| blob | 8b29cd4f01c89f8213d7dcc43cd1aaf6dd5799c7 |
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/seq_file.h>
25 #include <linux/ratelimit.h>
31 /*
32 * RAID10 provides a combination of RAID0 and RAID1 functionality.
33 * The layout of data is defined by
34 * chunk_size
35 * raid_disks
36 * near_copies (stored in low byte of layout)
37 * far_copies (stored in second byte of layout)
38 * far_offset (stored in bit 16 of layout )
39 *
40 * The data to be stored is divided into chunks using chunksize.
41 * Each device is divided into far_copies sections.
42 * In each section, chunks are laid out in a style similar to raid0, but
43 * near_copies copies of each chunk is stored (each on a different drive).
44 * The starting device for each section is offset near_copies from the starting
45 * device of the previous section.
46 * Thus they are (near_copies*far_copies) of each chunk, and each is on a different
47 * drive.
48 * near_copies and far_copies must be at least one, and their product is at most
49 * raid_disks.
50 *
51 * If far_offset is true, then the far_copies are handled a bit differently.
52 * The copies are still in different stripes, but instead of be very far apart
53 * on disk, there are adjacent stripes.
54 */
56 /*
57 * Number of guaranteed r10bios in case of extreme VM load:
58 */
59 #define NR_RAID10_BIOS 256
65 {
69 /* allocate a r10bio with room for raid_disks entries in the bios array */
71 }
74 {
76 }
78 /* Maximum size of each resync request */
79 #define RESYNC_BLOCK_SIZE (64*1024)
80 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
81 /* amount of memory to reserve for resync requests */
82 #define RESYNC_WINDOW (1024*1024)
83 /* maximum number of concurrent requests, memory permitting */
84 #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE)
86 /*
87 * When performing a resync, we need to read and compare, so
88 * we need as many pages are there are copies.
89 * When performing a recovery, we need 2 bios, one for read,
90 * one for write (we recover only one drive per r10buf)
91 *
92 */
94 {
108 else
111 /*
112 * Allocate bios.
113 */
119 }
120 /*
121 * Allocate RESYNC_PAGES data pages and attach them
122 * where needed.
123 */
129 /* we can share bv_page's during recovery */
139 }
140 }
144 out_free_pages:
151 out_free_bio:
156 }
159 {
171 }
173 }
174 }
176 }
179 {
187 }
188 }
191 {
196 }
199 {
205 }
208 {
218 /* wake up frozen array... */
222 }
224 /*
225 * raid_end_bio_io() is called when we have finished servicing a mirrored
226 * operation and are ready to return a success/failure code to the buffer
227 * cache layer.
228 */
230 {
247 /*
248 * Wake up any possible resync thread that waits for the device
249 * to go idle.
250 */
252 }
254 }
256 /*
257 * Update disk head position estimator based on IRQ completion info.
258 */
260 {
265 }
267 /*
268 * Find the disk number which triggered given bio
269 */
272 {
285 }
288 {
297 /*
298 * this branch is our 'one mirror IO has finished' event handler:
299 */
303 /*
304 * Set R10BIO_Uptodate in our master bio, so that
305 * we will return a good error code to the higher
306 * levels even if IO on some other mirrored buffer fails.
307 *
308 * The 'master' represents the composite IO operation to
309 * user-side. So if something waits for IO, then it will
310 * wait for the 'master' bio.
311 */
316 /*
317 * oops, read error - keep the refcount on the rdev
318 */
327 }
328 }
331 {
332 /* clear the bitmap if all writes complete successfully */
338 }
341 {
351 /*
352 * this branch is our 'one mirror IO has finished' event handler:
353 */
359 /*
360 * Set R10BIO_Uptodate in our master bio, so that
361 * we will return a good error code for to the higher
362 * levels even if IO on some other mirrored buffer fails.
363 *
364 * The 'master' represents the composite IO operation to
365 * user-side. So if something waits for IO, then it will
366 * wait for the 'master' bio.
