| blob | 63d42ae56a1ca28c295822b9fee86ef9cddc4d7e |
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 */
571 }
573 }
574 /* This is a reasonable device to use. It might
575 * even be best.
576 */
580 /* already have a better device */
583 /* cannot read here. If this is the 'primary'
584 * device, then we must not read beyond
585 * bad_sectors from another device..
586 */
598 }
601 }
613 }
614 /* Don't change to another disk for sequential reads */
621 /*
622 * If buffered sequential IO size exceeds optimal
623 * iosize, check if there is idle disk. If yes, choose
624 * the idle disk. read_balance could already choose an
625 * idle disk before noticing it's a sequential IO in
626 * this disk. This doesn't matter because this disk
627 * will idle, next time it will be utilized after the
628 * first disk has IO size exceeds optimal iosize. In
629 * this way, iosize of the first disk will be optimal
630 * iosize at least. iosize of the second disk might be
631 * small, but not a big deal since when the second disk
632 * starts IO, the first disk is likely still busy.
633 */
641 }
643 }
644 /* If device is idle, use it */
648 }
656 }
661 }
662 }
664 /*
665 * If all disks are rotational, choose the closest disk. If any disk is
666 * non-rotational, choose the disk with less pending request even the
667 * disk is rotational, which might/might not be optimal for raids with
668 * mixed ratation/non-rotational disks depending on workload.
669 */
673 else
675 }
683 /* cannot risk returning a device that failed
684 * before we inc'ed nr_pending
685 */
688 }
695 }
700 }
705 {
726 }
727 }
728 }
730 }
733 }
736 {
752 /* Note the '|| 1' - when read_balance prefers
753 * non-congested targets, it can be removed
754 */
757 else
759 }
760 }
763 }
767 {
772 }
775 {
776 /* Any writes that have been queued but are awaiting
777 * bitmap updates get flushed here.
778 */
786 /* flush any pending bitmap writes to
787 * disk before proceeding w/ I/O */
796 /* Just ignore it */
798 else
801 }
804 }
806 /* Barriers....
807 * Sometimes we need to suspend IO while we do something else,
808 * either some resync/recovery, or reconfigure the array.
809 * To do this we raise a 'barrier'.
810 * The 'barrier' is a counter that can be raised multiple times
811 * to count how many activities are happening which preclude
812 * normal IO.
813 * We can only raise the barrier if there is no pending IO.
814 * i.e. if nr_pending == 0.
815 * We choose only to raise the barrier if no-one is waiting for the
816 * barrier to go down. This means that as soon as an IO request
817 * is ready, no other operations which require a barrier will start
818 * until the IO request has had a chance.
819 *
820 * So: regular IO calls 'wait_barrier'. When that returns there
821 * is no backgroup IO happening, It must arrange to call
822 * allow_barrier when it has finished its IO.
823 * backgroup IO calls must call raise_barrier. Once that returns
824 * there is no normal IO happeing. It must arrange to call
825 * lower_barrier when the particular background IO completes.
826 */
827 #define RESYNC_DEPTH 32
830 {
833 /* Wait until no block IO is waiting */
837 /* block any new IO from starting */
840 /* Now wait for all pending IO to complete */
846 }
849 {
856 }
859 {
863 /* Wait for the barrier to drop.
864 * However if there are already pending
865 * requests (preventing the barrier from
866 * rising completely), and the
867 * pre-process bio queue isn't empty,
868 * then don't wait, as we need to empty
869 * that queue to get the nr_pending
870 * count down.
871 */
879 }
882 }
885 {
891 }
894 {
895 /* stop syncio and normal IO and wait for everything to
896 * go quite.
897 * We increment barrier and nr_waiting, and then
898 * wait until nr_pending match nr_queued+extra
899 * This is called in the context of one normal IO request
900 * that has failed. Thus any sync request that might be pending
901 * will be blocked by nr_pending, and we need to wait for
902 * pending IO requests to complete or be queued for re-try.
