target = sh->ops.target;
646 struct r5dev *tgt = &sh->dev[target];
647 struct page *xor_dest = tgt->page;
649 struct dma_async_tx_descriptor *tx;
652 pr_debug("%s: stripe %llu block: %d\n",
653 __func__, (unsigned long long)sh->sector, target);
654 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
656 for (i = disks; i--; )
658 xor_srcs[count++] = sh->dev[i].page;
660 atomic_inc(&sh->count);
662 if (unlikely(count == 1))
663 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE,
664 0, NULL, ops_complete_compute5, sh);
666 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
667 ASYNC_TX_XOR_ZERO_DST, NULL,
668 ops_complete_compute5, sh);
673 static void ops_complete_prexor(void *stripe_head_ref)
675 struct stripe_head *sh = stripe_head_ref;
677 pr_debug("%s: stripe %llu\n", __func__,
678 (unsigned long long)sh->sector);
681 static struct dma_async_tx_descriptor *
682 ops_run_prexor(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
684 /* kernel stack size limits the total number of disks */
685 int disks = sh->disks;
686 struct page *xor_srcs[disks];
687 int count = 0, pd_idx = sh->pd_idx, i;
689 /* existing parity data subtracted */
690 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
692 pr_debug("%s: stripe %llu\n", __func__,
693 (unsigned long long)sh->sector);
695 for (i = disks; i--; ) {
696 struct r5dev *dev = &sh->dev[i];
697 /* Only process blocks that are known to be uptodate */
698 if (test_bit(R5_Wantdrain, &dev->flags))
699 xor_srcs[count++] = dev->page;
702 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
703 ASYNC_TX_DEP_ACK | ASYNC_TX_XOR_DROP_DST, tx,
704 ops_complete_prexor, sh);
709 static struct dma_async_tx_descriptor *
710 ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
712 int disks = sh->disks;
715 pr_debug("%s: stripe %llu\n", __func__,
716 (unsigned long long)sh->sector);
718 for (i = disks; i--; ) {
719 struct r5dev *dev = &sh->dev[i];
722 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
725 spin_lock(&sh->lock);
726 chosen = dev->towrite;
728 BUG_ON(dev->written);
729 wbi = dev->written = chosen;
730 spin_unlock(&sh->lock);
732 while (wbi && wbi->bi_sector <
733 dev->sector + STRIPE_SECTORS) {
734 tx = async_copy_data(1, wbi, dev->page,
736 wbi = r5_next_bio(wbi, dev->sector);
744 static void ops_complete_postxor(void *stripe_head_ref)
746 struct stripe_head *sh = stripe_head_ref;
747 int disks = sh->disks, i, pd_idx = sh->pd_idx;
749 pr_debug("%s: stripe %llu\n", __func__,
750 (unsigned long long)sh->sector);
752 for (i = disks; i--; ) {
753 struct r5dev *dev = &sh->dev[i];
754 if (dev->written || i == pd_idx)
755 set_bit(R5_UPTODATE, &dev->flags);
758 if (sh->reconstruct_state == reconstruct_state_drain_run)
759 sh->reconstruct_state = reconstruct_state_drain_result;
760 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
761 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
763 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
764 sh->reconstruct_state = reconstruct_state_result;
767 set_bit(STRIPE_HANDLE, &sh->state);
772 ops_run_postxor(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
774 /* kernel stack size limits the total number of disks */
775 int disks = sh->disks;
776 struct page *xor_srcs[disks];
778 int count = 0, pd_idx = sh->pd_idx, i;
779 struct page *xor_dest;
783 pr_debug("%s: stripe %llu\n", __func__,
784 (unsigned long long)sh->sector);
786 /* check if prexor is active which means only process blocks
787 * that are part of a read-modify-write (written)
789 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
791 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
792 for (i = disks; i--; ) {
793 struct r5dev *dev = &sh->dev[i];
795 xor_srcs[count++] = dev->page;
798 xor_dest = sh->dev[pd_idx].page;
799 for (i = disks; i--; ) {
800 struct r5dev *dev = &sh->dev[i];
802 xor_srcs[count++] = dev->page;
806 /* 1/ if we prexor'd then the dest is reused as a source
807 * 2/ if we did not prexor then we are redoing the parity
808 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
809 * for the synchronous xor case
811 flags = ASYNC_TX_DEP_ACK | ASYNC_TX_ACK |
812 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
814 atomic_inc(&sh->count);
816 if (unlikely(count == 1)) {
817 flags &= ~(ASYNC_TX_XOR_DROP_DST | ASYNC_TX_XOR_ZERO_DST);
818 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE,
819 flags, tx, ops_complete_postxor, sh);
821 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
822 flags, tx, ops_complete_postxor, sh);
825 static void ops_complete_check(void *stripe_head_ref)
827 struct stripe_head *sh = stripe_head_ref;
829 pr_debug("%s: stripe %llu\n", __func__,
830 (unsigned long long)sh->sector);
832 sh->check_state = check_state_check_result;
833 set_bit(STRIPE_HANDLE, &sh->state);
837 static void ops_run_check(struct stripe_head *sh)
839 /* kernel stack size limits the total number of disks */
840 int disks = sh->disks;
841 struct page *xor_srcs[disks];
842 struct dma_async_tx_descriptor *tx;
844 int count = 0, pd_idx = sh->pd_idx, i;
845 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
847 pr_debug("%s: stripe %llu\n", __func__,
848 (unsigned long long)sh->sector);
850 for (i = disks; i--; ) {
851 struct r5dev *dev = &sh->dev[i];
853 xor_srcs[count++] = dev->page;
856 tx = async_xor_zero_sum(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
857 &sh->ops.zero_sum_result, 0, NULL, NULL, NULL);
859 atomic_inc(&sh->count);
860 tx = async_trigger_callback(ASYNC_TX_DEP_ACK | ASYNC_TX_ACK, tx,
861 ops_complete_check, sh);
864 static void raid5_run_ops(struct stripe_head *sh, unsigned long ops_request)
866 int overlap_clear = 0, i, disks = sh->disks;
867 struct dma_async_tx_descriptor *tx = NULL;
869 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
874 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
875 tx = ops_run_compute5(sh);
876 /* terminate the chain if postxor is not set to be run */
877 if (tx && !test_bit(STRIPE_OP_POSTXOR, &ops_request))
881 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
882 tx = ops_run_prexor(sh, tx);
884 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
885 tx = ops_run_biodrain(sh, tx);
889 if (test_bit(STRIPE_OP_POSTXOR, &ops_request))
890 ops_run_postxor(sh, tx);
892 if (test_bit(STRIPE_OP_CHECK, &ops_request))
896 for (i = disks; i--; ) {
897 struct r5dev *dev = &sh->dev[i];
898 if (test_and_clear_bit(R5_Overlap, &dev->flags))
899 wake_up(&sh->raid_conf->wait_for_overlap);
903 static int grow_one_stripe(raid5_conf_t *conf)
905 struct stripe_head *sh;
906 sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL);
909 memset(sh, 0, sizeof(*sh) + (conf->raid_disks-1)*sizeof(struct r5dev));
910 sh->raid_conf = conf;
911 spin_lock_init(&sh->lock);
913 if (grow_buffers(sh, conf->raid_disks)) {
914 shrink_buffers(sh, conf->raid_disks);
915 kmem_cache_free(conf->slab_cache, sh);
918 sh->disks = conf->raid_disks;
919 /* we just created an active stripe so... */
920 atomic_set(&sh->count, 1);
921 atomic_inc(&conf->active_stripes);
922 INIT_LIST_HEAD(&sh->lru);
927 static int grow_stripes(raid5_conf_t *conf, int num)
929 struct kmem_cache *sc;
930 int devs = conf->raid_disks;
932 sprintf(conf->cache_name[0],
933 "raid%d-%s", conf->level, mdname(conf->mddev));
934 sprintf(conf->cache_name[1],
935 "raid%d-%s-alt", conf->level, mdname(conf->mddev));
936 conf->active_name = 0;
937 sc = kmem_cache_create(conf->cache_name[conf->active_name],
938 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
942 conf->slab_cache = sc;
943 conf->pool_size = devs;
945 if (!grow_one_stripe(conf))
950 #ifdef CONFIG_MD_RAID5_RESHAPE
951 static int resize_stripes(raid5_conf_t *conf, int newsize)
953 /* Make all the stripes able to hold 'newsize' devices.
954 * New slots in each stripe get 'page' set to a new page.
956 * This happens in stages:
957 * 1/ create a new kmem_cache and allocate the required number of
959 * 2/ gather all the old stripe_heads and tranfer the pages across
960 * to the new stripe_heads. This will have the side effect of
961 * freezing the array as once all stripe_heads have been collected,
962 * no IO will be possible. Old stripe heads are freed once their
963 * pages have been transferred over, and the old kmem_cache is
964 * freed when all stripes are done.
965 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
966 * we simple return a failre status - no need to clean anything up.
967 * 4/ allocate new pages for the new slots in the new stripe_heads.
968 * If this fails, we don't bother trying the shrink the
969 * stripe_heads down again, we just leave them as they are.
970 * As each stripe_head is processed the new one is released into
973 * Once step2 is started, we cannot afford to wait for a write,
974 * so we use GFP_NOIO allocations.
976 struct stripe_head *osh, *nsh;
977 LIST_HEAD(newstripes);
978 struct disk_info *ndisks;
980 struct kmem_cache *sc;
983 if (newsize <= conf->pool_size)
984 return 0; /* never bother to shrink */
986 err = md_allow_write(conf->mddev);
991 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
992 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
997 for (i = conf->max_nr_stripes; i; i--) {
998 nsh = kmem_cache_alloc(sc, GFP_KERNEL);
1002 memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev));
1004 nsh->raid_conf = conf;
1005 spin_lock_init(&nsh->lock);
1007 list_add(&nsh->lru, &newstripes);
1010 /* didn't get enough, give up */
1011 while (!list_empty(&newstripes)) {
1012 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1013 list_del(&nsh->lru);
1014 kmem_cache_free(sc, nsh);
1016 kmem_cache_destroy(sc);
1019 /* Step 2 - Must use GFP_NOIO now.
1020 * OK, we have enough stripes, start collecting inactive
1021 * stripes and copying them over
1023 list_for_each_entry(nsh, &newstripes, lru) {
1024 spin_lock_irq(&conf->device_lock);
1025 wait_event_lock_irq(conf->wait_for_stripe,
1026 !list_empty(&conf->inactive_list),
1028 unplug_slaves(conf->mddev)
1030 osh = get_free_stripe(conf);
1031 spin_unlock_irq(&conf->device_lock);
1032 atomic_set(&nsh->count, 1);
1033 for(i=0; i<conf->pool_size; i++)
1034 nsh->dev[i].page = osh->dev[i].page;
1035 for( ; i<newsize; i++)
1036 nsh->dev[i].page = NULL;
1037 kmem_cache_free(conf->slab_cache, osh);
1039 kmem_cache_destroy(conf->slab_cache);
1042 * At this point, we are holding all the stripes so the array
1043 * is completely stalled, so now is a good time to resize
1046 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1048 for (i=0; i<conf->raid_disks; i++)
1049 ndisks[i] = conf->disks[i];
1051 conf->disks = ndisks;
1055 /* Step 4, return new stripes to service */
1056 while(!list_empty(&newstripes)) {
1057 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1058 list_del_init(&nsh->lru);
1059 for (i=conf->raid_disks; i < newsize; i++)
1060 if (nsh->dev[i].page == NULL) {
1061 struct page *p = alloc_page(GFP_NOIO);
1062 nsh->dev[i].page = p;
1066 release_stripe(nsh);
1068 /* critical section pass, GFP_NOIO no longer needed */
1070 conf->slab_cache = sc;
1071 conf->active_name = 1-conf->active_name;
1072 conf->pool_size = newsize;
1077 static int drop_one_stripe(raid5_conf_t *conf)
1079 struct stripe_head *sh;
1081 spin_lock_irq(&conf->device_lock);
1082 sh = get_free_stripe(conf);
1083 spin_unlock_irq(&conf->device_lock);
1086 BUG_ON(atomic_read(&sh->count));
1087 shrink_buffers(sh, conf->pool_size);
1088 kmem_cache_free(conf->slab_cache, sh);
1089 atomic_dec(&conf->active_stripes);
1093 static void shrink_stripes(raid5_conf_t *conf)
1095 while (drop_one_stripe(conf))
1098 if (conf->slab_cache)
1099 kmem_cache_destroy(conf->slab_cache);
1100 conf->slab_cache = NULL;
1103 static void raid5_end_read_request(struct bio * bi, int error)
1105 struct stripe_head *sh = bi->bi_private;
1106 raid5_conf_t *conf = sh->raid_conf;
1107 int disks = sh->disks, i;
1108 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
1109 char b[BDEVNAME_SIZE];
1113 for (i=0 ; i<disks; i++)
1114 if (bi == &sh->dev[i].req)
1117 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1118 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1126 set_bit(R5_UPTODATE, &sh->dev[i].flags);
1127 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
1128 rdev = conf->disks[i].rdev;
1129 printk_rl(KERN_INFO "raid5:%s: read error corrected"
1130 " (%lu sectors at %llu on %s)\n",
1131 mdname(conf->mddev), STRIPE_SECTORS,
1132 (unsigned long long)(sh->sector
1133 + rdev->data_offset),
1134 bdevname(rdev->bdev, b));
1135 clear_bit(R5_ReadError, &sh->dev[i].flags);
1136 clear_bit(R5_ReWrite, &sh->dev[i].flags);
1138 if (atomic_read(&conf->disks[i].rdev->read_errors))
1139 atomic_set(&conf->disks[i].rdev->read_errors, 0);
1141 const char *bdn = bdevname(conf->disks[i].rdev->bdev, b);
1143 rdev = conf->disks[i].rdev;
1145 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
1146 atomic_inc(&rdev->read_errors);
1147 if (conf->mddev->degraded)
1148 printk_rl(KERN_WARNING
1149 "raid5:%s: read error not correctable "
1150 "(sector %llu on %s).\n",
1151 mdname(conf->mddev),
1152 (unsigned long long)(sh->sector
1153 + rdev->data_offset),
1155 else if (test_bit(R5_ReWrite, &sh->dev[i].flags))
1157 printk_rl(KERN_WARNING
1158 "raid5:%s: read error NOT corrected!! "
1159 "(sector %llu on %s).\n",
1160 mdname(conf->mddev),
1161 (unsigned long long)(sh->sector
1162 + rdev->data_offset),
1164 else if (atomic_read(&rdev->read_errors)
1165 > conf->max_nr_stripes)
1167 "raid5:%s: Too many read errors, failing device %s.\n",
1168 mdname(conf->mddev), bdn);
1172 set_bit(R5_ReadError, &sh->dev[i].flags);
1174 clear_bit(R5_ReadError, &sh->dev[i].flags);
1175 clear_bit(R5_ReWrite, &sh->dev[i].flags);
1176 md_error(conf->mddev, rdev);
1179 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1180 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1181 set_bit(STRIPE_HANDLE, &sh->state);
1185 static void raid5_end_write_request(struct bio *bi, int error)
1187 struct stripe_head *sh = bi->bi_private;
1188 raid5_conf_t *conf = sh->raid_conf;
1189 int disks = sh->disks, i;
1190 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
1192 for (i=0 ; i<disks; i++)
1193 if (bi == &sh->dev[i].req)
1196 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
1197 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1205 md_error(conf->mddev, conf->disks[i].rdev);
1207 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1209 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1210 set_bit(STRIPE_HANDLE, &sh->state);
1215 static sector_t compute_blocknr(struct stripe_head *sh, int i);
1217 static void raid5_build_block(struct stripe_head *sh, int i)
1219 struct r5dev *dev = &sh->dev[i];
1221 bio_init(&dev->req);
1222 dev->req.bi_io_vec = &dev->vec;
1224 dev->req.bi_max_vecs++;
1225 dev->vec.bv_page = dev->page;
1226 dev->vec.bv_len = STRIPE_SIZE;
1227 dev->vec.bv_offset = 0;
1229 dev->req.bi_sector = sh->sector;
1230 dev->req.bi_private = sh;
1233 dev->sector = compute_blocknr(sh, i);
1236 static void error(mddev_t *mddev, mdk_rdev_t *rdev)
1238 char b[BDEVNAME_SIZE];
1239 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
1240 pr_debug("raid5: error called\n");
1242 if (!test_bit(Faulty, &rdev->flags)) {
1243 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1244 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1245 unsigned long flags;
1246 spin_lock_irqsave(&conf->device_lock, flags);
1248 spin_unlock_irqrestore(&conf->device_lock, flags);
1250 * if recovery was running, make sure it aborts.
