| blob | d34e238afa54c24ccaefbc7c6d58974dc2104be6 |
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 {
107 /*
108 * Allocate bios : 1 for reading, n-1 for writing
109 */
115 }
116 /*
117 * Allocate RESYNC_PAGES data pages and attach them to
118 * the first bio.
119 * If this is a user-requested check/repair, allocate
120 * RESYNC_PAGES for each bio.
121 */
124 else
132 }
133 /* If not user-requests, copy the page pointers to all bios */
139 }
145 out_free_pages:
151 }
153 out_free_bio:
158 }
161 {
172 }
177 }
180 {
188 }
189 }
192 {
197 }
200 {
208 }
213 }
216 {
228 }
230 /*
231 * raid_end_bio_io() is called when we have finished servicing a mirrored
232 * operation and are ready to return a success/failure code to the buffer
233 * cache layer.
234 */
236 {
249 /*
250 * make_request() might be waiting for
251 * bi_phys_segments to decrease
252 */
261 /*
262 * Wake up any possible resync thread that waits for the device
263 * to go idle.
264 */
266 }
267 }
270 {
273 /* if nobody has done the final endio yet, do it now */
281 }
283 }
285 /*
286 * Update disk head position estimator based on IRQ completion info.
287 */
289 {
294 }
296 /*
297 * Find the disk number which triggered given bio
298 */
300 {
313 }
316 {
323 /*
324 * this branch is our 'one mirror IO has finished' event handler:
325 */
331 /* If all other devices have failed, we want to return
332 * the error upwards rather than fail the last device.
333 * Here we redefine "uptodate" to mean "Don't want to retry"
334 */
342 }
348 /*
349 * oops, read error:
350 */
353 KERN_ERR "md/raid1:%s: %s: "
357 b),
361 /* don't drop the reference on read_disk yet */
362 }
363 }
366 {
367 /* it really is the end of this request */
369 /* free extra copy of the data pages */
375 }
376 /* clear the bitmap if all writes complete successfully */
382 }
385 {
395 else
397 }
398 }
401 {
410 /*
411 * 'one mirror IO has finished' event handler:
412 */
423 /*
424 * Set R1BIO_Uptodate in our master bio, so that we
425 * will return a good error code for to the higher
426 * levels even if IO on some other mirrored buffer
427 * fails.
428 *
429 * The 'master' represents the composite IO operation
430 * to user-side. So if something waits for IO, then it
431 * will wait for the 'master' bio.
432 */
433 sector_t first_bad;
438 /*
439 * Do not set R1BIO_Uptodate if the current device is
440 * rebuilding or Faulty. This is because we cannot use
441 * such device for properly reading the data back (we could
442 * potentially use it, if the current write would have felt
443 * before rdev->recovery_offset, but for simplicity we don't
444 * check this here.
445 */
450 /* Maybe we can clear some bad blocks. */
456 }
457 }
463 /*
464 * In behind mode, we ACK the master bio once the I/O
465 * has safely reached all non-writemostly
466 * disks. Setting the Returned bit ensures that this
467 * gets done only once -- we don't ever want to return
468 * -EIO here, instead we'll wait
469 */
472 /* Maybe we can return now */
480 }
481 }
482 }
487 /*
488 * Let's see if all mirrored write operations have finished
489 * already.
490 */
495 }
497 /*
498 * This routine returns the disk from which the requested read should
499 * be done. There is a per-array 'next expected sequential IO' sector
500 * number - if this matches on the next IO then we use the last disk.
501 * There is also a per-disk 'last know head position' sector that is
502 * maintained from IRQ contexts, both the normal and the resync IO
503 * completion handlers update this position correctly. If there is no
504 * perfect sequential match then we pick the disk whose head is closest.
505 *
506 * If there are 2 mirrors in the same 2 devices, performance degrades
507 * because position is mirror, not device based.
508 *
509 * The rdev for the device selected will have nr_pending incremented.
510 */
512 {
519 sector_t best_dist;
526 /*
527 * Check if we can balance. We can balance on the whole
528 * device if no resync is going on, or below the resync window.
529 * We take the first readable disk when above the resync window.
530 */
531 retry:
545 sector_t dist;
546 sector_t first_bad;
561 /* Don't balance among write-mostly, just
562 * use the first as a last resort */
567 /* Cannot use this */
574 }
576 }
577 /* This is a reasonable device to use. It might
578 * even be best.
579 */
583 /* already have a better device */
586 /* cannot read here. If this is the 'primary'
587 * device, then we must not read beyond
588 * bad_sectors from another device..
