| blob | d2fd2122410e029cd04732694b0de0310ef1a4c3 |
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
66 {
70 /* allocate a r10bio with room for raid_disks entries in the bios array */
72 }
75 {
77 }
79 /* Maximum size of each resync request */
80 #define RESYNC_BLOCK_SIZE (64*1024)
81 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
82 /* amount of memory to reserve for resync requests */
83 #define RESYNC_WINDOW (1024*1024)
84 /* maximum number of concurrent requests, memory permitting */
85 #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE)
87 /*
88 * When performing a resync, we need to read and compare, so
89 * we need as many pages are there are copies.
90 * When performing a recovery, we need 2 bios, one for read,
91 * one for write (we recover only one drive per r10buf)
92 *
93 */
95 {
109 else
112 /*
113 * Allocate bios.
114 */
120 }
121 /*
122 * Allocate RESYNC_PAGES data pages and attach them
123 * where needed.
124 */
130 /* we can share bv_page's during recovery */
140 }
141 }
145 out_free_pages:
152 out_free_bio:
157 }
160 {
172 }
174 }
175 }
177 }
180 {
188 }
189 }
192 {
197 }
200 {
206 }
209 {
219 /* wake up frozen array... */
223 }
225 /*
226 * raid_end_bio_io() is called when we have finished servicing a mirrored
227 * operation and are ready to return a success/failure code to the buffer
228 * cache layer.
229 */
231 {
248 /*
249 * Wake up any possible resync thread that waits for the device
250 * to go idle.
251 */
253 }
255 }
257 /*
258 * Update disk head position estimator based on IRQ completion info.
259 */
261 {
266 }
268 /*
269 * Find the disk number which triggered given bio
270 */
273 {
286 }
289 {
298 /*
299 * this branch is our 'one mirror IO has finished' event handler:
300 */
304 /*
305 * Set R10BIO_Uptodate in our master bio, so that
306 * we will return a good error code to the higher
307 * levels even if IO on some other mirrored buffer fails.
308 *
309 * The 'master' represents the composite IO operation to
310 * user-side. So if something waits for IO, then it will
311 * wait for the 'master' bio.
312 */
317 /*
318 * oops, read error - keep the refcount on the rdev
319 */
328 }
329 }
332 {
333 /* clear the bitmap if all writes complete successfully */
339 }
342 {
352 /*
353 * this branch is our 'one mirror IO has finished' event handler:
354 */
360 /*
361 * Set R10BIO_Uptodate in our master bio, so that
362 * we will return a good error code for to the higher
363 * levels even if IO on some other mirrored buffer fails.
364 *
365 * The 'master' represents the composite IO operation to
366 * user-side. So if something waits for IO, then it will
367 * wait for the 'master' bio.
368 */
369 sector_t first_bad;
374 /* Maybe we can clear some bad blocks. */
383 }
384 }
386 /*
387 *
388 * Let's see if all mirrored write operations have finished
389 * already.
390 */
398 else
400 }
401 }
404 }
407 /*
408 * RAID10 layout manager
409 * As well as the chunksize and raid_disks count, there are two
410 * parameters: near_copies and far_copies.
411 * near_copies * far_copies must be <= raid_disks.
412 * Normally one of these will be 1.
413 * If both are 1, we get raid0.
414 * If near_copies == raid_disks, we get raid1.
415 *
416 * Chunks are laid out in raid0 style with near_copies copies of the
417 * first chunk, followed by near_copies copies of the next chunk and
418 * so on.
419 * If far_copies > 1, then after 1/far_copies of the array has been assigned
420 * as described above, we start again with a device offset of near_copies.
421 * So we effectively have another copy of the whole array further down all
422 * the drives, but with blocks on different drives.
423 * With this layout, and block is never stored twice on the one device.
424 *
425 * raid10_find_phys finds the sector offset of a given virtual sector
426 * on each device that it is on.
427 *
428 * raid10_find_virt does the reverse mapping, from a device and a
429 * sector offset to a virtual address
430 */
433 {
435 sector_t sector;
436 sector_t chunk;
437 sector_t stripe;
442 /* now calculate first sector/dev */
454 /* and calculate all the others */
460 slot++;
469 slot++;
470 }
471 dev++;
475 }
476 }
478 }
481 {
497 else
499 }
501 }
505 }
507 /**
508 * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged
509 * @q: request queue
510 * @bvm: properties of new bio
511 * @biovec: the request that could be merged to it.