1252 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1254 set_bit(Faulty, &rdev->flags);
1256 "raid5: Disk failure on %s, disabling device.\n"
1257 "raid5: Operation continuing on %d devices.\n",
1258 bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded);
1263 * Input: a 'big' sector number,
1264 * Output: index of the data and parity disk, and the sector # in them.
1266 static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector,
1267 int previous, int *dd_idx,
1268 struct stripe_head *sh)
1271 unsigned long chunk_number;
1272 unsigned int chunk_offset;
1275 sector_t new_sector;
1276 int sectors_per_chunk = conf->chunk_size >> 9;
1277 int raid_disks = previous ? conf->previous_raid_disks
1279 int data_disks = raid_disks - conf->max_degraded;
1281 /* First compute the information on this sector */
1284 * Compute the chunk number and the sector offset inside the chunk
1286 chunk_offset = sector_div(r_sector, sectors_per_chunk);
1287 chunk_number = r_sector;
1288 BUG_ON(r_sector != chunk_number);
1291 * Compute the stripe number
1293 stripe = chunk_number / data_disks;
1296 * Compute the data disk and parity disk indexes inside the stripe
1298 *dd_idx = chunk_number % data_disks;
1301 * Select the parity disk based on the user selected algorithm.
1303 pd_idx = qd_idx = ~0;
1304 switch(conf->level) {
1306 pd_idx = data_disks;
1309 switch (conf->algorithm) {
1310 case ALGORITHM_LEFT_ASYMMETRIC:
1311 pd_idx = data_disks - stripe % raid_disks;
1312 if (*dd_idx >= pd_idx)
1315 case ALGORITHM_RIGHT_ASYMMETRIC:
1316 pd_idx = stripe % raid_disks;
1317 if (*dd_idx >= pd_idx)
1320 case ALGORITHM_LEFT_SYMMETRIC:
1321 pd_idx = data_disks - stripe % raid_disks;
1322 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1324 case ALGORITHM_RIGHT_SYMMETRIC:
1325 pd_idx = stripe % raid_disks;
1326 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1328 case ALGORITHM_PARITY_0:
1332 case ALGORITHM_PARITY_N:
1333 pd_idx = data_disks;
1336 printk(KERN_ERR "raid5: unsupported algorithm %d\n",
1343 switch (conf->algorithm) {
1344 case ALGORITHM_LEFT_ASYMMETRIC:
1345 pd_idx = raid_disks - 1 - (stripe % raid_disks);
1346 qd_idx = pd_idx + 1;
1347 if (pd_idx == raid_disks-1) {
1348 (*dd_idx)++; /* Q D D D P */
1350 } else if (*dd_idx >= pd_idx)
1351 (*dd_idx) += 2; /* D D P Q D */
1353 case ALGORITHM_RIGHT_ASYMMETRIC:
1354 pd_idx = stripe % raid_disks;
1355 qd_idx = pd_idx + 1;
1356 if (pd_idx == raid_disks-1) {
1357 (*dd_idx)++; /* Q D D D P */
1359 } else if (*dd_idx >= pd_idx)
1360 (*dd_idx) += 2; /* D D P Q D */
1362 case ALGORITHM_LEFT_SYMMETRIC:
1363 pd_idx = raid_disks - 1 - (stripe % raid_disks);
1364 qd_idx = (pd_idx + 1) % raid_disks;
1365 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
1367 case ALGORITHM_RIGHT_SYMMETRIC:
1368 pd_idx = stripe % raid_disks;
1369 qd_idx = (pd_idx + 1) % raid_disks;
1370 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
1373 case ALGORITHM_PARITY_0:
1378 case ALGORITHM_PARITY_N:
1379 pd_idx = data_disks;
1380 qd_idx = data_disks + 1;
1383 case ALGORITHM_ROTATING_ZERO_RESTART:
1384 /* Exactly the same as RIGHT_ASYMMETRIC, but or
1385 * of blocks for computing Q is different.
1387 pd_idx = stripe % raid_disks;
1388 qd_idx = pd_idx + 1;
1389 if (pd_idx == raid_disks-1) {
1390 (*dd_idx)++; /* Q D D D P */
1392 } else if (*dd_idx >= pd_idx)
1393 (*dd_idx) += 2; /* D D P Q D */
1397 case ALGORITHM_ROTATING_N_RESTART:
1398 /* Same a left_asymmetric, by first stripe is
1399 * D D D P Q rather than
1402 pd_idx = raid_disks - 1 - ((stripe + 1) % raid_disks);
1403 qd_idx = pd_idx + 1;
1404 if (pd_idx == raid_disks-1) {
1405 (*dd_idx)++; /* Q D D D P */
1407 } else if (*dd_idx >= pd_idx)
1408 (*dd_idx) += 2; /* D D P Q D */
1412 case ALGORITHM_ROTATING_N_CONTINUE:
1413 /* Same as left_symmetric but Q is before P */
1414 pd_idx = raid_disks - 1 - (stripe % raid_disks);
1415 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
1416 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
1420 case ALGORITHM_LEFT_ASYMMETRIC_6:
1421 /* RAID5 left_asymmetric, with Q on last device */
1422 pd_idx = data_disks - stripe % (raid_disks-1);
1423 if (*dd_idx >= pd_idx)
1425 qd_idx = raid_disks - 1;
1428 case ALGORITHM_RIGHT_ASYMMETRIC_6:
1429 pd_idx = stripe % (raid_disks-1);
1430 if (*dd_idx >= pd_idx)
1432 qd_idx = raid_disks - 1;
1435 case ALGORITHM_LEFT_SYMMETRIC_6:
1436 pd_idx = data_disks - stripe % (raid_disks-1);
1437 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1438 qd_idx = raid_disks - 1;
1441 case ALGORITHM_RIGHT_SYMMETRIC_6:
1442 pd_idx = stripe % (raid_disks-1);
1443 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1444 qd_idx = raid_disks - 1;
1447 case ALGORITHM_PARITY_0_6:
1450 qd_idx = raid_disks - 1;
1455 printk(KERN_CRIT "raid6: unsupported algorithm %d\n",
1463 sh->pd_idx = pd_idx;
1464 sh->qd_idx = qd_idx;
1465 sh->ddf_layout = ddf_layout;
1468 * Finally, compute the new sector number
1470 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
1475 static sector_t compute_blocknr(struct stripe_head *sh, int i)
1477 raid5_conf_t *conf = sh->raid_conf;
1478 int raid_disks = sh->disks;
1479 int data_disks = raid_disks - conf->max_degraded;
1480 sector_t new_sector = sh->sector, check;
1481 int sectors_per_chunk = conf->chunk_size >> 9;
1484 int chunk_number, dummy1, dd_idx = i;
1486 struct stripe_head sh2;
1489 chunk_offset = sector_div(new_sector, sectors_per_chunk);
1490 stripe = new_sector;
1491 BUG_ON(new_sector != stripe);
1493 if (i == sh->pd_idx)
1495 switch(conf->level) {
1498 switch (conf->algorithm) {
1499 case ALGORITHM_LEFT_ASYMMETRIC:
1500 case ALGORITHM_RIGHT_ASYMMETRIC:
1504 case ALGORITHM_LEFT_SYMMETRIC:
1505 case ALGORITHM_RIGHT_SYMMETRIC:
1508 i -= (sh->pd_idx + 1);
1510 case ALGORITHM_PARITY_0:
1513 case ALGORITHM_PARITY_N:
1516 printk(KERN_ERR "raid5: unsupported algorithm %d\n",
1522 if (i == sh->qd_idx)
1523 return 0; /* It is the Q disk */
1524 switch (conf->algorithm) {
1525 case ALGORITHM_LEFT_ASYMMETRIC:
1526 case ALGORITHM_RIGHT_ASYMMETRIC:
1527 case ALGORITHM_ROTATING_ZERO_RESTART:
1528 case ALGORITHM_ROTATING_N_RESTART:
1529 if (sh->pd_idx == raid_disks-1)
1530 i--; /* Q D D D P */
1531 else if (i > sh->pd_idx)
1532 i -= 2; /* D D P Q D */
1534 case ALGORITHM_LEFT_SYMMETRIC:
1535 case ALGORITHM_RIGHT_SYMMETRIC:
1536 if (sh->pd_idx == raid_disks-1)
1537 i--; /* Q D D D P */
1542 i -= (sh->pd_idx + 2);
1545 case ALGORITHM_PARITY_0:
1548 case ALGORITHM_PARITY_N:
1550 case ALGORITHM_ROTATING_N_CONTINUE:
1551 if (sh->pd_idx == 0)
1552 i--; /* P D D D Q */
1553 else if (i > sh->pd_idx)
1554 i -= 2; /* D D Q P D */
1556 case ALGORITHM_LEFT_ASYMMETRIC_6:
1557 case ALGORITHM_RIGHT_ASYMMETRIC_6:
1561 case ALGORITHM_LEFT_SYMMETRIC_6:
1562 case ALGORITHM_RIGHT_SYMMETRIC_6:
1564 i += data_disks + 1;
1565 i -= (sh->pd_idx + 1);
1567 case ALGORITHM_PARITY_0_6:
1571 printk(KERN_CRIT "raid6: unsupported algorithm %d\n",
1578 chunk_number = stripe * data_disks + i;
1579 r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset;
1581 check = raid5_compute_sector(conf, r_sector,
1582 (raid_disks != conf->raid_disks),
1584 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
1585 || sh2.qd_idx != sh->qd_idx) {
1586 printk(KERN_ERR "compute_blocknr: map not correct\n");
1595 * Copy data between a page in the stripe cache, and one or more bion
1596 * The page could align with the middle of the bio, or there could be
1597 * several bion, each with several bio_vecs, which cover part of the page
1598 * Multiple bion are linked together on bi_next. There may be extras
1599 * at the end of this list. We ignore them.
1601 static void copy_data(int frombio, struct bio *bio,
1605 char *pa = page_address(page);
1606 struct bio_vec *bvl;
1610 if (bio->bi_sector >= sector)
1611 page_offset = (signed)(bio->bi_sector - sector) * 512;
1613 page_offset = (signed)(sector - bio->bi_sector) * -512;
1614 bio_for_each_segment(bvl, bio, i) {
1615 int len = bio_iovec_idx(bio,i)->bv_len;
1619 if (page_offset < 0) {
1620 b_offset = -page_offset;
1621 page_offset += b_offset;
1625 if (len > 0 && page_offset + len > STRIPE_SIZE)
1626 clen = STRIPE_SIZE - page_offset;
1630 char *ba = __bio_kmap_atomic(bio, i, KM_USER0);
1632 memcpy(pa+page_offset, ba+b_offset, clen);
1634 memcpy(ba+b_offset, pa+page_offset, clen);
1635 __bio_kunmap_atomic(ba, KM_USER0);
1637 if (clen < len) /* hit end of page */
1643 #define check_xor() do { \
1644 if (count == MAX_XOR_BLOCKS) { \
1645 xor_blocks(count, STRIPE_SIZE, dest, ptr);\
1650 static void compute_parity6(struct stripe_head *sh, int method)
1652 raid5_conf_t *conf = sh->raid_conf;
1653 int i, pd_idx, qd_idx, d0_idx, disks = sh->disks, count;
1654 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
1656 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1657 void *ptrs[syndrome_disks+2];
1659 pd_idx = sh->pd_idx;
1660 qd_idx = sh->qd_idx;
1661 d0_idx = raid6_d0(sh);
1663 pr_debug("compute_parity, stripe %llu, method %d\n",
1664 (unsigned long long)sh->sector, method);
1667 case READ_MODIFY_WRITE:
1668 BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */
1669 case RECONSTRUCT_WRITE:
1670 for (i= disks; i-- ;)
1671 if ( i != pd_idx && i != qd_idx && sh->dev[i].towrite ) {
1672 chosen = sh->dev[i].towrite;
1673 sh->dev[i].towrite = NULL;
1675 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
1676 wake_up(&conf->wait_for_overlap);
1678 BUG_ON(sh->dev[i].written);
1679 sh->dev[i].written = chosen;
1683 BUG(); /* Not implemented yet */
1686 for (i = disks; i--;)
1687 if (sh->dev[i].written) {
1688 sector_t sector = sh->dev[i].sector;
1689 struct bio *wbi = sh->dev[i].written;
1690 while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) {
1691 copy_data(1, wbi, sh->dev[i].page, sector);
1692 wbi = r5_next_bio(wbi, sector);
1695 set_bit(R5_LOCKED, &sh->dev[i].flags);
1696 set_bit(R5_UPTODATE, &sh->dev[i].flags);
1699 /* Note that unlike RAID-5, the ordering of the disks matters greatly.*/
1701 for (i = 0; i < disks; i++)
1702 ptrs[i] = (void *)raid6_empty_zero_page;
1707 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1709 ptrs[slot] = page_address(sh->dev[i].page);
1710 if (slot < syndrome_disks &&
1711 !test_bit(R5_UPTODATE, &sh->dev[i].flags)) {
1712 printk(KERN_ERR "block %d/%d not uptodate "
1713 "on parity calc\n", i, count);
1717 i = raid6_next_disk(i, disks);
1718 } while (i != d0_idx);
1719 BUG_ON(count != syndrome_disks);
1721 raid6_call.gen_syndrome(syndrome_disks+2, STRIPE_SIZE, ptrs);
1724 case RECONSTRUCT_WRITE:
1725 set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
1726 set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags);
1727 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
1728 set_bit(R5_LOCKED, &sh->dev[qd_idx].flags);
1731 set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
1732 set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags);
1738 /* Compute one missing block */
1739 static void compute_block_1(struct stripe_head *sh, int dd_idx, int nozero)
1741 int i, count, disks = sh->disks;
1742 void *ptr[MAX_XOR_BLOCKS], *dest, *p;
1743 int qd_idx = sh->qd_idx;
1745 pr_debug("compute_block_1, stripe %llu, idx %d\n",
1746 (unsigned long long)sh->sector, dd_idx);
1748 if ( dd_idx == qd_idx ) {
1749 /* We're actually computing the Q drive */
1750 compute_parity6(sh, UPDATE_PARITY);
1752 dest = page_address(sh->dev[dd_idx].page);
1753 if (!nozero) memset(dest, 0, STRIPE_SIZE);
1755 for (i = disks ; i--; ) {
1756 if (i == dd_idx || i == qd_idx)
1758 p = page_address(sh->dev[i].page);
1759 if (test_bit(R5_UPTODATE, &sh->dev[i].flags))
1762 printk("compute_block() %d, stripe %llu, %d"
1763 " not present\n", dd_idx,
1764 (unsigned long long)sh->sector, i);
1769 xor_blocks(count, STRIPE_SIZE, dest, ptr);
1770 if (!nozero) set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags);
1771 else clear_bit(R5_UPTODATE, &sh->dev[dd_idx].flags);
1775 /* Compute two missing blocks */
1776 static void compute_block_2(struct stripe_head *sh, int dd_idx1, int dd_idx2)
1778 int i, count, disks = sh->disks;
1779 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1780 int d0_idx = raid6_d0(sh);
1781 int faila = -1, failb = -1;
1782 /**** FIX THIS: This could be very bad if disks is close to 256 ****/
1783 void *ptrs[syndrome_disks+2];
1785 for (i = 0; i < disks ; i++)
1786 ptrs[i] = (void *)raid6_empty_zero_page;
1790 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1792 ptrs[slot] = page_address(sh->dev[i].page);
1798 i = raid6_next_disk(i, disks);
1799 } while (i != d0_idx);
1800 BUG_ON(count != syndrome_disks);
1802 BUG_ON(faila == failb);
1803 if ( failb < faila ) { int tmp = faila; faila = failb; failb = tmp; }
1805 pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n",
1806 (unsigned long long)sh->sector, dd_idx1, dd_idx2,
1809 if (failb == syndrome_disks+1) {
1810 /* Q disk is one of the missing disks */
1811 if (faila == syndrome_disks) {
1812 /* Missing P+Q, just recompute */
1813 compute_parity6(sh, UPDATE_PARITY);
1816 /* We're missing D+Q; recompute D from P */
1817 compute_block_1(sh, ((dd_idx1 == sh->qd_idx) ?