589 */
601 }
604 }
616 }
617 /* Don't change to another disk for sequential reads */
624 /*
625 * If buffered sequential IO size exceeds optimal
626 * iosize, check if there is idle disk. If yes, choose
627 * the idle disk. read_balance could already choose an
628 * idle disk before noticing it's a sequential IO in
629 * this disk. This doesn't matter because this disk
630 * will idle, next time it will be utilized after the
631 * first disk has IO size exceeds optimal iosize. In
632 * this way, iosize of the first disk will be optimal
633 * iosize at least. iosize of the second disk might be
634 * small, but not a big deal since when the second disk
635 * starts IO, the first disk is likely still busy.
636 */
644 }
646 }
647 /* If device is idle, use it */
651 }
659 }
664 }
665 }
667 /*
668 * If all disks are rotational, choose the closest disk. If any disk is
669 * non-rotational, choose the disk with less pending request even the
670 * disk is rotational, which might/might not be optimal for raids with
671 * mixed ratation/non-rotational disks depending on workload.
672 */
676 else
678 }
686 /* cannot risk returning a device that failed
687 * before we inc'ed nr_pending
688 */
691 }
698 }
703 }
708 {
728 }
729 }
730 }
732 }
735 }
738 {
754 /* Note the '|| 1' - when read_balance prefers
755 * non-congested targets, it can be removed
756 */
759 else
761 }
762 }
765 }
768 {
769 /* Any writes that have been queued but are awaiting
770 * bitmap updates get flushed here.
771 */
779 /* flush any pending bitmap writes to
780 * disk before proceeding w/ I/O */
789 /* Just ignore it */
791 else
794 }
797 }
799 /* Barriers....
800 * Sometimes we need to suspend IO while we do something else,
801 * either some resync/recovery, or reconfigure the array.
802 * To do this we raise a 'barrier'.
803 * The 'barrier' is a counter that can be raised multiple times
804 * to count how many activities are happening which preclude
805 * normal IO.
806 * We can only raise the barrier if there is no pending IO.
807 * i.e. if nr_pending == 0.
808 * We choose only to raise the barrier if no-one is waiting for the
809 * barrier to go down. This means that as soon as an IO request
810 * is ready, no other operations which require a barrier will start
811 * until the IO request has had a chance.
812 *
813 * So: regular IO calls 'wait_barrier'. When that returns there
814 * is no backgroup IO happening, It must arrange to call
815 * allow_barrier when it has finished its IO.
816 * backgroup IO calls must call raise_barrier. Once that returns
817 * there is no normal IO happeing. It must arrange to call
818 * lower_barrier when the particular background IO completes.
819 */
821 {
824 /* Wait until no block IO is waiting */
828 /* block any new IO from starting */
832 /* For these conditions we must wait:
833 * A: while the array is in frozen state
834 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
835 * the max count which allowed.
836 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
837 * next resync will reach to the window which normal bios are
838 * handling.
839 * D: while there are any active requests in the current window.
840 */
851 }
854 {
862 }
865 {
876 else
878 }
881 }
884 {
890 /* Wait for the barrier to drop.
891 * However if there are already pending
892 * requests (preventing the barrier from
893 * rising completely), and the
894 * per-process bio queue isn't empty,
895 * then don't wait, as we need to empty
896 * that queue to allow conf->start_next_window
897 * to increase.
898 */
908 }
916 NEXT_NORMALIO_DISTANCE;
921 else
924 }
925 }
930 }
933 sector_t bi_sector)
934 {
944 else
955 NEXT_NORMALIO_DISTANCE;
958 }
959 }
962 }
965 {
966 /* stop syncio and normal IO and wait for everything to
967 * go quite.
968 * We wait until nr_pending match nr_queued+extra
969 * This is called in the context of one normal IO request
970 * that has failed. Thus any sync request that might be pending
971 * will be blocked by nr_pending, and we need to wait for
972 * pending IO requests to complete or be queued for re-try.
973 * Thus the number queued (nr_queued) plus this request (extra)
974 * must match the number of pending IOs (nr_pending) before
975 * we continue.
976 */
984 }
986 {
987 /* reverse the effect of the freeze */
992 }
994 /* duplicate the data pages for behind I/O
995 */
997 {
1001 GFP_NOIO);
1014 }
1020 do_sync_io:
1027 }
1033 };
1036 {
1038 cb);
1052 }
1054 /* we aren't scheduling, so we can do the write-out directly. */
1064 /* Just ignore it */
1066 else
1069 }
1071 }
1074 {
1094 sector_t start_next_window;
1096 /*
1097 * Register the new request and wait if the reconstruction
1098 * thread has put up a bar for new requests.
