| blob | 685ddf325ee43f4492466c34a50a4a836c766e65 |
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 {
795 );
797 }
800 }
803 {
809 }
812 {
813 /* stop syncio and normal IO and wait for everything to
814 * go quiet.
815 * We increment barrier and nr_waiting, and then
816 * wait until nr_pending match nr_queued+1
817 * This is called in the context of one normal IO request
818 * that has failed. Thus any sync request that might be pending
819 * will be blocked by nr_pending, and we need to wait for
820 * pending IO requests to complete or be queued for re-try.
821 * Thus the number queued (nr_queued) plus this request (1)
822 * must match the number of pending IOs (nr_pending) before
823 * we continue.
824 */
834 }
837 {
838 /* reverse the effect of the freeze */
844 }
847 {
866 }
868 /* If this request crosses a chunk boundary, we need to
869 * split it. This will only happen for 1 PAGE (or less) requests.
870 */
875 /* Sanity check -- queue functions should prevent this happening */
879 /* This is a one page bio that upper layers
880 * refuse to split for us, so we need to split it.
881 */
885 /* Each of these 'make_request' calls will call 'wait_barrier'.
886 * If the first succeeds but the second blocks due to the resync
887 * thread raising the barrier, we will deadlock because the
888 * IO to the underlying device will be queued in generic_make_request
889 * and will never complete, so will never reduce nr_pending.
890 * So increment nr_waiting here so no new raise_barriers will
891 * succeed, and so the second wait_barrier cannot block.
892 */
907 bad_map:
914 }
918 /*
919 * Register the new request and wait if the reconstruction
920 * thread has put up a bar for new requests.
921 * Continue immediately if no resync is active currently.
922 */
934 /* We might need to issue multiple reads to different
935 * devices if there are bad blocks around, so we keep
936 * track of the number of reads in bio->bi_phys_segments.
937 * If this is 0, there is only one r10_bio and no locking
938 * will be needed when the request completes. If it is
939 * non-zero, then it is the number of not-completed requests.
940 */
945 /*
946 * read balancing logic:
947 */
951 read_again:
957 }
962 max_sectors);
974 /* Could not read all from this device, so we will
975 * need another r10_bio.
976 */
983 else
986 /* Cannot call generic_make_request directly
987 * as that will be queued in __generic_make_request
988 * and subsequent mempool_alloc might block
989 * waiting for it. so hand bio over to raid10d.
990 */
1005 }
1007 /*
1008 * WRITE:
1009 */
1014 }
1015 /* first select target devices under rcu_lock and
1016 * inc refcount on their rdev. Record them by setting
1017 * bios[x] to bio
1018 * If there are known/acknowledged bad blocks on any device
1019 * on which we have seen a write error, we want to avoid
1020 * writing to those blocks. This potentially requires several
1021 * writes to write around the bad blocks. Each set of writes
1022 * gets its own r10_bio with a set of bios attached. The number
1023 * of r10_bios is recored in bio->bi_phys_segments just as with
1024 * the read case.
1025 */
1029 retry_write:
1041 }
1046 }
1048 sector_t first_bad;
1054 max_sectors,
1057 /* Mustn't write here until the bad block
1058 * is acknowledged
1059 */
1064 }
1066 /* Cannot write here at all */
1069 /* Mustn't write more than bad_sectors
1070 * to other devices yet
1071 */
1073 /* We don't set R10BIO_Degraded as that
1074 * only applies if the disk is missing,
1075 * so it might be re-added, and we want to
1076 * know to recover this chunk.
1077 * In this case the device is here, and the
1078 * fact that this chunk is not in-sync is
1079 * recorded in the bad block log.
1080 */
1082 }
1087 }
1088 }
1091 }
1095 /* Have to wait for this device to get unblocked, then retry */
1103 }
1108 }
1111 /* We are splitting this into multiple parts, so
1112 * we need to prepare for allocating another r10_bio.
1113 */
1118 else
1121 }
1135 max_sectors);
1150 }
1152 /* Don't remove the bias on 'remaining' (one_write_done) until
1153 * after checking if we need to go around again.
1154 */
1158 /* We need another r10_bio. It has already been counted
1159 * in bio->bi_phys_segments.
1160 */
1170 }
1173 /* In case raid10d snuck in to freeze_array */
1178 }
1181 {
1192 else
1194 }
1202 }
1204 /* check if there are enough drives for
1205 * every block to appear on atleast one.
1206 * Don't consider the device numbered 'ignore'
1207 * as we might be about to remove it.
1208 */
1210 {
1219 cnt++;
1221 }
1226 }
1229 {
1233 /*
1234 * If it is not operational, then we have already marked it as dead
1235 * else if it is the last working disks, ignore the error, let the
1236 * next level up know.
