| blob | 636bae0405e8167edcec328e5759b3f949dd0876 |
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 }
340 /*
341 * oops, read error:
342 */
345 KERN_ERR "md/raid1:%s: %s: "
349 b),
353 /* don't drop the reference on read_disk yet */
354 }
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 /* Just ignore it */
788 else
791 }
794 }
796 /* Barriers....
797 * Sometimes we need to suspend IO while we do something else,
798 * either some resync/recovery, or reconfigure the array.
799 * To do this we raise a 'barrier'.
800 * The 'barrier' is a counter that can be raised multiple times
801 * to count how many activities are happening which preclude
802 * normal IO.
803 * We can only raise the barrier if there is no pending IO.
804 * i.e. if nr_pending == 0.
805 * We choose only to raise the barrier if no-one is waiting for the
806 * barrier to go down. This means that as soon as an IO request
807 * is ready, no other operations which require a barrier will start
808 * until the IO request has had a chance.
809 *
810 * So: regular IO calls 'wait_barrier'. When that returns there
811 * is no backgroup IO happening, It must arrange to call
812 * allow_barrier when it has finished its IO.
813 * backgroup IO calls must call raise_barrier. Once that returns
814 * there is no normal IO happeing. It must arrange to call
815 * lower_barrier when the particular background IO completes.
816 */
817 #define RESYNC_DEPTH 32
820 {
823 /* Wait until no block IO is waiting */
827 /* block any new IO from starting */
830 /* Now wait for all pending IO to complete */
836 }
839 {
846 }
849 {
853 /* Wait for the barrier to drop.
854 * However if there are already pending
855 * requests (preventing the barrier from
856 * rising completely), and the
857 * pre-process bio queue isn't empty,
858 * then don't wait, as we need to empty
859 * that queue to get the nr_pending
860 * count down.
861 */
868 );
870 }
873 }
876 {
882 }
885 {
886 /* stop syncio and normal IO and wait for everything to
887 * go quite.
888 * We increment barrier and nr_waiting, and then
889 * wait until nr_pending match nr_queued+1
890 * This is called in the context of one normal IO request
891 * that has failed. Thus any sync request that might be pending
892 * will be blocked by nr_pending, and we need to wait for
893 * pending IO requests to complete or be queued for re-try.
894 * Thus the number queued (nr_queued) plus this request (1)
895 * must match the number of pending IOs (nr_pending) before
896 * we continue.
897 */
906 }
908 {
909 /* reverse the effect of the freeze */
915 }
918 /* duplicate the data pages for behind I/O
919 */
921 {
925 GFP_NOIO);
938 }
944 do_sync_io:
950 }
956 };
959 {
961 cb);
974 }
976 /* we aren't scheduling, so we can do the write-out directly. */
986 }
988 }
991 {
1011 /*
1012 * Register the new request and wait if the reconstruction
1013 * thread has put up a bar for new requests.
1014 * Continue immediately if no resync is active currently.
1015 */
1022 /* As the suspend_* range is controlled by
1023 * userspace, we want an interruptible
1024 * wait.
1025 */
1035 }
1037 }
1043 /*
1044 * make_request() can abort the operation when READA is being
1045 * used and no empty request is available.
1046 *
1047 */
1056 /* We might need to issue multiple reads to different
1057 * devices if there are bad blocks around, so we keep
1058 * track of the number of reads in bio->bi_phys_segments.
1059 * If this is 0, there is only one r1_bio and no locking
1060 * will be needed when requests complete. If it is
1061 * non-zero, then it is the number of not-completed requests.
1062 */
1067 /*
1068 * read balancing logic:
1069 */
1072 read_again:
1076 /* couldn't find anywhere to read from */
1079 }
1083 bitmap) {
1084 /* Reading from a write-mostly device must
1085 * take care not to over-take any writes
1086 * that are 'behind'
1087 */
1090 }
1095 max_sectors);
1106 /* could not read all from this device, so we will
1107 * need another r1_bio.
1108 */
1116 else
1119 /* Cannot call generic_make_request directly
1120 * as that will be queued in __make_request
1121 * and subsequent mempool_alloc might block waiting
1122 * for it. So hand bio over to raid1d.
1123 */
1137 }
1139 /*
1140 * WRITE:
1141 */
1146 }
1147 /* first select target devices under rcu_lock and
1148 * inc refcount on their rdev. Record them by setting
1149 * bios[x] to bio
1150 * If there are known/acknowledged bad blocks on any device on
1151 * which we have seen a write error, we want to avoid writing those
1152 * blocks.
