| blob | f6873fc8e5ee3103b94ceaf5fd379de682ee50cd |
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 {
349 else
351 }
352 }
353 }
356 {
366 /*
367 * this branch is our 'one mirror IO has finished' event handler:
368 */
374 /*
375 * Set R10BIO_Uptodate in our master bio, so that
376 * we will return a good error code for to the higher
377 * levels even if IO on some other mirrored buffer fails.
378 *
379 * The 'master' represents the composite IO operation to
380 * user-side. So if something waits for IO, then it will
381 * wait for the 'master' bio.
382 */
383 sector_t first_bad;
388 /* Maybe we can clear some bad blocks. */
397 }
398 }
400 /*
401 *
402 * Let's see if all mirrored write operations have finished
403 * already.
404 */
408 }
411 /*
412 * RAID10 layout manager
413 * As well as the chunksize and raid_disks count, there are two
414 * parameters: near_copies and far_copies.
415 * near_copies * far_copies must be <= raid_disks.
416 * Normally one of these will be 1.
417 * If both are 1, we get raid0.
418 * If near_copies == raid_disks, we get raid1.
419 *
420 * Chunks are laid out in raid0 style with near_copies copies of the
421 * first chunk, followed by near_copies copies of the next chunk and
422 * so on.
423 * If far_copies > 1, then after 1/far_copies of the array has been assigned
424 * as described above, we start again with a device offset of near_copies.
425 * So we effectively have another copy of the whole array further down all
426 * the drives, but with blocks on different drives.
427 * With this layout, and block is never stored twice on the one device.
428 *
429 * raid10_find_phys finds the sector offset of a given virtual sector
430 * on each device that it is on.
431 *
432 * raid10_find_virt does the reverse mapping, from a device and a
433 * sector offset to a virtual address
434 */
437 {
439 sector_t sector;
440 sector_t chunk;
441 sector_t stripe;
446 /* now calculate first sector/dev */
458 /* and calculate all the others */
464 slot++;
473 slot++;
474 }
475 dev++;
479 }
480 }
482 }
485 {
501 else
503 }
505 }
509 }
511 /**
512 * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged
513 * @q: request queue
514 * @bvm: properties of new bio
515 * @biovec: the request that could be merged to it.
516 *
517 * Return amount of bytes we can accept at this offset
518 * If near_copies == raid_disk, there are no striping issues,
519 * but in that case, the function isn't called at all.
520 */
524 {
535 else
537 }
539 /*
540 * This routine returns the disk from which the requested read should
541 * be done. There is a per-array 'next expected sequential IO' sector
542 * number - if this matches on the next IO then we use the last disk.
543 * There is also a per-disk 'last know head position' sector that is
544 * maintained from IRQ contexts, both the normal and the resync IO
545 * completion handlers update this position correctly. If there is no
546 * perfect sequential match then we pick the disk whose head is closest.
547 *
548 * If there are 2 mirrors in the same 2 devices, performance degrades
549 * because position is mirror, not device based.
550 *
551 * The rdev for the device selected will have nr_pending incremented.
552 */
554 /*
555 * FIXME: possibly should rethink readbalancing and do it differently
556 * depending on near_copies / far_copies geometry.
557 */
559 {
571 retry:
577 /*
578 * Check if we can balance. We can balance on the whole
579 * device if no resync is going on (recovery is ok), or below
580 * the resync window. We take the first readable disk when
581 * above the resync window.
582 */
588 sector_t first_bad;
590 sector_t dev_sector;
605 /* Already have a better slot */
608 /* Cannot read here. If this is the
609 * 'primary' device, then we must not read
610 * beyond 'bad_sectors' from another device.
611 */
618 sector_t good_sectors =
623 }
625 /* Must read from here */
627 }
635 /* This optimisation is debatable, and completely destroys
636 * sequential read speed for 'far copies' arrays. So only
637 * keep it for 'near' arrays, and review those later.