1820 compute_parity6(sh, UPDATE_PARITY); /* Is this necessary? */
1825 /* We're missing D+P or D+D; */
1826 if (failb == syndrome_disks) {
1827 /* We're missing D+P. */
1828 raid6_datap_recov(syndrome_disks+2, STRIPE_SIZE, faila, ptrs);
1830 /* We're missing D+D. */
1831 raid6_2data_recov(syndrome_disks+2, STRIPE_SIZE, faila, failb,
1835 /* Both the above update both missing blocks */
1836 set_bit(R5_UPTODATE, &sh->dev[dd_idx1].flags);
1837 set_bit(R5_UPTODATE, &sh->dev[dd_idx2].flags);
1841 schedule_reconstruction5(struct stripe_head *sh, struct stripe_head_state *s,
1842 int rcw, int expand)
1844 int i, pd_idx = sh->pd_idx, disks = sh->disks;
1847 /* if we are not expanding this is a proper write request, and
1848 * there will be bios with new data to be drained into the
1852 sh->reconstruct_state = reconstruct_state_drain_run;
1853 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
1855 sh->reconstruct_state = reconstruct_state_run;
1857 set_bit(STRIPE_OP_POSTXOR, &s->ops_request);
1859 for (i = disks; i--; ) {
1860 struct r5dev *dev = &sh->dev[i];
1863 set_bit(R5_LOCKED, &dev->flags);
1864 set_bit(R5_Wantdrain, &dev->flags);
1866 clear_bit(R5_UPTODATE, &dev->flags);
1870 if (s->locked + 1 == disks)
1871 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
1872 atomic_inc(&sh->raid_conf->pending_full_writes);
1874 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
1875 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
1877 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
1878 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
1879 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
1880 set_bit(STRIPE_OP_POSTXOR, &s->ops_request);
1882 for (i = disks; i--; ) {
1883 struct r5dev *dev = &sh->dev[i];
1888 (test_bit(R5_UPTODATE, &dev->flags) ||
1889 test_bit(R5_Wantcompute, &dev->flags))) {
1890 set_bit(R5_Wantdrain, &dev->flags);
1891 set_bit(R5_LOCKED, &dev->flags);
1892 clear_bit(R5_UPTODATE, &dev->flags);
1898 /* keep the parity disk locked while asynchronous operations
1901 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
1902 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
1905 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
1906 __func__, (unsigned long long)sh->sector,
1907 s->locked, s->ops_request);
1911 * Each stripe/dev can have one or more bion attached.
1912 * toread/towrite point to the first in a chain.
1913 * The bi_next chain must be in order.
1915 static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
1918 raid5_conf_t *conf = sh->raid_conf;
1921 pr_debug("adding bh b#%llu to stripe s#%llu\n",
1922 (unsigned long long)bi->bi_sector,
1923 (unsigned long long)sh->sector);
1926 spin_lock(&sh->lock);
1927 spin_lock_irq(&conf->device_lock);
1929 bip = &sh->dev[dd_idx].towrite;
1930 if (*bip == NULL && sh->dev[dd_idx].written == NULL)
1933 bip = &sh->dev[dd_idx].toread;
1934 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
1935 if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector)
1937 bip = & (*bip)->bi_next;
1939 if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9))
1942 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
1946 bi->bi_phys_segments++;
1947 spin_unlock_irq(&conf->device_lock);
1948 spin_unlock(&sh->lock);
1950 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
1951 (unsigned long long)bi->bi_sector,
1952 (unsigned long long)sh->sector, dd_idx);
1954 if (conf->mddev->bitmap && firstwrite) {
1955 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
1957 sh->bm_seq = conf->seq_flush+1;
1958 set_bit(STRIPE_BIT_DELAY, &sh->state);
1962 /* check if page is covered */
1963 sector_t sector = sh->dev[dd_idx].sector;
1964 for (bi=sh->dev[dd_idx].towrite;
1965 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
1966 bi && bi->bi_sector <= sector;
1967 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
1968 if (bi->bi_sector + (bi->bi_size>>9) >= sector)
1969 sector = bi->bi_sector + (bi->bi_size>>9);
1971 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
1972 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
1977 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
1978 spin_unlock_irq(&conf->device_lock);
1979 spin_unlock(&sh->lock);
1983 static void end_reshape(raid5_conf_t *conf);
1985 static int page_is_zero(struct page *p)
1987 char *a = page_address(p);
1988 return ((*(u32*)a) == 0 &&
1989 memcmp(a, a+4, STRIPE_SIZE-4)==0);
1992 static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
1993 struct stripe_head *sh)
1995 int sectors_per_chunk = conf->chunk_size >> 9;
1997 int chunk_offset = sector_div(stripe, sectors_per_chunk);
1998 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
2000 raid5_compute_sector(conf,
2001 stripe * (disks - conf->max_degraded)
2002 *sectors_per_chunk + chunk_offset,
2008 handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh,
2009 struct stripe_head_state *s, int disks,
2010 struct bio **return_bi)
2013 for (i = disks; i--; ) {
2017 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
2020 rdev = rcu_dereference(conf->disks[i].rdev);
2021 if (rdev && test_bit(In_sync, &rdev->flags))
2022 /* multiple read failures in one stripe */
2023 md_error(conf->mddev, rdev);
2026 spin_lock_irq(&conf->device_lock);
2027 /* fail all writes first */
2028 bi = sh->dev[i].towrite;
2029 sh->dev[i].towrite = NULL;
2035 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2036 wake_up(&conf->wait_for_overlap);
2038 while (bi && bi->bi_sector <
2039 sh->dev[i].sector + STRIPE_SECTORS) {
2040 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2041 clear_bit(BIO_UPTODATE, &bi->bi_flags);
2042 if (!raid5_dec_bi_phys_segments(bi)) {
2043 md_write_end(conf->mddev);
2044 bi->bi_next = *return_bi;
2049 /* and fail all 'written' */
2050 bi = sh->dev[i].written;
2051 sh->dev[i].written = NULL;
2052 if (bi) bitmap_end = 1;
2053 while (bi && bi->bi_sector <
2054 sh->dev[i].sector + STRIPE_SECTORS) {
2055 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2056 clear_bit(BIO_UPTODATE, &bi->bi_flags);
2057 if (!raid5_dec_bi_phys_segments(bi)) {
2058 md_write_end(conf->mddev);
2059 bi->bi_next = *return_bi;
2065 /* fail any reads if this device is non-operational and
2066 * the data has not reached the cache yet.
2068 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2069 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2070 test_bit(R5_ReadError, &sh->dev[i].flags))) {
2071 bi = sh->dev[i].toread;
2072 sh->dev[i].toread = NULL;
2073 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2074 wake_up(&conf->wait_for_overlap);
2075 if (bi) s->to_read--;
2076 while (bi && bi->bi_sector <
2077 sh->dev[i].sector + STRIPE_SECTORS) {
2078 struct bio *nextbi =
2079 r5_next_bio(bi, sh->dev[i].sector);
2080 clear_bit(BIO_UPTODATE, &bi->bi_flags);
2081 if (!raid5_dec_bi_phys_segments(bi)) {
2082 bi->bi_next = *return_bi;
2088 spin_unlock_irq(&conf->device_lock);
2090 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2091 STRIPE_SECTORS, 0, 0);
2094 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2095 if (atomic_dec_and_test(&conf->pending_full_writes))
2096 md_wakeup_thread(conf->mddev->thread);
2099 /* fetch_block5 - checks the given member device to see if its data needs
2100 * to be read or computed to satisfy a request.
2102 * Returns 1 when no more member devices need to be checked, otherwise returns
2103 * 0 to tell the loop in handle_stripe_fill5 to continue
2105 static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s,
2106 int disk_idx, int disks)
2108 struct r5dev *dev = &sh->dev[disk_idx];
2109 struct r5dev *failed_dev = &sh->dev[s->failed_num];
2111 /* is the data in this block needed, and can we get it? */
2112 if (!test_bit(R5_LOCKED, &dev->flags) &&
2113 !test_bit(R5_UPTODATE, &dev->flags) &&
2115 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2116 s->syncing || s->expanding ||
2118 (failed_dev->toread ||
2119 (failed_dev->towrite &&
2120 !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) {
2121 /* We would like to get this block, possibly by computing it,
2122 * otherwise read it if the backing disk is insync
2124 if ((s->uptodate == disks - 1) &&
2125 (s->failed && disk_idx == s->failed_num)) {
2126 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2127 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2128 set_bit(R5_Wantcompute, &dev->flags);
2129 sh->ops.target = disk_idx;
2131 /* Careful: from this point on 'uptodate' is in the eye
2132 * of raid5_run_ops which services 'compute' operations
2133 * before writes. R5_Wantcompute flags a block that will
2134 * be R5_UPTODATE by the time it is needed for a
2135 * subsequent operation.
2138 return 1; /* uptodate + compute == disks */
2139 } else if (test_bit(R5_Insync, &dev->flags)) {
2140 set_bit(R5_LOCKED, &dev->flags);
2141 set_bit(R5_Wantread, &dev->flags);
2143 pr_debug("Reading block %d (sync=%d)\n", disk_idx,
2152 * handle_stripe_fill5 - read or compute data to satisfy pending requests.
2154 static void handle_stripe_fill5(struct stripe_head *sh,
2155 struct stripe_head_state *s, int disks)
2159 /* look for blocks to read/compute, skip this if a compute
2160 * is already in flight, or if the stripe contents are in the
2161 * midst of changing due to a write
2163 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
2164 !sh->reconstruct_state)
2165 for (i = disks; i--; )
2166 if (fetch_block5(sh, s, i, disks))
2168 set_bit(STRIPE_HANDLE, &sh->state);
2171 static void handle_stripe_fill6(struct stripe_head *sh,
2172 struct stripe_head_state *s, struct r6_state *r6s,
2176 for (i = disks; i--; ) {
2177 struct r5dev *dev = &sh->dev[i];
2178 if (!test_bit(R5_LOCKED, &dev->flags) &&
2179 !test_bit(R5_UPTODATE, &dev->flags) &&
2180 (dev->toread || (dev->towrite &&
2181 !test_bit(R5_OVERWRITE, &dev->flags)) ||
2182 s->syncing || s->expanding ||
2184 (sh->dev[r6s->failed_num[0]].toread ||
2187 (sh->dev[r6s->failed_num[1]].toread ||
2189 /* we would like to get this block, possibly
2190 * by computing it, but we might not be able to
2192 if ((s->uptodate == disks - 1) &&
2193 (s->failed && (i == r6s->failed_num[0] ||
2194 i == r6s->failed_num[1]))) {
2195 pr_debug("Computing stripe %llu block %d\n",
2196 (unsigned long long)sh->sector, i);
2197 compute_block_1(sh, i, 0);
2199 } else if ( s->uptodate == disks-2 && s->failed >= 2 ) {
2200 /* Computing 2-failure is *very* expensive; only
2201 * do it if failed >= 2
2204 for (other = disks; other--; ) {
2207 if (!test_bit(R5_UPTODATE,
2208 &sh->dev[other].flags))
2212 pr_debug("Computing stripe %llu blocks %d,%d\n",
2213 (unsigned long long)sh->sector,
2215 compute_block_2(sh, i, other);
2217 } else if (test_bit(R5_Insync, &dev->flags)) {
2218 set_bit(R5_LOCKED, &dev->flags);
2219 set_bit(R5_Wantread, &dev->flags);
2221 pr_debug("Reading block %d (sync=%d)\n",
2226 set_bit(STRIPE_HANDLE, &sh->state);
2230 /* handle_stripe_clean_event
2231 * any written block on an uptodate or failed drive can be returned.
2232 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2233 * never LOCKED, so we don't need to test 'failed' directly.
2235 static void handle_stripe_clean_event(raid5_conf_t *conf,
2236 struct stripe_head *sh, int disks, struct bio **return_bi)
2241 for (i = disks; i--; )
2242 if (sh->dev[i].written) {
2244 if (!test_bit(R5_LOCKED, &dev->flags) &&
2245 test_bit(R5_UPTODATE, &dev->flags)) {
2246 /* We can return any write requests */
2247 struct bio *wbi, *wbi2;
2249 pr_debug("Return write for disc %d\n", i);
2250 spin_lock_irq(&conf->device_lock);
2252 dev->written = NULL;
2253 while (wbi && wbi->bi_sector <
2254 dev->sector + STRIPE_SECTORS) {
2255 wbi2 = r5_next_bio(wbi, dev->sector);
2256 if (!raid5_dec_bi_phys_segments(wbi)) {
2257 md_write_end(conf->mddev);
2258 wbi->bi_next = *return_bi;
2263 if (dev->towrite == NULL)
2265 spin_unlock_irq(&conf->device_lock);
2267 bitmap_endwrite(conf->mddev->bitmap,
2270 !test_bit(STRIPE_DEGRADED, &sh->state),
2275 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2276 if (atomic_dec_and_test(&conf->pending_full_writes))
2277 md_wakeup_thread(conf->mddev->thread);
2280 static void handle_stripe_dirtying5(raid5_conf_t *conf,
2281 struct stripe_head *sh, struct stripe_head_state *s, int disks)
2283 int rmw = 0, rcw = 0, i;
2284 for (i = disks; i--; ) {
2285 /* would I have to read this buffer for read_modify_write */
2286 struct r5dev *dev = &sh->dev[i];
2287 if ((dev->towrite || i == sh->pd_idx) &&
2288 !test_bit(R5_LOCKED, &dev->flags) &&
2289 !(test_bit(R5_UPTODATE, &dev->flags) ||
2290 test_bit(R5_Wantcompute, &dev->flags))) {
2291 if (test_bit(R5_Insync, &dev->flags))
2294 rmw += 2*disks; /* cannot read it */
2296 /* Would I have to read this buffer for reconstruct_write */
2297 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
2298 !test_bit(R5_LOCKED, &dev->flags) &&
2299 !(test_bit(R5_UPTODATE, &dev->flags) ||
2300 test_bit(R5_Wantcompute, &dev->flags))) {
2301 if (test_bit(R5_Insync, &dev->flags)) rcw++;
2306 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
2307 (unsigned long long)sh->sector, rmw, rcw);
2308 set_bit(STRIPE_HANDLE, &sh->state);
2309 if (rmw < rcw && rmw > 0)
2310 /* prefer read-modify-write, but need to get some data */
2311 for (i = disks; i--; ) {
2312 struct r5dev *dev = &sh->dev[i];
2313 if ((dev->towrite || i == sh->pd_idx) &&
2314 !test_bit(R5_LOCKED, &dev->flags) &&
2315 !(test_bit(R5_UPTODATE, &dev->flags) ||
2316 test_bit(R5_Wantcompute, &dev->flags)) &&
2317 test_bit(R5_Insync, &dev->flags)) {
2319 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2320 pr_debug("Read_old block "
2321 "%d for r-m-w\n", i);
2322 set_bit(R5_LOCKED, &dev->flags);
2323 set_bit(R5_Wantread, &dev->flags);
2326 set_bit(STRIPE_DELAYED, &sh->state);
2327 set_bit(STRIPE_HANDLE, &sh->state);
2331 if (rcw <= rmw && rcw > 0)
2332 /* want reconstruct write, but need to get some data */
2333 for (i = disks; i--; ) {
2334 struct r5dev *dev = &sh->dev[i];
2335 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2337 !test_bit(R5_LOCKED, &dev->flags) &&
2338 !(test_bit(R5_UPTODATE, &dev->flags) ||
2339 test_bit(R5_Wantcompute, &dev->flags)) &&
2340 test_bit(R5_Insync, &dev->flags)) {
2342 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2343 pr_debug("Read_old block "
2344 "%d for Reconstruct\n", i);
2345 set_bit(R5_LOCKED, &dev->flags);
2346 set_bit(R5_Wantread, &dev->flags);
2349 set_bit(STRIPE_DELAYED, &sh->state);
2350 set_bit(STRIPE_HANDLE, &sh->state);
2354 /* now if nothing is locked, and if we have enough data,
2355 * we can start a write request
2357 /* since handle_stripe can be called at any time we need to handle the
2358 * case where a compute block operation has been submitted and then a
2359 * subsequent call wants to start a write request. raid5_run_ops only
2360 * handles the case where compute block and postxor are requested
2361 * simultaneously. If this is not the case then new writes need to be
2362 * held off until the compute completes.
