| blob | 05bb49e13648267d21a02efb698db165d0180026 |
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
132 }
133 }
134 /* If not user-requests, copy the page pointers to all bios */
140 }
146 out_free_pages:
151 out_free_bio:
156 }
159 {
170 }
175 }
178 {
186 }
187 }
190 {
195 }
198 {
206 }
211 }
214 {
226 }
228 /*
229 * raid_end_bio_io() is called when we have finished servicing a mirrored
230 * operation and are ready to return a success/failure code to the buffer
231 * cache layer.
232 */
234 {
252 /*
253 * Wake up any possible resync thread that waits for the device
254 * to go idle.
255 */
257 }
258 }
261 {
264 /* if nobody has done the final endio yet, do it now */
273 }
275 }
277 /*
278 * Update disk head position estimator based on IRQ completion info.
279 */
281 {
286 }
288 /*
289 * Find the disk number which triggered given bio
290 */
292 {
305 }
308 {
315 /*
316 * this branch is our 'one mirror IO has finished' event handler:
317 */
323 /* If all other devices have failed, we want to return
324 * the error upwards rather than fail the last device.
325 * Here we redefine "uptodate" to mean "Don't want to retry"
326 */
334 }
339 /*
340 * oops, read error:
341 */
344 KERN_ERR "md/raid1:%s: %s: "
348 b),
352 }
355 }
358 {
359 /* it really is the end of this request */
361 /* free extra copy of the data pages */
367 }
368 /* clear the bitmap if all writes complete successfully */
374 }
377 {
387 else
389 }
390 }
393 {
402 /*
403 * 'one mirror IO has finished' event handler:
404 */
415 /*
416 * Set R1BIO_Uptodate in our master bio, so that we
417 * will return a good error code for to the higher
418 * levels even if IO on some other mirrored buffer
419 * fails.
420 *
421 * The 'master' represents the composite IO operation
422 * to user-side. So if something waits for IO, then it
423 * will wait for the 'master' bio.
424 */
425 sector_t first_bad;
432 /* Maybe we can clear some bad blocks. */
438 }
439 }
445 /*
446 * In behind mode, we ACK the master bio once the I/O
447 * has safely reached all non-writemostly
448 * disks. Setting the Returned bit ensures that this
449 * gets done only once -- we don't ever want to return
450 * -EIO here, instead we'll wait
451 */
454 /* Maybe we can return now */
463 }
464 }
465 }
470 /*
471 * Let's see if all mirrored write operations have finished
472 * already.
473 */
478 }
481 /*
482 * This routine returns the disk from which the requested read should
483 * be done. There is a per-array 'next expected sequential IO' sector
484 * number - if this matches on the next IO then we use the last disk.
485 * There is also a per-disk 'last know head position' sector that is
486 * maintained from IRQ contexts, both the normal and the resync IO
487 * completion handlers update this position correctly. If there is no
488 * perfect sequential match then we pick the disk whose head is closest.
489 *
490 * If there are 2 mirrors in the same 2 devices, performance degrades
491 * because position is mirror, not device based.
492 *
493 * The rdev for the device selected will have nr_pending incremented.
494 */
496 {
503 sector_t best_dist;
510 /*
511 * Check if we can balance. We can balance on the whole
512 * device if no resync is going on, or below the resync window.
513 * We take the first readable disk when above the resync window.
514 */
515 retry:
529 else
533 sector_t dist;
534 sector_t first_bad;
549 /* Don't balance among write-mostly, just
550 * use the first as a last resort */
555 /* Cannot use this */
561 }
563 }
564 /* This is a reasonable device to use. It might
565 * even be best.
566 */
570 /* already have a better device */
573 /* cannot read here. If this is the 'primary'
574 * device, then we must not read beyond
575 * bad_sectors from another device..
576 */
588 }
591 }
603 }
604 /* Don't change to another disk for sequential reads */
611 /*
612 * If buffered sequential IO size exceeds optimal
613 * iosize, check if there is idle disk. If yes, choose
614 * the idle disk. read_balance could already choose an
615 * idle disk before noticing it's a sequential IO in
616 * this disk. This doesn't matter because this disk
617 * will idle, next time it will be utilized after the
618 * first disk has IO size exceeds optimal iosize. In
619 * this way, iosize of the first disk will be optimal
620 * iosize at least. iosize of the second disk might be
621 * small, but not a big deal since when the second disk
622 * starts IO, the first disk is likely still busy.