638 */
642 /* for far > 1 always use the lowest address */
645 else
651 }
652 }
663 /* Cannot risk returning a device that failed
664 * before we inc'ed nr_pending
665 */
668 }
676 }
679 {
693 }
694 }
697 }
700 {
701 /* Any writes that have been queued but are awaiting
702 * bitmap updates get flushed here.
703 */
710 /* flush any pending bitmap writes to disk
711 * before proceeding w/ I/O */
719 }
722 }
724 /* Barriers....
725 * Sometimes we need to suspend IO while we do something else,
726 * either some resync/recovery, or reconfigure the array.
727 * To do this we raise a 'barrier'.
728 * The 'barrier' is a counter that can be raised multiple times
729 * to count how many activities are happening which preclude
730 * normal IO.
731 * We can only raise the barrier if there is no pending IO.
732 * i.e. if nr_pending == 0.
733 * We choose only to raise the barrier if no-one is waiting for the
734 * barrier to go down. This means that as soon as an IO request
735 * is ready, no other operations which require a barrier will start
736 * until the IO request has had a chance.
737 *
738 * So: regular IO calls 'wait_barrier'. When that returns there
739 * is no backgroup IO happening, It must arrange to call
740 * allow_barrier when it has finished its IO.
741 * backgroup IO calls must call raise_barrier. Once that returns
742 * there is no normal IO happeing. It must arrange to call
743 * lower_barrier when the particular background IO completes.
744 */
747 {
751 /* Wait until no block IO is waiting (unless 'force') */
755 /* block any new IO from starting */
758 /* Now wait for all pending IO to complete */
764 }
767 {
773 }
776 {
782 );
784 }
787 }
790 {
796 }
799 {
800 /* stop syncio and normal IO and wait for everything to
801 * go quiet.
802 * We increment barrier and nr_waiting, and then
803 * wait until nr_pending match nr_queued+1
804 * This is called in the context of one normal IO request
805 * that has failed. Thus any sync request that might be pending
806 * will be blocked by nr_pending, and we need to wait for
807 * pending IO requests to complete or be queued for re-try.
808 * Thus the number queued (nr_queued) plus this request (1)
809 * must match the number of pending IOs (nr_pending) before
810 * we continue.
811 */
821 }
824 {
825 /* reverse the effect of the freeze */
831 }
834 {
853 }
855 /* If this request crosses a chunk boundary, we need to
856 * split it. This will only happen for 1 PAGE (or less) requests.
857 */
862 /* Sanity check -- queue functions should prevent this happening */
866 /* This is a one page bio that upper layers
867 * refuse to split for us, so we need to split it.
868 */
872 /* Each of these 'make_request' calls will call 'wait_barrier'.
873 * If the first succeeds but the second blocks due to the resync
874 * thread raising the barrier, we will deadlock because the
875 * IO to the underlying device will be queued in generic_make_request
876 * and will never complete, so will never reduce nr_pending.
877 * So increment nr_waiting here so no new raise_barriers will
878 * succeed, and so the second wait_barrier cannot block.
879 */
896 bad_map:
903 }
907 /*
908 * Register the new request and wait if the reconstruction
909 * thread has put up a bar for new requests.
910 * Continue immediately if no resync is active currently.
911 */
923 /* We might need to issue multiple reads to different
924 * devices if there are bad blocks around, so we keep
925 * track of the number of reads in bio->bi_phys_segments.
926 * If this is 0, there is only one r10_bio and no locking
927 * will be needed when the request completes. If it is
928 * non-zero, then it is the number of not-completed requests.
929 */
934 /*
935 * read balancing logic:
936 */
940 read_again:
946 }
951 max_sectors);
963 /* Could not read all from this device, so we will
964 * need another r10_bio.
965 */
972 else
975 /* Cannot call generic_make_request directly
976 * as that will be queued in __generic_make_request
977 * and subsequent mempool_alloc might block
978 * waiting for it. so hand bio over to raid10d.
