| blob | 4a6ca1cb2e78539679b96a00b89542f6f0eab8f0 |
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 */
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 */
471 }
472 }
473 }
478 /*
479 * Let's see if all mirrored write operations have finished
480 * already.
481 */
486 }
489 /*
490 * This routine returns the disk from which the requested read should
491 * be done. There is a per-array 'next expected sequential IO' sector
492 * number - if this matches on the next IO then we use the last disk.
493 * There is also a per-disk 'last know head position' sector that is
494 * maintained from IRQ contexts, both the normal and the resync IO
495 * completion handlers update this position correctly. If there is no
496 * perfect sequential match then we pick the disk whose head is closest.
497 *
498 * If there are 2 mirrors in the same 2 devices, performance degrades
499 * because position is mirror, not device based.
500 *
501 * The rdev for the device selected will have nr_pending incremented.
502 */
504 {
511 sector_t best_dist;
518 /*
519 * Check if we can balance. We can balance on the whole
520 * device if no resync is going on, or below the resync window.
521 * We take the first readable disk when above the resync window.
522 */
523 retry:
537 else
541 sector_t dist;
542 sector_t first_bad;
557 /* Don't balance among write-mostly, just
558 * use the first as a last resort */
563 /* Cannot use this */
569 }
571 }
572 /* This is a reasonable device to use. It might
573 * even be best.
574 */
578 /* already have a better device */
581 /* cannot read here. If this is the 'primary'
582 * device, then we must not read beyond
583 * bad_sectors from another device..
584 */
596 }
599 }
611 }
612 /* Don't change to another disk for sequential reads */
619 /*
620 * If buffered sequential IO size exceeds optimal
621 * iosize, check if there is idle disk. If yes, choose
622 * the idle disk. read_balance could already choose an
623 * idle disk before noticing it's a sequential IO in
624 * this disk. This doesn't matter because this disk
625 * will idle, next time it will be utilized after the
626 * first disk has IO size exceeds optimal iosize. In
627 * this way, iosize of the first disk will be optimal
628 * iosize at least. iosize of the second disk might be
629 * small, but not a big deal since when the second disk
630 * starts IO, the first disk is likely still busy.
631 */
639 }
641 }
642 /* If device is idle, use it */
646 }
654 }
659 }
660 }
662 /*
663 * If all disks are rotational, choose the closest disk. If any disk is
664 * non-rotational, choose the disk with less pending request even the
665 * disk is rotational, which might/might not be optimal for raids with
666 * mixed ratation/non-rotational disks depending on workload.
667 */
671 else
673 }
681 /* cannot risk returning a device that failed
682 * before we inc'ed nr_pending
683 */
686 }
693 }
698 }
703 {
724 }
725 }
726 }
728 }
731 }
734 {
750 /* Note the '|| 1' - when read_balance prefers
751 * non-congested targets, it can be removed
752 */
755 else
757 }
758 }
761 }
765 {
770 }
773 {
774 /* Any writes that have been queued but are awaiting
775 * bitmap updates get flushed here.
776 */
784 /* flush any pending bitmap writes to
785 * disk before proceeding w/ I/O */
794 /* Just ignore it */
796 else
799 }
802 }
804 /* Barriers....
805 * Sometimes we need to suspend IO while we do something else,
806 * either some resync/recovery, or reconfigure the array.
807 * To do this we raise a 'barrier'.
808 * The 'barrier' is a counter that can be raised multiple times
809 * to count how many activities are happening which preclude
810 * normal IO.
811 * We can only raise the barrier if there is no pending IO.
812 * i.e. if nr_pending == 0.
813 * We choose only to raise the barrier if no-one is waiting for the
814 * barrier to go down. This means that as soon as an IO request
815 * is ready, no other operations which require a barrier will start
816 * until the IO request has had a chance.
817 *
818 * So: regular IO calls 'wait_barrier'. When that returns there
819 * is no backgroup IO happening, It must arrange to call
820 * allow_barrier when it has finished its IO.