512 *
513 * Return amount of bytes we can accept at this offset
514 * If near_copies == raid_disk, there are no striping issues,
515 * but in that case, the function isn't called at all.
516 */
520 {
531 else
533 }
535 /*
536 * This routine returns the disk from which the requested read should
537 * be done. There is a per-array 'next expected sequential IO' sector
538 * number - if this matches on the next IO then we use the last disk.
539 * There is also a per-disk 'last know head position' sector that is
540 * maintained from IRQ contexts, both the normal and the resync IO
541 * completion handlers update this position correctly. If there is no
542 * perfect sequential match then we pick the disk whose head is closest.
543 *
544 * If there are 2 mirrors in the same 2 devices, performance degrades
545 * because position is mirror, not device based.
546 *
547 * The rdev for the device selected will have nr_pending incremented.
548 */
550 /*
551 * FIXME: possibly should rethink readbalancing and do it differently
552 * depending on near_copies / far_copies geometry.
553 */
555 {
567 retry:
573 /*
574 * Check if we can balance. We can balance on the whole
575 * device if no resync is going on (recovery is ok), or below
576 * the resync window. We take the first readable disk when
577 * above the resync window.
578 */
584 sector_t first_bad;
586 sector_t dev_sector;
601 /* Already have a better slot */
604 /* Cannot read here. If this is the
605 * 'primary' device, then we must not read
606 * beyond 'bad_sectors' from another device.
607 */
614 sector_t good_sectors =
619 }
621 /* Must read from here */
623 }
631 /* This optimisation is debatable, and completely destroys
632 * sequential read speed for 'far copies' arrays. So only
633 * keep it for 'near' arrays, and review those later.
634 */
638 /* for far > 1 always use the lowest address */
641 else
647 }
648 }
659 /* Cannot risk returning a device that failed
660 * before we inc'ed nr_pending
661 */
664 }
672 }
675 {
689 }
690 }
693 }
696 {
697 /* Any writes that have been queued but are awaiting
698 * bitmap updates get flushed here.
699 */
706 /* flush any pending bitmap writes to disk
707 * before proceeding w/ I/O */
715 }
718 }
720 /* Barriers....
721 * Sometimes we need to suspend IO while we do something else,
722 * either some resync/recovery, or reconfigure the array.
723 * To do this we raise a 'barrier'.
724 * The 'barrier' is a counter that can be raised multiple times
725 * to count how many activities are happening which preclude
726 * normal IO.
727 * We can only raise the barrier if there is no pending IO.
728 * i.e. if nr_pending == 0.
729 * We choose only to raise the barrier if no-one is waiting for the
730 * barrier to go down. This means that as soon as an IO request
731 * is ready, no other operations which require a barrier will start
732 * until the IO request has had a chance.
733 *
734 * So: regular IO calls 'wait_barrier'. When that returns there
735 * is no backgroup IO happening, It must arrange to call
736 * allow_barrier when it has finished its IO.
737 * backgroup IO calls must call raise_barrier. Once that returns
738 * there is no normal IO happeing. It must arrange to call
739 * lower_barrier when the particular background IO completes.
740 */
743 {
747 /* Wait until no block IO is waiting (unless 'force') */
751 /* block any new IO from starting */
754 /* Now wait for all pending IO to complete */
760 }
763 {
769 }
772 {
778 );
780 }
783 }
786 {
792 }
795 {
796 /* stop syncio and normal IO and wait for everything to
797 * go quiet.
798 * We increment barrier and nr_waiting, and then
799 * wait until nr_pending match nr_queued+1
800 * This is called in the context of one normal IO request
801 * that has failed. Thus any sync request that might be pending
802 * will be blocked by nr_pending, and we need to wait for
803 * pending IO requests to complete or be queued for re-try.
804 * Thus the number queued (nr_queued) plus this request (1)
805 * must match the number of pending IOs (nr_pending) before
806 * we continue.
807 */
817 }
820 {
821 /* reverse the effect of the freeze */
827 }
830 {
849 }
851 /* If this request crosses a chunk boundary, we need to
852 * split it. This will only happen for 1 PAGE (or less) requests.
853 */
858 /* Sanity check -- queue functions should prevent this happening */
862 /* This is a one page bio that upper layers
863 * refuse to split for us, so we need to split it.