979 */
994 }
996 /*
997 * WRITE:
998 */
999 /* first select target devices under rcu_lock and
1000 * inc refcount on their rdev. Record them by setting
1001 * bios[x] to bio
1002 * If there are known/acknowledged bad blocks on any device
1003 * on which we have seen a write error, we want to avoid
1004 * writing to those blocks. This potentially requires several
1005 * writes to write around the bad blocks. Each set of writes
1006 * gets its own r10_bio with a set of bios attached. The number
1007 * of r10_bios is recored in bio->bi_phys_segments just as with
1008 * the read case.
1009 */
1013 retry_write:
1025 }
1030 }
1032 sector_t first_bad;
1038 max_sectors,
1041 /* Mustn't write here until the bad block
1042 * is acknowledged
1043 */
1048 }
1050 /* Cannot write here at all */
1053 /* Mustn't write more than bad_sectors
1054 * to other devices yet
1055 */
1057 /* We don't set R10BIO_Degraded as that
1058 * only applies if the disk is missing,
1059 * so it might be re-added, and we want to
1060 * know to recover this chunk.
1061 * In this case the device is here, and the
1062 * fact that this chunk is not in-sync is
1063 * recorded in the bad block log.
1064 */
1066 }
1071 }
1072 }
1075 }
1079 /* Have to wait for this device to get unblocked, then retry */
1087 }
1092 }
1095 /* We are splitting this into multiple parts, so
1096 * we need to prepare for allocating another r10_bio.
1097 */
1102 else
1105 }
1119 max_sectors);
1133 }
1135 /* Remove the bias on 'remaining' */
1138 /* In case raid10d snuck in to freeze_array */
1142 /* We need another r10_bio. It has already been counted
1143 * in bio->bi_phys_segments.
1144 */
1154 }
1159 }
1162 {
1173 else
1175 }
1183 }
1185 /* check if there are enough drives for
1186 * every block to appear on atleast one.
1187 * Don't consider the device numbered 'ignore'
1188 * as we might be about to remove it.
1189 */
1191 {
1200 cnt++;
1202 }
1207 }
1210 {
1214 /*
1215 * If it is not operational, then we have already marked it as dead
1216 * else if it is the last working disks, ignore the error, let the
1217 * next level up know.
1218 * else mark the drive as failed
1219 */
1222 /*
1223 * Don't fail the drive, just return an IO error.
1224 */
1231 /*
1232 * if recovery is running, make sure it aborts.
1233 */
1235 }
1244 }
1247 {
1255 }
1267 }
1268 }
1271 {
1277 }
1280 {
1287 /*
1288 * Find all non-in_sync disks within the RAID10 configuration
1289 * and mark them in_sync
1290 */
1296 count++;
1298 }
1299 }
1306 }
1310 {
1318 /* only hot-add to in-sync arrays, as recovery is
1319 * very different from resync
1320 */
1331 else
1342 /* as we don't honour merge_bvec_fn, we must
1343 * never risk violating it, so limit
1344 * ->max_segments to one lying with a single
1345 * page, as a one page request is never in
1346 * violation.
1347 */
1352 }
1361 }
1366 }
1369 {
1382 }
1383 /* Only remove faulty devices in recovery
1384 * is not possible.
1385 */
1391 }
1395 /* lost the race, try later */
1399 }
1401 }
1402 abort:
1406 }
1410 {
1419 else
1420 /* The write handler will notice the lack of
1421 * R10BIO_Uptodate and record any errors etc
1422 */
1426 /* for reconstruct, we always reschedule after a read.
1427 * for resync, only after all reads
1428 */
1432 /* we have read all the blocks,
1433 * do the comparison in process context in raid10d
1434 */
1436 }
1437 }
1440 {
1445 /* the primary of several recovery bios */
1450 else
1459 else
1462 }
1463 }
1464 }
1467 {
1473 sector_t first_bad;
1491 }
1493 /*
1494 * Note: sync and recover and handled very differently for raid10
1495 * This code is for resync.
1496 * For resync, we read through virtual addresses and read all blocks.
1497 * If there is any error, we schedule a write. The lowest numbered
1498 * drive is authoritative.