623 */
631 }
633 }
634 /* If device is idle, use it */
638 }
646 }
651 }
652 }
654 /*
655 * If all disks are rotational, choose the closest disk. If any disk is
656 * non-rotational, choose the disk with less pending request even the
657 * disk is rotational, which might/might not be optimal for raids with
658 * mixed ratation/non-rotational disks depending on workload.
659 */
663 else
665 }
673 /* cannot risk returning a device that failed
674 * before we inc'ed nr_pending
675 */
678 }
685 }
690 }
695 {
716 }
717 }
718 }
720 }
723 }
726 {
742 /* Note the '|| 1' - when read_balance prefers
743 * non-congested targets, it can be removed
744 */
747 else
749 }
750 }
753 }
757 {
762 }
765 {
766 /* Any writes that have been queued but are awaiting
767 * bitmap updates get flushed here.
768 */
776 /* flush any pending bitmap writes to
777 * disk before proceeding w/ I/O */
786 }
789 }
791 /* Barriers....
792 * Sometimes we need to suspend IO while we do something else,
793 * either some resync/recovery, or reconfigure the array.
794 * To do this we raise a 'barrier'.
795 * The 'barrier' is a counter that can be raised multiple times
796 * to count how many activities are happening which preclude
797 * normal IO.
798 * We can only raise the barrier if there is no pending IO.
799 * i.e. if nr_pending == 0.
800 * We choose only to raise the barrier if no-one is waiting for the
801 * barrier to go down. This means that as soon as an IO request
802 * is ready, no other operations which require a barrier will start
803 * until the IO request has had a chance.
804 *
805 * So: regular IO calls 'wait_barrier'. When that returns there
806 * is no backgroup IO happening, It must arrange to call
807 * allow_barrier when it has finished its IO.
808 * backgroup IO calls must call raise_barrier. Once that returns
809 * there is no normal IO happeing. It must arrange to call
810 * lower_barrier when the particular background IO completes.
811 */
812 #define RESYNC_DEPTH 32
815 {
818 /* Wait until no block IO is waiting */
822 /* block any new IO from starting */
825 /* Now wait for all pending IO to complete */
831 }
834 {
841 }
844 {
848 /* Wait for the barrier to drop.
849 * However if there are already pending
850 * requests (preventing the barrier from
851 * rising completely), and the
852 * pre-process bio queue isn't empty,
853 * then don't wait, as we need to empty
854 * that queue to get the nr_pending
855 * count down.
856 */
863 );
865 }
868 }
871 {
877 }
880 {
881 /* stop syncio and normal IO and wait for everything to
882 * go quite.
883 * We increment barrier and nr_waiting, and then
884 * wait until nr_pending match nr_queued+1
885 * This is called in the context of one normal IO request
886 * that has failed. Thus any sync request that might be pending
887 * will be blocked by nr_pending, and we need to wait for
888 * pending IO requests to complete or be queued for re-try.
889 * Thus the number queued (nr_queued) plus this request (1)
890 * must match the number of pending IOs (nr_pending) before
891 * we continue.
892 */
901 }
903 {
904 /* reverse the effect of the freeze */
910 }
913 /* duplicate the data pages for behind I/O
914 */
916 {
920 GFP_NOIO);
933 }
939 do_sync_io:
945 }
951 };
954 {
956 cb);
969 }
971 /* we aren't scheduling, so we can do the write-out directly. */
981 }
983 }
986 {
1004 /*
1005 * Register the new request and wait if the reconstruction
1006 * thread has put up a bar for new requests.
1007 * Continue immediately if no resync is active currently.
1008 */
1015 /* As the suspend_* range is controlled by
1016 * userspace, we want an interruptible
1017 * wait.
1018 */
1028 }
1030 }
1036 /*
1037 * make_request() can abort the operation when READA is being
1038 * used and no empty request is available.
1039 *
1040 */
1049 /* We might need to issue multiple reads to different
1050 * devices if there are bad blocks around, so we keep
1051 * track of the number of reads in bio->bi_phys_segments.
1052 * If this is 0, there is only one r1_bio and no locking
1053 * will be needed when requests complete. If it is
1054 * non-zero, then it is the number of not-completed requests.
1055 */
1060 /*
1061 * read balancing logic:
1062 */
1065 read_again:
1069 /* couldn't find anywhere to read from */
1072 }
1076 bitmap) {
1077 /* Reading from a write-mostly device must
1078 * take care not to over-take any writes
1079 * that are 'behind'
1080 */
1083 }
1088 max_sectors);
1099 /* could not read all from this device, so we will
1100 * need another r1_bio.
1101 */
1109 else
1112 /* Cannot call generic_make_request directly
1113 * as that will be queued in __make_request
1114 * and subsequent mempool_alloc might block waiting
1115 * for it. So hand bio over to raid1d.
