| blob | 7a9eef64c4f9355e8816721eddceefee66d91efd |
1 /*
2 * raid10.c : Multiple Devices driver for Linux
3 *
4 * Copyright (C) 2000-2004 Neil Brown
5 *
6 * RAID-10 support for md.
7 *
8 * Base on code in raid1.c. See raid1.c for further copyright information.
9 *
10 *
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2, or (at your option)
14 * any later version.
15 *
16 * You should have received a copy of the GNU General Public License
17 * (for example /usr/src/linux/COPYING); if not, write to the Free
18 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 */
21 #include <linux/slab.h>
22 #include <linux/delay.h>
23 #include <linux/blkdev.h>
24 #include <linux/module.h>
25 #include <linux/seq_file.h>
26 #include <linux/ratelimit.h>
32 /*
33 * RAID10 provides a combination of RAID0 and RAID1 functionality.
34 * The layout of data is defined by
35 * chunk_size
36 * raid_disks
37 * near_copies (stored in low byte of layout)
38 * far_copies (stored in second byte of layout)
39 * far_offset (stored in bit 16 of layout )
40 *
41 * The data to be stored is divided into chunks using chunksize.
42 * Each device is divided into far_copies sections.
43 * In each section, chunks are laid out in a style similar to raid0, but
44 * near_copies copies of each chunk is stored (each on a different drive).
45 * The starting device for each section is offset near_copies from the starting
46 * device of the previous section.
47 * Thus they are (near_copies*far_copies) of each chunk, and each is on a different
48 * drive.
49 * near_copies and far_copies must be at least one, and their product is at most
50 * raid_disks.
51 *
52 * If far_offset is true, then the far_copies are handled a bit differently.
53 * The copies are still in different stripes, but instead of be very far apart
54 * on disk, there are adjacent stripes.
55 */
57 /*
58 * Number of guaranteed r10bios in case of extreme VM load:
59 */
60 #define NR_RAID10_BIOS 256
62 /* When there are this many requests queue to be written by
63 * the raid10 thread, we become 'congested' to provide back-pressure
64 * for writeback.
65 */
72 {
76 /* allocate a r10bio with room for raid_disks entries in the bios array */
78 }
81 {
83 }
85 /* Maximum size of each resync request */
86 #define RESYNC_BLOCK_SIZE (64*1024)
87 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
88 /* amount of memory to reserve for resync requests */
89 #define RESYNC_WINDOW (1024*1024)
90 /* maximum number of concurrent requests, memory permitting */
91 #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE)
93 /*
94 * When performing a resync, we need to read and compare, so
95 * we need as many pages are there are copies.
96 * When performing a recovery, we need 2 bios, one for read,
97 * one for write (we recover only one drive per r10buf)
98 *
99 */
101 {
115 else
118 /*
119 * Allocate bios.
120 */
126 }
127 /*
128 * Allocate RESYNC_PAGES data pages and attach them
129 * where needed.
130 */
136 /* we can share bv_page's during recovery */
146 }
147 }
151 out_free_pages:
158 out_free_bio:
163 }
166 {
178 }
180 }
181 }
183 }
186 {
194 }
195 }
198 {
203 }
206 {
212 }
215 {
225 /* wake up frozen array... */
229 }
231 /*
232 * raid_end_bio_io() is called when we have finished servicing a mirrored
233 * operation and are ready to return a success/failure code to the buffer
234 * cache layer.
235 */
237 {
254 /*
255 * Wake up any possible resync thread that waits for the device
256 * to go idle.
257 */
259 }
261 }
263 /*
264 * Update disk head position estimator based on IRQ completion info.
265 */
267 {
272 }
274 /*
275 * Find the disk number which triggered given bio
276 */
279 {
292 }
295 {
304 /*
305 * this branch is our 'one mirror IO has finished' event handler:
306 */
310 /*
311 * Set R10BIO_Uptodate in our master bio, so that
312 * we will return a good error code to the higher
313 * levels even if IO on some other mirrored buffer fails.
314 *
315 * The 'master' represents the composite IO operation to
316 * user-side. So if something waits for IO, then it will
317 * wait for the 'master' bio.