2364 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2365 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
2366 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
2367 schedule_reconstruction5(sh, s, rcw == 0, 0);
2370 static void handle_stripe_dirtying6(raid5_conf_t *conf,
2371 struct stripe_head *sh, struct stripe_head_state *s,
2372 struct r6_state *r6s, int disks)
2374 int rcw = 0, must_compute = 0, pd_idx = sh->pd_idx, i;
2375 int qd_idx = sh->qd_idx;
2376 for (i = disks; i--; ) {
2377 struct r5dev *dev = &sh->dev[i];
2378 /* Would I have to read this buffer for reconstruct_write */
2379 if (!test_bit(R5_OVERWRITE, &dev->flags)
2380 && i != pd_idx && i != qd_idx
2381 && (!test_bit(R5_LOCKED, &dev->flags)
2383 !test_bit(R5_UPTODATE, &dev->flags)) {
2384 if (test_bit(R5_Insync, &dev->flags)) rcw++;
2386 pr_debug("raid6: must_compute: "
2387 "disk %d flags=%#lx\n", i, dev->flags);
2392 pr_debug("for sector %llu, rcw=%d, must_compute=%d\n",
2393 (unsigned long long)sh->sector, rcw, must_compute);
2394 set_bit(STRIPE_HANDLE, &sh->state);
2397 /* want reconstruct write, but need to get some data */
2398 for (i = disks; i--; ) {
2399 struct r5dev *dev = &sh->dev[i];
2400 if (!test_bit(R5_OVERWRITE, &dev->flags)
2401 && !(s->failed == 0 && (i == pd_idx || i == qd_idx))
2402 && !test_bit(R5_LOCKED, &dev->flags) &&
2403 !test_bit(R5_UPTODATE, &dev->flags) &&
2404 test_bit(R5_Insync, &dev->flags)) {
2406 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2407 pr_debug("Read_old stripe %llu "
2408 "block %d for Reconstruct\n",
2409 (unsigned long long)sh->sector, i);
2410 set_bit(R5_LOCKED, &dev->flags);
2411 set_bit(R5_Wantread, &dev->flags);
2414 pr_debug("Request delayed stripe %llu "
2415 "block %d for Reconstruct\n",
2416 (unsigned long long)sh->sector, i);
2417 set_bit(STRIPE_DELAYED, &sh->state);
2418 set_bit(STRIPE_HANDLE, &sh->state);
2422 /* now if nothing is locked, and if we have enough data, we can start a
2425 if (s->locked == 0 && rcw == 0 &&
2426 !test_bit(STRIPE_BIT_DELAY, &sh->state)) {
2427 if (must_compute > 0) {
2428 /* We have failed blocks and need to compute them */
2429 switch (s->failed) {
2433 compute_block_1(sh, r6s->failed_num[0], 0);
2436 compute_block_2(sh, r6s->failed_num[0],
2437 r6s->failed_num[1]);
2439 default: /* This request should have been failed? */
2444 pr_debug("Computing parity for stripe %llu\n",
2445 (unsigned long long)sh->sector);
2446 compute_parity6(sh, RECONSTRUCT_WRITE);
2447 /* now every locked buffer is ready to be written */
2448 for (i = disks; i--; )
2449 if (test_bit(R5_LOCKED, &sh->dev[i].flags)) {
2450 pr_debug("Writing stripe %llu block %d\n",
2451 (unsigned long long)sh->sector, i);
2453 set_bit(R5_Wantwrite, &sh->dev[i].flags);
2455 if (s->locked == disks)
2456 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
2457 atomic_inc(&conf->pending_full_writes);
2458 /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */
2459 set_bit(STRIPE_INSYNC, &sh->state);
2461 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2462 atomic_dec(&conf->preread_active_stripes);
2463 if (atomic_read(&conf->preread_active_stripes) <
2465 md_wakeup_thread(conf->mddev->thread);
2470 static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh,
2471 struct stripe_head_state *s, int disks)
2473 struct r5dev *dev = NULL;
2475 set_bit(STRIPE_HANDLE, &sh->state);
2477 switch (sh->check_state) {
2478 case check_state_idle:
2479 /* start a new check operation if there are no failures */
2480 if (s->failed == 0) {
2481 BUG_ON(s->uptodate != disks);
2482 sh->check_state = check_state_run;
2483 set_bit(STRIPE_OP_CHECK, &s->ops_request);
2484 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
2488 dev = &sh->dev[s->failed_num];
2490 case check_state_compute_result:
2491 sh->check_state = check_state_idle;
2493 dev = &sh->dev[sh->pd_idx];
2495 /* check that a write has not made the stripe insync */
2496 if (test_bit(STRIPE_INSYNC, &sh->state))
2499 /* either failed parity check, or recovery is happening */
2500 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
2501 BUG_ON(s->uptodate != disks);
2503 set_bit(R5_LOCKED, &dev->flags);
2505 set_bit(R5_Wantwrite, &dev->flags);
2507 clear_bit(STRIPE_DEGRADED, &sh->state);
2508 set_bit(STRIPE_INSYNC, &sh->state);
2510 case check_state_run:
2511 break; /* we will be called again upon completion */
2512 case check_state_check_result:
2513 sh->check_state = check_state_idle;
2515 /* if a failure occurred during the check operation, leave
2516 * STRIPE_INSYNC not set and let the stripe be handled again
2521 /* handle a successful check operation, if parity is correct
2522 * we are done. Otherwise update the mismatch count and repair
2523 * parity if !MD_RECOVERY_CHECK
2525 if (sh->ops.zero_sum_result == 0)
2526 /* parity is correct (on disc,
2527 * not in buffer any more)
2529 set_bit(STRIPE_INSYNC, &sh->state);
2531 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2532 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2533 /* don't try to repair!! */
2534 set_bit(STRIPE_INSYNC, &sh->state);
2536 sh->check_state = check_state_compute_run;
2537 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2538 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2539 set_bit(R5_Wantcompute,
2540 &sh->dev[sh->pd_idx].flags);
2541 sh->ops.target = sh->pd_idx;
2546 case check_state_compute_run:
2549 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2550 __func__, sh->check_state,
2551 (unsigned long long) sh->sector);
2557 static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
2558 struct stripe_head_state *s,
2559 struct r6_state *r6s, struct page *tmp_page,
2562 int update_p = 0, update_q = 0;
2564 int pd_idx = sh->pd_idx;
2565 int qd_idx = sh->qd_idx;
2567 set_bit(STRIPE_HANDLE, &sh->state);
2569 BUG_ON(s->failed > 2);
2570 BUG_ON(s->uptodate < disks);
2571 /* Want to check and possibly repair P and Q.
2572 * However there could be one 'failed' device, in which
2573 * case we can only check one of them, possibly using the
2574 * other to generate missing data
2577 /* If !tmp_page, we cannot do the calculations,
2578 * but as we have set STRIPE_HANDLE, we will soon be called
2579 * by stripe_handle with a tmp_page - just wait until then.
2582 if (s->failed == r6s->q_failed) {
2583 /* The only possible failed device holds 'Q', so it
2584 * makes sense to check P (If anything else were failed,
2585 * we would have used P to recreate it).
2587 compute_block_1(sh, pd_idx, 1);
2588 if (!page_is_zero(sh->dev[pd_idx].page)) {
2589 compute_block_1(sh, pd_idx, 0);
2593 if (!r6s->q_failed && s->failed < 2) {
2594 /* q is not failed, and we didn't use it to generate
2595 * anything, so it makes sense to check it
2597 memcpy(page_address(tmp_page),
2598 page_address(sh->dev[qd_idx].page),
2600 compute_parity6(sh, UPDATE_PARITY);
2601 if (memcmp(page_address(tmp_page),
2602 page_address(sh->dev[qd_idx].page),
2603 STRIPE_SIZE) != 0) {
2604 clear_bit(STRIPE_INSYNC, &sh->state);
2608 if (update_p || update_q) {
2609 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2610 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2611 /* don't try to repair!! */
2612 update_p = update_q = 0;
2615 /* now write out any block on a failed drive,
2616 * or P or Q if they need it
2619 if (s->failed == 2) {
2620 dev = &sh->dev[r6s->failed_num[1]];
2622 set_bit(R5_LOCKED, &dev->flags);
2623 set_bit(R5_Wantwrite, &dev->flags);
2625 if (s->failed >= 1) {
2626 dev = &sh->dev[r6s->failed_num[0]];
2628 set_bit(R5_LOCKED, &dev->flags);
2629 set_bit(R5_Wantwrite, &dev->flags);
2633 dev = &sh->dev[pd_idx];
2635 set_bit(R5_LOCKED, &dev->flags);
2636 set_bit(R5_Wantwrite, &dev->flags);
2639 dev = &sh->dev[qd_idx];
2641 set_bit(R5_LOCKED, &dev->flags);
2642 set_bit(R5_Wantwrite, &dev->flags);
2644 clear_bit(STRIPE_DEGRADED, &sh->state);
2646 set_bit(STRIPE_INSYNC, &sh->state);
2650 static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh,
2651 struct r6_state *r6s)
2655 /* We have read all the blocks in this stripe and now we need to
2656 * copy some of them into a target stripe for expand.
2658 struct dma_async_tx_descriptor *tx = NULL;
2659 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2660 for (i = 0; i < sh->disks; i++)
2661 if (i != sh->pd_idx && i != sh->qd_idx) {
2663 struct stripe_head *sh2;
2665 sector_t bn = compute_blocknr(sh, i);
2666 sector_t s = raid5_compute_sector(conf, bn, 0,
2668 sh2 = get_active_stripe(conf, s, 0, 1);
2670 /* so far only the early blocks of this stripe
2671 * have been requested. When later blocks
2672 * get requested, we will try again
2675 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
2676 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
2677 /* must have already done this block */
2678 release_stripe(sh2);
2682 /* place all the copies on one channel */
2683 tx = async_memcpy(sh2->dev[dd_idx].page,
2684 sh->dev[i].page, 0, 0, STRIPE_SIZE,
2685 ASYNC_TX_DEP_ACK, tx, NULL, NULL);
2687 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
2688 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
2689 for (j = 0; j < conf->raid_disks; j++)
2690 if (j != sh2->pd_idx &&
2691 (!r6s || j != sh2->qd_idx) &&
2692 !test_bit(R5_Expanded, &sh2->dev[j].flags))
2694 if (j == conf->raid_disks) {
2695 set_bit(STRIPE_EXPAND_READY, &sh2->state);
2696 set_bit(STRIPE_HANDLE, &sh2->state);
2698 release_stripe(sh2);
2701 /* done submitting copies, wait for them to complete */
2704 dma_wait_for_async_tx(tx);
2710 * handle_stripe - do things to a stripe.
2712 * We lock the stripe and then examine the state of various bits
2713 * to see what needs to be done.
2715 * return some read request which now have data
2716 * return some write requests which are safely on disc
2717 * schedule a read on some buffers
2718 * schedule a write of some buffers
2719 * return confirmation of parity correctness
2721 * buffers are taken off read_list or write_list, and bh_cache buffers
2722 * get BH_Lock set before the stripe lock is released.
2726 static bool handle_stripe5(struct stripe_head *sh)
2728 raid5_conf_t *conf = sh->raid_conf;
2729 int disks = sh->disks, i;
2730 struct bio *return_bi = NULL;
2731 struct stripe_head_state s;
2733 mdk_rdev_t *blocked_rdev = NULL;
2736 memset(&s, 0, sizeof(s));
2737 pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d "
2738 "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state,
2739 atomic_read(&sh->count), sh->pd_idx, sh->check_state,
2740 sh->reconstruct_state);
2742 spin_lock(&sh->lock);
2743 clear_bit(STRIPE_HANDLE, &sh->state);
2744 clear_bit(STRIPE_DELAYED, &sh->state);
2746 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
2747 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2748 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
2750 /* Now to look around and see what can be done */
2752 for (i=disks; i--; ) {
2754 struct r5dev *dev = &sh->dev[i];
2755 clear_bit(R5_Insync, &dev->flags);
2757 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
2758 "written %p\n", i, dev->flags, dev->toread, dev->read,
2759 dev->towrite, dev->written);
2761 /* maybe we can request a biofill operation
2763 * new wantfill requests are only permitted while
2764 * ops_complete_biofill is guaranteed to be inactive
2766 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
2767 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
2768 set_bit(R5_Wantfill, &dev->flags);
2770 /* now count some things */
2771 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
2772 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
2773 if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++;
2775 if (test_bit(R5_Wantfill, &dev->flags))
2777 else if (dev->toread)
2781 if (!test_bit(R5_OVERWRITE, &dev->flags))
2786 rdev = rcu_dereference(conf->disks[i].rdev);
2787 if (blocked_rdev == NULL &&
2788 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
2789 blocked_rdev = rdev;
2790 atomic_inc(&rdev->nr_pending);
2792 if (!rdev || !test_bit(In_sync, &rdev->flags)) {
2793 /* The ReadError flag will just be confusing now */
2794 clear_bit(R5_ReadError, &dev->flags);
2795 clear_bit(R5_ReWrite, &dev->flags);
2797 if (!rdev || !test_bit(In_sync, &rdev->flags)
2798 || test_bit(R5_ReadError, &dev->flags)) {
2802 set_bit(R5_Insync, &dev->flags);
2806 if (unlikely(blocked_rdev)) {
2807 if (s.syncing || s.expanding || s.expanded ||
2808 s.to_write || s.written) {
2809 set_bit(STRIPE_HANDLE, &sh->state);
2812 /* There is nothing for the blocked_rdev to block */
2813 rdev_dec_pending(blocked_rdev, conf->mddev);
2814 blocked_rdev = NULL;
2817 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
2818 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
2819 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
2822 pr_debug("locked=%d uptodate=%d to_read=%d"
2823 " to_write=%d failed=%d failed_num=%d\n",
2824 s.locked, s.uptodate, s.to_read, s.to_write,
2825 s.failed, s.failed_num);
2826 /* check if the array has lost two devices and, if so, some requests might
2829 if (s.failed > 1 && s.to_read+s.to_write+s.written)
2830 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
2831 if (s.failed > 1 && s.syncing) {
2832 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
2833 clear_bit(STRIPE_SYNCING, &sh->state);
2837 /* might be able to return some write requests if the parity block
2838 * is safe, or on a failed drive
2840 dev = &sh->dev[sh->pd_idx];
2842 ((test_bit(R5_Insync, &dev->flags) &&
2843 !test_bit(R5_LOCKED, &dev->flags) &&
2844 test_bit(R5_UPTODATE, &dev->flags)) ||
2845 (s.failed == 1 && s.failed_num == sh->pd_idx)))
2846 handle_stripe_clean_event(conf, sh, disks, &return_bi);
2848 /* Now we might consider reading some blocks, either to check/generate
2849 * parity, or to satisfy requests
2850 * or to load a block that is being partially written.
2852 if (s.to_read || s.non_overwrite ||
2853 (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
2854 handle_stripe_fill5(sh, &s, disks);
2856 /* Now we check to see if any write operations have recently
2860 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
2862 if (sh->reconstruct_state == reconstruct_state_drain_result ||
2863 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
2864 sh->reconstruct_state = reconstruct_state_idle;
2866 /* All the 'written' buffers and the parity block are ready to
2867 * be written back to disk
2869 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
2870 for (i = disks; i--; ) {
2872 if (test_bit(R5_LOCKED, &dev->flags) &&
2873 (i == sh->pd_idx || dev->written)) {
2874 pr_debug("Writing block %d\n", i);
2875 set_bit(R5_Wantwrite, &dev->flags);
2878 if (!test_bit(R5_Insync, &dev->flags) ||
2879 (i == sh->pd_idx && s.failed == 0))
2880 set_bit(STRIPE_INSYNC, &sh->state);
2883 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2884 atomic_dec(&conf->preread_active_stripes);
2885 if (atomic_read(&conf->preread_active_stripes) <
2887 md_wakeup_thread(conf->mddev->thread);
2891 /* Now to consider new write requests and what else, if anything
2892 * should be read. We do not handle new writes when:
2893 * 1/ A 'write' operation (copy+xor) is already in flight.
2894 * 2/ A 'check' operation is in flight, as it may clobber the parity
2897 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
2898 handle_stripe_dirtying5(conf, sh, &s, disks);
2900 /* maybe we need to check and possibly fix the parity for this stripe
2901 * Any reads will already have been scheduled, so we just see if enough
2902 * data is available. The parity check is held off while parity
2903 * dependent operations are in flight.