1116 */
1130 }
1132 /*
1133 * WRITE:
1134 */
1139 }
1140 /* first select target devices under rcu_lock and
1141 * inc refcount on their rdev. Record them by setting
1142 * bios[x] to bio
1143 * If there are known/acknowledged bad blocks on any device on
1144 * which we have seen a write error, we want to avoid writing those
1145 * blocks.
1146 * This potentially requires several writes to write around
1147 * the bad blocks. Each set of writes gets it's own r1bio
1148 * with a set of bios attached.
1149 */
1152 retry_write:
1162 }
1169 }
1173 sector_t first_bad;
1178 max_sectors,
1181 /* mustn't write here until the bad block is
1182 * acknowledged*/
1186 }
1188 /* Cannot write here at all */
1191 /* mustn't write more than bad_sectors
1192 * to other devices yet
1193 */
1196 /* We don't set R1BIO_Degraded as that
1197 * only applies if the disk is
1198 * missing, so it might be re-added,
1199 * and we want to know to recover this
1200 * chunk.
1201 * In this case the device is here,
1202 * and the fact that this chunk is not
1203 * in-sync is recorded in the bad
1204 * block log
1205 */
1207 }
1212 }
1213 }
1215 }
1219 /* Wait for this device to become unblocked */
1230 }
1233 /* We are splitting this write into multiple parts, so
1234 * we need to prepare for allocating another r1_bio.
1235 */
1240 else
1243 }
1259 /* do behind I/O ?
1260 * Not if there are too many, or cannot
1261 * allocate memory, or a reader on WriteMostly
1262 * is waiting for behind writes to flush */
1274 }
1279 /* Yes, I really want the '__' version so that
1280 * we clear any unused pointer in the io_vec, rather
1281 * than leave them unchanged. This is important
1282 * because when we come to free the pages, we won't
1283 * know the original bi_idx, so we just free
1284 * them all
1285 */
1290 }
1306 else
1315 }
1319 }
1320 /* Mustn't call r1_bio_write_done before this next test,
1321 * as it could result in the bio being freed.
1322 */
1325 /* We need another r1_bio. It has already been counted
1326 * in bio->bi_phys_segments
1327 */
1335 }
1339 /* In case raid1d snuck in to freeze_array */
1341 }
1344 {
1355 }
1358 }
1362 {
1366 /*
1367 * If it is not operational, then we have already marked it as dead
1368 * else if it is the last working disks, ignore the error, let the
1369 * next level up know.
1370 * else mark the drive as failed
1371 */
1374 /*
1375 * Don't fail the drive, act as though we were just a
1376 * normal single drive.
1377 * However don't try a recovery from this drive as
1378 * it is very likely to fail.
1379 */
1382 }
1390 /*
1391 * if recovery is running, make sure it aborts.
1392 */
1402 }
1405 {
1412 }
1425 }
1427 }
1430 {
1436 }
1439 {
1445 /*
1446 * Find all failed disks within the RAID1 configuration
1447 * and mark them readable.
1448 * Called under mddev lock, so rcu protection not needed.
1449 */
1457 /* replacement has just become active */
1460 count++;
1462 /* Replaced device not technically
1463 * faulty, but we need to be sure
1464 * it gets removed and never re-added
1465 */
1469 }
1470 }
1474 count++;
1476 }
1477 }
1484 }
1488 {
1506 }
1518 /* As all devices are equivalent, we don't need a full recovery
1519 * if this was recently any drive of the array
1520 */
1525 }
1528 /* Add this device as a replacement */
1536 }
1537 }
1539 /* Some requests might not have seen this new
1540 * merge_bvec_fn. We must wait for them to complete
1541 * before merging the device fully.
1542 * First we make sure any code which has tested
1543 * our function has submitted the request, then
1544 * we wait for all outstanding requests to complete.
1545 */
1550 }
1554 }
1557 {
1572 }
1573 /* Only remove non-faulty devices if recovery
1574 * is not possible.
1575 */
1581 }
1585 /* lost the race, try later */
1590 /* We just removed a device that is being replaced.
1591 * Move down the replacement. We drain all IO before
1592 * doing this to avoid confusion.
1593 */
1605 }
1606 abort:
1610 }
1614 {
1619 /*
1620 * we have read a block, now it needs to be re-written,
1621 * or re-read if the read failed.