1153 * This potentially requires several writes to write around
1154 * the bad blocks. Each set of writes gets it's own r1bio
1155 * with a set of bios attached.
1156 */
1159 retry_write:
1169 }
1176 }
1180 sector_t first_bad;
1185 max_sectors,
1188 /* mustn't write here until the bad block is
1189 * acknowledged*/
1193 }
1195 /* Cannot write here at all */
1198 /* mustn't write more than bad_sectors
1199 * to other devices yet
1200 */
1203 /* We don't set R1BIO_Degraded as that
1204 * only applies if the disk is
1205 * missing, so it might be re-added,
1206 * and we want to know to recover this
1207 * chunk.
1208 * In this case the device is here,
1209 * and the fact that this chunk is not
1210 * in-sync is recorded in the bad
1211 * block log
1212 */
1214 }
1219 }
1220 }
1222 }
1226 /* Wait for this device to become unblocked */
1237 }
1240 /* We are splitting this write into multiple parts, so
1241 * we need to prepare for allocating another r1_bio.
1242 */
1247 else
1250 }
1266 /* do behind I/O ?
1267 * Not if there are too many, or cannot
1268 * allocate memory, or a reader on WriteMostly
1269 * is waiting for behind writes to flush */
1281 }
1286 /* Yes, I really want the '__' version so that
1287 * we clear any unused pointer in the io_vec, rather
1288 * than leave them unchanged. This is important
1289 * because when we come to free the pages, we won't
1290 * know the original bi_idx, so we just free
1291 * them all
1292 */
1297 }
1313 else
1322 }
1326 }
1327 /* Mustn't call r1_bio_write_done before this next test,
1328 * as it could result in the bio being freed.
1329 */
1332 /* We need another r1_bio. It has already been counted
1333 * in bio->bi_phys_segments
1334 */
1342 }
1346 /* In case raid1d snuck in to freeze_array */
1348 }
1351 {
1362 }
1365 }
1369 {
1373 /*
1374 * If it is not operational, then we have already marked it as dead
1375 * else if it is the last working disks, ignore the error, let the
1376 * next level up know.
1377 * else mark the drive as failed
1378 */
1381 /*
1382 * Don't fail the drive, act as though we were just a
1383 * normal single drive.
1384 * However don't try a recovery from this drive as
1385 * it is very likely to fail.
1386 */
1389 }
1397 /*
1398 * if recovery is running, make sure it aborts.
1399 */
1409 }
1412 {
1419 }
1432 }
1434 }
1437 {
1443 }
1446 {
1452 /*
1453 * Find all failed disks within the RAID1 configuration
1454 * and mark them readable.
1455 * Called under mddev lock, so rcu protection not needed.
1456 */
1464 /* replacement has just become active */
1467 count++;
1469 /* Replaced device not technically
1470 * faulty, but we need to be sure
1471 * it gets removed and never re-added
1472 */
1476 }
1477 }
1481 count++;
1483 }
1484 }
1491 }
1495 {
1513 }
1525 /* As all devices are equivalent, we don't need a full recovery
1526 * if this was recently any drive of the array
1527 */
1532 }
1535 /* Add this device as a replacement */
1543 }
1544 }
1546 /* Some requests might not have seen this new
1547 * merge_bvec_fn. We must wait for them to complete
1548 * before merging the device fully.
1549 * First we make sure any code which has tested
1550 * our function has submitted the request, then
1551 * we wait for all outstanding requests to complete.
1552 */
1557 }
1563 }
1566 {
1581 }
1582 /* Only remove non-faulty devices if recovery
1583 * is not possible.
1584 */
1590 }
1594 /* lost the race, try later */
1599 /* We just removed a device that is being replaced.
1600 * Move down the replacement. We drain all IO before
1601 * doing this to avoid confusion.
1602 */
1614 }
1615 abort:
1619 }
1623 {
1628 /*
1629 * we have read a block, now it needs to be re-written,
1630 * or re-read if the read failed.
1631 * We don't do much here, just schedule handling by raid1d
1632 */
1638 }
1641 {
1647 sector_t first_bad;
1656 /* make sure these bits doesn't get cleared. */
1678 )
1689 }
1690 }
1691 }
1695 {
1697 /* success */
1705 }
1706 /* need to record an error - either for the block or the device */
1710 }
1713 {
1714 /* Try some synchronous reads of other devices to get
1715 * good data, much like with normal read errors. Only
1716 * read into the pages we already have so we don't
1717 * need to re-issue the read request.
