| blob | 66c4aee20c72afc9f3318ce0ac62db9e39ad88be |
1 /*
2 * raid1.c : Multiple Devices driver for Linux
3 *
4 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
5 *
6 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
7 *
8 * RAID-1 management functions.
9 *
10 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
11 *
12 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
13 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
14 *
15 * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support
16 * bitmapped intelligence in resync:
17 *
18 * - bitmap marked during normal i/o
19 * - bitmap used to skip nondirty blocks during sync
20 *
21 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
22 * - persistent bitmap code
23 *
24 * This program is free software; you can redistribute it and/or modify
25 * it under the terms of the GNU General Public License as published by
26 * the Free Software Foundation; either version 2, or (at your option)
27 * any later version.
28 *
29 * You should have received a copy of the GNU General Public License
30 * (for example /usr/src/linux/COPYING); if not, write to the Free
31 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
32 */
34 #include <linux/slab.h>
35 #include <linux/delay.h>
36 #include <linux/blkdev.h>
37 #include <linux/module.h>
38 #include <linux/seq_file.h>
39 #include <linux/ratelimit.h>
44 /*
45 * Number of guaranteed r1bios in case of extreme VM load:
46 */
47 #define NR_RAID1_BIOS 256
49 /* when we get a read error on a read-only array, we redirect to another
50 * device without failing the first device, or trying to over-write to
51 * correct the read error. To keep track of bad blocks on a per-bio
52 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
53 */
54 #define IO_BLOCKED ((struct bio *)1)
55 /* When we successfully write to a known bad-block, we need to remove the
56 * bad-block marking which must be done from process context. So we record
57 * the success by setting devs[n].bio to IO_MADE_GOOD
58 */
59 #define IO_MADE_GOOD ((struct bio *)2)
61 #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
63 /* When there are this many requests queue to be written by
64 * the raid1 thread, we become 'congested' to provide back-pressure
65 * for writeback.
66 */
73 {
77 /* allocate a r1bio with room for raid_disks entries in the bios array */
79 }
82 {
84 }
86 #define RESYNC_BLOCK_SIZE (64*1024)
87 //#define RESYNC_BLOCK_SIZE PAGE_SIZE
88 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
89 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
90 #define RESYNC_WINDOW (2048*1024)
93 {
104 /*
105 * Allocate bios : 1 for reading, n-1 for writing
106 */
112 }
113 /*
114 * Allocate RESYNC_PAGES data pages and attach them to
115 * the first bio.
116 * If this is a user-requested check/repair, allocate
117 * RESYNC_PAGES for each bio.
118 */
121 else
129 }
130 /* If not user-requests, copy the page pointers to all bios */
136 }
142 out_free_pages:
148 }
150 out_free_bio:
155 }
158 {
169 }
174 }
177 {
185 }
186 }
189 {
194 }
197 {
205 }
210 }
213 {
225 }
227 /*
228 * raid_end_bio_io() is called when we have finished servicing a mirrored
229 * operation and are ready to return a success/failure code to the buffer
230 * cache layer.
231 */
233 {
251 /*
252 * Wake up any possible resync thread that waits for the device
253 * to go idle.
254 */
256 }
257 }
260 {
263 /* if nobody has done the final endio yet, do it now */
272 }
274 }
276 /*
277 * Update disk head position estimator based on IRQ completion info.
278 */
280 {
285 }
287 /*
288 * Find the disk number which triggered given bio
289 */
291 {
304 }
307 {
314 /*
315 * this branch is our 'one mirror IO has finished' event handler:
316 */
322 /* If all other devices have failed, we want to return
323 * the error upwards rather than fail the last device.
324 * Here we redefine "uptodate" to mean "Don't want to retry"
325 */
333 }
339 /*
340 * oops, read error:
341 */
344 KERN_ERR "md/raid1:%s: %s: "
348 b),
352 /* don't drop the reference on read_disk yet */
353 }
354 }
357 {
358 /* it really is the end of this request */
360 /* free extra copy of the data pages */
366 }
367 /* clear the bitmap if all writes complete successfully */
373 }
376 {
386 else
388 }
389 }
392 {
401 /*
402 * 'one mirror IO has finished' event handler:
403 */
414 /*
415 * Set R1BIO_Uptodate in our master bio, so that we
416 * will return a good error code for to the higher
417 * levels even if IO on some other mirrored buffer
418 * fails.