1622 * We don't do much here, just schedule handling by raid1d
1623 */
1629 }
1632 {
1638 sector_t first_bad;
1647 /* make sure these bits doesn't get cleared. */
1669 )
1680 }
1681 }
1682 }
1686 {
1688 /* success */
1696 }
1697 /* need to record an error - either for the block or the device */
1701 }
1704 {
1705 /* Try some synchronous reads of other devices to get
1706 * good data, much like with normal read errors. Only
1707 * read into the pages we already have so we don't
1708 * need to re-issue the read request.
1709 * We don't need to freeze the array, because being in an
1710 * active sync request, there is no normal IO, and
1711 * no overlapping syncs.
1712 * We don't need to check is_badblock() again as we
1713 * made sure that anything with a bad block in range
1714 * will have bi_end_io clear.
1715 */
1734 /* No rcu protection needed here devices
1735 * can only be removed when no resync is
1736 * active, and resync is currently active
1737 */
1744 }
1745 }
1746 d++;
1754 /* Cannot read from anywhere, this block is lost.
1755 * Record a bad block on each device. If that doesn't
1756 * work just disable and interrupt the recovery.
1757 * Don't fail devices as that won't really help.
1758 */
1770 }
1778 }
1779 /* Try next page */
1782 idx++;
1784 }
1787 /* write it back and re-read */
1791 d--;
1800 }
1801 }
1806 d--;
1814 }
1817 idx ++;
1818 }
1822 }
1825 {
1826 /* We have read all readable devices. If we haven't
1827 * got the block, then there is no hope left.
1828 * If we have, then we want to do a comparison
1829 * and skip the write if everything is the same.
1830 * If any blocks failed to read, then we need to
1831 * attempt an over-write
1832 */
1845 }
1866 }
1873 /* No need to write to this device. */
1877 }
1878 /* fixup the bio for reuse */
1896 else
1901 PAGE_SIZE);
1902 }
1903 }
1905 }
1908 {
1917 /* ouch - failed to read all of that. */
1924 /*
1925 * schedule writes
1926 */
1942 }
1945 /* if we're here, all write(s) have completed, so clean up */
1953 }
1954 }
1955 }
1957 /*
1958 * This is a kernel thread which:
1959 *
1960 * 1. Retries failed read operations on working mirrors.
1961 * 2. Updates the raid superblock when problems encounter.
1962 * 3. Performs writes following reads for array synchronising.
1963 */
1967 {
1980 /* Note: no rcu protection needed here
1981 * as this is synchronous in the raid1d thread
1982 * which is the thread that might remove
1983 * a device. If raid1d ever becomes multi-threaded....
1984 */
1985 sector_t first_bad;
1999 d++;
2002 }
2006 /* Cannot read from anywhere - mark it bad */
2011 }
2012 /* write it back and re-read */
2017 d--;
2023 }
2029 d--;
2037 "md/raid1:%s: read error corrected "
2043 }
2044 }
2045 }
2048 }
2049 }
2052 {
2054 }
2057 {
2068 }
2071 {
2078 /* bio has the data to be written to device 'i' where
2079 * we just recently had a write error.
2080 * We repeatedly clone the bio and trim down to one block,
2081 * then try the write. Where the write fails we record
2082 * a bad block.
2083 * It is conceivable that the bio doesn't exactly align with
2084 * blocks. We must handle this somehow.
2085 *
2086 * We currently own a reference on the rdev.
2087 */
2090 sector_t sector;
2109 idx++;
2114 }
2119 /* Write at 'sector' for 'sectors'*/
2133 /* failure! */
2142 }
2144 }
2147 {
2158 }
2163 }
2164 }
2167 }
2170 {
2180 /* This drive got a write error. We need to
2181 * narrow down and record precise write
2182 * errors.
2183 */
2187 /* an I/O failed, we can't clear the bitmap */
2189 }
2192 }
2196 }
2199 {
2208 /* we got a read error. Maybe the drive is bad. Maybe just
2209 * the block and we can fix it.
2210 * We freeze all other IO, and try reading the block from
2211 * other devices. When we find one, we re-write
2212 * and check it that fixes the read error.
2213 * This is all done synchronously while the array is
2214 * frozen
2215 */
2226 read_more:
2234 const unsigned long do_sync
2240 }
2247 "md/raid1:%s: redirecting sector %llu"
2258 /* Drat - have to split this up more */
2266 else
2285 }
2286 }
2289 {
2307 }
2319 else
2326 else
2327 /* just a partial read to be scheduled from separate
2328 * context
2329 */
2335 }
2337 }
2341 {
2352 }
2354 /*
2355 * perform a "sync" on one "block"
2356 *
2357 * We need to make sure that no normal I/O request - particularly write
2358 * requests - conflict with active sync requests.
2359 *
2360 * This is achieved by tracking pending requests and a 'barrier' concept
2361 * that can be installed to exclude normal IO requests.