1718 * We don't need to freeze the array, because being in an
1719 * active sync request, there is no normal IO, and
1720 * no overlapping syncs.
1721 * We don't need to check is_badblock() again as we
1722 * made sure that anything with a bad block in range
1723 * will have bi_end_io clear.
1724 */
1743 /* No rcu protection needed here devices
1744 * can only be removed when no resync is
1745 * active, and resync is currently active
1746 */
1753 }
1754 }
1755 d++;
1763 /* Cannot read from anywhere, this block is lost.
1764 * Record a bad block on each device. If that doesn't
1765 * work just disable and interrupt the recovery.
1766 * Don't fail devices as that won't really help.
1767 */
1779 }
1787 }
1788 /* Try next page */
1791 idx++;
1793 }
1796 /* write it back and re-read */
1800 d--;
1809 }
1810 }
1815 d--;
1823 }
1826 idx ++;
1827 }
1831 }
1834 {
1835 /* We have read all readable devices. If we haven't
1836 * got the block, then there is no hope left.
1837 * If we have, then we want to do a comparison
1838 * and skip the write if everything is the same.
1839 * If any blocks failed to read, then we need to
1840 * attempt an over-write
1841 */
1854 }
1875 }
1882 /* No need to write to this device. */
1886 }
1887 /* fixup the bio for reuse */
1905 else
1910 PAGE_SIZE);
1911 }
1912 }
1914 }
1917 {
1926 /* ouch - failed to read all of that. */
1933 /*
1934 * schedule writes
1935 */
1951 }
1954 /* if we're here, all write(s) have completed, so clean up */
1962 }
1963 }
1964 }
1966 /*
1967 * This is a kernel thread which:
1968 *
1969 * 1. Retries failed read operations on working mirrors.
1970 * 2. Updates the raid superblock when problems encounter.
1971 * 3. Performs writes following reads for array synchronising.
1972 */
1976 {
1989 /* Note: no rcu protection needed here
1990 * as this is synchronous in the raid1d thread
1991 * which is the thread that might remove
1992 * a device. If raid1d ever becomes multi-threaded....
1993 */
1994 sector_t first_bad;
2008 d++;
2011 }
2015 /* Cannot read from anywhere - mark it bad */
2020 }
2021 /* write it back and re-read */
2026 d--;
2032 }
2038 d--;
2046 "md/raid1:%s: read error corrected "
2052 }
2053 }
2054 }
2057 }
2058 }
2061 {
2063 }
2066 {
2077 }
2080 {
2087 /* bio has the data to be written to device 'i' where
2088 * we just recently had a write error.
2089 * We repeatedly clone the bio and trim down to one block,
2090 * then try the write. Where the write fails we record
2091 * a bad block.
2092 * It is conceivable that the bio doesn't exactly align with
2093 * blocks. We must handle this somehow.
2094 *
2095 * We currently own a reference on the rdev.
2096 */
2099 sector_t sector;
2118 idx++;
2123 }
2128 /* Write at 'sector' for 'sectors'*/
2142 /* failure! */
2151 }
2153 }
2156 {
2167 }
2172 }
2173 }
2176 }
2179 {
2189 /* This drive got a write error. We need to
2190 * narrow down and record precise write
2191 * errors.
2192 */
2196 /* an I/O failed, we can't clear the bitmap */
2198 }
2201 }
2205 }
2208 {
2217 /* we got a read error. Maybe the drive is bad. Maybe just
2218 * the block and we can fix it.
2219 * We freeze all other IO, and try reading the block from
2220 * other devices. When we find one, we re-write
2221 * and check it that fixes the read error.
2222 * This is all done synchronously while the array is
2223 * frozen
2224 */
2236 read_more:
2244 const unsigned long do_sync
2250 }
2257 "md/raid1:%s: redirecting sector %llu"
2268 /* Drat - have to split this up more */
2276 else
2295 }
2296 }
2299 {
2318 }
2330 else
2337 else
2338 /* just a partial read to be scheduled from separate
2339 * context
2340 */
2346 }
2348 }
2352 {
2363 }
2365 /*
2366 * perform a "sync" on one "block"
2367 *
2368 * We need to make sure that no normal I/O request - particularly write
2369 * requests - conflict with active sync requests.
2370 *
2371 * This is achieved by tracking pending requests and a 'barrier' concept
2372 * that can be installed to exclude normal IO requests.