318 */
323 /*
324 * oops, read error - keep the refcount on the rdev
325 */
334 }
335 }
338 {
339 /* clear the bitmap if all writes complete successfully */
345 }
348 {
356 else
358 }
359 }
360 }
363 {
373 /*
374 * this branch is our 'one mirror IO has finished' event handler:
375 */
381 /*
382 * Set R10BIO_Uptodate in our master bio, so that
383 * we will return a good error code for to the higher
384 * levels even if IO on some other mirrored buffer fails.
385 *
386 * The 'master' represents the composite IO operation to
387 * user-side. So if something waits for IO, then it will
388 * wait for the 'master' bio.
389 */
390 sector_t first_bad;
395 /* Maybe we can clear some bad blocks. */
404 }
405 }
407 /*
408 *
409 * Let's see if all mirrored write operations have finished
410 * already.
411 */
415 }
418 /*
419 * RAID10 layout manager
420 * As well as the chunksize and raid_disks count, there are two
421 * parameters: near_copies and far_copies.
422 * near_copies * far_copies must be <= raid_disks.
423 * Normally one of these will be 1.
424 * If both are 1, we get raid0.
425 * If near_copies == raid_disks, we get raid1.
426 *
427 * Chunks are laid out in raid0 style with near_copies copies of the
428 * first chunk, followed by near_copies copies of the next chunk and
429 * so on.
430 * If far_copies > 1, then after 1/far_copies of the array has been assigned
431 * as described above, we start again with a device offset of near_copies.
432 * So we effectively have another copy of the whole array further down all
433 * the drives, but with blocks on different drives.
434 * With this layout, and block is never stored twice on the one device.
435 *
436 * raid10_find_phys finds the sector offset of a given virtual sector
437 * on each device that it is on.
438 *
439 * raid10_find_virt does the reverse mapping, from a device and a
440 * sector offset to a virtual address
441 */
444 {
446 sector_t sector;
447 sector_t chunk;
448 sector_t stripe;
453 /* now calculate first sector/dev */
465 /* and calculate all the others */
471 slot++;
480 slot++;
481 }
482 dev++;
486 }
487 }
489 }
492 {
508 else
510 }
512 }
516 }
518 /**
519 * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged
520 * @q: request queue
521 * @bvm: properties of new bio
522 * @biovec: the request that could be merged to it.
523 *
524 * Return amount of bytes we can accept at this offset
525 * If near_copies == raid_disk, there are no striping issues,
526 * but in that case, the function isn't called at all.
527 */
531 {
542 else
544 }
546 /*
547 * This routine returns the disk from which the requested read should
548 * be done. There is a per-array 'next expected sequential IO' sector
549 * number - if this matches on the next IO then we use the last disk.
550 * There is also a per-disk 'last know head position' sector that is
551 * maintained from IRQ contexts, both the normal and the resync IO
552 * completion handlers update this position correctly. If there is no
553 * perfect sequential match then we pick the disk whose head is closest.
554 *
555 * If there are 2 mirrors in the same 2 devices, performance degrades
556 * because position is mirror, not device based.
557 *
558 * The rdev for the device selected will have nr_pending incremented.
559 */
561 /*
562 * FIXME: possibly should rethink readbalancing and do it differently
563 * depending on near_copies / far_copies geometry.
564 */
566 {
578 retry:
584 /*
585 * Check if we can balance. We can balance on the whole
586 * device if no resync is going on (recovery is ok), or below
587 * the resync window. We take the first readable disk when
588 * above the resync window.
589 */
595 sector_t first_bad;
597 sector_t dev_sector;
612 /* Already have a better slot */
615 /* Cannot read here. If this is the
616 * 'primary' device, then we must not read
617 * beyond 'bad_sectors' from another device.
618 */
625 sector_t good_sectors =
630 }
632 /* Must read from here */
634 }
642 /* This optimisation is debatable, and completely destroys
643 * sequential read speed for 'far copies' arrays. So only
644 * keep it for 'near' arrays, and review those later.