419 *
420 * The 'master' represents the composite IO operation
421 * to user-side. So if something waits for IO, then it
422 * will wait for the 'master' bio.
423 */
424 sector_t first_bad;
429 /*
430 * Do not set R1BIO_Uptodate if the current device is
431 * rebuilding or Faulty. This is because we cannot use
432 * such device for properly reading the data back (we could
433 * potentially use it, if the current write would have felt
434 * before rdev->recovery_offset, but for simplicity we don't
435 * check this here.
436 */
441 /* Maybe we can clear some bad blocks. */
447 }
448 }
454 /*
455 * In behind mode, we ACK the master bio once the I/O
456 * has safely reached all non-writemostly
457 * disks. Setting the Returned bit ensures that this
458 * gets done only once -- we don't ever want to return
459 * -EIO here, instead we'll wait
460 */
463 /* Maybe we can return now */
472 }
473 }
474 }
479 /*
480 * Let's see if all mirrored write operations have finished
481 * already.
482 */
487 }
490 /*
491 * This routine returns the disk from which the requested read should
492 * be done. There is a per-array 'next expected sequential IO' sector
493 * number - if this matches on the next IO then we use the last disk.
494 * There is also a per-disk 'last know head position' sector that is
495 * maintained from IRQ contexts, both the normal and the resync IO
496 * completion handlers update this position correctly. If there is no
497 * perfect sequential match then we pick the disk whose head is closest.
498 *
499 * If there are 2 mirrors in the same 2 devices, performance degrades
500 * because position is mirror, not device based.
501 *
502 * The rdev for the device selected will have nr_pending incremented.
503 */
505 {
512 sector_t best_dist;
519 /*
520 * Check if we can balance. We can balance on the whole
521 * device if no resync is going on, or below the resync window.
522 * We take the first readable disk when above the resync window.
523 */
524 retry:
538 else
542 sector_t dist;
543 sector_t first_bad;
558 /* Don't balance among write-mostly, just
559 * use the first as a last resort */
564 /* Cannot use this */
570 }
572 }
573 /* This is a reasonable device to use. It might
574 * even be best.
575 */
579 /* already have a better device */
582 /* cannot read here. If this is the 'primary'
583 * device, then we must not read beyond
584 * bad_sectors from another device..
585 */
597 }
600 }
612 }
613 /* Don't change to another disk for sequential reads */
620 /*
621 * If buffered sequential IO size exceeds optimal
622 * iosize, check if there is idle disk. If yes, choose
623 * the idle disk. read_balance could already choose an
624 * idle disk before noticing it's a sequential IO in
625 * this disk. This doesn't matter because this disk
626 * will idle, next time it will be utilized after the
627 * first disk has IO size exceeds optimal iosize. In
628 * this way, iosize of the first disk will be optimal
629 * iosize at least. iosize of the second disk might be
630 * small, but not a big deal since when the second disk
631 * starts IO, the first disk is likely still busy.
632 */
640 }
642 }
643 /* If device is idle, use it */
647 }
655 }
660 }
661 }
663 /*
664 * If all disks are rotational, choose the closest disk. If any disk is
665 * non-rotational, choose the disk with less pending request even the
666 * disk is rotational, which might/might not be optimal for raids with
667 * mixed ratation/non-rotational disks depending on workload.
668 */
672 else
674 }
682 /* cannot risk returning a device that failed
683 * before we inc'ed nr_pending
684 */
687 }
694 }
699 }
704 {
725 }
726 }
727 }
729 }
732 }
735 {
751 /* Note the '|| 1' - when read_balance prefers
752 * non-congested targets, it can be removed
753 */
756 else
758 }
759 }
762 }
766 {
771 }
774 {
775 /* Any writes that have been queued but are awaiting
776 * bitmap updates get flushed here.
777 */
785 /* flush any pending bitmap writes to
786 * disk before proceeding w/ I/O */
795 /* Just ignore it */
797 else
800 }
803 }
805 /* Barriers....
806 * Sometimes we need to suspend IO while we do something else,
807 * either some resync/recovery, or reconfigure the array.
808 * To do this we raise a 'barrier'.
809 * The 'barrier' is a counter that can be raised multiple times
810 * to count how many activities are happening which preclude
811 * normal IO.