2905 if (sh->check_state ||
2906 (s.syncing && s.locked == 0 &&
2907 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
2908 !test_bit(STRIPE_INSYNC, &sh->state)))
2909 handle_parity_checks5(conf, sh, &s, disks);
2911 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
2912 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
2913 clear_bit(STRIPE_SYNCING, &sh->state);
2916 /* If the failed drive is just a ReadError, then we might need to progress
2917 * the repair/check process
2919 if (s.failed == 1 && !conf->mddev->ro &&
2920 test_bit(R5_ReadError, &sh->dev[s.failed_num].flags)
2921 && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags)
2922 && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags)
2924 dev = &sh->dev[s.failed_num];
2925 if (!test_bit(R5_ReWrite, &dev->flags)) {
2926 set_bit(R5_Wantwrite, &dev->flags);
2927 set_bit(R5_ReWrite, &dev->flags);
2928 set_bit(R5_LOCKED, &dev->flags);
2931 /* let's read it back */
2932 set_bit(R5_Wantread, &dev->flags);
2933 set_bit(R5_LOCKED, &dev->flags);
2938 /* Finish reconstruct operations initiated by the expansion process */
2939 if (sh->reconstruct_state == reconstruct_state_result) {
2940 sh->reconstruct_state = reconstruct_state_idle;
2941 clear_bit(STRIPE_EXPANDING, &sh->state);
2942 for (i = conf->raid_disks; i--; ) {
2943 set_bit(R5_Wantwrite, &sh->dev[i].flags);
2944 set_bit(R5_LOCKED, &sh->dev[i].flags);
2949 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
2950 !sh->reconstruct_state) {
2951 /* Need to write out all blocks after computing parity */
2952 sh->disks = conf->raid_disks;
2953 stripe_set_idx(sh->sector, conf, 0, sh);
2954 schedule_reconstruction5(sh, &s, 1, 1);
2955 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
2956 clear_bit(STRIPE_EXPAND_READY, &sh->state);
2957 atomic_dec(&conf->reshape_stripes);
2958 wake_up(&conf->wait_for_overlap);
2959 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
2962 if (s.expanding && s.locked == 0 &&
2963 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
2964 handle_stripe_expansion(conf, sh, NULL);
2967 spin_unlock(&sh->lock);
2969 /* wait for this device to become unblocked */
2970 if (unlikely(blocked_rdev))
2971 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
2974 raid5_run_ops(sh, s.ops_request);
2978 return_io(return_bi);
2980 return blocked_rdev == NULL;
2983 static bool handle_stripe6(struct stripe_head *sh, struct page *tmp_page)
2985 raid5_conf_t *conf = sh->raid_conf;
2986 int disks = sh->disks;
2987 struct bio *return_bi = NULL;
2988 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx;
2989 struct stripe_head_state s;
2990 struct r6_state r6s;
2991 struct r5dev *dev, *pdev, *qdev;
2992 mdk_rdev_t *blocked_rdev = NULL;
2994 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
2995 "pd_idx=%d, qd_idx=%d\n",
2996 (unsigned long long)sh->sector, sh->state,
2997 atomic_read(&sh->count), pd_idx, qd_idx);
2998 memset(&s, 0, sizeof(s));
3000 spin_lock(&sh->lock);
3001 clear_bit(STRIPE_HANDLE, &sh->state);
3002 clear_bit(STRIPE_DELAYED, &sh->state);
3004 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
3005 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3006 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
3007 /* Now to look around and see what can be done */
3010 for (i=disks; i--; ) {
3013 clear_bit(R5_Insync, &dev->flags);
3015 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
3016 i, dev->flags, dev->toread, dev->towrite, dev->written);
3017 /* maybe we can reply to a read */
3018 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) {
3019 struct bio *rbi, *rbi2;
3020 pr_debug("Return read for disc %d\n", i);
3021 spin_lock_irq(&conf->device_lock);
3024 if (test_and_clear_bit(R5_Overlap, &dev->flags))
3025 wake_up(&conf->wait_for_overlap);
3026 spin_unlock_irq(&conf->device_lock);
3027 while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) {
3028 copy_data(0, rbi, dev->page, dev->sector);
3029 rbi2 = r5_next_bio(rbi, dev->sector);
3030 spin_lock_irq(&conf->device_lock);
3031 if (!raid5_dec_bi_phys_segments(rbi)) {
3032 rbi->bi_next = return_bi;
3035 spin_unlock_irq(&conf->device_lock);
3040 /* now count some things */
3041 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
3042 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
3049 if (!test_bit(R5_OVERWRITE, &dev->flags))
3054 rdev = rcu_dereference(conf->disks[i].rdev);
3055 if (blocked_rdev == NULL &&
3056 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
3057 blocked_rdev = rdev;
3058 atomic_inc(&rdev->nr_pending);
3060 if (!rdev || !test_bit(In_sync, &rdev->flags)) {
3061 /* The ReadError flag will just be confusing now */
3062 clear_bit(R5_ReadError, &dev->flags);
3063 clear_bit(R5_ReWrite, &dev->flags);
3065 if (!rdev || !test_bit(In_sync, &rdev->flags)
3066 || test_bit(R5_ReadError, &dev->flags)) {
3068 r6s.failed_num[s.failed] = i;
3071 set_bit(R5_Insync, &dev->flags);
3075 if (unlikely(blocked_rdev)) {
3076 if (s.syncing || s.expanding || s.expanded ||
3077 s.to_write || s.written) {
3078 set_bit(STRIPE_HANDLE, &sh->state);
3081 /* There is nothing for the blocked_rdev to block */
3082 rdev_dec_pending(blocked_rdev, conf->mddev);
3083 blocked_rdev = NULL;
3086 pr_debug("locked=%d uptodate=%d to_read=%d"
3087 " to_write=%d failed=%d failed_num=%d,%d\n",
3088 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
3089 r6s.failed_num[0], r6s.failed_num[1]);
3090 /* check if the array has lost >2 devices and, if so, some requests
3091 * might need to be failed
3093 if (s.failed > 2 && s.to_read+s.to_write+s.written)
3094 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
3095 if (s.failed > 2 && s.syncing) {
3096 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3097 clear_bit(STRIPE_SYNCING, &sh->state);
3102 * might be able to return some write requests if the parity blocks
3103 * are safe, or on a failed drive
3105 pdev = &sh->dev[pd_idx];
3106 r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx)
3107 || (s.failed >= 2 && r6s.failed_num[1] == pd_idx);
3108 qdev = &sh->dev[qd_idx];
3109 r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == qd_idx)
3110 || (s.failed >= 2 && r6s.failed_num[1] == qd_idx);
3113 ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
3114 && !test_bit(R5_LOCKED, &pdev->flags)
3115 && test_bit(R5_UPTODATE, &pdev->flags)))) &&
3116 ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
3117 && !test_bit(R5_LOCKED, &qdev->flags)
3118 && test_bit(R5_UPTODATE, &qdev->flags)))))
3119 handle_stripe_clean_event(conf, sh, disks, &return_bi);
3121 /* Now we might consider reading some blocks, either to check/generate
3122 * parity, or to satisfy requests
3123 * or to load a block that is being partially written.
3125 if (s.to_read || s.non_overwrite || (s.to_write && s.failed) ||
3126 (s.syncing && (s.uptodate < disks)) || s.expanding)
3127 handle_stripe_fill6(sh, &s, &r6s, disks);
3129 /* now to consider writing and what else, if anything should be read */
3131 handle_stripe_dirtying6(conf, sh, &s, &r6s, disks);
3133 /* maybe we need to check and possibly fix the parity for this stripe
3134 * Any reads will already have been scheduled, so we just see if enough
3137 if (s.syncing && s.locked == 0 && !test_bit(STRIPE_INSYNC, &sh->state))
3138 handle_parity_checks6(conf, sh, &s, &r6s, tmp_page, disks);
3140 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
3141 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3142 clear_bit(STRIPE_SYNCING, &sh->state);
3145 /* If the failed drives are just a ReadError, then we might need
3146 * to progress the repair/check process
3148 if (s.failed <= 2 && !conf->mddev->ro)
3149 for (i = 0; i < s.failed; i++) {
3150 dev = &sh->dev[r6s.failed_num[i]];
3151 if (test_bit(R5_ReadError, &dev->flags)
3152 && !test_bit(R5_LOCKED, &dev->flags)
3153 && test_bit(R5_UPTODATE, &dev->flags)
3155 if (!test_bit(R5_ReWrite, &dev->flags)) {
3156 set_bit(R5_Wantwrite, &dev->flags);
3157 set_bit(R5_ReWrite, &dev->flags);
3158 set_bit(R5_LOCKED, &dev->flags);
3160 /* let's read it back */
3161 set_bit(R5_Wantread, &dev->flags);
3162 set_bit(R5_LOCKED, &dev->flags);
3167 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state)) {
3168 /* Need to write out all blocks after computing P&Q */
3169 sh->disks = conf->raid_disks;
3170 stripe_set_idx(sh->sector, conf, 0, sh);
3171 compute_parity6(sh, RECONSTRUCT_WRITE);
3172 for (i = conf->raid_disks ; i-- ; ) {
3173 set_bit(R5_LOCKED, &sh->dev[i].flags);
3175 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3177 clear_bit(STRIPE_EXPANDING, &sh->state);
3178 } else if (s.expanded) {
3179 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3180 atomic_dec(&conf->reshape_stripes);
3181 wake_up(&conf->wait_for_overlap);
3182 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3185 if (s.expanding && s.locked == 0 &&
3186 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
3187 handle_stripe_expansion(conf, sh, &r6s);
3190 spin_unlock(&sh->lock);
3192 /* wait for this device to become unblocked */
3193 if (unlikely(blocked_rdev))
3194 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
3198 return_io(return_bi);
3200 return blocked_rdev == NULL;
3203 /* returns true if the stripe was handled */
3204 static bool handle_stripe(struct stripe_head *sh, struct page *tmp_page)
3206 if (sh->raid_conf->level == 6)
3207 return handle_stripe6(sh, tmp_page);
3209 return handle_stripe5(sh);
3214 static void raid5_activate_delayed(raid5_conf_t *conf)
3216 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
3217 while (!list_empty(&conf->delayed_list)) {
3218 struct list_head *l = conf->delayed_list.next;
3219 struct stripe_head *sh;
3220 sh = list_entry(l, struct stripe_head, lru);
3222 clear_bit(STRIPE_DELAYED, &sh->state);
3223 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3224 atomic_inc(&conf->preread_active_stripes);
3225 list_add_tail(&sh->lru, &conf->hold_list);
3228 blk_plug_device(conf->mddev->queue);
3231 static void activate_bit_delay(raid5_conf_t *conf)
3233 /* device_lock is held */
3234 struct list_head head;
3235 list_add(&head, &conf->bitmap_list);
3236 list_del_init(&conf->bitmap_list);
3237 while (!list_empty(&head)) {
3238 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
3239 list_del_init(&sh->lru);
3240 atomic_inc(&sh->count);
3241 __release_stripe(conf, sh);
3245 static void unplug_slaves(mddev_t *mddev)
3247 raid5_conf_t *conf = mddev_to_conf(mddev);
3251 for (i=0; i<mddev->raid_disks; i++) {
3252 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
3253 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
3254 struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
3256 atomic_inc(&rdev->nr_pending);
3259 blk_unplug(r_queue);
3261 rdev_dec_pending(rdev, mddev);
3268 static void raid5_unplug_device(struct request_queue *q)
3270 mddev_t *mddev = q->queuedata;
3271 raid5_conf_t *conf = mddev_to_conf(mddev);
3272 unsigned long flags;
3274 spin_lock_irqsave(&conf->device_lock, flags);
3276 if (blk_remove_plug(q)) {
3278 raid5_activate_delayed(conf);
3280 md_wakeup_thread(mddev->thread);
3282 spin_unlock_irqrestore(&conf->device_lock, flags);
3284 unplug_slaves(mddev);
3287 static int raid5_congested(void *data, int bits)
3289 mddev_t *mddev = data;
3290 raid5_conf_t *conf = mddev_to_conf(mddev);
3292 /* No difference between reads and writes. Just check
3293 * how busy the stripe_cache is
3295 if (conf->inactive_blocked)
3299 if (list_empty_careful(&conf->inactive_list))
3305 /* We want read requests to align with chunks where possible,
3306 * but write requests don't need to.
3308 static int raid5_mergeable_bvec(struct request_queue *q,
3309 struct bvec_merge_data *bvm,
3310 struct bio_vec *biovec)
3312 mddev_t *mddev = q->queuedata;
3313 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
3315 unsigned int chunk_sectors = mddev->chunk_size >> 9;
3316 unsigned int bio_sectors = bvm->bi_size >> 9;
3318 if ((bvm->bi_rw & 1) == WRITE)
3319 return biovec->bv_len; /* always allow writes to be mergeable */
3321 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
3322 if (max < 0) max = 0;
3323 if (max <= biovec->bv_len && bio_sectors == 0)
3324 return biovec->bv_len;
3330 static int in_chunk_boundary(mddev_t *mddev, struct bio *bio)
3332 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
3333 unsigned int chunk_sectors = mddev->chunk_size >> 9;
3334 unsigned int bio_sectors = bio->bi_size >> 9;
3336 return chunk_sectors >=
3337 ((sector & (chunk_sectors - 1)) + bio_sectors);
3341 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3342 * later sampled by raid5d.
3344 static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf)
3346 unsigned long flags;
3348 spin_lock_irqsave(&conf->device_lock, flags);
3350 bi->bi_next = conf->retry_read_aligned_list;
3351 conf->retry_read_aligned_list = bi;
3353 spin_unlock_irqrestore(&conf->device_lock, flags);
3354 md_wakeup_thread(conf->mddev->thread);
3358 static struct bio *remove_bio_from_retry(raid5_conf_t *conf)
3362 bi = conf->retry_read_aligned;
3364 conf->retry_read_aligned = NULL;
3367 bi = conf->retry_read_aligned_list;
3369 conf->retry_read_aligned_list = bi->bi_next;
3372 * this sets the active strip count to 1 and the processed
3373 * strip count to zero (upper 8 bits)
3375 bi->bi_phys_segments = 1; /* biased count of active stripes */
3383 * The "raid5_align_endio" should check if the read succeeded and if it
3384 * did, call bio_endio on the original bio (having bio_put the new bio
3386 * If the read failed..