645 */
649 /* for far > 1 always use the lowest address */
652 else
658 }
659 }
670 /* Cannot risk returning a device that failed
671 * before we inc'ed nr_pending
672 */
675 }
683 }
686 {
704 }
705 }
708 }
711 {
712 /* Any writes that have been queued but are awaiting
713 * bitmap updates get flushed here.
714 */
722 /* flush any pending bitmap writes to disk
723 * before proceeding w/ I/O */
732 }
735 }
737 /* Barriers....
738 * Sometimes we need to suspend IO while we do something else,
739 * either some resync/recovery, or reconfigure the array.
740 * To do this we raise a 'barrier'.
741 * The 'barrier' is a counter that can be raised multiple times
742 * to count how many activities are happening which preclude
743 * normal IO.
744 * We can only raise the barrier if there is no pending IO.
745 * i.e. if nr_pending == 0.
746 * We choose only to raise the barrier if no-one is waiting for the
747 * barrier to go down. This means that as soon as an IO request
748 * is ready, no other operations which require a barrier will start
749 * until the IO request has had a chance.
750 *
751 * So: regular IO calls 'wait_barrier'. When that returns there
752 * is no backgroup IO happening, It must arrange to call
753 * allow_barrier when it has finished its IO.
754 * backgroup IO calls must call raise_barrier. Once that returns
755 * there is no normal IO happeing. It must arrange to call
756 * lower_barrier when the particular background IO completes.
757 */
760 {
764 /* Wait until no block IO is waiting (unless 'force') */
768 /* block any new IO from starting */
771 /* Now wait for all pending IO to complete */
777 }
780 {
786 }
789 {
793 /* Wait for the barrier to drop.
794 * However if there are already pending
795 * requests (preventing the barrier from
796 * rising completely), and the
797 * pre-process bio queue isn't empty,
798 * then don't wait, as we need to empty
799 * that queue to get the nr_pending
800 * count down.
801 */
808 );
810 }
813 }
816 {
822 }
825 {
826 /* stop syncio and normal IO and wait for everything to
827 * go quiet.
828 * We increment barrier and nr_waiting, and then
829 * wait until nr_pending match nr_queued+1
830 * This is called in the context of one normal IO request
831 * that has failed. Thus any sync request that might be pending
832 * will be blocked by nr_pending, and we need to wait for
833 * pending IO requests to complete or be queued for re-try.
834 * Thus the number queued (nr_queued) plus this request (1)
835 * must match the number of pending IOs (nr_pending) before
836 * we continue.
837 */
847 }
850 {
851 /* reverse the effect of the freeze */
857 }
860 {
879 }
881 /* If this request crosses a chunk boundary, we need to
882 * split it. This will only happen for 1 PAGE (or less) requests.
883 */
888 /* Sanity check -- queue functions should prevent this happening */
892 /* This is a one page bio that upper layers
893 * refuse to split for us, so we need to split it.
894 */
898 /* Each of these 'make_request' calls will call 'wait_barrier'.
899 * If the first succeeds but the second blocks due to the resync
900 * thread raising the barrier, we will deadlock because the
901 * IO to the underlying device will be queued in generic_make_request
902 * and will never complete, so will never reduce nr_pending.
903 * So increment nr_waiting here so no new raise_barriers will
904 * succeed, and so the second wait_barrier cannot block.
905 */
920 bad_map:
927 }
931 /*
932 * Register the new request and wait if the reconstruction
933 * thread has put up a bar for new requests.
934 * Continue immediately if no resync is active currently.
935 */
947 /* We might need to issue multiple reads to different
948 * devices if there are bad blocks around, so we keep
949 * track of the number of reads in bio->bi_phys_segments.
950 * If this is 0, there is only one r10_bio and no locking
951 * will be needed when the request completes. If it is
952 * non-zero, then it is the number of not-completed requests.
953 */
958 /*
959 * read balancing logic:
960 */
964 read_again:
970 }
975 max_sectors);
987 /* Could not read all from this device, so we will
988 * need another r10_bio.
989 */
996 else
999 /* Cannot call generic_make_request directly
1000 * as that will be queued in __generic_make_request
1001 * and subsequent mempool_alloc might block
1002 * waiting for it. so hand bio over to raid10d.
