| blob | 1e5a540995e932852df5ff484a96bfcc8636a432 |
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 */
70 sector_t bi_sector);
74 {
78 /* allocate a r1bio with room for raid_disks entries in the bios array */
80 }
83 {
85 }
87 #define RESYNC_BLOCK_SIZE (64*1024)
88 #define RESYNC_DEPTH 32
89 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
90 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
91 #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH)
92 #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9)
93 #define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS)
96 {
106 /*
107 * Allocate bios : 1 for reading, n-1 for writing
108 */
114 }
115 /*
116 * Allocate RESYNC_PAGES data pages and attach them to
117 * the first bio.
118 * If this is a user-requested check/repair, allocate
119 * RESYNC_PAGES for each bio.
120 */
123 else
131 }
132 /* If not user-requests, copy the page pointers to all bios */
138 }
144 out_free_bio:
149 }
152 {
163 }
168 }
171 {
179 }
180 }
183 {
188 }
191 {
199 }
204 }
207 {
219 }
221 /*
222 * raid_end_bio_io() is called when we have finished servicing a mirrored
223 * operation and are ready to return a success/failure code to the buffer
224 * cache layer.
225 */
227 {
240 /*
241 * make_request() might be waiting for
242 * bi_phys_segments to decrease
243 */
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;
430 /*
431 * Do not set R1BIO_Uptodate if the current device is
432 * rebuilding or Faulty. This is because we cannot use
433 * such device for properly reading the data back (we could
434 * potentially use it, if the current write would have felt
435 * before rdev->recovery_offset, but for simplicity we don't
436 * check this here.
437 */
442 /* Maybe we can clear some bad blocks. */
448 }
449 }
455 /*
456 * In behind mode, we ACK the master bio once the I/O
457 * has safely reached all non-writemostly
458 * disks. Setting the Returned bit ensures that this
459 * gets done only once -- we don't ever want to return
460 * -EIO here, instead we'll wait
461 */
464 /* Maybe we can return now */
473 }
474 }
475 }
480 /*
481 * Let's see if all mirrored write operations have finished
482 * already.
483 */
488 }
491 /*
492 * This routine returns the disk from which the requested read should
493 * be done. There is a per-array 'next expected sequential IO' sector
494 * number - if this matches on the next IO then we use the last disk.
495 * There is also a per-disk 'last know head position' sector that is
496 * maintained from IRQ contexts, both the normal and the resync IO
497 * completion handlers update this position correctly. If there is no
498 * perfect sequential match then we pick the disk whose head is closest.
499 *
500 * If there are 2 mirrors in the same 2 devices, performance degrades
501 * because position is mirror, not device based.
502 *
503 * The rdev for the device selected will have nr_pending incremented.
504 */
506 {
513 sector_t best_dist;
520 /*
521 * Check if we can balance. We can balance on the whole
522 * device if no resync is going on, or below the resync window.
523 * We take the first readable disk when above the resync window.
524 */
525 retry:
539 else
543 sector_t dist;
544 sector_t first_bad;
559 /* Don't balance among write-mostly, just
560 * use the first as a last resort */
565 /* 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 */
828 {
831 /* Wait until no block IO is waiting */
835 /* block any new IO from starting */
838 /* For these conditions we must wait:
839 * A: while the array is in frozen state
840 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
841 * the max count which allowed.
842 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
843 * next resync will reach to the window which normal bios are
844 * handling.
845 */
854 }
857 {
864 }
867 {
880 else
882 }
885 }
888 {
894 /* Wait for the barrier to drop.
895 * However if there are already pending
896 * requests (preventing the barrier from
897 * rising completely), and the
898 * pre-process bio queue isn't empty,
899 * then don't wait, as we need to empty
900 * that queue to get the nr_pending
901 * count down.
902 */
912 }
920 NEXT_NORMALIO_DISTANCE;
925 else
927 }
930 }
935 }
938 sector_t bi_sector)
939 {
949 else
960 NEXT_NORMALIO_DISTANCE;
963 }
964 }
967 }
970 {
971 /* stop syncio and normal IO and wait for everything to
972 * go quite.