1237 * else mark the drive as failed
1238 */
1241 /*
1242 * Don't fail the drive, just return an IO error.
1243 */
1250 /*
1251 * if recovery is running, make sure it aborts.
1252 */
1254 }
1263 }
1266 {
1274 }
1286 }
1287 }
1290 {
1296 }
1299 {
1306 /*
1307 * Find all non-in_sync disks within the RAID10 configuration
1308 * and mark them in_sync
1309 */
1315 count++;
1317 }
1318 }
1325 }
1329 {
1337 /* only hot-add to in-sync arrays, as recovery is
1338 * very different from resync
1339 */
1350 else
1361 /* as we don't honour merge_bvec_fn, we must
1362 * never risk violating it, so limit
1363 * ->max_segments to one lying with a single
1364 * page, as a one page request is never in
1365 * violation.
1366 */
1371 }
1381 }
1386 }
1389 {
1402 }
1403 /* Only remove faulty devices in recovery
1404 * is not possible.
1405 */
1411 }
1415 /* lost the race, try later */
1419 }
1421 }
1422 abort:
1426 }
1430 {
1439 else
1440 /* The write handler will notice the lack of
1441 * R10BIO_Uptodate and record any errors etc
1442 */
1446 /* for reconstruct, we always reschedule after a read.
1447 * for resync, only after all reads
1448 */
1452 /* we have read all the blocks,
1453 * do the comparison in process context in raid10d
1454 */
1456 }
1457 }
1460 {
1465 /* the primary of several recovery bios */
1470 else
1479 else
1482 }
1483 }
1484 }
1487 {
1493 sector_t first_bad;
1511 }
1513 /*
1514 * Note: sync and recover and handled very differently for raid10
1515 * This code is for resync.
1516 * For resync, we read through virtual addresses and read all blocks.
1517 * If there is any error, we schedule a write. The lowest numbered
1518 * drive is authoritative.
1519 * However requests come for physical address, so we need to map.
1520 * For every physical address there are raid_disks/copies virtual addresses,
1521 * which is always are least one, but is not necessarly an integer.
1522 * This means that a physical address can span multiple chunks, so we may
1523 * have to submit multiple io requests for a single sync request.
1524 */
1525 /*
1526 * We check if all blocks are in-sync and only write to blocks that
1527 * aren't in sync
1528 */
1530 {
1537 /* find the first device with a block */
1548 /* now find blocks with errors */
1560 /* We know that the bi_io_vec layout is the same for
1561 * both 'first' and 'i', so we just compare them.
1562 * All vec entries are PAGE_SIZE;
1563 */
1567 PAGE_SIZE))
1573 /* Don't fix anything. */
1575 }
1576 /* Ok, we need to write this bio, either to correct an
1577 * inconsistency or to correct an unreadable block.
1578 * First we need to fixup bv_offset, bv_len and
1579 * bi_vecs, as the read request might have corrupted these
1580 */
1598 PAGE_SIZE);
1599 }
1610 }
1612 done:
1616 }
1617 }
1619 /*
1620 * Now for the recovery code.
1621 * Recovery happens across physical sectors.
1622 * We recover all non-is_sync drives by finding the virtual address of
1623 * each, and then choose a working drive that also has that virt address.
1624 * There is a separate r10_bio for each non-in_sync drive.
1625 * Only the first two slots are in use. The first for reading,
1626 * The second for writing.
1627 *
1628 */
1630 {
1631 /* We got a read error during recovery.
1632 * We repeat the read in smaller page-sized sections.
1633 * If a read succeeds, write it to the new device or record
1634 * a bad block if we cannot.
1635 * If a read fails, record a bad block on both old and
1636 * new devices.
1637 */
1650 sector_t addr;
1659 addr,
1667 addr,
1673 }
1675 /* We don't worry if we cannot set a bad block -
1676 * it really is bad so there is no loss in not
1677 * recording it yet
1678 */
1682 /* need bad block on destination too */
1687 /* just abort the recovery */
1689 "md/raid10:%s: recovery aborted"
1698 }
1699 }
1700 }
1704 idx++;
1705 }
1706 }
1709 {
1718 }
1720 /*
1721 * share the pages with the first bio
1722 * and submit the write request
1723 */
1730 }
1733 /*
1734 * Used by fix_read_error() to decay the per rdev read_errors.
1735 * We halve the read error count for every hour that has elapsed
1736 * since the last recorded read error.
1737 *
1738 */
1740 {
1749 /* first time we've seen a read error */
1752 }
1759 /*
1760 * if hours_since_last is > the number of bits in read_errors
1761 * just set read errors to 0. We do this to avoid
1762 * overflowing the shift of read_errors by hours_since_last.