1003 */
1018 }
1020 /*
1021 * WRITE:
1022 */
1027 }
1028 /* first select target devices under rcu_lock and
1029 * inc refcount on their rdev. Record them by setting
1030 * bios[x] to bio
1031 * If there are known/acknowledged bad blocks on any device
1032 * on which we have seen a write error, we want to avoid
1033 * writing to those blocks. This potentially requires several
1034 * writes to write around the bad blocks. Each set of writes
1035 * gets its own r10_bio with a set of bios attached. The number
1036 * of r10_bios is recored in bio->bi_phys_segments just as with
1037 * the read case.
1038 */
1042 retry_write:
1054 }
1059 }
1061 sector_t first_bad;
1067 max_sectors,
1070 /* Mustn't write here until the bad block
1071 * is acknowledged
1072 */
1077 }
1079 /* Cannot write here at all */
1082 /* Mustn't write more than bad_sectors
1083 * to other devices yet
1084 */
1086 /* We don't set R10BIO_Degraded as that
1087 * only applies if the disk is missing,
1088 * so it might be re-added, and we want to
1089 * know to recover this chunk.
1090 * In this case the device is here, and the
1091 * fact that this chunk is not in-sync is
1092 * recorded in the bad block log.
1093 */
1095 }
1100 }
1101 }
1104 }
1108 /* Have to wait for this device to get unblocked, then retry */
1116 }
1121 }
1124 /* We are splitting this into multiple parts, so
1125 * we need to prepare for allocating another r10_bio.
1126 */
1131 else
1134 }
1148 max_sectors);
1163 }
1165 /* Don't remove the bias on 'remaining' (one_write_done) until
1166 * after checking if we need to go around again.
1167 */
1171 /* We need another r10_bio. It has already been counted
1172 * in bio->bi_phys_segments.
1173 */
1183 }
1186 /* In case raid10d snuck in to freeze_array */
1191 }
1194 {
1205 else
1207 }
1215 }
1217 /* check if there are enough drives for
1218 * every block to appear on atleast one.
1219 * Don't consider the device numbered 'ignore'
1220 * as we might be about to remove it.
1221 */
1223 {
1232 cnt++;
1234 }
1239 }
1242 {
1246 /*
1247 * If it is not operational, then we have already marked it as dead
1248 * else if it is the last working disks, ignore the error, let the
1249 * next level up know.
1250 * else mark the drive as failed
1251 */
1254 /*
1255 * Don't fail the drive, just return an IO error.
1256 */
1263 /*
1264 * if recovery is running, make sure it aborts.
1265 */
1267 }
1276 }
1279 {
1287 }
1299 }
1300 }
1303 {
1309 }
1312 {
1319 /*
1320 * Find all non-in_sync disks within the RAID10 configuration
1321 * and mark them in_sync
1322 */
1328 count++;
1330 }
1331 }
1338 }
1342 {
1350 /* only hot-add to in-sync arrays, as recovery is
1351 * very different from resync
1352 */
1363 else
1374 /* as we don't honour merge_bvec_fn, we must
1375 * never risk violating it, so limit
1376 * ->max_segments to one lying with a single
1377 * page, as a one page request is never in
1378 * violation.
1379 */
1384 }
1394 }
1399 }
1402 {
1415 }
1416 /* Only remove faulty devices in recovery
1417 * is not possible.
1418 */
1424 }
1428 /* lost the race, try later */
1432 }
1434 }
1435 abort:
1439 }
1443 {
1452 else
1453 /* The write handler will notice the lack of
1454 * R10BIO_Uptodate and record any errors etc
1455 */
1459 /* for reconstruct, we always reschedule after a read.
1460 * for resync, only after all reads
1461 */
1465 /* we have read all the blocks,
1466 * do the comparison in process context in raid10d
1467 */
1469 }
1470 }
1473 {
1478 /* the primary of several recovery bios */
1483 else
1492 else
1495 }
1496 }
1497 }
1500 {
1506 sector_t first_bad;
1524 }
1526 /*
1527 * Note: sync and recover and handled very differently for raid10
1528 * This code is for resync.