864 */
868 /* Each of these 'make_request' calls will call 'wait_barrier'.
869 * If the first succeeds but the second blocks due to the resync
870 * thread raising the barrier, we will deadlock because the
871 * IO to the underlying device will be queued in generic_make_request
872 * and will never complete, so will never reduce nr_pending.
873 * So increment nr_waiting here so no new raise_barriers will
874 * succeed, and so the second wait_barrier cannot block.
875 */
892 bad_map:
899 }
903 /*
904 * Register the new request and wait if the reconstruction
905 * thread has put up a bar for new requests.
906 * Continue immediately if no resync is active currently.
907 */
919 /* We might need to issue multiple reads to different
920 * devices if there are bad blocks around, so we keep
921 * track of the number of reads in bio->bi_phys_segments.
922 * If this is 0, there is only one r10_bio and no locking
923 * will be needed when the request completes. If it is
924 * non-zero, then it is the number of not-completed requests.
925 */
930 /*
931 * read balancing logic:
932 */
936 read_again:
942 }
947 max_sectors);
959 /* Could not read all from this device, so we will
960 * need another r10_bio.
961 */
968 else
971 /* Cannot call generic_make_request directly
972 * as that will be queued in __generic_make_request
973 * and subsequent mempool_alloc might block
974 * waiting for it. so hand bio over to raid10d.
975 */
990 }
992 /*
993 * WRITE:
994 */
995 /* first select target devices under rcu_lock and
996 * inc refcount on their rdev. Record them by setting
997 * bios[x] to bio
998 * If there are known/acknowledged bad blocks on any device
999 * on which we have seen a write error, we want to avoid
1000 * writing to those blocks. This potentially requires several
1001 * writes to write around the bad blocks. Each set of writes
1002 * gets its own r10_bio with a set of bios attached. The number
1003 * of r10_bios is recored in bio->bi_phys_segments just as with
1004 * the read case.
1005 */
1009 retry_write:
1021 }
1026 }
1028 sector_t first_bad;
1034 max_sectors,
1037 /* Mustn't write here until the bad block
1038 * is acknowledged
1039 */
1044 }
1046 /* Cannot write here at all */
1049 /* Mustn't write more than bad_sectors
1050 * to other devices yet
1051 */
1053 /* We don't set R10BIO_Degraded as that
1054 * only applies if the disk is missing,
1055 * so it might be re-added, and we want to
1056 * know to recover this chunk.
1057 * In this case the device is here, and the
1058 * fact that this chunk is not in-sync is
1059 * recorded in the bad block log.
1060 */
1062 }
1067 }
1068 }
1071 }
1075 /* Have to wait for this device to get unblocked, then retry */
1083 }
1088 }
1091 /* We are splitting this into multiple parts, so
1092 * we need to prepare for allocating another r10_bio.
1093 */
1098 else
1101 }
1115 max_sectors);
1129 }
1132 /* This matches the end of raid10_end_write_request() */
1139 }
1141 /* In case raid10d snuck in to freeze_array */
1145 /* We need another r10_bio. It has already been counted
1146 * in bio->bi_phys_segments.
1147 */
1157 }
1162 }
1165 {
1176 else
1178 }
1186 }
1188 /* check if there are enough drives for
1189 * every block to appear on atleast one.
1190 * Don't consider the device numbered 'ignore'
1191 * as we might be about to remove it.
1192 */
1194 {
1203 cnt++;
1205 }
1210 }
1213 {
1217 /*
1218 * If it is not operational, then we have already marked it as dead
1219 * else if it is the last working disks, ignore the error, let the
1220 * next level up know.
1221 * else mark the drive as failed
1222 */
1225 /*
1226 * Don't fail the drive, just return an IO error.
1227 */
1234 /*
1235 * if recovery is running, make sure it aborts.
1236 */
1238 }
1247 }
1250 {
1258 }
1270 }
1271 }
1274 {
1280 }
1283 {
1290 /*
1291 * Find all non-in_sync disks within the RAID10 configuration
1292 * and mark them in_sync
1293 */
1299 count++;
1301 }
1302 }
1309 }
1313 {
1321 /* only hot-add to in-sync arrays, as recovery is
1322 * very different from resync
1323 */
1334 else
1345 /* as we don't honour merge_bvec_fn, we must
1346 * never risk violating it, so limit
1347 * ->max_segments to one lying with a single
1348 * page, as a one page request is never in
1349 * violation.