812 * We can only raise the barrier if there is no pending IO.
813 * i.e. if nr_pending == 0.
814 * We choose only to raise the barrier if no-one is waiting for the
815 * barrier to go down. This means that as soon as an IO request
816 * is ready, no other operations which require a barrier will start
817 * until the IO request has had a chance.
818 *
819 * So: regular IO calls 'wait_barrier'. When that returns there
820 * is no backgroup IO happening, It must arrange to call
821 * allow_barrier when it has finished its IO.
822 * backgroup IO calls must call raise_barrier. Once that returns
823 * there is no normal IO happeing. It must arrange to call
824 * lower_barrier when the particular background IO completes.
825 */
826 #define RESYNC_DEPTH 32
829 {
832 /* Wait until no block IO is waiting */
836 /* block any new IO from starting */
839 /* Now wait for all pending IO to complete */
845 }
848 {
855 }
858 {
862 /* Wait for the barrier to drop.
863 * However if there are already pending
864 * requests (preventing the barrier from
865 * rising completely), and the
866 * pre-process bio queue isn't empty,
867 * then don't wait, as we need to empty
868 * that queue to get the nr_pending
869 * count down.
870 */
878 }
881 }
884 {
890 }
893 {
894 /* stop syncio and normal IO and wait for everything to
895 * go quite.
896 * We increment barrier and nr_waiting, and then
897 * wait until nr_pending match nr_queued+extra
898 * This is called in the context of one normal IO request
899 * that has failed. Thus any sync request that might be pending
900 * will be blocked by nr_pending, and we need to wait for
901 * pending IO requests to complete or be queued for re-try.
902 * Thus the number queued (nr_queued) plus this request (extra)
903 * must match the number of pending IOs (nr_pending) before
904 * we continue.
905 */
914 }
916 {
917 /* reverse the effect of the freeze */
923 }
926 /* duplicate the data pages for behind I/O
927 */
929 {
933 GFP_NOIO);
946 }
952 do_sync_io:
958 }
964 };
967 {
969 cb);
983 }
985 /* we aren't scheduling, so we can do the write-out directly. */
995 /* Just ignore it */
997 else
1000 }
1002 }
1005 {
1026 /*
1027 * Register the new request and wait if the reconstruction
1028 * thread has put up a bar for new requests.
1029 * Continue immediately if no resync is active currently.
1030 */
1037 /* As the suspend_* range is controlled by
1038 * userspace, we want an interruptible
1039 * wait.
1040 */
1050 }
1052 }
1058 /*
1059 * make_request() can abort the operation when READA is being
1060 * used and no empty request is available.
1061 *
1062 */
1071 /* We might need to issue multiple reads to different
1072 * devices if there are bad blocks around, so we keep
1073 * track of the number of reads in bio->bi_phys_segments.
1074 * If this is 0, there is only one r1_bio and no locking
1075 * will be needed when requests complete. If it is
1076 * non-zero, then it is the number of not-completed requests.
1077 */
1082 /*
1083 * read balancing logic:
1084 */
1087 read_again:
1091 /* couldn't find anywhere to read from */
1094 }
1098 bitmap) {
1099 /* Reading from a write-mostly device must
1100 * take care not to over-take any writes
1101 * that are 'behind'
1102 */
1105 }
1110 max_sectors);
1121 /* could not read all from this device, so we will
1122 * need another r1_bio.
1123 */
1131 else
1134 /* Cannot call generic_make_request directly
1135 * as that will be queued in __make_request
1136 * and subsequent mempool_alloc might block waiting
1137 * for it. So hand bio over to raid1d.
1138 */
1152 }
1154 /*
1155 * WRITE:
1156 */
1161 }
1162 /* first select target devices under rcu_lock and
1163 * inc refcount on their rdev. Record them by setting
1164 * bios[x] to bio
1165 * If there are known/acknowledged bad blocks on any device on
1166 * which we have seen a write error, we want to avoid writing those
1167 * blocks.
1168 * This potentially requires several writes to write around
1169 * the bad blocks. Each set of writes gets it's own r1bio
1170 * with a set of bios attached.