1529 * For resync, we read through virtual addresses and read all blocks.
1530 * If there is any error, we schedule a write. The lowest numbered
1531 * drive is authoritative.
1532 * However requests come for physical address, so we need to map.
1533 * For every physical address there are raid_disks/copies virtual addresses,
1534 * which is always are least one, but is not necessarly an integer.
1535 * This means that a physical address can span multiple chunks, so we may
1536 * have to submit multiple io requests for a single sync request.
1537 */
1538 /*
1539 * We check if all blocks are in-sync and only write to blocks that
1540 * aren't in sync
1541 */
1543 {
1550 /* find the first device with a block */
1561 /* now find blocks with errors */
1573 /* We know that the bi_io_vec layout is the same for
1574 * both 'first' and 'i', so we just compare them.
1575 * All vec entries are PAGE_SIZE;
1576 */
1580 PAGE_SIZE))
1586 /* Don't fix anything. */
1588 }
1589 /* Ok, we need to write this bio, either to correct an
1590 * inconsistency or to correct an unreadable block.
1591 * First we need to fixup bv_offset, bv_len and
1592 * bi_vecs, as the read request might have corrupted these
1593 */
1611 PAGE_SIZE);
1612 }
1623 }
1625 done:
1629 }
1630 }
1632 /*
1633 * Now for the recovery code.
1634 * Recovery happens across physical sectors.
1635 * We recover all non-is_sync drives by finding the virtual address of
1636 * each, and then choose a working drive that also has that virt address.
1637 * There is a separate r10_bio for each non-in_sync drive.
1638 * Only the first two slots are in use. The first for reading,
1639 * The second for writing.
1640 *
1641 */
1643 {
1644 /* We got a read error during recovery.
1645 * We repeat the read in smaller page-sized sections.
1646 * If a read succeeds, write it to the new device or record
1647 * a bad block if we cannot.
1648 * If a read fails, record a bad block on both old and
1649 * new devices.
1650 */
1663 sector_t addr;
1672 addr,
1680 addr,
1686 }
1688 /* We don't worry if we cannot set a bad block -
1689 * it really is bad so there is no loss in not
1690 * recording it yet
1691 */
1695 /* need bad block on destination too */
1700 /* just abort the recovery */
1702 "md/raid10:%s: recovery aborted"
1711 }
1712 }
1713 }
1717 idx++;
1718 }
1719 }
1722 {
1731 }
1733 /*
1734 * share the pages with the first bio
1735 * and submit the write request
1736 */
1743 }
1746 /*
1747 * Used by fix_read_error() to decay the per rdev read_errors.
1748 * We halve the read error count for every hour that has elapsed
1749 * since the last recorded read error.
1750 *
1751 */
1753 {
1762 /* first time we've seen a read error */
1765 }
1772 /*
1773 * if hours_since_last is > the number of bits in read_errors
1774 * just set read errors to 0. We do this to avoid
1775 * overflowing the shift of read_errors by hours_since_last.
1776 */
1779 else
1781 }
1785 {
1786 sector_t first_bad;
1793 /* success */
1797 /* need to record an error - either for the block or the device */
1801 }
1803 /*
1804 * This is a kernel thread which:
1805 *
1806 * 1. Retries failed read operations on working mirrors.
1807 * 2. Updates the raid superblock when problems encounter.
1808 * 3. Performs writes following reads for array synchronising.
1809 */
1812 {
1819 /* still own a reference to this rdev, so it cannot
1820 * have been cleared recently.
1821 */
1825 /* drive has already been failed, just ignore any
1826 more fix_read_error() attempts */
1836 "md/raid10:%s: %s: Raid device exceeded "
1845 }
1858 sector_t first_bad;
1871 sect,
1878 }
1879 sl++;
1886 /* Cannot read from anywhere, just mark the block
1887 * as bad on the first device to discourage future
1888 * reads.