1099 * Continue immediately if no resync is active currently.
1100 */
1107 /* As the suspend_* range is controlled by
1108 * userspace, we want an interruptible
1109 * wait.
1110 */
1120 }
1122 }
1128 /*
1129 * make_request() can abort the operation when READA is being
1130 * used and no empty request is available.
1131 *
1132 */
1141 /* We might need to issue multiple reads to different
1142 * devices if there are bad blocks around, so we keep
1143 * track of the number of reads in bio->bi_phys_segments.
1144 * If this is 0, there is only one r1_bio and no locking
1145 * will be needed when requests complete. If it is
1146 * non-zero, then it is the number of not-completed requests.
1147 */
1152 /*
1153 * read balancing logic:
1154 */
1157 read_again:
1161 /* couldn't find anywhere to read from */
1164 }
1168 bitmap) {
1169 /* Reading from a write-mostly device must
1170 * take care not to over-take any writes
1171 * that are 'behind'
1172 */
1175 }
1181 max_sectors);
1193 /* could not read all from this device, so we will
1194 * need another r1_bio.
1195 */
1203 else
1206 /* Cannot call generic_make_request directly
1207 * as that will be queued in __make_request
1208 * and subsequent mempool_alloc might block waiting
1209 * for it. So hand bio over to raid1d.
1210 */
1220 sectors_handled;
1225 }
1227 /*
1228 * WRITE:
1229 */
1234 }
1235 /* first select target devices under rcu_lock and
1236 * inc refcount on their rdev. Record them by setting
1237 * bios[x] to bio
1238 * If there are known/acknowledged bad blocks on any device on
1239 * which we have seen a write error, we want to avoid writing those
1240 * blocks.
1241 * This potentially requires several writes to write around
1242 * the bad blocks. Each set of writes gets it's own r1bio
1243 * with a set of bios attached.
1244 */
1247 retry_write:
1258 }
1265 }
1269 sector_t first_bad;
1274 max_sectors,
1277 /* mustn't write here until the bad block is
1278 * acknowledged*/
1282 }
1284 /* Cannot write here at all */
1287 /* mustn't write more than bad_sectors
1288 * to other devices yet
1289 */
1292 /* We don't set R1BIO_Degraded as that
1293 * only applies if the disk is
1294 * missing, so it might be re-added,
1295 * and we want to know to recover this
1296 * chunk.
1297 * In this case the device is here,
1298 * and the fact that this chunk is not
1299 * in-sync is recorded in the bad
1300 * block log
1301 */
1303 }
1308 }
1309 }
1311 }
1315 /* Wait for this device to become unblocked */
1326 /*
1327 * We must make sure the multi r1bios of bio have
1328 * the same value of bi_phys_segments
1329 */
1332 /* Wait for the former r1bio(s) to complete */
1336 }
1339 /* We are splitting this write into multiple parts, so
1340 * we need to prepare for allocating another r1_bio.
1341 */
1346 else
1349 }
1365 /* do behind I/O ?
1366 * Not if there are too many, or cannot
1367 * allocate memory, or a reader on WriteMostly
1368 * is waiting for behind writes to flush */
1380 }
1385 /*
1386 * We trimmed the bio, so _all is legit
1387 */
1392 }
1409 else
1418 }
1422 }
1423 /* Mustn't call r1_bio_write_done before this next test,
1424 * as it could result in the bio being freed.
1425 */
1428 /* We need another r1_bio. It has already been counted
1429 * in bio->bi_phys_segments
1430 */
1438 }
1442 /* In case raid1d snuck in to freeze_array */
1444 }
1447 {
1458 }
1461 }
1464 {
1468 /*
1469 * If it is not operational, then we have already marked it as dead
1470 * else if it is the last working disks, ignore the error, let the
1471 * next level up know.
1472 * else mark the drive as failed
1473 */
1476 /*
1477 * Don't fail the drive, act as though we were just a
1478 * normal single drive.
1479 * However don't try a recovery from this drive as
1480 * it is very likely to fail.
1481 */
1484 }
1494 /*
1495 * if recovery is running, make sure it aborts.
1496 */
1504 }
1507 {
1514 }
1527 }
1529 }
1532 {
1546 }
1549 {
1555 /*
1556 * Find all failed disks within the RAID1 configuration
1557 * and mark them readable.