903 * Thus the number queued (nr_queued) plus this request (extra)
904 * must match the number of pending IOs (nr_pending) before
905 * we continue.
906 */
915 }
917 {
918 /* reverse the effect of the freeze */
924 }
927 /* duplicate the data pages for behind I/O
928 */
930 {
934 GFP_NOIO);
947 }
953 do_sync_io:
959 }
965 };
968 {
970 cb);
984 }
986 /* we aren't scheduling, so we can do the write-out directly. */
996 /* Just ignore it */
998 else
1001 }
1003 }
1006 {
1027 /*
1028 * Register the new request and wait if the reconstruction
1029 * thread has put up a bar for new requests.
1030 * Continue immediately if no resync is active currently.
1031 */
1038 /* As the suspend_* range is controlled by
1039 * userspace, we want an interruptible
1040 * wait.
1041 */
1051 }
1053 }
1059 /*
1060 * make_request() can abort the operation when READA is being
1061 * used and no empty request is available.
1062 *
1063 */
1072 /* We might need to issue multiple reads to different
1073 * devices if there are bad blocks around, so we keep
1074 * track of the number of reads in bio->bi_phys_segments.
1075 * If this is 0, there is only one r1_bio and no locking
1076 * will be needed when requests complete. If it is
1077 * non-zero, then it is the number of not-completed requests.
1078 */
1083 /*
1084 * read balancing logic:
1085 */
1088 read_again:
1092 /* couldn't find anywhere to read from */
1095 }
1099 bitmap) {
1100 /* Reading from a write-mostly device must
1101 * take care not to over-take any writes
1102 * that are 'behind'
1103 */
1106 }
1111 max_sectors);
1122 /* could not read all from this device, so we will
1123 * need another r1_bio.
1124 */
1132 else
1135 /* Cannot call generic_make_request directly
1136 * as that will be queued in __make_request
1137 * and subsequent mempool_alloc might block waiting
1138 * for it. So hand bio over to raid1d.
1139 */
1153 }
1155 /*
1156 * WRITE:
1157 */
1162 }
1163 /* first select target devices under rcu_lock and
1164 * inc refcount on their rdev. Record them by setting
1165 * bios[x] to bio
1166 * If there are known/acknowledged bad blocks on any device on
1167 * which we have seen a write error, we want to avoid writing those
1168 * blocks.
1169 * This potentially requires several writes to write around
1170 * the bad blocks. Each set of writes gets it's own r1bio
1171 * with a set of bios attached.
1172 */
1175 retry_write:
1185 }
1192 }
1196 sector_t first_bad;
1201 max_sectors,
1204 /* mustn't write here until the bad block is
1205 * acknowledged*/
1209 }
1211 /* Cannot write here at all */
1214 /* mustn't write more than bad_sectors
1215 * to other devices yet
1216 */
1219 /* We don't set R1BIO_Degraded as that
1220 * only applies if the disk is
1221 * missing, so it might be re-added,
1222 * and we want to know to recover this
1223 * chunk.
1224 * In this case the device is here,
1225 * and the fact that this chunk is not
1226 * in-sync is recorded in the bad
1227 * block log
1228 */
1230 }
1235 }
1236 }
1238 }
1242 /* Wait for this device to become unblocked */
1253 }
1256 /* We are splitting this write into multiple parts, so
1257 * we need to prepare for allocating another r1_bio.
1258 */
1263 else
1266 }
1282 /* do behind I/O ?
1283 * Not if there are too many, or cannot
1284 * allocate memory, or a reader on WriteMostly
1285 * is waiting for behind writes to flush */
1297 }
1302 /*
1303 * We trimmed the bio, so _all is legit
1304 */
1309 }
1326 else
1335 }
1339 }
1340 /* Mustn't call r1_bio_write_done before this next test,
1341 * as it could result in the bio being freed.