1889 */
1894 rdev,
1900 }
1903 /* write it back and re-read */
1910 sl--;
1921 sect,
1924 /* Well, this device is dead */
1926 "md/raid10:%s: read correction "
1927 "write failed"
1937 }
1940 }
1947 sl--;
1958 sect,
1960 READ)) {
1962 /* Well, this device is dead */
1964 "md/raid10:%s: unable to read back "
1965 "corrected sectors"
1978 "md/raid10:%s: read error corrected"
1985 }
1989 }
1994 }
1995 }
1998 {
2000 }
2003 {
2014 }
2017 {
2022 /* bio has the data to be written to slot 'i' where
2023 * we just recently had a write error.
2024 * We repeatedly clone the bio and trim down to one block,
2025 * then try the write. Where the write fails we record
2026 * a bad block.
2027 * It is conceivable that the bio doesn't exactly align with
2028 * blocks. We must handle this.
2029 *
2030 * We currently own a reference to the rdev.
2031 */
2034 sector_t sector;
2052 /* Write at 'sector' for 'sectors' */
2060 /* Failure! */
2069 }
2071 }
2074 {
2084 /* we got a read error. Maybe the drive is bad. Maybe just
2085 * the block and we can fix it.
2086 * We freeze all other IO, and try reading the block from
2087 * other devices. When we find one, we re-write
2088 * and check it that fixes the read error.
2089 * This is all done synchronously while the array is
2090 * frozen.
2091 */
2096 }
2103 read_more:
2113 }
2121 KERN_ERR
2122 "md/raid10:%s: %s: redirecting "
2131 max_sectors);
2140 /* Drat - have to split this up more */
2149 else
2156 GFP_NOIO);
2170 }
2173 {
2174 /* Some sort of write request has finished and it
2175 * succeeded in writing where we thought there was a
2176 * bad block. So forget the bad block.
2177 * Or possibly if failed and we need to record
2178 * a bad block.
2179 */
2193 rdev,
2198 rdev,
2202 }
2203 }
2212 rdev,
2222 }
2224 }
2225 }
2230 }
2231 }
2234 {
2252 }
2270 /* just a partial read to be scheduled from a
2271 * separate context
2272 */
2275 }
2280 }
2282 }
2286 {
2296 }
2298 /*
2299 * perform a "sync" on one "block"
2300 *
2301 * We need to make sure that no normal I/O request - particularly write
2302 * requests - conflict with active sync requests.
2303 *
2304 * This is achieved by tracking pending requests and a 'barrier' concept
2305 * that can be installed to exclude normal IO requests.
2306 *
2307 * Resync and recovery are handled very differently.
2308 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
2309 *
2310 * For resync, we iterate over virtual addresses, read all copies,
2311 * and update if there are differences. If only one copy is live,
2312 * skip it.
2313 * For recovery, we iterate over physical addresses, read a good
2314 * value for each non-in_sync drive, and over-write.
2315 *
2316 * So, for recovery we may have several outstanding complex requests for a
2317 * given address, one for each out-of-sync device. We model this by allocating
2318 * a number of r10_bio structures, one for each out-of-sync device.
2319 * As we setup these structures, we collect all bio's together into a list
2320 * which we then process collectively to add pages, and then process again
2321 * to pass to generic_make_request.
2322 *
2323 * The r10_bio structures are linked using a borrowed master_bio pointer.
2324 * This link is counted in ->remaining. When the r10_bio that points to NULL
2325 * has its remaining count decremented to 0, the whole complex operation
2326 * is complete.
2327 *
2328 */
2332 {
2339 sector_t sync_blocks;
2347 skipped:
2352 /* If we aborted, we need to abort the
2353 * sync on the 'current' bitmap chucks (there can
2354 * be several when recovering multiple devices).
2355 * as we may have started syncing it but not finished.
2356 * We can find the current address in
2357 * mddev->curr_resync, but for recovery,
2358 * we need to convert that to several
2359 * virtual addresses.
2360 */
2366 sector_t sect =
2370 }
2378 }
2380 /* if there has been nothing to do on any drive,
2381 * then there is nothing to do at all..
2382 */
2385 }
2390 /* make sure whole request will fit in a chunk - if chunks
2391 * are meaningful
2392 */
2396 /*
2397 * If there is non-resync activity waiting for us then
2398 * put in a delay to throttle resync.