1350 */
1355 }
1364 }
1369 }
1372 {
1385 }
1386 /* Only remove faulty devices in recovery
1387 * is not possible.
1388 */
1394 }
1398 /* lost the race, try later */
1402 }
1404 }
1405 abort:
1409 }
1413 {
1422 else
1423 /* The write handler will notice the lack of
1424 * R10BIO_Uptodate and record any errors etc
1425 */
1429 /* for reconstruct, we always reschedule after a read.
1430 * for resync, only after all reads
1431 */
1435 /* we have read all the blocks,
1436 * do the comparison in process context in raid10d
1437 */
1439 }
1440 }
1443 {
1448 /* the primary of several recovery bios */
1453 else
1462 else
1465 }
1466 }
1467 }
1470 {
1476 sector_t first_bad;
1494 }
1496 /*
1497 * Note: sync and recover and handled very differently for raid10
1498 * This code is for resync.
1499 * For resync, we read through virtual addresses and read all blocks.
1500 * If there is any error, we schedule a write. The lowest numbered
1501 * drive is authoritative.
1502 * However requests come for physical address, so we need to map.
1503 * For every physical address there are raid_disks/copies virtual addresses,
1504 * which is always are least one, but is not necessarly an integer.
1505 * This means that a physical address can span multiple chunks, so we may
1506 * have to submit multiple io requests for a single sync request.
1507 */
1508 /*
1509 * We check if all blocks are in-sync and only write to blocks that
1510 * aren't in sync
1511 */
1513 {
1520 /* find the first device with a block */
1531 /* now find blocks with errors */
1543 /* We know that the bi_io_vec layout is the same for
1544 * both 'first' and 'i', so we just compare them.
1545 * All vec entries are PAGE_SIZE;
1546 */
1550 PAGE_SIZE))
1556 /* Don't fix anything. */
1558 }
1559 /* Ok, we need to write this bio, either to correct an
1560 * inconsistency or to correct an unreadable block.
1561 * First we need to fixup bv_offset, bv_len and
1562 * bi_vecs, as the read request might have corrupted these
1563 */
1581 PAGE_SIZE);
1582 }
1593 }
1595 done:
1599 }
1600 }
1602 /*
1603 * Now for the recovery code.
1604 * Recovery happens across physical sectors.
1605 * We recover all non-is_sync drives by finding the virtual address of
1606 * each, and then choose a working drive that also has that virt address.
1607 * There is a separate r10_bio for each non-in_sync drive.
1608 * Only the first two slots are in use. The first for reading,
1609 * The second for writing.
1610 *
1611 */
1613 {
1614 /* We got a read error during recovery.
1615 * We repeat the read in smaller page-sized sections.
1616 * If a read succeeds, write it to the new device or record
1617 * a bad block if we cannot.
1618 * If a read fails, record a bad block on both old and
1619 * new devices.
1620 */
1633 sector_t addr;
1642 addr,
1650 addr,
1656 }
1658 /* We don't worry if we cannot set a bad block -
1659 * it really is bad so there is no loss in not
1660 * recording it yet
1661 */
1665 /* need bad block on destination too */
1670 /* just abort the recovery */
1672 "md/raid10:%s: recovery aborted"
1681 }
1682 }
1683 }
1687 idx++;
1688 }
1689 }
1692 {
1701 }
1703 /*
1704 * share the pages with the first bio
1705 * and submit the write request
1706 */
1713 }
1716 /*
1717 * Used by fix_read_error() to decay the per rdev read_errors.
1718 * We halve the read error count for every hour that has elapsed
1719 * since the last recorded read error.
1720 *
1721 */
1723 {
1732 /* first time we've seen a read error */
1735 }
1742 /*
1743 * if hours_since_last is > the number of bits in read_errors
1744 * just set read errors to 0. We do this to avoid
1745 * overflowing the shift of read_errors by hours_since_last.
1746 */
1749 else
1751 }
1755 {
1756 sector_t first_bad;
1763 /* success */
1767 /* need to record an error - either for the block or the device */
1771 }
1773 /*
1774 * This is a kernel thread which:
1775 *
1776 * 1. Retries failed read operations on working mirrors.
1777 * 2. Updates the raid superblock when problems encounter.
1778 * 3. Performs writes following reads for array synchronising.