3388 static void raid5_align_endio(struct bio *bi, int error)
3390 struct bio* raid_bi = bi->bi_private;
3393 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3398 mddev = raid_bi->bi_bdev->bd_disk->queue->queuedata;
3399 conf = mddev_to_conf(mddev);
3400 rdev = (void*)raid_bi->bi_next;
3401 raid_bi->bi_next = NULL;
3403 rdev_dec_pending(rdev, conf->mddev);
3405 if (!error && uptodate) {
3406 bio_endio(raid_bi, 0);
3407 if (atomic_dec_and_test(&conf->active_aligned_reads))
3408 wake_up(&conf->wait_for_stripe);
3413 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
3415 add_bio_to_retry(raid_bi, conf);
3418 static int bio_fits_rdev(struct bio *bi)
3420 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
3422 if ((bi->bi_size>>9) > q->max_sectors)
3424 blk_recount_segments(q, bi);
3425 if (bi->bi_phys_segments > q->max_phys_segments)
3428 if (q->merge_bvec_fn)
3429 /* it's too hard to apply the merge_bvec_fn at this stage,
3438 static int chunk_aligned_read(struct request_queue *q, struct bio * raid_bio)
3440 mddev_t *mddev = q->queuedata;
3441 raid5_conf_t *conf = mddev_to_conf(mddev);
3442 unsigned int dd_idx;
3443 struct bio* align_bi;
3446 if (!in_chunk_boundary(mddev, raid_bio)) {
3447 pr_debug("chunk_aligned_read : non aligned\n");
3451 * use bio_clone to make a copy of the bio
3453 align_bi = bio_clone(raid_bio, GFP_NOIO);
3457 * set bi_end_io to a new function, and set bi_private to the
3460 align_bi->bi_end_io = raid5_align_endio;
3461 align_bi->bi_private = raid_bio;
3465 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
3470 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
3471 if (rdev && test_bit(In_sync, &rdev->flags)) {
3472 atomic_inc(&rdev->nr_pending);
3474 raid_bio->bi_next = (void*)rdev;
3475 align_bi->bi_bdev = rdev->bdev;
3476 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
3477 align_bi->bi_sector += rdev->data_offset;
3479 if (!bio_fits_rdev(align_bi)) {
3480 /* too big in some way */
3482 rdev_dec_pending(rdev, mddev);
3486 spin_lock_irq(&conf->device_lock);
3487 wait_event_lock_irq(conf->wait_for_stripe,
3489 conf->device_lock, /* nothing */);
3490 atomic_inc(&conf->active_aligned_reads);
3491 spin_unlock_irq(&conf->device_lock);
3493 generic_make_request(align_bi);
3502 /* __get_priority_stripe - get the next stripe to process
3504 * Full stripe writes are allowed to pass preread active stripes up until
3505 * the bypass_threshold is exceeded. In general the bypass_count
3506 * increments when the handle_list is handled before the hold_list; however, it
3507 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3508 * stripe with in flight i/o. The bypass_count will be reset when the
3509 * head of the hold_list has changed, i.e. the head was promoted to the
3512 static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf)
3514 struct stripe_head *sh;
3516 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3518 list_empty(&conf->handle_list) ? "empty" : "busy",
3519 list_empty(&conf->hold_list) ? "empty" : "busy",
3520 atomic_read(&conf->pending_full_writes), conf->bypass_count);
3522 if (!list_empty(&conf->handle_list)) {
3523 sh = list_entry(conf->handle_list.next, typeof(*sh), lru);
3525 if (list_empty(&conf->hold_list))
3526 conf->bypass_count = 0;
3527 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
3528 if (conf->hold_list.next == conf->last_hold)
3529 conf->bypass_count++;
3531 conf->last_hold = conf->hold_list.next;
3532 conf->bypass_count -= conf->bypass_threshold;
3533 if (conf->bypass_count < 0)
3534 conf->bypass_count = 0;
3537 } else if (!list_empty(&conf->hold_list) &&
3538 ((conf->bypass_threshold &&
3539 conf->bypass_count > conf->bypass_threshold) ||
3540 atomic_read(&conf->pending_full_writes) == 0)) {
3541 sh = list_entry(conf->hold_list.next,
3543 conf->bypass_count -= conf->bypass_threshold;
3544 if (conf->bypass_count < 0)
3545 conf->bypass_count = 0;
3549 list_del_init(&sh->lru);
3550 atomic_inc(&sh->count);
3551 BUG_ON(atomic_read(&sh->count) != 1);
3555 static int make_request(struct request_queue *q, struct bio * bi)
3557 mddev_t *mddev = q->queuedata;
3558 raid5_conf_t *conf = mddev_to_conf(mddev);
3560 sector_t new_sector;
3561 sector_t logical_sector, last_sector;
3562 struct stripe_head *sh;
3563 const int rw = bio_data_dir(bi);
3566 if (unlikely(bio_barrier(bi))) {
3567 bio_endio(bi, -EOPNOTSUPP);
3571 md_write_start(mddev, bi);
3573 cpu = part_stat_lock();
3574 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
3575 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
3580 mddev->reshape_position == MaxSector &&
3581 chunk_aligned_read(q,bi))
3584 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
3585 last_sector = bi->bi_sector + (bi->bi_size>>9);
3587 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
3589 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
3591 int disks, data_disks;
3596 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
3597 if (likely(conf->reshape_progress == MaxSector))
3598 disks = conf->raid_disks;
3600 /* spinlock is needed as reshape_progress may be
3601 * 64bit on a 32bit platform, and so it might be
3602 * possible to see a half-updated value
3603 * Ofcourse reshape_progress could change after
3604 * the lock is dropped, so once we get a reference
3605 * to the stripe that we think it is, we will have
3608 spin_lock_irq(&conf->device_lock);
3609 disks = conf->raid_disks;
3610 if (mddev->delta_disks < 0
3611 ? logical_sector < conf->reshape_progress
3612 : logical_sector >= conf->reshape_progress) {
3613 disks = conf->previous_raid_disks;
3616 if (mddev->delta_disks < 0
3617 ? logical_sector < conf->reshape_safe
3618 : logical_sector >= conf->reshape_safe) {
3619 spin_unlock_irq(&conf->device_lock);
3624 spin_unlock_irq(&conf->device_lock);
3626 data_disks = disks - conf->max_degraded;
3628 new_sector = raid5_compute_sector(conf, logical_sector,
3631 pr_debug("raid5: make_request, sector %llu logical %llu\n",
3632 (unsigned long long)new_sector,
3633 (unsigned long long)logical_sector);
3635 sh = get_active_stripe(conf, new_sector, previous,
3636 (bi->bi_rw&RWA_MASK));
3638 if (unlikely(conf->reshape_progress != MaxSector)) {
3639 /* expansion might have moved on while waiting for a
3640 * stripe, so we must do the range check again.
3641 * Expansion could still move past after this
3642 * test, but as we are holding a reference to
3643 * 'sh', we know that if that happens,
3644 * STRIPE_EXPANDING will get set and the expansion
3645 * won't proceed until we finish with the stripe.
3648 spin_lock_irq(&conf->device_lock);
3649 if ((mddev->delta_disks < 0
3650 ? logical_sector >= conf->reshape_progress
3651 : logical_sector < conf->reshape_progress)
3653 /* mismatch, need to try again */
3655 spin_unlock_irq(&conf->device_lock);
3661 /* FIXME what if we get a false positive because these
3662 * are being updated.
3664 if (logical_sector >= mddev->suspend_lo &&
3665 logical_sector < mddev->suspend_hi) {
3671 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
3672 !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) {
3673 /* Stripe is busy expanding or
3674 * add failed due to overlap. Flush everything
3677 raid5_unplug_device(mddev->queue);
3682 finish_wait(&conf->wait_for_overlap, &w);
3683 set_bit(STRIPE_HANDLE, &sh->state);
3684 clear_bit(STRIPE_DELAYED, &sh->state);
3687 /* cannot get stripe for read-ahead, just give-up */
3688 clear_bit(BIO_UPTODATE, &bi->bi_flags);
3689 finish_wait(&conf->wait_for_overlap, &w);
3694 spin_lock_irq(&conf->device_lock);
3695 remaining = raid5_dec_bi_phys_segments(bi);
3696 spin_unlock_irq(&conf->device_lock);
3697 if (remaining == 0) {
3700 md_write_end(mddev);
3707 static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks);
3709 static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped)
3711 /* reshaping is quite different to recovery/resync so it is
3712 * handled quite separately ... here.
3714 * On each call to sync_request, we gather one chunk worth of
3715 * destination stripes and flag them as expanding.
3716 * Then we find all the source stripes and request reads.
3717 * As the reads complete, handle_stripe will copy the data
3718 * into the destination stripe and release that stripe.
3720 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
3721 struct stripe_head *sh;
3722 sector_t first_sector, last_sector;
3723 int raid_disks = conf->previous_raid_disks;
3724 int data_disks = raid_disks - conf->max_degraded;
3725 int new_data_disks = conf->raid_disks - conf->max_degraded;
3728 sector_t writepos, safepos, gap;
3729 sector_t stripe_addr;
3731 if (sector_nr == 0) {
3732 /* If restarting in the middle, skip the initial sectors */
3733 if (mddev->delta_disks < 0 &&
3734 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
3735 sector_nr = raid5_size(mddev, 0, 0)
3736 - conf->reshape_progress;
3737 } else if (mddev->delta_disks > 0 &&
3738 conf->reshape_progress > 0)
3739 sector_nr = conf->reshape_progress;
3740 sector_div(sector_nr, new_data_disks);
3747 /* we update the metadata when there is more than 3Meg
3748 * in the block range (that is rather arbitrary, should
3749 * probably be time based) or when the data about to be
3750 * copied would over-write the source of the data at
3751 * the front of the range.
3752 * i.e. one new_stripe along from reshape_progress new_maps
3753 * to after where reshape_safe old_maps to
3755 writepos = conf->reshape_progress;
3756 sector_div(writepos, new_data_disks);
3757 safepos = conf->reshape_safe;
3758 sector_div(safepos, data_disks);
3759 if (mddev->delta_disks < 0) {
3760 writepos -= conf->chunk_size/512;
3761 safepos += conf->chunk_size/512;
3762 gap = conf->reshape_safe - conf->reshape_progress;
3764 writepos += conf->chunk_size/512;
3765 safepos -= conf->chunk_size/512;
3766 gap = conf->reshape_progress - conf->reshape_safe;
3769 if ((mddev->delta_disks < 0
3770 ? writepos < safepos
3771 : writepos > safepos) ||
3772 gap > (new_data_disks)*3000*2 /*3Meg*/) {
3773 /* Cannot proceed until we've updated the superblock... */
3774 wait_event(conf->wait_for_overlap,
3775 atomic_read(&conf->reshape_stripes)==0);
3776 mddev->reshape_position = conf->reshape_progress;
3777 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3778 md_wakeup_thread(mddev->thread);
3779 wait_event(mddev->sb_wait, mddev->flags == 0 ||
3780 kthread_should_stop());
3781 spin_lock_irq(&conf->device_lock);
3782 conf->reshape_safe = mddev->reshape_position;
3783 spin_unlock_irq(&conf->device_lock);
3784 wake_up(&conf->wait_for_overlap);
3787 if (mddev->delta_disks < 0) {
3788 BUG_ON(conf->reshape_progress == 0);
3789 stripe_addr = writepos;
3790 BUG_ON((mddev->dev_sectors &
3791 ~((sector_t)mddev->chunk_size / 512 - 1))
3792 - (conf->chunk_size / 512) - stripe_addr
3795 BUG_ON(writepos != sector_nr + conf->chunk_size / 512);
3796 stripe_addr = sector_nr;
3798 for (i=0; i < conf->chunk_size/512; i+= STRIPE_SECTORS) {
3801 sh = get_active_stripe(conf, stripe_addr+i, 0, 0);
3802 set_bit(STRIPE_EXPANDING, &sh->state);
3803 atomic_inc(&conf->reshape_stripes);
3804 /* If any of this stripe is beyond the end of the old
3805 * array, then we need to zero those blocks
3807 for (j=sh->disks; j--;) {
3809 if (j == sh->pd_idx)
3811 if (conf->level == 6 &&
3814 s = compute_blocknr(sh, j);
3815 if (s < raid5_size(mddev, 0, 0)) {
3819 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
3820 set_bit(R5_Expanded, &sh->dev[j].flags);
3821 set_bit(R5_UPTODATE, &sh->dev[j].flags);
3824 set_bit(STRIPE_EXPAND_READY, &sh->state);
3825 set_bit(STRIPE_HANDLE, &sh->state);
3829 spin_lock_irq(&conf->device_lock);
3830 if (mddev->delta_disks < 0)
3831 conf->reshape_progress -= i * new_data_disks;
3833 conf->reshape_progress += i * new_data_disks;
3834 spin_unlock_irq(&conf->device_lock);
3835 /* Ok, those stripe are ready. We can start scheduling
3836 * reads on the source stripes.
3837 * The source stripes are determined by mapping the first and last
3838 * block on the destination stripes.
3841 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
3844 raid5_compute_sector(conf, ((stripe_addr+conf->chunk_size/512)
3845 *(new_data_disks) - 1),
3847 if (last_sector >= mddev->dev_sectors)
3848 last_sector = mddev->dev_sectors - 1;
3849 while (first_sector <= last_sector) {
3850 sh = get_active_stripe(conf, first_sector, 1, 0);
3851 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3852 set_bit(STRIPE_HANDLE, &sh->state);
3854 first_sector += STRIPE_SECTORS;
3856 /* If this takes us to the resync_max point where we have to pause,
3857 * then we need to write out the superblock.
3859 sector_nr += conf->chunk_size>>9;
3860 if (sector_nr >= mddev->resync_max) {
3861 /* Cannot proceed until we've updated the superblock... */
3862 wait_event(conf->wait_for_overlap,
3863 atomic_read(&conf->reshape_stripes) == 0);
3864 mddev->reshape_position = conf->reshape_progress;
3865 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3866 md_wakeup_thread(mddev->thread);
3867 wait_event(mddev->sb_wait,
3868 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
3869 || kthread_should_stop());
3870 spin_lock_irq(&conf->device_lock);
3871 conf->reshape_safe = mddev->reshape_position;
3872 spin_unlock_irq(&conf->device_lock);
3873 wake_up(&conf->wait_for_overlap);
3875 return conf->chunk_size>>9;
3878 /* FIXME go_faster isn't used */
3879 static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
3881 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
3882 struct stripe_head *sh;
3883 sector_t max_sector = mddev->dev_sectors;
3885 int still_degraded = 0;
3888 if (sector_nr >= max_sector) {
3889 /* just being told to finish up .. nothing much to do */
3890 unplug_slaves(mddev);
3892 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
3897 if (mddev->curr_resync < max_sector) /* aborted */
3898 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
3900 else /* completed sync */
3902 bitmap_close_sync(mddev->bitmap);
3907 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3908 return reshape_request(mddev, sector_nr, skipped);
3910 /* No need to check resync_max as we never do more than one
3911 * stripe, and as resync_max will always be on a chunk boundary,
3912 * if the check in md_do_sync didn't fire, there is no chance
3913 * of overstepping resync_max here
3916 /* if there is too many failed drives and we are trying
3917 * to resync, then assert that we are finished, because there is
3918 * nothing we can do.
3920 if (mddev->degraded >= conf->max_degraded &&
3921 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3922 sector_t rv = mddev->dev_sectors - sector_nr;
3926 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
3927 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
3928 !conf->fullsync && sync_blocks >= STRIPE_SECTORS) {
3929 /* we can skip this block, and probably more */
3930 sync_blocks /= STRIPE_SECTORS;
3932 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
3936 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
3938 sh = get_active_stripe(conf, sector_nr, 0, 1);
3940 sh = get_active_stripe(conf, sector_nr, 0, 0);
3941 /* make sure we don't swamp the stripe cache if someone else
3942 * is trying to get access
3944 schedule_timeout_uninterruptible(1);
3946 /* Need to check if array will still be degraded after recovery/resync
3947 * We don't need to check the 'failed' flag as when that gets set,
3950 for (i=0; i<mddev->raid_disks; i++)
3951 if (conf->disks[i].rdev == NULL)
3954 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
3956 spin_lock(&sh->lock);
3957 set_bit(STRIPE_SYNCING, &sh->state);
3958 clear_bit(STRIPE_INSYNC, &sh->state);
3959 spin_unlock(&sh->lock);
3961 /* wait for any blocked device to be handled */
3962 while(unlikely(!handle_stripe(sh, NULL)))
3966 return STRIPE_SECTORS;
3969 static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
3971 /* We may not be able to submit a whole bio at once as there
3972 * may not be enough stripe_heads available.
3973 * We cannot pre-allocate enough stripe_heads as we may need
3974 * more than exist in the cache (if we allow ever large chunks).
3975 * So we do one stripe head at a time and record in
3976 * ->bi_hw_segments how many have been done.
3978 * We *know* that this entire raid_bio is in one chunk, so
3979 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
3981 struct stripe_head *sh;
3983 sector_t sector, logical_sector, last_sector;
3988 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
3989 sector = raid5_compute_sector(conf, logical_sector,
3991 last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9);
3993 for (; logical_sector < last_sector;
3994 logical_sector += STRIPE_SECTORS,
3995 sector += STRIPE_SECTORS,
3998 if (scnt < raid5_bi_hw_segments(raid_bio))
3999 /* already done this stripe */
4002 sh = get_active_stripe(conf, sector, 0, 1);
4005 /* failed to get a stripe - must wait */
4006 raid5_set_bi_hw_segments(raid_bio, scnt);
4007 conf->retry_read_aligned = raid_bio;
4011 set_bit(R5_ReadError, &sh->dev[dd_idx].flags);
4012 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
4014 raid5_set_bi_hw_segments(raid_bio, scnt);
4015 conf->retry_read_aligned = raid_bio;
4019 handle_stripe(sh, NULL);
4023 spin_lock_irq(&conf->device_lock);
4024 remaining = raid5_dec_bi_phys_segments(raid_bio);
4025 spin_unlock_irq(&conf->device_lock);
4027 bio_endio(raid_bio, 0);
4028 if (atomic_dec_and_test(&conf->active_aligned_reads))
4029 wake_up(&conf->wait_for_stripe);
4036 * This is our raid5 kernel thread.
4038 * We scan the hash table for stripes which can be handled now.
4039 * During the scan, completed stripes are saved for us by the interrupt
4040 * handler, so that they will not have to wait for our next wakeup.