1171 */
1174 retry_write:
1184 }
1191 }
1195 sector_t first_bad;
1200 max_sectors,
1203 /* mustn't write here until the bad block is
1204 * acknowledged*/
1208 }
1210 /* Cannot write here at all */
1213 /* mustn't write more than bad_sectors
1214 * to other devices yet
1215 */
1218 /* We don't set R1BIO_Degraded as that
1219 * only applies if the disk is
1220 * missing, so it might be re-added,
1221 * and we want to know to recover this
1222 * chunk.
1223 * In this case the device is here,
1224 * and the fact that this chunk is not
1225 * in-sync is recorded in the bad
1226 * block log
1227 */
1229 }
1234 }
1235 }
1237 }
1241 /* Wait for this device to become unblocked */
1252 }
1255 /* We are splitting this write into multiple parts, so
1256 * we need to prepare for allocating another r1_bio.
1257 */
1262 else
1265 }
1281 /* do behind I/O ?
1282 * Not if there are too many, or cannot
1283 * allocate memory, or a reader on WriteMostly
1284 * is waiting for behind writes to flush */
1296 }
1301 /*
1302 * We trimmed the bio, so _all is legit
1303 */
1308 }
1325 else
1334 }
1338 }
1339 /* Mustn't call r1_bio_write_done before this next test,
1340 * as it could result in the bio being freed.
1341 */
1344 /* We need another r1_bio. It has already been counted
1345 * in bio->bi_phys_segments
1346 */
1354 }
1358 /* In case raid1d snuck in to freeze_array */
1360 }
1363 {
1374 }
1377 }
1381 {
1385 /*
1386 * If it is not operational, then we have already marked it as dead
1387 * else if it is the last working disks, ignore the error, let the
1388 * next level up know.
1389 * else mark the drive as failed
1390 */
1393 /*
1394 * Don't fail the drive, act as though we were just a
1395 * normal single drive.
1396 * However don't try a recovery from this drive as
1397 * it is very likely to fail.
1398 */
1401 }
1409 /*
1410 * if recovery is running, make sure it aborts.
1411 */
1421 }
1424 {
1431 }
1444 }
1446 }
1449 {
1455 }
1458 {
1464 /*
1465 * Find all failed disks within the RAID1 configuration
1466 * and mark them readable.
1467 * Called under mddev lock, so rcu protection not needed.
1468 */
1476 /* replacement has just become active */
1479 count++;
1481 /* Replaced device not technically
1482 * faulty, but we need to be sure
1483 * it gets removed and never re-added
1484 */
1488 }
1489 }
1494 count++;
1496 }
1497 }
1504 }
1508 {
1526 }
1539 /* As all devices are equivalent, we don't need a full recovery
1540 * if this was recently any drive of the array
1541 */
1546 }
1549 /* Add this device as a replacement */
1557 }
1558 }
1560 /* Some requests might not have seen this new
1561 * merge_bvec_fn. We must wait for them to complete
1562 * before merging the device fully.
1563 * First we make sure any code which has tested
1564 * our function has submitted the request, then
1565 * we wait for all outstanding requests to complete.
1566 */
1571 }
1577 }
1580 {
1595 }
1596 /* Only remove non-faulty devices if recovery
1597 * is not possible.
1598 */
1604 }
1608 /* lost the race, try later */
1613 /* We just removed a device that is being replaced.
1614 * Move down the replacement. We drain all IO before
1615 * doing this to avoid confusion.
1616 */
1628 }
1629 abort:
1633 }
1637 {
1642 /*
1643 * we have read a block, now it needs to be re-written,
1644 * or re-read if the read failed.
1645 * We don't do much here, just schedule handling by raid1d
1646 */
1652 }
1655 {
1661 sector_t first_bad;
1670 /* make sure these bits doesn't get cleared. */
1692 )
1703 }
1704 }
1705 }
1709 {
1711 /* success */
1719 }
1720 /* need to record an error - either for the block or the device */
1724 }
1727 {
1728 /* Try some synchronous reads of other devices to get
1729 * good data, much like with normal read errors. Only
1730 * read into the pages we already have so we don't
1731 * need to re-issue the read request.
1732 * We don't need to freeze the array, because being in an
1733 * active sync request, there is no normal IO, and
1734 * no overlapping syncs.
1735 * We don't need to check is_badblock() again as we
1736 * made sure that anything with a bad block in range
1737 * will have bi_end_io clear.
1738 */
1757 /* No rcu protection needed here devices
1758 * can only be removed when no resync is
1759 * active, and resync is currently active
1760 */
1767 }
1768 }
1769 d++;
1777 /* Cannot read from anywhere, this block is lost.