1499 * However requests come for physical address, so we need to map.
1500 * For every physical address there are raid_disks/copies virtual addresses,
1501 * which is always are least one, but is not necessarly an integer.
1502 * This means that a physical address can span multiple chunks, so we may
1503 * have to submit multiple io requests for a single sync request.
1504 */
1505 /*
1506 * We check if all blocks are in-sync and only write to blocks that
1507 * aren't in sync
1508 */
1510 {
1517 /* find the first device with a block */
1528 /* now find blocks with errors */
1540 /* We know that the bi_io_vec layout is the same for
1541 * both 'first' and 'i', so we just compare them.
1542 * All vec entries are PAGE_SIZE;
1543 */
1547 PAGE_SIZE))
1553 /* Don't fix anything. */
1555 }
1556 /* Ok, we need to write this bio, either to correct an
1557 * inconsistency or to correct an unreadable block.
1558 * First we need to fixup bv_offset, bv_len and
1559 * bi_vecs, as the read request might have corrupted these
1560 */
1578 PAGE_SIZE);
1579 }
1590 }
1592 done:
1596 }
1597 }
1599 /*
1600 * Now for the recovery code.
1601 * Recovery happens across physical sectors.
1602 * We recover all non-is_sync drives by finding the virtual address of
1603 * each, and then choose a working drive that also has that virt address.
1604 * There is a separate r10_bio for each non-in_sync drive.
1605 * Only the first two slots are in use. The first for reading,
1606 * The second for writing.
1607 *
1608 */
1610 {
1611 /* We got a read error during recovery.
1612 * We repeat the read in smaller page-sized sections.
1613 * If a read succeeds, write it to the new device or record
1614 * a bad block if we cannot.
1615 * If a read fails, record a bad block on both old and
1616 * new devices.
1617 */
1630 sector_t addr;
1639 addr,
1647 addr,
1653 }
1655 /* We don't worry if we cannot set a bad block -
1656 * it really is bad so there is no loss in not
1657 * recording it yet
1658 */
1662 /* need bad block on destination too */
1667 /* just abort the recovery */
1669 "md/raid10:%s: recovery aborted"
1678 }
1679 }
1680 }
1684 idx++;
1685 }
1686 }
1689 {
1698 }
1700 /*
1701 * share the pages with the first bio
1702 * and submit the write request
1703 */
1710 }
1713 /*
1714 * Used by fix_read_error() to decay the per rdev read_errors.
1715 * We halve the read error count for every hour that has elapsed
1716 * since the last recorded read error.
1717 *
1718 */
1720 {
1729 /* first time we've seen a read error */
1732 }
1739 /*
1740 * if hours_since_last is > the number of bits in read_errors
1741 * just set read errors to 0. We do this to avoid
1742 * overflowing the shift of read_errors by hours_since_last.
1743 */
1746 else
1748 }
1752 {
1753 sector_t first_bad;
1760 /* success */
1764 /* need to record an error - either for the block or the device */
1768 }
1770 /*
1771 * This is a kernel thread which:
1772 *
1773 * 1. Retries failed read operations on working mirrors.
1774 * 2. Updates the raid superblock when problems encounter.
1775 * 3. Performs writes following reads for array synchronising.
1776 */
1779 {
1786 /* still own a reference to this rdev, so it cannot
1787 * have been cleared recently.
1788 */
1792 /* drive has already been failed, just ignore any
1793 more fix_read_error() attempts */
1803 "md/raid10:%s: %s: Raid device exceeded "
1812 }
1825 sector_t first_bad;
1838 sect,
1845 }
1846 sl++;
1853 /* Cannot read from anywhere, just mark the block
1854 * as bad on the first device to discourage future
1855 * reads.