367 */
368 sector_t first_bad;
373 /* Maybe we can clear some bad blocks. */
382 }
383 }
385 /*
386 *
387 * Let's see if all mirrored write operations have finished
388 * already.
389 */
397 else
399 }
400 }
403 }
406 /*
407 * RAID10 layout manager
408 * As well as the chunksize and raid_disks count, there are two
409 * parameters: near_copies and far_copies.
410 * near_copies * far_copies must be <= raid_disks.
411 * Normally one of these will be 1.
412 * If both are 1, we get raid0.
413 * If near_copies == raid_disks, we get raid1.
414 *
415 * Chunks are laid out in raid0 style with near_copies copies of the
416 * first chunk, followed by near_copies copies of the next chunk and
417 * so on.
418 * If far_copies > 1, then after 1/far_copies of the array has been assigned
419 * as described above, we start again with a device offset of near_copies.
420 * So we effectively have another copy of the whole array further down all
421 * the drives, but with blocks on different drives.
422 * With this layout, and block is never stored twice on the one device.
423 *
424 * raid10_find_phys finds the sector offset of a given virtual sector
425 * on each device that it is on.
426 *
427 * raid10_find_virt does the reverse mapping, from a device and a
428 * sector offset to a virtual address
429 */
432 {
434 sector_t sector;
435 sector_t chunk;
436 sector_t stripe;
441 /* now calculate first sector/dev */
453 /* and calculate all the others */
459 slot++;
468 slot++;
469 }
470 dev++;
474 }
475 }
477 }
480 {
496 else
498 }
500 }
504 }
506 /**
507 * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged
508 * @q: request queue
509 * @bvm: properties of new bio
510 * @biovec: the request that could be merged to it.
511 *
512 * Return amount of bytes we can accept at this offset
513 * If near_copies == raid_disk, there are no striping issues,
514 * but in that case, the function isn't called at all.
515 */
519 {
530 else
532 }
534 /*
535 * This routine returns the disk from which the requested read should
536 * be done. There is a per-array 'next expected sequential IO' sector
537 * number - if this matches on the next IO then we use the last disk.
538 * There is also a per-disk 'last know head position' sector that is
539 * maintained from IRQ contexts, both the normal and the resync IO
540 * completion handlers update this position correctly. If there is no
541 * perfect sequential match then we pick the disk whose head is closest.
542 *
543 * If there are 2 mirrors in the same 2 devices, performance degrades
544 * because position is mirror, not device based.
545 *
546 * The rdev for the device selected will have nr_pending incremented.
547 */
549 /*
550 * FIXME: possibly should rethink readbalancing and do it differently
551 * depending on near_copies / far_copies geometry.
552 */
554 {
566 retry:
572 /*
573 * Check if we can balance. We can balance on the whole
574 * device if no resync is going on (recovery is ok), or below
575 * the resync window. We take the first readable disk when
576 * above the resync window.
577 */
583 sector_t first_bad;
585 sector_t dev_sector;
600 /* Already have a better slot */
603 /* Cannot read here. If this is the
604 * 'primary' device, then we must not read
605 * beyond 'bad_sectors' from another device.
606 */
613 sector_t good_sectors =
618 }
620 /* Must read from here */
622 }
630 /* This optimisation is debatable, and completely destroys
631 * sequential read speed for 'far copies' arrays. So only
632 * keep it for 'near' arrays, and review those later.
633 */
637 /* for far > 1 always use the lowest address */
640 else
646 }
647 }
658 /* Cannot risk returning a device that failed
659 * before we inc'ed nr_pending
660 */
663 }
671 }
674 {
688 }
689 }
692 }
695 {
696 /* Any writes that have been queued but are awaiting
697 * bitmap updates get flushed here.
698 */
705 /* flush any pending bitmap writes to disk
706 * before proceeding w/ I/O */
714 }
717 }
719 /* Barriers....