2399 */
2403 /* Again, very different code for resync and recovery.
2404 * Both must result in an r10bio with a list of bios that
2405 * have bi_end_io, bi_sector, bi_bdev set,
2406 * and bi_private set to the r10bio.
2407 * For recovery, we may actually create several r10bios
2408 * with 2 bios in each, that correspond to the bios in the main one.
2409 * In this case, the subordinate r10bios link back through a
2410 * borrowed master_bio pointer, and the counter in the master
2411 * includes a ref from each subordinate.
2412 */
2413 /* First, we decide what to do and set ->bi_end_io
2414 * To end_sync_read if we want to read, and
2415 * end_sync_write if we will want to write.
2416 */
2420 /* recovery... the complicated one */
2427 sector_t sect;
2436 /* want to reconstruct this device */
2440 /* last stripe is not complete - don't
2441 * try to recover this sector.
2442 */
2444 }
2445 /* Unless we are doing a full sync, we only need
2446 * to recover the block if it is set in the bitmap
2447 */
2454 /* yep, skip the sync_blocks here, but don't assume
2455 * that there will never be anything to do here
2456 */
2459 }
2474 /* Need to check if the array will still be
2475 * degraded
2476 */
2482 }
2498 /* This is where we read from */
2508 bad_sectors -= (sector
2513 }
2514 }
2527 /* and we write to 'i' */
2550 }
2552 /* Cannot recover, so abort the recovery or
2553 * record a bad block */
2559 /* problem is that there are bad blocks
2560 * on other device(s)
2561 */
2571 }
2580 }
2582 }
2583 }
2590 }
2592 }
2594 /* resync. Schedule a read for every block at this virt offset */
2603 /* We can skip this block */
2606 }
2644 }
2645 }
2656 count++;
2657 }
2664 mddev);
2665 }
2669 }
2670 }
2681 }
2699 /* stop here */
2704 /* remove last page from this bio */
2708 }
2710 }
2714 bio_full:
2728 }
2729 }
2732 /* pretend they weren't skipped, it makes
2733 * no important difference in this case
2734 */
2738 giveup:
2739 /* There is nowhere to write, so all non-sync
2740 * drives must be failed or in resync, all drives
2741 * have a bad block, so try the next chunk...
2742 */
2747 chunks_skipped ++;
2750 }
2752 static sector_t
2754 {
2755 sector_t size;
2769 }
2773 {
2785 }
2796 }
2804 GFP_KERNEL);
2830 /* 'size' is now the number of chunks in the array */
2831 /* calculate "used chunks per device" in 'stride' */
2834 /* We need to round up when dividing by raid_disks to
2835 * get the stride size.
2836 */
2844 else
2862 out:
2871 }
2873 }
2876 {
2881 sector_t size;
2883 /*
2884 * copy the already verified devices into our private RAID10
2885 * bookkeeping area. [whatever we allocate in run(),
2886 * should be freed in stop()]
2887 */
2894 }
2906 else
2921 /* as we don't honour merge_bvec_fn, we must never risk
2922 * violating it, so limit max_segments to 1 lying
2923 * within a single page.
2924 */
2929 }
2932 }
2933 /* need to check that every block has at least one working mirror */
2938 }
2951 }
2953 }
2963 /*
2964 * Ok, everything is just fine now
2965 */
2974 /* Calculate max read-ahead size.
2975 * We need to readahead at least twice a whole stripe....
2976 * maybe...
2977 */
2978 {
2984 }
2994 out_free_conf:
3002 out:
3004 }
3007 {
3021 }
3024 {
3034 }
3035 }
3038 {
3046 }
3048 /* Set new parameters */
3050 /* new layout: far_copies = 1, near_copies = 2 */
3055 /* make sure it will be not marked as dirty */
3064 }
3067 }
3070 {
3073 /* raid10 can take over:
3074 * raid0 - providing it has only two drives
3075 */
3077 /* for raid0 takeover only one zone is supported */
3084 }
3086 }
3088 }
3091 {
3107 };
3110 {
3112 }
3115 {
3117 }