2373 */
2375 static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
2376 {
2385 sector_t sync_blocks;
2396 /* If we aborted, we need to abort the
2397 * sync on the 'current' bitmap chunk (there will
2398 * only be one in raid1 resync.
2399 * We can find the current addess in mddev->curr_resync
2400 */
2410 }
2418 }
2419 /* before building a request, check if we can skip these blocks..
2420 * This call the bitmap_start_sync doesn't actually record anything
2421 */
2424 /* We can skip this block, and probably several more */
2427 }
2428 /*
2429 * If there is non-resync activity waiting for a turn,
2430 * and resync is going fast enough,
2431 * then let it though before starting on this new sync request.
2432 */
2443 /*
2444 * If we get a correctably read error during resync or recovery,
2445 * we might want to read from a different device. So we
2446 * flag all drives that could conceivably be read from for READ,
2447 * and any others (which will be non-In_sync devices) for WRITE.
2448 * If a read fails, we try reading from something else for which READ
2449 * is OK.
2450 */
2461 /* take from bio_init */
2481 write_targets ++;
2483 /* may need to read from here */
2496 }
2497 }
2505 }
2508 read_targets++;
2512 /*
2513 * The device is suitable for reading (InSync),
2514 * but has bad block(s) here. Let's try to correct them,
2515 * if we are doing resync or repair. Otherwise, leave
2516 * this device alone for this sync request.
2517 */
2520 write_targets++;
2521 }
2522 }
2528 }
2529 }
2536 /* These sectors are bad on all InSync devices, so we
2537 * need to mark them bad on all write targets
2538 */
2546 }
2552 /* Cannot record the badblocks, so need to
2553 * abort the resync.
2554 * If there are multiple read targets, could just
2555 * fail the really bad ones ???
2556 */
2563 }
2565 /* only resync enough to reach the next bad->good
2566 * transition */
2568 }
2571 /* extra read targets are also write targets */
2575 /* There is nowhere to write, so all non-sync
2576 * drives must be failed - so we are finished
2577 */
2578 sector_t rv;
2585 }
2609 }
2616 /* stop here */
2619 i--;
2623 /* remove last page from this bio */
2627 }
2629 }
2630 }
2631 }
2636 bio_full:
2639 /* For a user-requested sync, we read all readable devices and do a
2640 * compare
2641 */
2647 read_targets--;
2650 }
2651 }
2658 }
2660 }
2663 {
2668 }
2671 {
2684 GFP_KERNEL);
2697 r1bio_pool_free,
2714 else
2726 }
2745 /* This slot has a replacement. */
2747 /* No original, just make the replacement
2748 * a recovering spare
2749 */
2754 /* Original is not in_sync - bad */
2756 }
2764 }
2765 }
2774 }
2778 abort:
2786 }
2788 }
2792 {
2803 }
2808 }
2809 /*
2810 * copy the already verified devices into our private RAID1
2811 * bookkeeping area. [whatever we allocate in run(),
2812 * should be freed in stop()]
2813 */
2816 else
2829 }
2850 /*
2851 * Ok, everything is just fine now
2852 */
2867 else
2870 }
2876 }
2879 {
2883 /* wait for behind writes to complete */
2887 /* need to kick something here to make sure I/O goes? */
2890 }
2903 }
2906 {
2907 /* no resync is happening, and there is enough space
2908 * on all devices, so we can resize.
2909 * We need to make sure resync covers any new space.
2910 * If the array is shrinking we should possibly wait until
2911 * any io in the removed space completes, but it hardly seems
2912 * worth it.
2913 */
2922 }
2930 }
2934 }
2937 {
2938 /* We need to:
2939 * 1/ resize the r1bio_pool
2940 * 2/ resize conf->mirrors
2941 *
2942 * We allocate a new r1bio_pool if we can.
2943 * Then raise a device barrier and wait until all IO stops.
2944 * Then resize conf->mirrors and swap in the new r1bio pool.
2945 *
2946 * At the same time, we "pack" the devices so that all the missing
2947 * devices have the higher raid_disk numbers.
2948 */
2957 /* Cannot change chunk_size, layout, or level */
2965 }
2977 cnt++;
2980 }
2993 }
2995 GFP_KERNEL);
3000 }
3004 /* ok, everything is stopped */
3018 }
3021 }
3040 }
3043 {
3056 }
3057 }
3060 {
3061 /* raid1 can take over:
3062 * raid5 with 2 devices, any layout or chunk size
3063 */
3073 }
3075 }
3078 {
3096 };
3099 {
3101 }
3104 {
3106 }