720 * Sometimes we need to suspend IO while we do something else,
721 * either some resync/recovery, or reconfigure the array.
722 * To do this we raise a 'barrier'.
723 * The 'barrier' is a counter that can be raised multiple times
724 * to count how many activities are happening which preclude
725 * normal IO.
726 * We can only raise the barrier if there is no pending IO.
727 * i.e. if nr_pending == 0.
728 * We choose only to raise the barrier if no-one is waiting for the
729 * barrier to go down. This means that as soon as an IO request
730 * is ready, no other operations which require a barrier will start
731 * until the IO request has had a chance.
732 *
733 * So: regular IO calls 'wait_barrier'. When that returns there
734 * is no backgroup IO happening, It must arrange to call
735 * allow_barrier when it has finished its IO.
736 * backgroup IO calls must call raise_barrier. Once that returns
737 * there is no normal IO happeing. It must arrange to call
738 * lower_barrier when the particular background IO completes.
739 */
742 {
746 /* Wait until no block IO is waiting (unless 'force') */
750 /* block any new IO from starting */
753 /* Now wait for all pending IO to complete */
759 }
762 {
768 }
771 {
777 );
779 }
782 }
785 {
791 }
794 {
795 /* stop syncio and normal IO and wait for everything to
796 * go quiet.
797 * We increment barrier and nr_waiting, and then
798 * wait until nr_pending match nr_queued+1
799 * This is called in the context of one normal IO request
800 * that has failed. Thus any sync request that might be pending
801 * will be blocked by nr_pending, and we need to wait for
802 * pending IO requests to complete or be queued for re-try.
803 * Thus the number queued (nr_queued) plus this request (1)
804 * must match the number of pending IOs (nr_pending) before
805 * we continue.
806 */
816 }
819 {
820 /* reverse the effect of the freeze */
826 }
829 {
848 }
850 /* If this request crosses a chunk boundary, we need to
851 * split it. This will only happen for 1 PAGE (or less) requests.
852 */
857 /* Sanity check -- queue functions should prevent this happening */
861 /* This is a one page bio that upper layers
862 * refuse to split for us, so we need to split it.
863 */
867 /* Each of these 'make_request' calls will call 'wait_barrier'.
868 * If the first succeeds but the second blocks due to the resync
869 * thread raising the barrier, we will deadlock because the
870 * IO to the underlying device will be queued in generic_make_request
871 * and will never complete, so will never reduce nr_pending.
872 * So increment nr_waiting here so no new raise_barriers will
873 * succeed, and so the second wait_barrier cannot block.
874 */
891 bad_map:
898 }
902 /*
903 * Register the new request and wait if the reconstruction
904 * thread has put up a bar for new requests.
905 * Continue immediately if no resync is active currently.
906 */
918 /* We might need to issue multiple reads to different
919 * devices if there are bad blocks around, so we keep
920 * track of the number of reads in bio->bi_phys_segments.
921 * If this is 0, there is only one r10_bio and no locking
922 * will be needed when the request completes. If it is
923 * non-zero, then it is the number of not-completed requests.
924 */
929 /*
930 * read balancing logic:
931 */
935 read_again:
941 }
946 max_sectors);
958 /* Could not read all from this device, so we will
959 * need another r10_bio.
960 */
967 else
970 /* Cannot call generic_make_request directly
971 * as that will be queued in __generic_make_request
972 * and subsequent mempool_alloc might block
973 * waiting for it. so hand bio over to raid10d.
974 */
989 }
991 /*
992 * WRITE:
993 */
994 /* first select target devices under rcu_lock and
995 * inc refcount on their rdev. Record them by setting
996 * bios[x] to bio
997 * If there are known/acknowledged bad blocks on any device
998 * on which we have seen a write error, we want to avoid
999 * writing to those blocks. This potentially requires several
1000 * writes to write around the bad blocks. Each set of writes
1001 * gets its own r10_bio with a set of bios attached. The number
1002 * of r10_bios is recored in bio->bi_phys_segments just as with
1003 * the read case.