1558 * Called under mddev lock, so rcu protection not needed.
1559 */
1567 /* replacement has just become active */
1570 count++;
1572 /* Replaced device not technically
1573 * faulty, but we need to be sure
1574 * it gets removed and never re-added
1575 */
1579 }
1580 }
1585 count++;
1587 }
1588 }
1595 }
1598 {
1616 }
1629 /* As all devices are equivalent, we don't need a full recovery
1630 * if this was recently any drive of the array
1631 */
1636 }
1639 /* Add this device as a replacement */
1647 }
1648 }
1650 /* Some requests might not have seen this new
1651 * merge_bvec_fn. We must wait for them to complete
1652 * before merging the device fully.
1653 * First we make sure any code which has tested
1654 * our function has submitted the request, then
1655 * we wait for all outstanding requests to complete.
1656 */
1661 }
1667 }
1670 {
1685 }
1686 /* Only remove non-faulty devices if recovery
1687 * is not possible.
1688 */
1694 }
1698 /* lost the race, try later */
1703 /* We just removed a device that is being replaced.
1704 * Move down the replacement. We drain all IO before
1705 * doing this to avoid confusion.
1706 */
1718 }
1719 abort:
1723 }
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 */
1960 /* fixup the bio for reuse, but preserve BIO_UPTODATE */
1980 else
1983 }
1984 }
1991 }
2001 /* Now we can 'fixup' the BIO_UPTODATE flag */
2013 }
2020 /* No need to write to this device. */
2024 }
2027 }
2028 }
2031 {
2040 /* ouch - failed to read all of that. */
2047 /*
2048 * schedule writes
2049 */
2065 }
2068 /* if we're here, all write(s) have completed, so clean up */
2076 }
2077 }
2078 }
2080 /*
2081 * This is a kernel thread which:
2082 *
2083 * 1. Retries failed read operations on working mirrors.
2084 * 2. Updates the raid superblock when problems encounter.
2085 * 3. Performs writes following reads for array synchronising.
2086 */
2090 {
2103 /* Note: no rcu protection needed here
2104 * as this is synchronous in the raid1d thread
2105 * which is the thread that might remove
2106 * a device. If raid1d ever becomes multi-threaded....
2107 */
2108 sector_t first_bad;
2122 d++;
2125 }
2129 /* Cannot read from anywhere - mark it bad */
2134 }
2135 /* write it back and re-read */
2140 d--;
2146 }
2152 d--;
2160 "md/raid1:%s: read error corrected "
2166 }
2167 }
2168 }
2171 }
2172 }
2175 {
2180 /* bio has the data to be written to device 'i' where
2181 * we just recently had a write error.
2182 * We repeatedly clone the bio and trim down to one block,
2183 * then try the write. Where the write fails we record
2184 * a bad block.
2185 * It is conceivable that the bio doesn't exactly align with
2186 * blocks. We must handle this somehow.
2187 *
2188 * We currently own a reference on the rdev.
2189 */
2192 sector_t sector;
2211 /* Write at 'sector' for 'sectors'*/
2218 vec++;
2219 vcnt--;
2220 }
2228 }
2238 /* failure! */
2247 }
2249 }
2252 {
2263 }
2268 }
2269 }
2272 }
2275 {
2285 /* This drive got a write error. We need to
2286 * narrow down and record precise write
2287 * errors.
2288 */
2292 /* an I/O failed, we can't clear the bitmap */
2294 }
2297 }
2301 }
2304 {
2313 /* we got a read error. Maybe the drive is bad. Maybe just
2314 * the block and we can fix it.
2315 * We freeze all other IO, and try reading the block from
2316 * other devices. When we find one, we re-write
2317 * and check it that fixes the read error.
2318 * This is all done synchronously while the array is
2319 * frozen
2320 */
2332 read_more:
2340 const unsigned long do_sync
2346 }
2350 max_sectors);
2354 "md/raid1:%s: redirecting sector %llu"
2365 /* Drat - have to split this up more */
2373 else
2387 sectors_handled;
2392 }
2393 }
2396 {
2415 }
2427 else
2434 else
2435 /* just a partial read to be scheduled from separate
2436 * context
2437 */
2443 }
2445 }
2448 {
2459 }
2461 /*
2462 * perform a "sync" on one "block"
2463 *
2464 * We need to make sure that no normal I/O request - particularly write
2465 * requests - conflict with active sync requests.
2466 *
2467 * This is achieved by tracking pending requests and a 'barrier' concept
2468 * that can be installed to exclude normal IO requests.