1763 */
1766 else
1768 }
1772 {
1773 sector_t first_bad;
1780 /* success */
1784 /* need to record an error - either for the block or the device */
1788 }
1790 /*
1791 * This is a kernel thread which:
1792 *
1793 * 1. Retries failed read operations on working mirrors.
1794 * 2. Updates the raid superblock when problems encounter.
1795 * 3. Performs writes following reads for array synchronising.
1796 */
1799 {
1806 /* still own a reference to this rdev, so it cannot
1807 * have been cleared recently.
1808 */
1812 /* drive has already been failed, just ignore any
1813 more fix_read_error() attempts */
1823 "md/raid10:%s: %s: Raid device exceeded "
1832 }
1845 sector_t first_bad;
1858 sect,
1865 }
1866 sl++;
1873 /* Cannot read from anywhere, just mark the block
1874 * as bad on the first device to discourage future
1875 * reads.
1876 */
1881 rdev,
1887 }
1890 /* write it back and re-read */
1897 sl--;
1908 sect,
1911 /* Well, this device is dead */
1913 "md/raid10:%s: read correction "
1914 "write failed"
1924 }
1927 }
1934 sl--;
1945 sect,
1947 READ)) {
1949 /* Well, this device is dead */
1951 "md/raid10:%s: unable to read back "
1952 "corrected sectors"
1965 "md/raid10:%s: read error corrected"
1972 }
1976 }
1981 }
1982 }
1985 {
1987 }
1990 {
2001 }
2004 {
2009 /* bio has the data to be written to slot 'i' where
2010 * we just recently had a write error.
2011 * We repeatedly clone the bio and trim down to one block,
2012 * then try the write. Where the write fails we record
2013 * a bad block.
2014 * It is conceivable that the bio doesn't exactly align with
2015 * blocks. We must handle this.
2016 *
2017 * We currently own a reference to the rdev.
2018 */
2021 sector_t sector;
2039 /* Write at 'sector' for 'sectors' */
2047 /* Failure! */
2056 }
2058 }
2061 {
2071 /* we got a read error. Maybe the drive is bad. Maybe just
2072 * the block and we can fix it.
2073 * We freeze all other IO, and try reading the block from
2074 * other devices. When we find one, we re-write
2075 * and check it that fixes the read error.
2076 * This is all done synchronously while the array is
2077 * frozen.
2078 */
2083 }
2090 read_more:
2100 }
2108 KERN_ERR
2109 "md/raid10:%s: %s: redirecting"
2118 max_sectors);
2127 /* Drat - have to split this up more */
2136 else
2143 GFP_NOIO);
2157 }
2160 {
2161 /* Some sort of write request has finished and it
2162 * succeeded in writing where we thought there was a
2163 * bad block. So forget the bad block.
2164 * Or possibly if failed and we need to record
2165 * a bad block.
2166 */
2180 rdev,
2185 rdev,
2189 }
2190 }
2199 rdev,
2209 }
2211 }
2212 }
2217 }
2218 }
2221 {
2239 }
2257 /* just a partial read to be scheduled from a
2258 * separate context
2259 */
2262 }
2267 }
2269 }
2273 {
2283 }
2285 /*
2286 * perform a "sync" on one "block"
2287 *
2288 * We need to make sure that no normal I/O request - particularly write
2289 * requests - conflict with active sync requests.
2290 *
2291 * This is achieved by tracking pending requests and a 'barrier' concept
2292 * that can be installed to exclude normal IO requests.
2293 *
2294 * Resync and recovery are handled very differently.
2295 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
2296 *
2297 * For resync, we iterate over virtual addresses, read all copies,
2298 * and update if there are differences. If only one copy is live,
2299 * skip it.
2300 * For recovery, we iterate over physical addresses, read a good
2301 * value for each non-in_sync drive, and over-write.
2302 *
2303 * So, for recovery we may have several outstanding complex requests for a
2304 * given address, one for each out-of-sync device. We model this by allocating
2305 * a number of r10_bio structures, one for each out-of-sync device.
2306 * As we setup these structures, we collect all bio's together into a list
2307 * which we then process collectively to add pages, and then process again
2308 * to pass to generic_make_request.
2309 *
2310 * The r10_bio structures are linked using a borrowed master_bio pointer.
2311 * This link is counted in ->remaining. When the r10_bio that points to NULL
2312 * has its remaining count decremented to 0, the whole complex operation
2313 * is complete.