1004 */
1008 retry_write:
1020 }
1025 }
1027 sector_t first_bad;
1033 max_sectors,
1036 /* Mustn't write here until the bad block
1037 * is acknowledged
1038 */
1043 }
1045 /* Cannot write here at all */
1048 /* Mustn't write more than bad_sectors
1049 * to other devices yet
1050 */
1052 /* We don't set R10BIO_Degraded as that
1053 * only applies if the disk is missing,
1054 * so it might be re-added, and we want to
1055 * know to recover this chunk.
1056 * In this case the device is here, and the
1057 * fact that this chunk is not in-sync is
1058 * recorded in the bad block log.
1059 */
1061 }
1066 }
1067 }
1070 }
1074 /* Have to wait for this device to get unblocked, then retry */
1082 }
1087 }
1090 /* We are splitting this into multiple parts, so
1091 * we need to prepare for allocating another r10_bio.
1092 */
1097 else
1100 }
1114 max_sectors);
1128 }
1131 /* This matches the end of raid10_end_write_request() */
1138 }
1140 /* In case raid10d snuck in to freeze_array */
1144 /* We need another r10_bio. It has already been counted
1145 * in bio->bi_phys_segments.
1146 */
1156 }
1161 }
1164 {
1175 else
1177 }
1185 }
1187 /* check if there are enough drives for
1188 * every block to appear on atleast one.
1189 * Don't consider the device numbered 'ignore'
1190 * as we might be about to remove it.
1191 */
1193 {
1202 cnt++;
1204 }
1209 }
1212 {
1216 /*
1217 * If it is not operational, then we have already marked it as dead
1218 * else if it is the last working disks, ignore the error, let the
1219 * next level up know.
1220 * else mark the drive as failed
1221 */
1224 /*
1225 * Don't fail the drive, just return an IO error.
1226 */
1233 /*
1234 * if recovery is running, make sure it aborts.
1235 */
1237 }
1246 }
1249 {
1257 }
1269 }
1270 }
1273 {
1279 }
1282 {
1289 /*
1290 * Find all non-in_sync disks within the RAID10 configuration
1291 * and mark them in_sync
1292 */
1298 count++;
1300 }
1301 }
1308 }
1312 {
1320 /* only hot-add to in-sync arrays, as recovery is
1321 * very different from resync
1322 */
1333 else
1344 /* as we don't honour merge_bvec_fn, we must
1345 * never risk violating it, so limit
1346 * ->max_segments to one lying with a single
1347 * page, as a one page request is never in
1348 * violation.
1349 */
1354 }
1363 }
1368 }
1371 {
1384 }
1385 /* Only remove faulty devices in recovery
1386 * is not possible.
1387 */
1393 }
1397 /* lost the race, try later */
1401 }
1403 }
1404 abort:
1408 }
1412 {
1421 else
1422 /* The write handler will notice the lack of
1423 * R10BIO_Uptodate and record any errors etc
1424 */
1428 /* for reconstruct, we always reschedule after a read.
1429 * for resync, only after all reads
1430 */
1434 /* we have read all the blocks,
1435 * do the comparison in process context in raid10d
1436 */
1438 }
1439 }
1442 {
1447 /* the primary of several recovery bios */
1452 else
1461 else
1464 }
1465 }
1466 }
1469 {
1475 sector_t first_bad;
1493 }
1495 /*
1496 * Note: sync and recover and handled very differently for raid10
1497 * This code is for resync.
1498 * For resync, we read through virtual addresses and read all blocks.
1499 * If there is any error, we schedule a write. The lowest numbered
1500 * drive is authoritative.
1501 * However requests come for physical address, so we need to map.
1502 * For every physical address there are raid_disks/copies virtual addresses,
1503 * which is always are least one, but is not necessarly an integer.