1342 */
1345 /* We need another r1_bio. It has already been counted
1346 * in bio->bi_phys_segments
1347 */
1355 }
1359 /* In case raid1d snuck in to freeze_array */
1361 }
1364 {
1375 }
1378 }
1382 {
1387 /*
1388 * If it is not operational, then we have already marked it as dead
1389 * else if it is the last working disks, ignore the error, let the
1390 * next level up know.
1391 * else mark the drive as failed
1392 */
1395 /*
1396 * Don't fail the drive, act as though we were just a
1397 * normal single drive.
1398 * However don't try a recovery from this drive as
1399 * it is very likely to fail.
1400 */
1403 }
1412 /*
1413 * if recovery is running, make sure it aborts.
1414 */
1422 }
1425 {
1432 }
1445 }
1447 }
1450 {
1456 }
1459 {
1465 /*
1466 * Find all failed disks within the RAID1 configuration
1467 * and mark them readable.
1468 * Called under mddev lock, so rcu protection not needed.
1469 * device_lock used to avoid races with raid1_end_read_request
1470 * which expects 'In_sync' flags and ->degraded to be consistent.
1471 */
1480 /* replacement has just become active */
1483 count++;
1485 /* Replaced device not technically
1486 * faulty, but we need to be sure
1487 * it gets removed and never re-added
1488 */
1492 }
1493 }
1498 count++;
1500 }
1501 }
1507 }
1511 {
1529 }
1541 /* As all devices are equivalent, we don't need a full recovery
1542 * if this was recently any drive of the array
1543 */
1548 }
1551 /* Add this device as a replacement */
1559 }
1560 }
1562 /* Some requests might not have seen this new
1563 * merge_bvec_fn. We must wait for them to complete
1564 * before merging the device fully.
1565 * First we make sure any code which has tested
1566 * our function has submitted the request, then
1567 * we wait for all outstanding requests to complete.
1568 */
1573 }
1579 }
1582 {
1597 }
1598 /* Only remove non-faulty devices if recovery
1599 * is not possible.
1600 */
1606 }
1610 /* lost the race, try later */
1615 /* We just removed a device that is being replaced.
1616 * Move down the replacement. We drain all IO before
1617 * doing this to avoid confusion.
1618 */
1630 }
1631 abort:
1635 }
1639 {
1644 /*
1645 * we have read a block, now it needs to be re-written,
1646 * or re-read if the read failed.
1647 * We don't do much here, just schedule handling by raid1d
1648 */
1654 }
1657 {
1663 sector_t first_bad;
1672 /* make sure these bits doesn't get cleared. */
1694 )
1705 }
1706 }
1707 }
1711 {
1713 /* success */
1721 }
1722 /* need to record an error - either for the block or the device */
1726 }
1729 {
1730 /* Try some synchronous reads of other devices to get
1731 * good data, much like with normal read errors. Only
1732 * read into the pages we already have so we don't
1733 * need to re-issue the read request.
1734 * We don't need to freeze the array, because being in an
1735 * active sync request, there is no normal IO, and
1736 * no overlapping syncs.
1737 * We don't need to check is_badblock() again as we
1738 * made sure that anything with a bad block in range
1739 * will have bi_end_io clear.
1740 */
1759 /* No rcu protection needed here devices
1760 * can only be removed when no resync is
1761 * active, and resync is currently active
1762 */
1769 }
1770 }
1771 d++;
1779 /* Cannot read from anywhere, this block is lost.
1780 * Record a bad block on each device. If that doesn't
1781 * work just disable and interrupt the recovery.
1782 * Don't fail devices as that won't really help.
1783 */
1795 }
1803 }
1804 /* Try next page */
1807 idx++;
1809 }
1812 /* write it back and re-read */
1816 d--;
1825 }
1826 }
1831 d--;
1839 }
1842 idx ++;
1843 }
1847 }
1850 {
1851 /* We have read all readable devices. If we haven't
1852 * got the block, then there is no hope left.