4042 static void raid5d(mddev_t *mddev)
4044 struct stripe_head *sh;
4045 raid5_conf_t *conf = mddev_to_conf(mddev);
4048 pr_debug("+++ raid5d active\n");
4050 md_check_recovery(mddev);
4053 spin_lock_irq(&conf->device_lock);
4057 if (conf->seq_flush != conf->seq_write) {
4058 int seq = conf->seq_flush;
4059 spin_unlock_irq(&conf->device_lock);
4060 bitmap_unplug(mddev->bitmap);
4061 spin_lock_irq(&conf->device_lock);
4062 conf->seq_write = seq;
4063 activate_bit_delay(conf);
4066 while ((bio = remove_bio_from_retry(conf))) {
4068 spin_unlock_irq(&conf->device_lock);
4069 ok = retry_aligned_read(conf, bio);
4070 spin_lock_irq(&conf->device_lock);
4076 sh = __get_priority_stripe(conf);
4080 spin_unlock_irq(&conf->device_lock);
4083 handle_stripe(sh, conf->spare_page);
4086 spin_lock_irq(&conf->device_lock);
4088 pr_debug("%d stripes handled\n", handled);
4090 spin_unlock_irq(&conf->device_lock);
4092 async_tx_issue_pending_all();
4093 unplug_slaves(mddev);
4095 pr_debug("--- raid5d inactive\n");
4099 raid5_show_stripe_cache_size(mddev_t *mddev, char *page)
4101 raid5_conf_t *conf = mddev_to_conf(mddev);
4103 return sprintf(page, "%d\n", conf->max_nr_stripes);
4109 raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
4111 raid5_conf_t *conf = mddev_to_conf(mddev);
4115 if (len >= PAGE_SIZE)
4120 if (strict_strtoul(page, 10, &new))
4122 if (new <= 16 || new > 32768)
4124 while (new < conf->max_nr_stripes) {
4125 if (drop_one_stripe(conf))
4126 conf->max_nr_stripes--;
4130 err = md_allow_write(mddev);
4133 while (new > conf->max_nr_stripes) {
4134 if (grow_one_stripe(conf))
4135 conf->max_nr_stripes++;
4141 static struct md_sysfs_entry
4142 raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
4143 raid5_show_stripe_cache_size,
4144 raid5_store_stripe_cache_size);
4147 raid5_show_preread_threshold(mddev_t *mddev, char *page)
4149 raid5_conf_t *conf = mddev_to_conf(mddev);
4151 return sprintf(page, "%d\n", conf->bypass_threshold);
4157 raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len)
4159 raid5_conf_t *conf = mddev_to_conf(mddev);
4161 if (len >= PAGE_SIZE)
4166 if (strict_strtoul(page, 10, &new))
4168 if (new > conf->max_nr_stripes)
4170 conf->bypass_threshold = new;
4174 static struct md_sysfs_entry
4175 raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
4177 raid5_show_preread_threshold,
4178 raid5_store_preread_threshold);
4181 stripe_cache_active_show(mddev_t *mddev, char *page)
4183 raid5_conf_t *conf = mddev_to_conf(mddev);
4185 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
4190 static struct md_sysfs_entry
4191 raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
4193 static struct attribute *raid5_attrs[] = {
4194 &raid5_stripecache_size.attr,
4195 &raid5_stripecache_active.attr,
4196 &raid5_preread_bypass_threshold.attr,
4199 static struct attribute_group raid5_attrs_group = {
4201 .attrs = raid5_attrs,
4205 raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks)
4207 raid5_conf_t *conf = mddev_to_conf(mddev);
4210 sectors = mddev->dev_sectors;
4212 /* size is defined by the smallest of previous and new size */
4213 if (conf->raid_disks < conf->previous_raid_disks)
4214 raid_disks = conf->raid_disks;
4216 raid_disks = conf->previous_raid_disks;
4219 sectors &= ~((sector_t)mddev->chunk_size/512 - 1);
4220 return sectors * (raid_disks - conf->max_degraded);
4223 static raid5_conf_t *setup_conf(mddev_t *mddev)
4226 int raid_disk, memory;
4228 struct disk_info *disk;
4230 if (mddev->new_level != 5
4231 && mddev->new_level != 4
4232 && mddev->new_level != 6) {
4233 printk(KERN_ERR "raid5: %s: raid level not set to 4/5/6 (%d)\n",
4234 mdname(mddev), mddev->new_level);
4235 return ERR_PTR(-EIO);
4237 if ((mddev->new_level == 5
4238 && !algorithm_valid_raid5(mddev->new_layout)) ||
4239 (mddev->new_level == 6
4240 && !algorithm_valid_raid6(mddev->new_layout))) {
4241 printk(KERN_ERR "raid5: %s: layout %d not supported\n",
4242 mdname(mddev), mddev->new_layout);
4243 return ERR_PTR(-EIO);
4245 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
4246 printk(KERN_ERR "raid6: not enough configured devices for %s (%d, minimum 4)\n",
4247 mdname(mddev), mddev->raid_disks);
4248 return ERR_PTR(-EINVAL);
4251 if (!mddev->new_chunk || mddev->new_chunk % PAGE_SIZE) {
4252 printk(KERN_ERR "raid5: invalid chunk size %d for %s\n",
4253 mddev->new_chunk, mdname(mddev));
4254 return ERR_PTR(-EINVAL);
4257 conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL);
4261 conf->raid_disks = mddev->raid_disks;
4262 if (mddev->reshape_position == MaxSector)
4263 conf->previous_raid_disks = mddev->raid_disks;
4265 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
4267 conf->disks = kzalloc(conf->raid_disks * sizeof(struct disk_info),
4272 conf->mddev = mddev;
4274 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
4277 if (mddev->new_level == 6) {
4278 conf->spare_page = alloc_page(GFP_KERNEL);
4279 if (!conf->spare_page)
4282 spin_lock_init(&conf->device_lock);
4283 init_waitqueue_head(&conf->wait_for_stripe);
4284 init_waitqueue_head(&conf->wait_for_overlap);
4285 INIT_LIST_HEAD(&conf->handle_list);
4286 INIT_LIST_HEAD(&conf->hold_list);
4287 INIT_LIST_HEAD(&conf->delayed_list);
4288 INIT_LIST_HEAD(&conf->bitmap_list);
4289 INIT_LIST_HEAD(&conf->inactive_list);
4290 atomic_set(&conf->active_stripes, 0);
4291 atomic_set(&conf->preread_active_stripes, 0);
4292 atomic_set(&conf->active_aligned_reads, 0);
4293 conf->bypass_threshold = BYPASS_THRESHOLD;
4295 pr_debug("raid5: run(%s) called.\n", mdname(mddev));
4297 list_for_each_entry(rdev, &mddev->disks, same_set) {
4298 raid_disk = rdev->raid_disk;
4299 if (raid_disk >= conf->raid_disks
4302 disk = conf->disks + raid_disk;
4306 if (test_bit(In_sync, &rdev->flags)) {
4307 char b[BDEVNAME_SIZE];
4308 printk(KERN_INFO "raid5: device %s operational as raid"
4309 " disk %d\n", bdevname(rdev->bdev,b),
4312 /* Cannot rely on bitmap to complete recovery */
4316 conf->chunk_size = mddev->new_chunk;
4317 conf->level = mddev->new_level;
4318 if (conf->level == 6)
4319 conf->max_degraded = 2;
4321 conf->max_degraded = 1;
4322 conf->algorithm = mddev->new_layout;
4323 conf->max_nr_stripes = NR_STRIPES;
4324 conf->reshape_progress = mddev->reshape_position;
4326 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
4327 conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
4328 if (grow_stripes(conf, conf->max_nr_stripes)) {
4330 "raid5: couldn't allocate %dkB for buffers\n", memory);
4333 printk(KERN_INFO "raid5: allocated %dkB for %s\n",
4334 memory, mdname(mddev));
4336 conf->thread = md_register_thread(raid5d, mddev, "%s_raid5");
4337 if (!conf->thread) {
4339 "raid5: couldn't allocate thread for %s\n",
4348 shrink_stripes(conf);
4349 safe_put_page(conf->spare_page);
4351 kfree(conf->stripe_hashtbl);
4353 return ERR_PTR(-EIO);
4355 return ERR_PTR(-ENOMEM);
4358 static int run(mddev_t *mddev)
4361 int working_disks = 0;
4364 if (mddev->reshape_position != MaxSector) {
4365 /* Check that we can continue the reshape.
4366 * Currently only disks can change, it must
4367 * increase, and we must be past the point where
4368 * a stripe over-writes itself
4370 sector_t here_new, here_old;
4372 int max_degraded = (mddev->level == 6 ? 2 : 1);
4374 if (mddev->new_level != mddev->level ||
4375 mddev->new_layout != mddev->layout ||
4376 mddev->new_chunk != mddev->chunk_size) {
4377 printk(KERN_ERR "raid5: %s: unsupported reshape "
4378 "required - aborting.\n",
4382 old_disks = mddev->raid_disks - mddev->delta_disks;
4383 /* reshape_position must be on a new-stripe boundary, and one
4384 * further up in new geometry must map after here in old
4387 here_new = mddev->reshape_position;
4388 if (sector_div(here_new, (mddev->chunk_size>>9)*
4389 (mddev->raid_disks - max_degraded))) {
4390 printk(KERN_ERR "raid5: reshape_position not "
4391 "on a stripe boundary\n");
4394 /* here_new is the stripe we will write to */
4395 here_old = mddev->reshape_position;
4396 sector_div(here_old, (mddev->chunk_size>>9)*
4397 (old_disks-max_degraded));
4398 /* here_old is the first stripe that we might need to read
4400 if (here_new >= here_old) {
4401 /* Reading from the same stripe as writing to - bad */
4402 printk(KERN_ERR "raid5: reshape_position too early for "
4403 "auto-recovery - aborting.\n");
4406 printk(KERN_INFO "raid5: reshape will continue\n");
4407 /* OK, we should be able to continue; */
4409 BUG_ON(mddev->level != mddev->new_level);
4410 BUG_ON(mddev->layout != mddev->new_layout);
4411 BUG_ON(mddev->chunk_size != mddev->new_chunk);
4412 BUG_ON(mddev->delta_disks != 0);
4415 if (mddev->private == NULL)
4416 conf = setup_conf(mddev);
4418 conf = mddev->private;
4421 return PTR_ERR(conf);
4423 mddev->thread = conf->thread;
4424 conf->thread = NULL;
4425 mddev->private = conf;
4428 * 0 for a fully functional array, 1 or 2 for a degraded array.
4430 list_for_each_entry(rdev, &mddev->disks, same_set)
4431 if (rdev->raid_disk >= 0 &&
4432 test_bit(In_sync, &rdev->flags))
4435 mddev->degraded = conf->raid_disks - working_disks;
4437 if (mddev->degraded > conf->max_degraded) {
4438 printk(KERN_ERR "raid5: not enough operational devices for %s"
4439 " (%d/%d failed)\n",
4440 mdname(mddev), mddev->degraded, conf->raid_disks);
4444 /* device size must be a multiple of chunk size */
4445 mddev->dev_sectors &= ~(mddev->chunk_size / 512 - 1);
4446 mddev->resync_max_sectors = mddev->dev_sectors;
4448 if (mddev->degraded > 0 &&
4449 mddev->recovery_cp != MaxSector) {
4450 if (mddev->ok_start_degraded)
4452 "raid5: starting dirty degraded array: %s"
4453 "- data corruption possible.\n",
4457 "raid5: cannot start dirty degraded array for %s\n",
4463 if (mddev->degraded == 0)
4464 printk("raid5: raid level %d set %s active with %d out of %d"
4465 " devices, algorithm %d\n", conf->level, mdname(mddev),
4466 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
4469 printk(KERN_ALERT "raid5: raid level %d set %s active with %d"
4470 " out of %d devices, algorithm %d\n", conf->level,
4471 mdname(mddev), mddev->raid_disks - mddev->degraded,
4472 mddev->raid_disks, conf->algorithm);
4474 print_raid5_conf(conf);
4476 if (conf->reshape_progress != MaxSector) {
4477 printk("...ok start reshape thread\n");
4478 conf->reshape_safe = conf->reshape_progress;
4479 atomic_set(&conf->reshape_stripes, 0);
4480 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4481 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4482 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
4483 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4484 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
4488 /* read-ahead size must cover two whole stripes, which is
4489 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4492 int data_disks = conf->previous_raid_disks - conf->max_degraded;
4493 int stripe = data_disks *
4494 (mddev->chunk_size / PAGE_SIZE);
4495 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
4496 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
4499 /* Ok, everything is just fine now */
4500 if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
4502 "raid5: failed to create sysfs attributes for %s\n",
4505 mddev->queue->queue_lock = &conf->device_lock;
4507 mddev->queue->unplug_fn = raid5_unplug_device;
4508 mddev->queue->backing_dev_info.congested_data = mddev;
4509 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
4511 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
4513 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
4517 md_unregister_thread(mddev->thread);
4518 mddev->thread = NULL;
4520 shrink_stripes(conf);
4521 print_raid5_conf(conf);
4522 safe_put_page(conf->spare_page);
4524 kfree(conf->stripe_hashtbl);
4527 mddev->private = NULL;
4528 printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev));
4534 static int stop(mddev_t *mddev)
4536 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
4538 md_unregister_thread(mddev->thread);
4539 mddev->thread = NULL;
4540 shrink_stripes(conf);
4541 kfree(conf->stripe_hashtbl);
4542 mddev->queue->backing_dev_info.congested_fn = NULL;
4543 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
4544 sysfs_remove_group(&mddev->kobj, &raid5_attrs_group);
4547 mddev->private = NULL;
4552 static void print_sh(struct seq_file *seq, struct stripe_head *sh)
4556 seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n",
4557 (unsigned long long)sh->sector, sh->pd_idx, sh->state);
4558 seq_printf(seq, "sh %llu, count %d.\n",
4559 (unsigned long long)sh->sector, atomic_read(&sh->count));
4560 seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector);
4561 for (i = 0; i < sh->disks; i++) {
4562 seq_printf(seq, "(cache%d: %p %ld) ",
4563 i, sh->dev[i].page, sh->dev[i].flags);
4565 seq_printf(seq, "\n");
4568 static void printall(struct seq_file *seq, raid5_conf_t *conf)
4570 struct stripe_head *sh;
4571 struct hlist_node *hn;
4574 spin_lock_irq(&conf->device_lock);
4575 for (i = 0; i < NR_HASH; i++) {
4576 hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) {
4577 if (sh->raid_conf != conf)
4582 spin_unlock_irq(&conf->device_lock);
4586 static void status(struct seq_file *seq, mddev_t *mddev)
4588 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
4591 seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout);
4592 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
4593 for (i = 0; i < conf->raid_disks; i++)
4594 seq_printf (seq, "%s",
4595 conf->disks[i].rdev &&
4596 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
4597 seq_printf (seq, "]");
4599 seq_printf (seq, "\n");
4600 printall(seq, conf);
4604 static void print_raid5_conf (raid5_conf_t *conf)
4607 struct disk_info *tmp;
4609 printk("RAID5 conf printout:\n");
4611 printk("(conf==NULL)\n");
4614 printk(" --- rd:%d wd:%d\n", conf->raid_disks,
4615 conf->raid_disks - conf->mddev->degraded);
4617 for (i = 0; i < conf->raid_disks; i++) {
4618 char b[BDEVNAME_SIZE];
4619 tmp = conf->disks + i;
4621 printk(" disk %d, o:%d, dev:%s\n",
4622 i, !test_bit(Faulty, &tmp->rdev->flags),
4623 bdevname(tmp->rdev->bdev,b));
4627 static int raid5_spare_active(mddev_t *mddev)
4630 raid5_conf_t *conf = mddev->private;
4631 struct disk_info *tmp;
4633 for (i = 0; i < conf->raid_disks; i++) {
4634 tmp = conf->disks + i;
4636 && !test_bit(Faulty, &tmp->rdev->flags)
4637 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
4638 unsigned long flags;
4639 spin_lock_irqsave(&conf->device_lock, flags);
4641 spin_unlock_irqrestore(&conf->device_lock, flags);
4644 print_raid5_conf(conf);
4648 static int raid5_remove_disk(mddev_t *mddev, int number)
4650 raid5_conf_t *conf = mddev->private;
4653 struct disk_info *p = conf->disks + number;
4655 print_raid5_conf(conf);
4658 if (number >= conf->raid_disks &&
4659 conf->reshape_progress == MaxSector)
4660 clear_bit(In_sync, &rdev->flags);
4662 if (test_bit(In_sync, &rdev->flags) ||
4663 atomic_read(&rdev->nr_pending)) {
4667 /* Only remove non-faulty devices if recovery
4670 if (!test_bit(Faulty, &rdev->flags) &&
4671 mddev->degraded <= conf->max_degraded &&
4672 number < conf->raid_disks) {
4678 if (atomic_read(&rdev->nr_pending)) {
4679 /* lost the race, try later */
4686 print_raid5_conf(conf);
4690 static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
4692 raid5_conf_t *conf = mddev->private;
4695 struct disk_info *p;
4697 int last = conf->raid_disks - 1;
4699 if (mddev->degraded > conf->max_degraded)
4700 /* no point adding a device */
4703 if (rdev->raid_disk >= 0)
4704 first = last = rdev->raid_disk;
4707 * find the disk ... but prefer rdev->saved_raid_disk
4710 if (rdev->saved_raid_disk >= 0 &&
4711 rdev->saved_raid_disk >= first &&
4712 conf->disks[rdev->saved_raid_disk].rdev == NULL)
4713 disk = rdev->saved_raid_disk;
4716 for ( ; disk <= last ; disk++)
4717 if ((p=conf->disks + disk)->rdev == NULL) {
4718 clear_bit(In_sync, &rdev->flags);
4719 rdev->raid_disk = disk;
4721 if (rdev->saved_raid_disk != disk)
4723 rcu_assign_pointer(p->rdev, rdev);
4726 print_raid5_conf(conf);
4730 static int raid5_resize(mddev_t *mddev, sector_t sectors)
4732 /* no resync is happening, and there is enough space
4733 * on all devices, so we can resize.