1504 * This means that a physical address can span multiple chunks, so we may
1505 * have to submit multiple io requests for a single sync request.
1506 */
1507 /*
1508 * We check if all blocks are in-sync and only write to blocks that
1509 * aren't in sync
1510 */
1512 {
1519 /* find the first device with a block */
1530 /* now find blocks with errors */
1542 /* We know that the bi_io_vec layout is the same for
1543 * both 'first' and 'i', so we just compare them.
1544 * All vec entries are PAGE_SIZE;
1545 */
1549 PAGE_SIZE))
1555 /* Don't fix anything. */
1557 }
1558 /* Ok, we need to write this bio, either to correct an
1559 * inconsistency or to correct an unreadable block.
1560 * First we need to fixup bv_offset, bv_len and
1561 * bi_vecs, as the read request might have corrupted these
1562 */
1580 PAGE_SIZE);
1581 }
1592 }
1594 done:
1598 }
1599 }
1601 /*
1602 * Now for the recovery code.
1603 * Recovery happens across physical sectors.
1604 * We recover all non-is_sync drives by finding the virtual address of
1605 * each, and then choose a working drive that also has that virt address.
1606 * There is a separate r10_bio for each non-in_sync drive.
1607 * Only the first two slots are in use. The first for reading,
1608 * The second for writing.
1609 *
1610 */
1612 {
1613 /* We got a read error during recovery.
1614 * We repeat the read in smaller page-sized sections.
1615 * If a read succeeds, write it to the new device or record
1616 * a bad block if we cannot.
1617 * If a read fails, record a bad block on both old and
1618 * new devices.
1619 */
1632 sector_t addr;
1641 addr,
1649 addr,
1655 }
1657 /* We don't worry if we cannot set a bad block -
1658 * it really is bad so there is no loss in not
1659 * recording it yet
1660 */
1664 /* need bad block on destination too */
1669 /* just abort the recovery */
1671 "md/raid10:%s: recovery aborted"
1680 }
1681 }
1682 }
1686 idx++;
1687 }
1688 }
1691 {
1700 }
1702 /*
1703 * share the pages with the first bio
1704 * and submit the write request
1705 */
1712 }
1715 /*
1716 * Used by fix_read_error() to decay the per rdev read_errors.
1717 * We halve the read error count for every hour that has elapsed
1718 * since the last recorded read error.
1719 *
1720 */
1722 {
1731 /* first time we've seen a read error */
1734 }
1741 /*
1742 * if hours_since_last is > the number of bits in read_errors
1743 * just set read errors to 0. We do this to avoid
1744 * overflowing the shift of read_errors by hours_since_last.
1745 */
1748 else
1750 }
1754 {
1755 sector_t first_bad;
1762 /* success */
1766 /* need to record an error - either for the block or the device */
1770 }
1772 /*
1773 * This is a kernel thread which:
1774 *
1775 * 1. Retries failed read operations on working mirrors.
1776 * 2. Updates the raid superblock when problems encounter.
1777 * 3. Performs writes following reads for array synchronising.
1778 */
1781 {
1788 /* still own a reference to this rdev, so it cannot
1789 * have been cleared recently.
1790 */
1794 /* drive has already been failed, just ignore any
1795 more fix_read_error() attempts */
1805 "md/raid10:%s: %s: Raid device exceeded "
1814 }
1827 sector_t first_bad;
1840 sect,
1847 }
1848 sl++;
1855 /* Cannot read from anywhere, just mark the block
1856 * as bad on the first device to discourage future
1857 * reads.
1858 */
1863 rdev,
1869 }
1872 /* write it back and re-read */
1879 sl--;
1890 sect,
1893 /* Well, this device is dead */
1895 "md/raid10:%s: read correction "
1896 "write failed"
1906 }
1909 }
1916 sl--;
1927 sect,
1929 READ)) {
1931 /* Well, this device is dead */
1933 "md/raid10:%s: unable to read back "
1934 "corrected sectors"
1947 "md/raid10:%s: read error corrected"
1954 }
1958 }
1963 }
1964 }
1967 {
1969 }
1972 {
1983 }
1986 {
1991 /* bio has the data to be written to slot 'i' where
1992 * we just recently had a write error.