1779 */
1782 {
1789 /* still own a reference to this rdev, so it cannot
1790 * have been cleared recently.
1791 */
1795 /* drive has already been failed, just ignore any
1796 more fix_read_error() attempts */
1806 "md/raid10:%s: %s: Raid device exceeded "
1815 }
1828 sector_t first_bad;
1841 sect,
1848 }
1849 sl++;
1856 /* Cannot read from anywhere, just mark the block
1857 * as bad on the first device to discourage future
1858 * reads.
1859 */
1864 rdev,
1870 }
1873 /* write it back and re-read */
1880 sl--;
1891 sect,
1894 /* Well, this device is dead */
1896 "md/raid10:%s: read correction "
1897 "write failed"
1907 }
1910 }
1917 sl--;
1928 sect,
1930 READ)) {
1932 /* Well, this device is dead */
1934 "md/raid10:%s: unable to read back "
1935 "corrected sectors"
1948 "md/raid10:%s: read error corrected"
1955 }
1959 }
1964 }
1965 }
1968 {
1970 }
1973 {
1984 }
1987 {
1992 /* bio has the data to be written to slot 'i' where
1993 * we just recently had a write error.
1994 * We repeatedly clone the bio and trim down to one block,
1995 * then try the write. Where the write fails we record
1996 * a bad block.
1997 * It is conceivable that the bio doesn't exactly align with
1998 * blocks. We must handle this.
1999 *
2000 * We currently own a reference to the rdev.
2001 */
2004 sector_t sector;
2022 /* Write at 'sector' for 'sectors' */
2030 /* Failure! */
2039 }
2041 }
2044 {
2054 /* we got a read error. Maybe the drive is bad. Maybe just
2055 * the block and we can fix it.
2056 * We freeze all other IO, and try reading the block from
2057 * other devices. When we find one, we re-write
2058 * and check it that fixes the read error.
2059 * This is all done synchronously while the array is
2060 * frozen.
2061 */
2066 }
2073 read_more:
2083 }
2091 KERN_ERR
2092 "md/raid10:%s: %s: redirecting"
2101 max_sectors);
2110 /* Drat - have to split this up more */
2119 else
2126 GFP_NOIO);
2140 }
2143 {
2144 /* Some sort of write request has finished and it
2145 * succeeded in writing where we thought there was a
2146 * bad block. So forget the bad block.
2147 * Or possibly if failed and we need to record
2148 * a bad block.
2149 */
2163 rdev,
2168 rdev,
2172 }
2173 }
2182 rdev,
2192 }
2194 }
2195 }
2200 }
2201 }
2204 {
2222 }
2240 /* just a partial read to be scheduled from a
2241 * separate context
2242 */
2245 }
2250 }
2252 }
2256 {
2266 }
2268 /*
2269 * perform a "sync" on one "block"
2270 *
2271 * We need to make sure that no normal I/O request - particularly write
2272 * requests - conflict with active sync requests.
2273 *
2274 * This is achieved by tracking pending requests and a 'barrier' concept
2275 * that can be installed to exclude normal IO requests.
2276 *
2277 * Resync and recovery are handled very differently.
2278 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
2279 *
2280 * For resync, we iterate over virtual addresses, read all copies,
2281 * and update if there are differences. If only one copy is live,
2282 * skip it.
2283 * For recovery, we iterate over physical addresses, read a good
2284 * value for each non-in_sync drive, and over-write.
2285 *
2286 * So, for recovery we may have several outstanding complex requests for a
2287 * given address, one for each out-of-sync device. We model this by allocating
2288 * a number of r10_bio structures, one for each out-of-sync device.
2289 * As we setup these structures, we collect all bio's together into a list
2290 * which we then process collectively to add pages, and then process again
2291 * to pass to generic_make_request.
2292 *
2293 * The r10_bio structures are linked using a borrowed master_bio pointer.
2294 * This link is counted in ->remaining. When the r10_bio that points to NULL
2295 * has its remaining count decremented to 0, the whole complex operation
2296 * is complete.
2297 *
2298 */
2302 {
2309 sector_t sync_blocks;
2317 skipped:
2322 /* If we aborted, we need to abort the
2323 * sync on the 'current' bitmap chucks (there can
2324 * be several when recovering multiple devices).
2325 * as we may have started syncing it but not finished.
2326 * We can find the current address in
2327 * mddev->curr_resync, but for recovery,
2328 * we need to convert that to several
2329 * virtual addresses.