973 * We wait until nr_pending match nr_queued+extra
974 * This is called in the context of one normal IO request
975 * that has failed. Thus any sync request that might be pending
976 * will be blocked by nr_pending, and we need to wait for
977 * pending IO requests to complete or be queued for re-try.
978 * Thus the number queued (nr_queued) plus this request (extra)
979 * must match the number of pending IOs (nr_pending) before
980 * we continue.
981 */
989 }
991 {
992 /* reverse the effect of the freeze */
997 }
1000 /* duplicate the data pages for behind I/O
1001 */
1003 {
1007 GFP_NOIO);
1020 }
1026 do_sync_io:
1032 }
1038 };
1041 {
1043 cb);
1057 }
1059 /* we aren't scheduling, so we can do the write-out directly. */
1069 /* Just ignore it */
1071 else
1074 }
1076 }
1079 {
1099 sector_t start_next_window;
1101 /*
1102 * Register the new request and wait if the reconstruction
1103 * thread has put up a bar for new requests.
1104 * Continue immediately if no resync is active currently.
1105 */
1112 /* As the suspend_* range is controlled by
1113 * userspace, we want an interruptible
1114 * wait.
1115 */
1125 }
1127 }
1133 /*
1134 * make_request() can abort the operation when READA is being
1135 * used and no empty request is available.
1136 *
1137 */
1146 /* We might need to issue multiple reads to different
1147 * devices if there are bad blocks around, so we keep
1148 * track of the number of reads in bio->bi_phys_segments.
1149 * If this is 0, there is only one r1_bio and no locking
1150 * will be needed when requests complete. If it is
1151 * non-zero, then it is the number of not-completed requests.
1152 */
1157 /*
1158 * read balancing logic:
1159 */
1162 read_again:
1166 /* couldn't find anywhere to read from */
1169 }
1173 bitmap) {
1174 /* Reading from a write-mostly device must
1175 * take care not to over-take any writes
1176 * that are 'behind'
1177 */
1180 }
1185 max_sectors);
1196 /* could not read all from this device, so we will
1197 * need another r1_bio.
1198 */
1206 else
1209 /* Cannot call generic_make_request directly
1210 * as that will be queued in __make_request
1211 * and subsequent mempool_alloc might block waiting
1212 * for it. So hand bio over to raid1d.
1213 */
1227 }
1229 /*
1230 * WRITE:
1231 */
1236 }
1237 /* first select target devices under rcu_lock and
1238 * inc refcount on their rdev. Record them by setting
1239 * bios[x] to bio
1240 * If there are known/acknowledged bad blocks on any device on
1241 * which we have seen a write error, we want to avoid writing those
1242 * blocks.
1243 * This potentially requires several writes to write around
1244 * the bad blocks. Each set of writes gets it's own r1bio
1245 * with a set of bios attached.
1246 */
1249 retry_write:
1260 }
1267 }
1271 sector_t first_bad;
1276 max_sectors,
1279 /* mustn't write here until the bad block is
1280 * acknowledged*/
1284 }
1286 /* Cannot write here at all */
1289 /* mustn't write more than bad_sectors
1290 * to other devices yet
1291 */
1294 /* We don't set R1BIO_Degraded as that
1295 * only applies if the disk is
1296 * missing, so it might be re-added,
1297 * and we want to know to recover this
1298 * chunk.
1299 * In this case the device is here,
1300 * and the fact that this chunk is not
1301 * in-sync is recorded in the bad
1302 * block log
1303 */
1305 }
1310 }
1311 }
1313 }
1317 /* Wait for this device to become unblocked */
1328 /*
1329 * We must make sure the multi r1bios of bio have
1330 * the same value of bi_phys_segments
1331 */
1334 /* Wait for the former r1bio(s) to complete */
1338 }
1341 /* We are splitting this write into multiple parts, so
1342 * we need to prepare for allocating another r1_bio.
1343 */
1348 else
1351 }
1367 /* do behind I/O ?