2314 *
2315 */
2319 {
2326 sector_t sync_blocks;
2334 skipped:
2339 /* If we aborted, we need to abort the
2340 * sync on the 'current' bitmap chucks (there can
2341 * be several when recovering multiple devices).
2342 * as we may have started syncing it but not finished.
2343 * We can find the current address in
2344 * mddev->curr_resync, but for recovery,
2345 * we need to convert that to several
2346 * virtual addresses.
2347 */
2353 sector_t sect =
2357 }
2365 }
2367 /* if there has been nothing to do on any drive,
2368 * then there is nothing to do at all..
2369 */
2372 }
2377 /* make sure whole request will fit in a chunk - if chunks
2378 * are meaningful
2379 */
2383 /*
2384 * If there is non-resync activity waiting for us then
2385 * put in a delay to throttle resync.
2386 */
2390 /* Again, very different code for resync and recovery.
2391 * Both must result in an r10bio with a list of bios that
2392 * have bi_end_io, bi_sector, bi_bdev set,
2393 * and bi_private set to the r10bio.
2394 * For recovery, we may actually create several r10bios
2395 * with 2 bios in each, that correspond to the bios in the main one.
2396 * In this case, the subordinate r10bios link back through a
2397 * borrowed master_bio pointer, and the counter in the master
2398 * includes a ref from each subordinate.
2399 */
2400 /* First, we decide what to do and set ->bi_end_io
2401 * To end_sync_read if we want to read, and
2402 * end_sync_write if we will want to write.
2403 */
2407 /* recovery... the complicated one */
2414 sector_t sect;
2423 /* want to reconstruct this device */
2426 /* Unless we are doing a full sync, we only need
2427 * to recover the block if it is set in the bitmap
2428 */
2435 /* yep, skip the sync_blocks here, but don't assume
2436 * that there will never be anything to do here
2437 */
2440 }
2455 /* Need to check if the array will still be
2456 * degraded
2457 */
2463 }
2479 /* This is where we read from */
2489 bad_sectors -= (sector
2494 }
2495 }
2508 /* and we write to 'i' */
2531 }
2533 /* Cannot recover, so abort the recovery or
2534 * record a bad block */
2540 /* problem is that there are bad blocks
2541 * on other device(s)
2542 */
2552 }
2561 }
2563 }
2564 }
2571 }
2573 }
2575 /* resync. Schedule a read for every block at this virt offset */
2584 /* We can skip this block */
2587 }
2625 }
2626 }
2637 count++;
2638 }
2645 mddev);
2646 }
2650 }
2651 }
2662 }
2680 /* stop here */
2685 /* remove last page from this bio */
2689 }
2691 }
2695 bio_full:
2709 }
2710 }
2713 /* pretend they weren't skipped, it makes
2714 * no important difference in this case
2715 */
2719 giveup:
2720 /* There is nowhere to write, so all non-sync
2721 * drives must be failed or in resync, all drives
2722 * have a bad block, so try the next chunk...
2723 */
2728 chunks_skipped ++;
2731 }
2733 static sector_t
2735 {
2736 sector_t size;
2750 }
2754 {
2766 }
2777 }
2785 GFP_KERNEL);
2811 /* 'size' is now the number of chunks in the array */
2812 /* calculate "used chunks per device" in 'stride' */
2815 /* We need to round up when dividing by raid_disks to
2816 * get the stride size.
2817 */
2825 else
2843 out:
2852 }
2854 }
2857 {
2862 sector_t size;
2864 /*
2865 * copy the already verified devices into our private RAID10
2866 * bookkeeping area. [whatever we allocate in run(),
2867 * should be freed in stop()]
2868 */
2875 }
2887 else
2902 /* as we don't honour merge_bvec_fn, we must never risk
2903 * violating it, so limit max_segments to 1 lying
2904 * within a single page.
2905 */
2910 }
2913 }
2914 /* need to check that every block has at least one working mirror */
2919 }
2932 }
2934 }
2944 /*
2945 * Ok, everything is just fine now
2946 */
2955 /* Calculate max read-ahead size.
2956 * We need to readahead at least twice a whole stripe....
2957 * maybe...
2958 */
2959 {
2965 }
2975 out_free_conf:
2983 out:
2985 }
2988 {
3002 }
3005 {
3015 }
3016 }
3019 {
3027 }
3029 /* Set new parameters */
3031 /* new layout: far_copies = 1, near_copies = 2 */
3036 /* make sure it will be not marked as dirty */
3045 }
3048 }
3051 {
3054 /* raid10 can take over:
3055 * raid0 - providing it has only two drives
3056 */
3058 /* for raid0 takeover only one zone is supported */
3065 }
3067 }
3069 }
3072 {
3088 };
3091 {
3093 }
3096 {
3098 }