1856 */
1861 rdev,
1867 }
1870 /* write it back and re-read */
1877 sl--;
1888 sect,
1891 /* Well, this device is dead */
1893 "md/raid10:%s: read correction "
1894 "write failed"
1904 }
1907 }
1914 sl--;
1925 sect,
1927 READ)) {
1929 /* Well, this device is dead */
1931 "md/raid10:%s: unable to read back "
1932 "corrected sectors"
1945 "md/raid10:%s: read error corrected"
1952 }
1956 }
1961 }
1962 }
1965 {
1967 }
1970 {
1981 }
1984 {
1989 /* bio has the data to be written to slot 'i' where
1990 * we just recently had a write error.
1991 * We repeatedly clone the bio and trim down to one block,
1992 * then try the write. Where the write fails we record
1993 * a bad block.
1994 * It is conceivable that the bio doesn't exactly align with
1995 * blocks. We must handle this.
1996 *
1997 * We currently own a reference to the rdev.
1998 */
2001 sector_t sector;
2019 /* Write at 'sector' for 'sectors' */
2027 /* Failure! */
2036 }
2038 }
2041 {
2051 /* we got a read error. Maybe the drive is bad. Maybe just
2052 * the block and we can fix it.
2053 * We freeze all other IO, and try reading the block from
2054 * other devices. When we find one, we re-write
2055 * and check it that fixes the read error.
2056 * This is all done synchronously while the array is
2057 * frozen.
2058 */
2063 }
2070 read_more:
2080 }
2088 KERN_ERR
2089 "md/raid10:%s: %s: redirecting"
2098 max_sectors);
2107 /* Drat - have to split this up more */
2116 else
2123 GFP_NOIO);
2137 }
2140 {
2141 /* Some sort of write request has finished and it
2142 * succeeded in writing where we thought there was a
2143 * bad block. So forget the bad block.
2144 * Or possibly if failed and we need to record
2145 * a bad block.
2146 */
2160 rdev,
2165 rdev,
2169 }
2170 }
2179 rdev,
2189 }
2191 }
2192 }
2197 }
2198 }
2201 {
2219 }
2237 /* just a partial read to be scheduled from a
2238 * separate context
2239 */
2242 }
2247 }
2249 }
2253 {
2263 }
2265 /*
2266 * perform a "sync" on one "block"
2267 *
2268 * We need to make sure that no normal I/O request - particularly write
2269 * requests - conflict with active sync requests.
2270 *
2271 * This is achieved by tracking pending requests and a 'barrier' concept
2272 * that can be installed to exclude normal IO requests.
2273 *
2274 * Resync and recovery are handled very differently.
2275 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
2276 *
2277 * For resync, we iterate over virtual addresses, read all copies,
2278 * and update if there are differences. If only one copy is live,
2279 * skip it.
2280 * For recovery, we iterate over physical addresses, read a good
2281 * value for each non-in_sync drive, and over-write.
2282 *
2283 * So, for recovery we may have several outstanding complex requests for a
2284 * given address, one for each out-of-sync device. We model this by allocating
2285 * a number of r10_bio structures, one for each out-of-sync device.
2286 * As we setup these structures, we collect all bio's together into a list
2287 * which we then process collectively to add pages, and then process again
2288 * to pass to generic_make_request.
2289 *
2290 * The r10_bio structures are linked using a borrowed master_bio pointer.
2291 * This link is counted in ->remaining. When the r10_bio that points to NULL
2292 * has its remaining count decremented to 0, the whole complex operation
2293 * is complete.
2294 *
2295 */
2299 {
2306 sector_t sync_blocks;
2314 skipped:
2319 /* If we aborted, we need to abort the
2320 * sync on the 'current' bitmap chucks (there can
2321 * be several when recovering multiple devices).
2322 * as we may have started syncing it but not finished.
2323 * We can find the current address in
2324 * mddev->curr_resync, but for recovery,
2325 * we need to convert that to several
2326 * virtual addresses.
2327 */
2333 sector_t sect =
2337 }
2345 }
2347 /* if there has been nothing to do on any drive,
2348 * then there is nothing to do at all..
2349 */
2352 }
2357 /* make sure whole request will fit in a chunk - if chunks
2358 * are meaningful
2359 */
2363 /*
2364 * If there is non-resync activity waiting for us then
2365 * put in a delay to throttle resync.