1853 * If we have, then we want to do a comparison
1854 * and skip the write if everything is the same.
1855 * If any blocks failed to read, then we need to
1856 * attempt an over-write
1857 */
1864 /* Fix variable parts of all bios */
1873 /* fixup the bio for reuse, but preserve BIO_UPTODATE */
1893 else
1896 }
1897 }
1904 }
1914 /* Now we can 'fixup' the BIO_UPTODATE flag */
1926 }
1933 /* No need to write to this device. */
1937 }
1940 }
1942 }
1945 {
1954 /* ouch - failed to read all of that. */
1961 /*
1962 * schedule writes
1963 */
1979 }
1982 /* if we're here, all write(s) have completed, so clean up */
1990 }
1991 }
1992 }
1994 /*
1995 * This is a kernel thread which:
1996 *
1997 * 1. Retries failed read operations on working mirrors.
1998 * 2. Updates the raid superblock when problems encounter.
1999 * 3. Performs writes following reads for array synchronising.
2000 */
2004 {
2017 /* Note: no rcu protection needed here
2018 * as this is synchronous in the raid1d thread
2019 * which is the thread that might remove
2020 * a device. If raid1d ever becomes multi-threaded....
2021 */
2022 sector_t first_bad;
2036 d++;
2039 }
2043 /* Cannot read from anywhere - mark it bad */
2048 }
2049 /* write it back and re-read */
2054 d--;
2060 }
2066 d--;
2074 "md/raid1:%s: read error corrected "
2080 }
2081 }
2082 }
2085 }
2086 }
2089 {
2094 /* bio has the data to be written to device 'i' where
2095 * we just recently had a write error.
2096 * We repeatedly clone the bio and trim down to one block,
2097 * then try the write. Where the write fails we record
2098 * a bad block.
2099 * It is conceivable that the bio doesn't exactly align with
2100 * blocks. We must handle this somehow.
2101 *
2102 * We currently own a reference on the rdev.
2103 */
2106 sector_t sector;
2124 /* Write at 'sector' for 'sectors'*/
2131 vec++;
2132 vcnt--;
2133 }
2141 }
2151 /* failure! */
2160 }
2162 }
2165 {
2176 }
2181 }
2182 }
2185 }
2188 {
2198 /* This drive got a write error. We need to
2199 * narrow down and record precise write
2200 * errors.
2201 */
2205 /* an I/O failed, we can't clear the bitmap */
2207 }
2210 }
2214 }
2217 {
2226 /* we got a read error. Maybe the drive is bad. Maybe just
2227 * the block and we can fix it.
2228 * We freeze all other IO, and try reading the block from
2229 * other devices. When we find one, we re-write
2230 * and check it that fixes the read error.
2231 * This is all done synchronously while the array is
2232 * frozen
2233 */
2245 read_more:
2253 const unsigned long do_sync
2259 }
2266 "md/raid1:%s: redirecting sector %llu"
2277 /* Drat - have to split this up more */
2285 else
2303 }
2304 }
2307 {
2326 }
2338 else
2345 else
2346 /* just a partial read to be scheduled from separate
2347 * context
2348 */
2354 }
2356 }
2360 {
2371 }
2373 /*
2374 * perform a "sync" on one "block"
2375 *
2376 * We need to make sure that no normal I/O request - particularly write
2377 * requests - conflict with active sync requests.
2378 *
2379 * This is achieved by tracking pending requests and a 'barrier' concept
2380 * that can be installed to exclude normal IO requests.
2381 */
2383 static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
2384 {
2393 sector_t sync_blocks;
2404 /* If we aborted, we need to abort the
2405 * sync on the 'current' bitmap chunk (there will
2406 * only be one in raid1 resync.
2407 * We can find the current addess in mddev->curr_resync
2408 */
2418 }
2426 }
2427 /* before building a request, check if we can skip these blocks..