2469 */
2471 static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
2472 {
2481 sector_t sync_blocks;
2492 /* If we aborted, we need to abort the
2493 * sync on the 'current' bitmap chunk (there will
2494 * only be one in raid1 resync.
2495 * We can find the current addess in mddev->curr_resync
2496 */
2506 }
2514 }
2515 /* before building a request, check if we can skip these blocks..
2516 * This call the bitmap_start_sync doesn't actually record anything
2517 */
2520 /* We can skip this block, and probably several more */
2523 }
2524 /*
2525 * If there is non-resync activity waiting for a turn,
2526 * and resync is going fast enough,
2527 * then let it though before starting on this new sync request.
2528 */
2538 /*
2539 * If we get a correctably read error during resync or recovery,
2540 * we might want to read from a different device. So we
2541 * flag all drives that could conceivably be read from for READ,
2542 * and any others (which will be non-In_sync devices) for WRITE.
2543 * If a read fails, we try reading from something else for which READ
2544 * is OK.
2545 */
2565 write_targets ++;
2567 /* may need to read from here */
2580 }
2581 }
2589 }
2592 read_targets++;
2596 /*
2597 * The device is suitable for reading (InSync),
2598 * but has bad block(s) here. Let's try to correct them,
2599 * if we are doing resync or repair. Otherwise, leave
2600 * this device alone for this sync request.
2601 */
2604 write_targets++;
2605 }
2606 }
2612 }
2613 }
2620 /* These sectors are bad on all InSync devices, so we
2621 * need to mark them bad on all write targets
2622 */
2630 }
2636 /* Cannot record the badblocks, so need to
2637 * abort the resync.
2638 * If there are multiple read targets, could just
2639 * fail the really bad ones ???
2640 */
2647 }
2649 /* only resync enough to reach the next bad->good
2650 * transition */
2652 }
2655 /* extra read targets are also write targets */
2659 /* There is nowhere to write, so all non-sync
2660 * drives must be failed - so we are finished
2661 */
2662 sector_t rv;
2669 }
2693 }
2700 /* stop here */
2703 i--;
2707 /* remove last page from this bio */
2711 }
2713 }
2714 }
2715 }
2720 bio_full:
2723 /* For a user-requested sync, we read all readable devices and do a
2724 * compare
2725 */
2731 read_targets--;
2734 }
2735 }
2742 }
2744 }
2747 {
2752 }
2755 {
2768 GFP_KERNEL);
2781 r1bio_pool_free,
2798 else
2810 }
2832 /* This slot has a replacement. */
2834 /* No original, just make the replacement
2835 * a recovering spare
2836 */
2841 /* Original is not in_sync - bad */
2843 }
2851 }
2852 }
2861 }
2865 abort:
2873 }
2875 }
2879 {
2890 }
2895 }
2896 /*
2897 * copy the already verified devices into our private RAID1
2898 * bookkeeping area. [whatever we allocate in run(),
2899 * should be freed in raid1_free()]
2900 */
2903 else
2919 }
2940 /*
2941 * Ok, everything is just fine now
2942 */
2953 else
2956 }
2962 }
2964 }
2967 {
2976 }
2979 {
2980 /* no resync is happening, and there is enough space
2981 * on all devices, so we can resize.
2982 * We need to make sure resync covers any new space.
2983 * If the array is shrinking we should possibly wait until
2984 * any io in the removed space completes, but it hardly seems
2985 * worth it.
2986 */
2995 }
3003 }
3007 }
3010 {
3011 /* We need to:
3012 * 1/ resize the r1bio_pool
3013 * 2/ resize conf->mirrors
3014 *
3015 * We allocate a new r1bio_pool if we can.
3016 * Then raise a device barrier and wait until all IO stops.
3017 * Then resize conf->mirrors and swap in the new r1bio pool.
3018 *
3019 * At the same time, we "pack" the devices so that all the missing
3020 * devices have the higher raid_disk numbers.
3021 */
3030 /* Cannot change chunk_size, layout, or level */
3038 }
3050 cnt++;
3053 }
3066 }
3068 GFP_KERNEL);
3073 }
3077 /* ok, everything is stopped */
3091 }
3094 }
3113 }
3116 {
3129 }
3130 }
3133 {
3134 /* raid1 can take over:
3135 * raid5 with 2 devices, any layout or chunk size
3136 */
3144 /* Array must appear to be quiesced */
3147 }
3149 }
3152 {
3172 };
3175 {
3177 }
3180 {
3182 }