1368 * Not if there are too many, or cannot
1369 * allocate memory, or a reader on WriteMostly
1370 * is waiting for behind writes to flush */
1382 }
1387 /*
1388 * We trimmed the bio, so _all is legit
1389 */
1394 }
1411 else
1420 }
1424 }
1425 /* Mustn't call r1_bio_write_done before this next test,
1426 * as it could result in the bio being freed.
1427 */
1430 /* We need another r1_bio. It has already been counted
1431 * in bio->bi_phys_segments
1432 */
1440 }
1444 /* In case raid1d snuck in to freeze_array */
1446 }
1449 {
1460 }
1463 }
1467 {
1471 /*
1472 * If it is not operational, then we have already marked it as dead
1473 * else if it is the last working disks, ignore the error, let the
1474 * next level up know.
1475 * else mark the drive as failed
1476 */
1479 /*
1480 * Don't fail the drive, act as though we were just a
1481 * normal single drive.
1482 * However don't try a recovery from this drive as
1483 * it is very likely to fail.
1484 */
1487 }
1495 /*
1496 * if recovery is running, make sure it aborts.
1497 */
1507 }
1510 {
1517 }
1530 }
1532 }
1535 {
1544 }
1547 {
1553 /*
1554 * Find all failed disks within the RAID1 configuration
1555 * and mark them readable.
1556 * Called under mddev lock, so rcu protection not needed.
1557 */
1565 /* replacement has just become active */
1568 count++;
1570 /* Replaced device not technically
1571 * faulty, but we need to be sure
1572 * it gets removed and never re-added
1573 */
1577 }
1578 }
1583 count++;
1585 }
1586 }
1593 }
1597 {
1615 }
1628 /* As all devices are equivalent, we don't need a full recovery
1629 * if this was recently any drive of the array
1630 */
1635 }
1638 /* Add this device as a replacement */
1646 }
1647 }
1649 /* Some requests might not have seen this new
1650 * merge_bvec_fn. We must wait for them to complete
1651 * before merging the device fully.
1652 * First we make sure any code which has tested
1653 * our function has submitted the request, then
1654 * we wait for all outstanding requests to complete.
1655 */
1660 }
1666 }
1669 {
1684 }
1685 /* Only remove non-faulty devices if recovery
1686 * is not possible.
1687 */
1693 }
1697 /* lost the race, try later */
1702 /* We just removed a device that is being replaced.
1703 * Move down the replacement. We drain all IO before
1704 * doing this to avoid confusion.
1705 */
1717 }
1718 abort:
1722 }
1726 {
1731 /*
1732 * we have read a block, now it needs to be re-written,
1733 * or re-read if the read failed.
1734 * We don't do much here, just schedule handling by raid1d
1735 */
1741 }
1744 {
1750 sector_t first_bad;
1759 /* make sure these bits doesn't get cleared. */
1781 )
1792 }
1793 }
1794 }
1798 {
1800 /* success */
1808 }
1809 /* need to record an error - either for the block or the device */
1813 }
1816 {
1817 /* Try some synchronous reads of other devices to get
1818 * good data, much like with normal read errors. Only
1819 * read into the pages we already have so we don't
1820 * need to re-issue the read request.
1821 * We don't need to freeze the array, because being in an
1822 * active sync request, there is no normal IO, and
1823 * no overlapping syncs.
1824 * We don't need to check is_badblock() again as we
1825 * made sure that anything with a bad block in range
1826 * will have bi_end_io clear.
1827 */
1846 /* No rcu protection needed here devices
1847 * can only be removed when no resync is
1848 * active, and resync is currently active
1849 */
1856 }
1857 }
1858 d++;
1866 /* Cannot read from anywhere, this block is lost.
1867 * Record a bad block on each device. If that doesn't
1868 * work just disable and interrupt the recovery.
1869 * Don't fail devices as that won't really help.
1870 */
1882 }
1890 }
1891 /* Try next page */
1894 idx++;
1896 }
1899 /* write it back and re-read */
1903 d--;
1912 }
1913 }
1918 d--;
1926 }
1929 idx ++;
1930 }
1934 }
1937 {
1938 /* We have read all readable devices. If we haven't
1939 * got the block, then there is no hope left.