2366 */
2370 /* Again, very different code for resync and recovery.
2371 * Both must result in an r10bio with a list of bios that
2372 * have bi_end_io, bi_sector, bi_bdev set,
2373 * and bi_private set to the r10bio.
2374 * For recovery, we may actually create several r10bios
2375 * with 2 bios in each, that correspond to the bios in the main one.
2376 * In this case, the subordinate r10bios link back through a
2377 * borrowed master_bio pointer, and the counter in the master
2378 * includes a ref from each subordinate.
2379 */
2380 /* First, we decide what to do and set ->bi_end_io
2381 * To end_sync_read if we want to read, and
2382 * end_sync_write if we will want to write.
2383 */
2387 /* recovery... the complicated one */
2394 sector_t sect;
2403 /* want to reconstruct this device */
2406 /* Unless we are doing a full sync, we only need
2407 * to recover the block if it is set in the bitmap
2408 */
2415 /* yep, skip the sync_blocks here, but don't assume
2416 * that there will never be anything to do here
2417 */
2420 }
2435 /* Need to check if the array will still be
2436 * degraded
2437 */
2443 }
2459 /* This is where we read from */
2469 bad_sectors -= (sector
2474 }
2475 }
2488 /* and we write to 'i' */
2511 }
2513 /* Cannot recover, so abort the recovery or
2514 * record a bad block */
2520 /* problem is that there are bad blocks
2521 * on other device(s)
2522 */
2532 }
2541 }
2543 }
2544 }
2551 }
2553 }
2555 /* resync. Schedule a read for every block at this virt offset */
2564 /* We can skip this block */
2567 }
2605 }
2606 }
2617 count++;
2618 }
2625 mddev);
2626 }
2630 }
2631 }
2642 }
2660 /* stop here */
2665 /* remove last page from this bio */
2669 }
2671 }
2675 bio_full:
2689 }
2690 }
2693 /* pretend they weren't skipped, it makes
2694 * no important difference in this case
2695 */
2699 giveup:
2700 /* There is nowhere to write, so all non-sync
2701 * drives must be failed or in resync, all drives
2702 * have a bad block, so try the next chunk...
2703 */
2708 chunks_skipped ++;
2711 }
2713 static sector_t
2715 {
2716 sector_t size;
2730 }
2734 {
2746 }
2757 }
2765 GFP_KERNEL);
2791 /* 'size' is now the number of chunks in the array */
2792 /* calculate "used chunks per device" in 'stride' */
2795 /* We need to round up when dividing by raid_disks to
2796 * get the stride size.
2797 */
2805 else
2823 out:
2832 }
2834 }
2837 {
2842 sector_t size;
2844 /*
2845 * copy the already verified devices into our private RAID10
2846 * bookkeeping area. [whatever we allocate in run(),
2847 * should be freed in stop()]
2848 */
2855 }
2867 else
2882 /* as we don't honour merge_bvec_fn, we must never risk
2883 * violating it, so limit max_segments to 1 lying
2884 * within a single page.
2885 */
2890 }
2893 }
2894 /* need to check that every block has at least one working mirror */
2899 }
2912 }
2913 }
2923 /*
2924 * Ok, everything is just fine now
2925 */
2934 /* Calculate max read-ahead size.
2935 * We need to readahead at least twice a whole stripe....
2936 * maybe...
2937 */
2938 {
2944 }
2954 out_free_conf:
2962 out:
2964 }
2967 {
2982 }
2985 {
2995 }
2996 }
2999 {
3007 }
3009 /* Set new parameters */
3011 /* new layout: far_copies = 1, near_copies = 2 */
3016 /* make sure it will be not marked as dirty */
3025 }
3028 }
3031 {
3034 /* raid10 can take over:
3035 * raid0 - providing it has only two drives
3036 */
3038 /* for raid0 takeover only one zone is supported */
3045 }
3047 }
3049 }
3052 {
3068 };
3071 {
3073 }
3076 {
3078 }