2330 */
2336 sector_t sect =
2340 }
2348 }
2350 /* if there has been nothing to do on any drive,
2351 * then there is nothing to do at all..
2352 */
2355 }
2360 /* make sure whole request will fit in a chunk - if chunks
2361 * are meaningful
2362 */
2366 /*
2367 * If there is non-resync activity waiting for us then
2368 * put in a delay to throttle resync.
2369 */
2373 /* Again, very different code for resync and recovery.
2374 * Both must result in an r10bio with a list of bios that
2375 * have bi_end_io, bi_sector, bi_bdev set,
2376 * and bi_private set to the r10bio.
2377 * For recovery, we may actually create several r10bios
2378 * with 2 bios in each, that correspond to the bios in the main one.
2379 * In this case, the subordinate r10bios link back through a
2380 * borrowed master_bio pointer, and the counter in the master
2381 * includes a ref from each subordinate.
2382 */
2383 /* First, we decide what to do and set ->bi_end_io
2384 * To end_sync_read if we want to read, and
2385 * end_sync_write if we will want to write.
2386 */
2390 /* recovery... the complicated one */
2397 sector_t sect;
2406 /* want to reconstruct this device */
2409 /* Unless we are doing a full sync, we only need
2410 * to recover the block if it is set in the bitmap
2411 */
2418 /* yep, skip the sync_blocks here, but don't assume
2419 * that there will never be anything to do here
2420 */
2423 }
2438 /* Need to check if the array will still be
2439 * degraded
2440 */
2446 }
2462 /* This is where we read from */
2472 bad_sectors -= (sector
2477 }
2478 }
2491 /* and we write to 'i' */
2514 }
2516 /* Cannot recover, so abort the recovery or
2517 * record a bad block */
2523 /* problem is that there are bad blocks
2524 * on other device(s)
2525 */
2535 }
2544 }
2546 }
2547 }
2554 }
2556 }
2558 /* resync. Schedule a read for every block at this virt offset */
2567 /* We can skip this block */
2570 }
2608 }
2609 }
2620 count++;
2621 }
2628 mddev);
2629 }
2633 }
2634 }
2645 }
2663 /* stop here */
2668 /* remove last page from this bio */
2672 }
2674 }
2678 bio_full:
2692 }
2693 }
2696 /* pretend they weren't skipped, it makes
2697 * no important difference in this case
2698 */
2702 giveup:
2703 /* There is nowhere to write, so all non-sync
2704 * drives must be failed or in resync, all drives
2705 * have a bad block, so try the next chunk...
2706 */
2711 chunks_skipped ++;
2714 }
2716 static sector_t
2718 {
2719 sector_t size;
2733 }
2737 {
2749 }
2760 }
2768 GFP_KERNEL);
2794 /* 'size' is now the number of chunks in the array */
2795 /* calculate "used chunks per device" in 'stride' */
2798 /* We need to round up when dividing by raid_disks to
2799 * get the stride size.
2800 */
2808 else
2826 out:
2835 }
2837 }
2840 {
2845 sector_t size;
2847 /*
2848 * copy the already verified devices into our private RAID10
2849 * bookkeeping area. [whatever we allocate in run(),
2850 * should be freed in stop()]
2851 */
2858 }
2870 else
2885 /* as we don't honour merge_bvec_fn, we must never risk
2886 * violating it, so limit max_segments to 1 lying
2887 * within a single page.
2888 */
2893 }
2896 }
2897 /* need to check that every block has at least one working mirror */
2902 }
2915 }
2916 }
2926 /*
2927 * Ok, everything is just fine now
2928 */
2937 /* Calculate max read-ahead size.
2938 * We need to readahead at least twice a whole stripe....
2939 * maybe...
2940 */
2941 {
2947 }
2957 out_free_conf:
2965 out:
2967 }
2970 {
2985 }
2988 {
2998 }
2999 }
3002 {
3010 }
3012 /* Set new parameters */
3014 /* new layout: far_copies = 1, near_copies = 2 */
3019 /* make sure it will be not marked as dirty */
3028 }
3031 }
3034 {
3037 /* raid10 can take over:
3038 * raid0 - providing it has only two drives
3039 */
3041 /* for raid0 takeover only one zone is supported */
3048 }
3050 }
3052 }
3055 {
3071 };
3074 {
3076 }
3079 {
3081 }