1778 * Record a bad block on each device. If that doesn't
1779 * work just disable and interrupt the recovery.
1780 * Don't fail devices as that won't really help.
1781 */
1793 }
1801 }
1802 /* Try next page */
1805 idx++;
1807 }
1810 /* write it back and re-read */
1814 d--;
1823 }
1824 }
1829 d--;
1837 }
1840 idx ++;
1841 }
1845 }
1848 {
1849 /* We have read all readable devices. If we haven't
1850 * got the block, then there is no hope left.
1851 * If we have, then we want to do a comparison
1852 * and skip the write if everything is the same.
1853 * If any blocks failed to read, then we need to
1854 * attempt an over-write
1855 */
1862 /* Fix variable parts of all bios */
1871 /* fixup the bio for reuse, but preserve BIO_UPTODATE */
1891 else
1894 }
1895 }
1902 }
1912 /* Now we can 'fixup' the BIO_UPTODATE flag */
1924 }
1931 /* No need to write to this device. */
1935 }
1938 }
1940 }
1943 {
1952 /* ouch - failed to read all of that. */
1959 /*
1960 * schedule writes
1961 */
1977 }
1980 /* if we're here, all write(s) have completed, so clean up */
1988 }
1989 }
1990 }
1992 /*
1993 * This is a kernel thread which:
1994 *
1995 * 1. Retries failed read operations on working mirrors.
1996 * 2. Updates the raid superblock when problems encounter.
1997 * 3. Performs writes following reads for array synchronising.
1998 */
2002 {
2015 /* Note: no rcu protection needed here
2016 * as this is synchronous in the raid1d thread
2017 * which is the thread that might remove
2018 * a device. If raid1d ever becomes multi-threaded....
2019 */
2020 sector_t first_bad;
2034 d++;
2037 }
2041 /* Cannot read from anywhere - mark it bad */
2046 }
2047 /* write it back and re-read */
2052 d--;
2058 }
2064 d--;
2072 "md/raid1:%s: read error corrected "
2078 }
2079 }
2080 }
2083 }
2084 }
2087 {
2092 /* bio has the data to be written to device 'i' where
2093 * we just recently had a write error.
2094 * We repeatedly clone the bio and trim down to one block,
2095 * then try the write. Where the write fails we record
2096 * a bad block.
2097 * It is conceivable that the bio doesn't exactly align with
2098 * blocks. We must handle this somehow.
2099 *
2100 * We currently own a reference on the rdev.
2101 */
2104 sector_t sector;
2122 /* Write at 'sector' for 'sectors'*/
2129 vec++;
2130 vcnt--;
2131 }
2139 }
2149 /* failure! */
2158 }
2160 }
2163 {
2174 }
2179 }
2180 }
2183 }
2186 {
2196 /* This drive got a write error. We need to
2197 * narrow down and record precise write
2198 * errors.
2199 */
2203 /* an I/O failed, we can't clear the bitmap */
2205 }
2208 }
2212 }
2215 {
2224 /* we got a read error. Maybe the drive is bad. Maybe just
2225 * the block and we can fix it.
2226 * We freeze all other IO, and try reading the block from
2227 * other devices. When we find one, we re-write
2228 * and check it that fixes the read error.
2229 * This is all done synchronously while the array is
2230 * frozen
2231 */
2243 read_more:
2251 const unsigned long do_sync
2257 }
2264 "md/raid1:%s: redirecting sector %llu"
2275 /* Drat - have to split this up more */
2283 else
2301 }
2302 }
2305 {
2324 }
2336 else
2343 else
2344 /* just a partial read to be scheduled from separate
2345 * context
2346 */
2352 }
2354 }
2358 {
2369 }
2371 /*
2372 * perform a "sync" on one "block"
2373 *
2374 * We need to make sure that no normal I/O request - particularly write
2375 * requests - conflict with active sync requests.
2376 *
2377 * This is achieved by tracking pending requests and a 'barrier' concept
2378 * that can be installed to exclude normal IO requests.
2379 */
2381 static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
2382 {
2391 sector_t sync_blocks;
2402 /* If we aborted, we need to abort the
2403 * sync on the 'current' bitmap chunk (there will
2404 * only be one in raid1 resync.