1993 * We repeatedly clone the bio and trim down to one block,
1994 * then try the write. Where the write fails we record
1995 * a bad block.
1996 * It is conceivable that the bio doesn't exactly align with
1997 * blocks. We must handle this.
1998 *
1999 * We currently own a reference to the rdev.
2000 */
2003 sector_t sector;
2021 /* Write at 'sector' for 'sectors' */
2029 /* Failure! */
2038 }
2040 }
2043 {
2053 /* we got a read error. Maybe the drive is bad. Maybe just
2054 * the block and we can fix it.
2055 * We freeze all other IO, and try reading the block from
2056 * other devices. When we find one, we re-write
2057 * and check it that fixes the read error.
2058 * This is all done synchronously while the array is
2059 * frozen.
2060 */
2065 }
2072 read_more:
2082 }
2090 KERN_ERR
2091 "md/raid10:%s: %s: redirecting"
2100 max_sectors);
2109 /* Drat - have to split this up more */
2118 else
2125 GFP_NOIO);
2139 }
2142 {
2143 /* Some sort of write request has finished and it
2144 * succeeded in writing where we thought there was a
2145 * bad block. So forget the bad block.
2146 * Or possibly if failed and we need to record
2147 * a bad block.
2148 */
2162 rdev,
2167 rdev,
2171 }
2172 }
2181 rdev,
2191 }
2193 }
2194 }
2199 }
2200 }
2203 {
2221 }
2239 /* just a partial read to be scheduled from a
2240 * separate context
2241 */
2244 }
2249 }
2251 }
2255 {
2265 }
2267 /*
2268 * perform a "sync" on one "block"
2269 *
2270 * We need to make sure that no normal I/O request - particularly write
2271 * requests - conflict with active sync requests.
2272 *
2273 * This is achieved by tracking pending requests and a 'barrier' concept
2274 * that can be installed to exclude normal IO requests.
2275 *
2276 * Resync and recovery are handled very differently.
2277 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
2278 *
2279 * For resync, we iterate over virtual addresses, read all copies,
2280 * and update if there are differences. If only one copy is live,
2281 * skip it.
2282 * For recovery, we iterate over physical addresses, read a good
2283 * value for each non-in_sync drive, and over-write.
2284 *
2285 * So, for recovery we may have several outstanding complex requests for a
2286 * given address, one for each out-of-sync device. We model this by allocating
2287 * a number of r10_bio structures, one for each out-of-sync device.
2288 * As we setup these structures, we collect all bio's together into a list
2289 * which we then process collectively to add pages, and then process again
2290 * to pass to generic_make_request.
2291 *
2292 * The r10_bio structures are linked using a borrowed master_bio pointer.
2293 * This link is counted in ->remaining. When the r10_bio that points to NULL
2294 * has its remaining count decremented to 0, the whole complex operation
2295 * is complete.
2296 *
2297 */
2301 {
2308 sector_t sync_blocks;
2316 skipped:
2321 /* If we aborted, we need to abort the
2322 * sync on the 'current' bitmap chucks (there can
2323 * be several when recovering multiple devices).
2324 * as we may have started syncing it but not finished.
2325 * We can find the current address in
2326 * mddev->curr_resync, but for recovery,
2327 * we need to convert that to several
2328 * virtual addresses.
2329 */
2335 sector_t sect =
2339 }
2347 }
2349 /* if there has been nothing to do on any drive,
2350 * then there is nothing to do at all..
2351 */
2354 }
2359 /* make sure whole request will fit in a chunk - if chunks
2360 * are meaningful
2361 */
2365 /*
2366 * If there is non-resync activity waiting for us then
2367 * put in a delay to throttle resync.