1940 * If we have, then we want to do a comparison
1941 * and skip the write if everything is the same.
1942 * If any blocks failed to read, then we need to
1943 * attempt an over-write
1944 */
1951 /* Fix variable parts of all bios */
1959 /* fixup the bio for reuse */
1976 else
1979 }
1980 }
1987 }
2006 }
2013 /* No need to write to this device. */
2017 }
2020 }
2022 }
2025 {
2034 /* ouch - failed to read all of that. */
2041 /*
2042 * schedule writes
2043 */
2059 }
2062 /* if we're here, all write(s) have completed, so clean up */
2070 }
2071 }
2072 }
2074 /*
2075 * This is a kernel thread which:
2076 *
2077 * 1. Retries failed read operations on working mirrors.
2078 * 2. Updates the raid superblock when problems encounter.
2079 * 3. Performs writes following reads for array synchronising.
2080 */
2084 {
2097 /* Note: no rcu protection needed here
2098 * as this is synchronous in the raid1d thread
2099 * which is the thread that might remove
2100 * a device. If raid1d ever becomes multi-threaded....
2101 */
2102 sector_t first_bad;
2116 d++;
2119 }
2123 /* Cannot read from anywhere - mark it bad */
2128 }
2129 /* write it back and re-read */
2134 d--;
2140 }
2146 d--;
2154 "md/raid1:%s: read error corrected "
2160 }
2161 }
2162 }
2165 }
2166 }
2169 {
2174 /* bio has the data to be written to device 'i' where
2175 * we just recently had a write error.
2176 * We repeatedly clone the bio and trim down to one block,
2177 * then try the write. Where the write fails we record
2178 * a bad block.
2179 * It is conceivable that the bio doesn't exactly align with
2180 * blocks. We must handle this somehow.
2181 *
2182 * We currently own a reference on the rdev.
2183 */
2186 sector_t sector;
2204 /* Write at 'sector' for 'sectors'*/
2211 vec++;
2212 vcnt--;
2213 }
2221 }
2231 /* failure! */
2240 }
2242 }
2245 {
2256 }
2261 }
2262 }
2265 }
2268 {
2278 /* This drive got a write error. We need to
2279 * narrow down and record precise write
2280 * errors.
2281 */
2285 /* an I/O failed, we can't clear the bitmap */
2287 }
2290 }
2294 }
2297 {
2306 /* we got a read error. Maybe the drive is bad. Maybe just
2307 * the block and we can fix it.
2308 * We freeze all other IO, and try reading the block from
2309 * other devices. When we find one, we re-write
2310 * and check it that fixes the read error.
2311 * This is all done synchronously while the array is
2312 * frozen
2313 */
2325 read_more:
2333 const unsigned long do_sync
2339 }
2346 "md/raid1:%s: redirecting sector %llu"
2357 /* Drat - have to split this up more */
2365 else
2383 }
2384 }
2387 {
2406 }
2418 else
2425 else
2426 /* just a partial read to be scheduled from separate
2427 * context
2428 */
2434 }
2436 }
2440 {
2451 }
2453 /*
2454 * perform a "sync" on one "block"
2455 *
2456 * We need to make sure that no normal I/O request - particularly write
2457 * requests - conflict with active sync requests.
2458 *
2459 * This is achieved by tracking pending requests and a 'barrier' concept
2460 * that can be installed to exclude normal IO requests.
2461 */
2463 static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
2464 {
2473 sector_t sync_blocks;
2484 /* If we aborted, we need to abort the
2485 * sync on the 'current' bitmap chunk (there will
2486 * only be one in raid1 resync.
2487 * We can find the current addess in mddev->curr_resync
2488 */
2498 }
2506 }
2507 /* before building a request, check if we can skip these blocks..
2508 * This call the bitmap_start_sync doesn't actually record anything
2509 */
2512 /* We can skip this block, and probably several more */
2515 }
2516 /*
2517 * If there is non-resync activity waiting for a turn,
2518 * and resync is going fast enough,
2519 * then let it though before starting on this new sync request.