2405 * We can find the current addess in mddev->curr_resync
2406 */
2416 }
2424 }
2425 /* before building a request, check if we can skip these blocks..
2426 * This call the bitmap_start_sync doesn't actually record anything
2427 */
2430 /* We can skip this block, and probably several more */
2433 }
2434 /*
2435 * If there is non-resync activity waiting for a turn,
2436 * and resync is going fast enough,
2437 * then let it though before starting on this new sync request.
2438 */
2449 /*
2450 * If we get a correctably read error during resync or recovery,
2451 * we might want to read from a different device. So we
2452 * flag all drives that could conceivably be read from for READ,
2453 * and any others (which will be non-In_sync devices) for WRITE.
2454 * If a read fails, we try reading from something else for which READ
2455 * is OK.
2456 */
2476 write_targets ++;
2478 /* may need to read from here */
2491 }
2492 }
2500 }
2503 read_targets++;
2507 /*
2508 * The device is suitable for reading (InSync),
2509 * but has bad block(s) here. Let's try to correct them,
2510 * if we are doing resync or repair. Otherwise, leave
2511 * this device alone for this sync request.
2512 */
2515 write_targets++;
2516 }
2517 }
2523 }
2524 }
2531 /* These sectors are bad on all InSync devices, so we
2532 * need to mark them bad on all write targets
2533 */
2541 }
2547 /* Cannot record the badblocks, so need to
2548 * abort the resync.
2549 * If there are multiple read targets, could just
2550 * fail the really bad ones ???
2551 */
2558 }
2560 /* only resync enough to reach the next bad->good
2561 * transition */
2563 }
2566 /* extra read targets are also write targets */
2570 /* There is nowhere to write, so all non-sync
2571 * drives must be failed - so we are finished
2572 */
2573 sector_t rv;
2580 }
2604 }
2611 /* stop here */
2614 i--;
2618 /* remove last page from this bio */
2622 }
2624 }
2625 }
2626 }
2631 bio_full:
2634 /* For a user-requested sync, we read all readable devices and do a
2635 * compare
2636 */
2642 read_targets--;
2645 }
2646 }
2653 }
2655 }
2658 {
2663 }
2666 {
2679 GFP_KERNEL);
2692 r1bio_pool_free,
2709 else
2721 }
2740 /* This slot has a replacement. */
2742 /* No original, just make the replacement
2743 * a recovering spare
2744 */
2749 /* Original is not in_sync - bad */
2751 }
2759 }
2760 }
2769 }
2773 abort:
2781 }
2783 }
2787 {
2798 }
2803 }
2804 /*
2805 * copy the already verified devices into our private RAID1
2806 * bookkeeping area. [whatever we allocate in run(),
2807 * should be freed in stop()]
2808 */
2811 else
2827 }
2848 /*
2849 * Ok, everything is just fine now
2850 */
2865 else
2868 }
2874 }
2877 {
2881 /* wait for behind writes to complete */
2885 /* need to kick something here to make sure I/O goes? */
2888 }
2902 }
2905 {
2906 /* no resync is happening, and there is enough space
2907 * on all devices, so we can resize.
2908 * We need to make sure resync covers any new space.
2909 * If the array is shrinking we should possibly wait until
2910 * any io in the removed space completes, but it hardly seems
2911 * worth it.
2912 */
2921 }
2929 }
2933 }
2936 {
2937 /* We need to:
2938 * 1/ resize the r1bio_pool
2939 * 2/ resize conf->mirrors
2940 *
2941 * We allocate a new r1bio_pool if we can.
2942 * Then raise a device barrier and wait until all IO stops.
2943 * Then resize conf->mirrors and swap in the new r1bio pool.
2944 *
2945 * At the same time, we "pack" the devices so that all the missing
2946 * devices have the higher raid_disk numbers.
2947 */
2956 /* Cannot change chunk_size, layout, or level */
2964 }
2976 cnt++;
2979 }
2992 }
2994 GFP_KERNEL);
2999 }
3003 /* ok, everything is stopped */
3017 }
3020 }
3039 }
3042 {
3055 }
3056 }
3059 {
3060 /* raid1 can take over:
3061 * raid5 with 2 devices, any layout or chunk size
3062 */
3072 }
3074 }
3077 {
3095 };
3098 {
3100 }
3103 {
3105 }