2368 */
2372 /* Again, very different code for resync and recovery.
2373 * Both must result in an r10bio with a list of bios that
2374 * have bi_end_io, bi_sector, bi_bdev set,
2375 * and bi_private set to the r10bio.
2376 * For recovery, we may actually create several r10bios
2377 * with 2 bios in each, that correspond to the bios in the main one.
2378 * In this case, the subordinate r10bios link back through a
2379 * borrowed master_bio pointer, and the counter in the master
2380 * includes a ref from each subordinate.
2381 */
2382 /* First, we decide what to do and set ->bi_end_io
2383 * To end_sync_read if we want to read, and
2384 * end_sync_write if we will want to write.
2385 */
2389 /* recovery... the complicated one */
2396 sector_t sect;
2405 /* want to reconstruct this device */
2408 /* Unless we are doing a full sync, we only need
2409 * to recover the block if it is set in the bitmap
2410 */
2417 /* yep, skip the sync_blocks here, but don't assume
2418 * that there will never be anything to do here
2419 */
2422 }
2437 /* Need to check if the array will still be
2438 * degraded
2439 */
2445 }
2461 /* This is where we read from */
2471 bad_sectors -= (sector
2476 }
2477 }
2490 /* and we write to 'i' */
2513 }
2515 /* Cannot recover, so abort the recovery or
2516 * record a bad block */
2522 /* problem is that there are bad blocks
2523 * on other device(s)
2524 */
2534 }
2543 }
2545 }
2546 }
2553 }
2555 }
2557 /* resync. Schedule a read for every block at this virt offset */
2566 /* We can skip this block */
2569 }
2607 }
2608 }
2619 count++;
2620 }
2627 mddev);
2628 }
2632 }
2633 }
2644 }
2662 /* stop here */
2667 /* remove last page from this bio */
2671 }
2673 }
2677 bio_full:
2691 }
2692 }
2695 /* pretend they weren't skipped, it makes
2696 * no important difference in this case
2697 */
2701 giveup:
2702 /* There is nowhere to write, so all non-sync
2703 * drives must be failed or in resync, all drives
2704 * have a bad block, so try the next chunk...
2705 */
2710 chunks_skipped ++;
2713 }
2715 static sector_t
2717 {
2718 sector_t size;
2732 }
2736 {
2748 }
2759 }
2767 GFP_KERNEL);
2793 /* 'size' is now the number of chunks in the array */
2794 /* calculate "used chunks per device" in 'stride' */
2797 /* We need to round up when dividing by raid_disks to
2798 * get the stride size.
2799 */
2807 else
2825 out:
2834 }
2836 }
2839 {
2844 sector_t size;
2846 /*
2847 * copy the already verified devices into our private RAID10
2848 * bookkeeping area. [whatever we allocate in run(),
2849 * should be freed in stop()]
2850 */
2857 }
2869 else
2884 /* as we don't honour merge_bvec_fn, we must never risk
2885 * violating it, so limit max_segments to 1 lying
2886 * within a single page.
2887 */
2892 }
2895 }
2896 /* need to check that every block has at least one working mirror */
2901 }
2914 }
2915 }
2925 /*
2926 * Ok, everything is just fine now
2927 */
2936 /* Calculate max read-ahead size.
2937 * We need to readahead at least twice a whole stripe....
2938 * maybe...
2939 */
2940 {
2946 }
2956 out_free_conf:
2964 out:
2966 }
2969 {
2984 }
2987 {
2997 }
2998 }
3001 {
3009 }
3011 /* Set new parameters */
3013 /* new layout: far_copies = 1, near_copies = 2 */
3018 /* make sure it will be not marked as dirty */
3027 }
3030 }
3033 {
3036 /* raid10 can take over:
3037 * raid0 - providing it has only two drives
3038 */
3040 /* for raid0 takeover only one zone is supported */
3047 }
3049 }
3051 }
3054 {
3070 };
3073 {
3075 }
3078 {
3080 }