4734 * We need to make sure resync covers any new space.
4735 * If the array is shrinking we should possibly wait until
4736 * any io in the removed space completes, but it hardly seems
4739 sectors &= ~((sector_t)mddev->chunk_size/512 - 1);
4740 md_set_array_sectors(mddev, raid5_size(mddev, sectors,
4741 mddev->raid_disks));
4742 if (mddev->array_sectors >
4743 raid5_size(mddev, sectors, mddev->raid_disks))
4745 set_capacity(mddev->gendisk, mddev->array_sectors);
4747 if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) {
4748 mddev->recovery_cp = mddev->dev_sectors;
4749 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4751 mddev->dev_sectors = sectors;
4752 mddev->resync_max_sectors = sectors;
4756 #ifdef CONFIG_MD_RAID5_RESHAPE
4757 static int raid5_check_reshape(mddev_t *mddev)
4759 raid5_conf_t *conf = mddev_to_conf(mddev);
4761 if (mddev->delta_disks == 0)
4762 return 0; /* nothing to do */
4764 /* Cannot grow a bitmap yet */
4766 if (mddev->degraded > conf->max_degraded)
4768 if (mddev->delta_disks < 0) {
4769 /* We might be able to shrink, but the devices must
4770 * be made bigger first.
4771 * For raid6, 4 is the minimum size.
4772 * Otherwise 2 is the minimum
4775 if (mddev->level == 6)
4777 if (mddev->raid_disks + mddev->delta_disks < min)
4781 /* Can only proceed if there are plenty of stripe_heads.
4782 * We need a minimum of one full stripe,, and for sensible progress
4783 * it is best to have about 4 times that.
4784 * If we require 4 times, then the default 256 4K stripe_heads will
4785 * allow for chunk sizes up to 256K, which is probably OK.
4786 * If the chunk size is greater, user-space should request more
4787 * stripe_heads first.
4789 if ((mddev->chunk_size / STRIPE_SIZE) * 4 > conf->max_nr_stripes ||
4790 (mddev->new_chunk / STRIPE_SIZE) * 4 > conf->max_nr_stripes) {
4791 printk(KERN_WARNING "raid5: reshape: not enough stripes. Needed %lu\n",
4792 (mddev->chunk_size / STRIPE_SIZE)*4);
4796 return resize_stripes(conf, conf->raid_disks + mddev->delta_disks);
4799 static int raid5_start_reshape(mddev_t *mddev)
4801 raid5_conf_t *conf = mddev_to_conf(mddev);
4804 int added_devices = 0;
4805 unsigned long flags;
4807 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4810 list_for_each_entry(rdev, &mddev->disks, same_set)
4811 if (rdev->raid_disk < 0 &&
4812 !test_bit(Faulty, &rdev->flags))
4815 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
4816 /* Not enough devices even to make a degraded array
4821 /* Refuse to reduce size of the array. Any reductions in
4822 * array size must be through explicit setting of array_size
4825 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
4826 < mddev->array_sectors) {
4827 printk(KERN_ERR "md: %s: array size must be reduced "
4828 "before number of disks\n", mdname(mddev));
4832 atomic_set(&conf->reshape_stripes, 0);
4833 spin_lock_irq(&conf->device_lock);
4834 conf->previous_raid_disks = conf->raid_disks;
4835 conf->raid_disks += mddev->delta_disks;
4836 if (mddev->delta_disks < 0)
4837 conf->reshape_progress = raid5_size(mddev, 0, 0);
4839 conf->reshape_progress = 0;
4840 conf->reshape_safe = conf->reshape_progress;
4842 spin_unlock_irq(&conf->device_lock);
4844 /* Add some new drives, as many as will fit.
4845 * We know there are enough to make the newly sized array work.
4847 list_for_each_entry(rdev, &mddev->disks, same_set)
4848 if (rdev->raid_disk < 0 &&
4849 !test_bit(Faulty, &rdev->flags)) {
4850 if (raid5_add_disk(mddev, rdev) == 0) {
4852 set_bit(In_sync, &rdev->flags);
4854 rdev->recovery_offset = 0;
4855 sprintf(nm, "rd%d", rdev->raid_disk);
4856 if (sysfs_create_link(&mddev->kobj,
4859 "raid5: failed to create "
4860 " link %s for %s\n",
4866 if (mddev->delta_disks > 0) {
4867 spin_lock_irqsave(&conf->device_lock, flags);
4868 mddev->degraded = (conf->raid_disks - conf->previous_raid_disks)
4870 spin_unlock_irqrestore(&conf->device_lock, flags);
4872 mddev->raid_disks = conf->raid_disks;
4873 mddev->reshape_position = 0;
4874 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4876 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4877 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4878 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
4879 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4880 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
4882 if (!mddev->sync_thread) {
4883 mddev->recovery = 0;
4884 spin_lock_irq(&conf->device_lock);
4885 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
4886 conf->reshape_progress = MaxSector;
4887 spin_unlock_irq(&conf->device_lock);
4890 md_wakeup_thread(mddev->sync_thread);
4891 md_new_event(mddev);
4896 /* This is called from the reshape thread and should make any
4897 * changes needed in 'conf'
4899 static void end_reshape(raid5_conf_t *conf)
4902 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
4904 spin_lock_irq(&conf->device_lock);
4905 conf->previous_raid_disks = conf->raid_disks;
4906 conf->reshape_progress = MaxSector;
4907 spin_unlock_irq(&conf->device_lock);
4909 /* read-ahead size must cover two whole stripes, which is
4910 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4913 int data_disks = conf->raid_disks - conf->max_degraded;
4914 int stripe = data_disks * (conf->chunk_size
4916 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
4917 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
4922 /* This is called from the raid5d thread with mddev_lock held.
4923 * It makes config changes to the device.
4925 static void raid5_finish_reshape(mddev_t *mddev)
4927 struct block_device *bdev;
4929 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4931 if (mddev->delta_disks > 0) {
4932 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
4933 set_capacity(mddev->gendisk, mddev->array_sectors);
4936 bdev = bdget_disk(mddev->gendisk, 0);
4938 mutex_lock(&bdev->bd_inode->i_mutex);
4939 i_size_write(bdev->bd_inode,
4940 (loff_t)mddev->array_sectors << 9);
4941 mutex_unlock(&bdev->bd_inode->i_mutex);
4946 raid5_conf_t *conf = mddev_to_conf(mddev);
4947 mddev->degraded = conf->raid_disks;
4948 for (d = 0; d < conf->raid_disks ; d++)
4949 if (conf->disks[d].rdev &&
4951 &conf->disks[d].rdev->flags))
4953 for (d = conf->raid_disks ;
4954 d < conf->raid_disks - mddev->delta_disks;
4956 raid5_remove_disk(mddev, d);
4958 mddev->reshape_position = MaxSector;
4959 mddev->delta_disks = 0;
4963 static void raid5_quiesce(mddev_t *mddev, int state)
4965 raid5_conf_t *conf = mddev_to_conf(mddev);
4968 case 2: /* resume for a suspend */
4969 wake_up(&conf->wait_for_overlap);
4972 case 1: /* stop all writes */
4973 spin_lock_irq(&conf->device_lock);
4975 wait_event_lock_irq(conf->wait_for_stripe,
4976 atomic_read(&conf->active_stripes) == 0 &&
4977 atomic_read(&conf->active_aligned_reads) == 0,
4978 conf->device_lock, /* nothing */);
4979 spin_unlock_irq(&conf->device_lock);
4982 case 0: /* re-enable writes */
4983 spin_lock_irq(&conf->device_lock);
4985 wake_up(&conf->wait_for_stripe);
4986 wake_up(&conf->wait_for_overlap);
4987 spin_unlock_irq(&conf->device_lock);
4993 static void *raid5_takeover_raid1(mddev_t *mddev)
4997 if (mddev->raid_disks != 2 ||
4998 mddev->degraded > 1)
4999 return ERR_PTR(-EINVAL);
5001 /* Should check if there are write-behind devices? */
5003 chunksect = 64*2; /* 64K by default */
5005 /* The array must be an exact multiple of chunksize */
5006 while (chunksect && (mddev->array_sectors & (chunksect-1)))
5009 if ((chunksect<<9) < STRIPE_SIZE)
5010 /* array size does not allow a suitable chunk size */
5011 return ERR_PTR(-EINVAL);
5013 mddev->new_level = 5;
5014 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
5015 mddev->new_chunk = chunksect << 9;
5017 return setup_conf(mddev);
5020 static void *raid5_takeover_raid6(mddev_t *mddev)
5024 switch (mddev->layout) {
5025 case ALGORITHM_LEFT_ASYMMETRIC_6:
5026 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
5028 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5029 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
5031 case ALGORITHM_LEFT_SYMMETRIC_6:
5032 new_layout = ALGORITHM_LEFT_SYMMETRIC;
5034 case ALGORITHM_RIGHT_SYMMETRIC_6:
5035 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
5037 case ALGORITHM_PARITY_0_6:
5038 new_layout = ALGORITHM_PARITY_0;
5040 case ALGORITHM_PARITY_N:
5041 new_layout = ALGORITHM_PARITY_N;
5044 return ERR_PTR(-EINVAL);
5046 mddev->new_level = 5;
5047 mddev->new_layout = new_layout;
5048 mddev->delta_disks = -1;
5049 mddev->raid_disks -= 1;
5050 return setup_conf(mddev);
5054 static int raid5_reconfig(mddev_t *mddev, int new_layout, int new_chunk)
5056 /* Currently the layout and chunk size can only be changed
5057 * for a 2-drive raid array, as in that case no data shuffling
5059 * Later we might validate these and set new_* so a reshape
5060 * can complete the change.
5062 raid5_conf_t *conf = mddev_to_conf(mddev);
5064 if (new_layout >= 0 && !algorithm_valid_raid5(new_layout))
5066 if (new_chunk > 0) {
5067 if (new_chunk & (new_chunk-1))
5068 /* not a power of 2 */
5070 if (new_chunk < PAGE_SIZE)
5072 if (mddev->array_sectors & ((new_chunk>>9)-1))
5073 /* not factor of array size */
5077 /* They look valid */
5079 if (mddev->raid_disks != 2)
5082 if (new_layout >= 0) {
5083 conf->algorithm = new_layout;
5084 mddev->layout = mddev->new_layout = new_layout;
5086 if (new_chunk > 0) {
5087 conf->chunk_size = new_chunk;
5088 mddev->chunk_size = mddev->new_chunk = new_chunk;
5090 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5091 md_wakeup_thread(mddev->thread);
5095 static void *raid5_takeover(mddev_t *mddev)
5097 /* raid5 can take over:
5098 * raid0 - if all devices are the same - make it a raid4 layout
5099 * raid1 - if there are two drives. We need to know the chunk size
5100 * raid4 - trivial - just use a raid4 layout.
5101 * raid6 - Providing it is a *_6 layout
5103 * For now, just do raid1
5106 if (mddev->level == 1)
5107 return raid5_takeover_raid1(mddev);
5108 if (mddev->level == 4) {
5109 mddev->new_layout = ALGORITHM_PARITY_N;
5110 mddev->new_level = 5;
5111 return setup_conf(mddev);
5113 if (mddev->level == 6)
5114 return raid5_takeover_raid6(mddev);
5116 return ERR_PTR(-EINVAL);
5120 static struct mdk_personality raid5_personality;
5122 static void *raid6_takeover(mddev_t *mddev)
5124 /* Currently can only take over a raid5. We map the
5125 * personality to an equivalent raid6 personality
5126 * with the Q block at the end.
5130 if (mddev->pers != &raid5_personality)
5131 return ERR_PTR(-EINVAL);
5132 if (mddev->degraded > 1)
5133 return ERR_PTR(-EINVAL);
5134 if (mddev->raid_disks > 253)
5135 return ERR_PTR(-EINVAL);
5136 if (mddev->raid_disks < 3)
5137 return ERR_PTR(-EINVAL);
5139 switch (mddev->layout) {
5140 case ALGORITHM_LEFT_ASYMMETRIC:
5141 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
5143 case ALGORITHM_RIGHT_ASYMMETRIC:
5144 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
5146 case ALGORITHM_LEFT_SYMMETRIC:
5147 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
5149 case ALGORITHM_RIGHT_SYMMETRIC:
5150 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
5152 case ALGORITHM_PARITY_0:
5153 new_layout = ALGORITHM_PARITY_0_6;
5155 case ALGORITHM_PARITY_N:
5156 new_layout = ALGORITHM_PARITY_N;
5159 return ERR_PTR(-EINVAL);
5161 mddev->new_level = 6;
5162 mddev->new_layout = new_layout;
5163 mddev->delta_disks = 1;
5164 mddev->raid_disks += 1;
5165 return setup_conf(mddev);
5169 static struct mdk_personality raid6_personality =
5173 .owner = THIS_MODULE,
5174 .make_request = make_request,
5178 .error_handler = error,
5179 .hot_add_disk = raid5_add_disk,
5180 .hot_remove_disk= raid5_remove_disk,
5181 .spare_active = raid5_spare_active,
5182 .sync_request = sync_request,
5183 .resize = raid5_resize,
5185 #ifdef CONFIG_MD_RAID5_RESHAPE
5186 .check_reshape = raid5_check_reshape,
5187 .start_reshape = raid5_start_reshape,
5188 .finish_reshape = raid5_finish_reshape,
5190 .quiesce = raid5_quiesce,
5191 .takeover = raid6_takeover,
5193 static struct mdk_personality raid5_personality =
5197 .owner = THIS_MODULE,
5198 .make_request = make_request,
5202 .error_handler = error,
5203 .hot_add_disk = raid5_add_disk,
5204 .hot_remove_disk= raid5_remove_disk,
5205 .spare_active = raid5_spare_active,
5206 .sync_request = sync_request,
5207 .resize = raid5_resize,
5209 #ifdef CONFIG_MD_RAID5_RESHAPE
5210 .check_reshape = raid5_check_reshape,
5211 .start_reshape = raid5_start_reshape,
5212 .finish_reshape = raid5_finish_reshape,
5214 .quiesce = raid5_quiesce,
5215 .takeover = raid5_takeover,
5216 .reconfig = raid5_reconfig,
5219 static struct mdk_personality raid4_personality =
5223 .owner = THIS_MODULE,
5224 .make_request = make_request,
5228 .error_handler = error,
5229 .hot_add_disk = raid5_add_disk,
5230 .hot_remove_disk= raid5_remove_disk,
5231 .spare_active = raid5_spare_active,
5232 .sync_request = sync_request,
5233 .resize = raid5_resize,
5235 #ifdef CONFIG_MD_RAID5_RESHAPE
5236 .check_reshape = raid5_check_reshape,
5237 .start_reshape = raid5_start_reshape,
5238 .finish_reshape = raid5_finish_reshape,
5240 .quiesce = raid5_quiesce,
5243 static int __init raid5_init(void)
5245 register_md_personality(&raid6_personality);
5246 register_md_personality(&raid5_personality);
5247 register_md_personality(&raid4_personality);
5251 static void raid5_exit(void)
5253 unregister_md_personality(&raid6_personality);
5254 unregister_md_personality(&raid5_personality);
5255 unregister_md_personality(&raid4_personality);
5258 module_init(raid5_init);
5259 module_exit(raid5_exit);
5260 MODULE_LICENSE("GPL");
5261 MODULE_ALIAS("md-personality-4"); /* RAID5 */
5262 MODULE_ALIAS("md-raid5");
5263 MODULE_ALIAS("md-raid4");
5264 MODULE_ALIAS("md-level-5");
5265 MODULE_ALIAS("md-level-4");
5266 MODULE_ALIAS("md-personality-8"); /* RAID6 */
5267 MODULE_ALIAS("md-raid6");
5268 MODULE_ALIAS("md-level-6");
5270 /* This used to be two separate modules, they were: */
5271 MODULE_ALIAS("raid5");
5272 MODULE_ALIAS("raid6");