2428 * This call the bitmap_start_sync doesn't actually record anything
2429 */
2432 /* We can skip this block, and probably several more */
2435 }
2436 /*
2437 * If there is non-resync activity waiting for a turn,
2438 * and resync is going fast enough,
2439 * then let it though before starting on this new sync request.
2440 */
2451 /*
2452 * If we get a correctably read error during resync or recovery,
2453 * we might want to read from a different device. So we
2454 * flag all drives that could conceivably be read from for READ,
2455 * and any others (which will be non-In_sync devices) for WRITE.
2456 * If a read fails, we try reading from something else for which READ
2457 * is OK.
2458 */
2478 write_targets ++;
2480 /* may need to read from here */
2493 }
2494 }
2502 }
2505 read_targets++;
2509 /*
2510 * The device is suitable for reading (InSync),
2511 * but has bad block(s) here. Let's try to correct them,
2512 * if we are doing resync or repair. Otherwise, leave
2513 * this device alone for this sync request.
2514 */
2517 write_targets++;
2518 }
2519 }
2525 }
2526 }
2533 /* These sectors are bad on all InSync devices, so we
2534 * need to mark them bad on all write targets
2535 */
2543 }
2549 /* Cannot record the badblocks, so need to
2550 * abort the resync.
2551 * If there are multiple read targets, could just
2552 * fail the really bad ones ???
2553 */
2560 }
2562 /* only resync enough to reach the next bad->good
2563 * transition */
2565 }
2568 /* extra read targets are also write targets */
2572 /* There is nowhere to write, so all non-sync
2573 * drives must be failed - so we are finished
2574 */
2575 sector_t rv;
2582 }
2606 }
2613 /* stop here */
2616 i--;
2620 /* remove last page from this bio */
2624 }
2626 }
2627 }
2628 }
2633 bio_full:
2636 /* For a user-requested sync, we read all readable devices and do a
2637 * compare
2638 */
2644 read_targets--;
2647 }
2648 }
2655 }
2657 }
2660 {
2665 }
2668 {
2681 GFP_KERNEL);
2694 r1bio_pool_free,
2711 else
2723 }
2742 /* This slot has a replacement. */
2744 /* No original, just make the replacement
2745 * a recovering spare
2746 */
2751 /* Original is not in_sync - bad */
2753 }
2761 }
2762 }
2771 }
2775 abort:
2783 }
2785 }
2789 {
2800 }
2805 }
2806 /*
2807 * copy the already verified devices into our private RAID1
2808 * bookkeeping area. [whatever we allocate in run(),
2809 * should be freed in stop()]
2810 */
2813 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 }
2904 }
2907 {
2908 /* no resync is happening, and there is enough space
2909 * on all devices, so we can resize.
2910 * We need to make sure resync covers any new space.
2911 * If the array is shrinking we should possibly wait until
2912 * any io in the removed space completes, but it hardly seems
2913 * worth it.
2914 */
2923 }
2931 }
2935 }
2938 {
2939 /* We need to:
2940 * 1/ resize the r1bio_pool
2941 * 2/ resize conf->mirrors
2942 *
2943 * We allocate a new r1bio_pool if we can.
2944 * Then raise a device barrier and wait until all IO stops.
2945 * Then resize conf->mirrors and swap in the new r1bio pool.
2946 *
2947 * At the same time, we "pack" the devices so that all the missing
2948 * devices have the higher raid_disk numbers.
2949 */
2958 /* Cannot change chunk_size, layout, or level */
2966 }
2978 cnt++;
2981 }
2994 }
2996 GFP_KERNEL);
3001 }
3005 /* ok, everything is stopped */
3019 }
3022 }
3041 }
3044 {
3057 }
3058 }
3061 {
3062 /* raid1 can take over:
3063 * raid5 with 2 devices, any layout or chunk size
3064 */
3074 }
3076 }
3079 {
3097 };
3100 {
3102 }
3105 {
3107 }