2362 */
2364 static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
2365 {
2374 sector_t sync_blocks;
2385 /* If we aborted, we need to abort the
2386 * sync on the 'current' bitmap chunk (there will
2387 * only be one in raid1 resync.
2388 * We can find the current addess in mddev->curr_resync
2389 */
2399 }
2407 }
2408 /* before building a request, check if we can skip these blocks..
2409 * This call the bitmap_start_sync doesn't actually record anything
2410 */
2413 /* We can skip this block, and probably several more */
2416 }
2417 /*
2418 * If there is non-resync activity waiting for a turn,
2419 * and resync is going fast enough,
2420 * then let it though before starting on this new sync request.
2421 */
2432 /*
2433 * If we get a correctably read error during resync or recovery,
2434 * we might want to read from a different device. So we
2435 * flag all drives that could conceivably be read from for READ,
2436 * and any others (which will be non-In_sync devices) for WRITE.
2437 * If a read fails, we try reading from something else for which READ
2438 * is OK.
2439 */
2450 /* take from bio_init */
2470 write_targets ++;
2472 /* may need to read from here */
2485 }
2486 }
2494 }
2497 read_targets++;
2501 /*
2502 * The device is suitable for reading (InSync),
2503 * but has bad block(s) here. Let's try to correct them,
2504 * if we are doing resync or repair. Otherwise, leave
2505 * this device alone for this sync request.
2506 */
2509 write_targets++;
2510 }
2511 }
2517 }
2518 }
2525 /* These sectors are bad on all InSync devices, so we
2526 * need to mark them bad on all write targets
2527 */
2535 }
2541 /* Cannot record the badblocks, so need to
2542 * abort the resync.
2543 * If there are multiple read targets, could just
2544 * fail the really bad ones ???
2545 */
2552 }
2554 /* only resync enough to reach the next bad->good
2555 * transition */
2557 }
2560 /* extra read targets are also write targets */
2564 /* There is nowhere to write, so all non-sync
2565 * drives must be failed - so we are finished
2566 */
2567 sector_t rv;
2574 }
2598 }
2605 /* stop here */
2608 i--;
2612 /* remove last page from this bio */
2616 }
2618 }
2619 }
2620 }
2625 bio_full:
2628 /* For a user-requested sync, we read all readable devices and do a
2629 * compare
2630 */
2636 read_targets--;
2639 }
2640 }
2647 }
2649 }
2652 {
2657 }
2660 {
2673 GFP_KERNEL);
2686 r1bio_pool_free,
2703 else
2715 }
2734 /* This slot has a replacement. */
2736 /* No original, just make the replacement
2737 * a recovering spare
2738 */
2743 /* Original is not in_sync - bad */
2745 }
2753 }
2754 }
2763 }
2767 abort:
2775 }
2777 }
2781 {
2791 }
2796 }
2797 /*
2798 * copy the already verified devices into our private RAID1
2799 * bookkeeping area. [whatever we allocate in run(),
2800 * should be freed in stop()]
2801 */
2804 else
2815 }
2836 /*
2837 * Ok, everything is just fine now
2838 */
2849 }
2855 }
2858 {
2862 /* wait for behind writes to complete */
2866 /* need to kick something here to make sure I/O goes? */
2869 }
2882 }
2885 {
2886 /* no resync is happening, and there is enough space
2887 * on all devices, so we can resize.
2888 * We need to make sure resync covers any new space.
2889 * If the array is shrinking we should possibly wait until
2890 * any io in the removed space completes, but it hardly seems
2891 * worth it.
2892 */
2901 }
2909 }
2913 }
2916 {
2917 /* We need to:
2918 * 1/ resize the r1bio_pool
2919 * 2/ resize conf->mirrors
2920 *
2921 * We allocate a new r1bio_pool if we can.
2922 * Then raise a device barrier and wait until all IO stops.
2923 * Then resize conf->mirrors and swap in the new r1bio pool.
2924 *
2925 * At the same time, we "pack" the devices so that all the missing
2926 * devices have the higher raid_disk numbers.
2927 */
2936 /* Cannot change chunk_size, layout, or level */
2944 }
2956 cnt++;
2959 }
2972 }
2974 GFP_KERNEL);
2979 }
2983 /* ok, everything is stopped */
2997 }
3000 }
3019 }
3022 {
3035 }
3036 }
3039 {
3040 /* raid1 can take over:
3041 * raid5 with 2 devices, any layout or chunk size
3042 */
3052 }
3054 }
3057 {
3075 };
3078 {
3080 }
3083 {
3085 }