2520 */
2531 /*
2532 * If we get a correctably read error during resync or recovery,
2533 * we might want to read from a different device. So we
2534 * flag all drives that could conceivably be read from for READ,
2535 * and any others (which will be non-In_sync devices) for WRITE.
2536 * If a read fails, we try reading from something else for which READ
2537 * is OK.
2538 */
2558 write_targets ++;
2560 /* may need to read from here */
2573 }
2574 }
2582 }
2585 read_targets++;
2589 /*
2590 * The device is suitable for reading (InSync),
2591 * but has bad block(s) here. Let's try to correct them,
2592 * if we are doing resync or repair. Otherwise, leave
2593 * this device alone for this sync request.
2594 */
2597 write_targets++;
2598 }
2599 }
2605 }
2606 }
2613 /* These sectors are bad on all InSync devices, so we
2614 * need to mark them bad on all write targets
2615 */
2623 }
2629 /* Cannot record the badblocks, so need to
2630 * abort the resync.
2631 * If there are multiple read targets, could just
2632 * fail the really bad ones ???
2633 */
2640 }
2642 /* only resync enough to reach the next bad->good
2643 * transition */
2645 }
2648 /* extra read targets are also write targets */
2652 /* There is nowhere to write, so all non-sync
2653 * drives must be failed - so we are finished
2654 */
2655 sector_t rv;
2662 }
2686 }
2693 /* stop here */
2696 i--;
2700 /* remove last page from this bio */
2704 }
2706 }
2707 }
2708 }
2713 bio_full:
2716 /* For a user-requested sync, we read all readable devices and do a
2717 * compare
2718 */
2724 read_targets--;
2727 }
2728 }
2735 }
2737 }
2740 {
2745 }
2748 {
2761 GFP_KERNEL);
2774 r1bio_pool_free,
2791 else
2803 }
2825 /* This slot has a replacement. */
2827 /* No original, just make the replacement
2828 * a recovering spare
2829 */
2834 /* Original is not in_sync - bad */
2836 }
2844 }
2845 }
2854 }
2858 abort:
2866 }
2868 }
2872 {
2883 }
2888 }
2889 /*
2890 * copy the already verified devices into our private RAID1
2891 * bookkeeping area. [whatever we allocate in run(),
2892 * should be freed in stop()]
2893 */
2896 else
2912 }
2933 /*
2934 * Ok, everything is just fine now
2935 */
2950 else
2953 }
2959 }
2962 {
2966 /* wait for behind writes to complete */
2970 /* need to kick something here to make sure I/O goes? */
2973 }
2987 }
2990 {
2991 /* no resync is happening, and there is enough space
2992 * on all devices, so we can resize.
2993 * We need to make sure resync covers any new space.
2994 * If the array is shrinking we should possibly wait until
2995 * any io in the removed space completes, but it hardly seems
2996 * worth it.
2997 */
3006 }
3014 }
3018 }
3021 {
3022 /* We need to:
3023 * 1/ resize the r1bio_pool
3024 * 2/ resize conf->mirrors
3025 *
3026 * We allocate a new r1bio_pool if we can.
3027 * Then raise a device barrier and wait until all IO stops.
3028 * Then resize conf->mirrors and swap in the new r1bio pool.
3029 *
3030 * At the same time, we "pack" the devices so that all the missing
3031 * devices have the higher raid_disk numbers.
3032 */
3041 /* Cannot change chunk_size, layout, or level */
3049 }
3061 cnt++;
3064 }
3077 }
3079 GFP_KERNEL);
3084 }
3088 /* ok, everything is stopped */
3102 }
3105 }
3124 }
3127 {
3140 }
3141 }
3144 {
3145 /* raid1 can take over:
3146 * raid5 with 2 devices, any layout or chunk size
3147 */
3155 /* Array must appear to be quiesced */
3158 }
3160 }
3163 {
3181 };
3184 {
3186 }
3189 {
3191 }