821 * backgroup IO calls must call raise_barrier. Once that returns
822 * there is no normal IO happeing. It must arrange to call
823 * lower_barrier when the particular background IO completes.
824 */
826 {
829 /* Wait until no block IO is waiting */
833 /* block any new IO from starting */
836 /* For these conditions we must wait:
837 * A: while the array is in frozen state
838 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
839 * the max count which allowed.
840 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
841 * next resync will reach to the window which normal bios are
842 * handling.
843 */
852 }
855 {
862 }
865 {
878 else
880 }
883 }
886 {
892 /* Wait for the barrier to drop.
893 * However if there are already pending
894 * requests (preventing the barrier from
895 * rising completely), and the
896 * pre-process bio queue isn't empty,
897 * then don't wait, as we need to empty
898 * that queue to get the nr_pending
899 * count down.
900 */
910 }
918 NEXT_NORMALIO_DISTANCE;
923 else
926 }
927 }
932 }
935 sector_t bi_sector)
936 {
946 else
957 NEXT_NORMALIO_DISTANCE;
960 }
961 }
964 }
967 {
968 /* stop syncio and normal IO and wait for everything to
969 * go quite.
970 * We wait until nr_pending match nr_queued+extra
971 * This is called in the context of one normal IO request
972 * that has failed. Thus any sync request that might be pending
973 * will be blocked by nr_pending, and we need to wait for
974 * pending IO requests to complete or be queued for re-try.
975 * Thus the number queued (nr_queued) plus this request (extra)
976 * must match the number of pending IOs (nr_pending) before
977 * we continue.
978 */
986 }
988 {
989 /* reverse the effect of the freeze */
994 }
997 /* duplicate the data pages for behind I/O
998 */
1000 {
1004 GFP_NOIO);
1017 }
1023 do_sync_io:
1030 }
1036 };
1039 {
1041 cb);
1055 }
1057 /* we aren't scheduling, so we can do the write-out directly. */
1067 /* Just ignore it */
1069 else
1072 }
1074 }
1077 {
1097 sector_t start_next_window;
1099 /*
1100 * Register the new request and wait if the reconstruction
1101 * thread has put up a bar for new requests.
1102 * Continue immediately if no resync is active currently.
1103 */
1110 /* As the suspend_* range is controlled by
1111 * userspace, we want an interruptible
1112 * wait.
1113 */
1123 }
1125 }
1131 /*
1132 * make_request() can abort the operation when READA is being
1133 * used and no empty request is available.
1134 *
1135 */
1144 /* We might need to issue multiple reads to different
1145 * devices if there are bad blocks around, so we keep
1146 * track of the number of reads in bio->bi_phys_segments.
1147 * If this is 0, there is only one r1_bio and no locking
1148 * will be needed when requests complete. If it is
1149 * non-zero, then it is the number of not-completed requests.
1150 */
1155 /*
1156 * read balancing logic:
1157 */
1160 read_again:
1164 /* couldn't find anywhere to read from */
1167 }
1171 bitmap) {
1172 /* Reading from a write-mostly device must
1173 * take care not to over-take any writes
1174 * that are 'behind'
1175 */
1178 }
1183 max_sectors);
1195 /* could not read all from this device, so we will
1196 * need another r1_bio.
1197 */
1205 else
1208 /* Cannot call generic_make_request directly
1209 * as that will be queued in __make_request
1210 * and subsequent mempool_alloc might block waiting
1211 * for it. So hand bio over to raid1d.
1212 */
1222 sectors_handled;
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 */
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 }
2029 }
2032 {
2041 /* ouch - failed to read all of that. */
2048 /*
2049 * schedule writes
2050 */
2066 }
2069 /* if we're here, all write(s) have completed, so clean up */
2077 }
2078 }
2079 }
2081 /*
2082 * This is a kernel thread which:
2083 *
2084 * 1. Retries failed read operations on working mirrors.
2085 * 2. Updates the raid superblock when problems encounter.
2086 * 3. Performs writes following reads for array synchronising.
2087 */
2091 {
2104 /* Note: no rcu protection needed here
2105 * as this is synchronous in the raid1d thread
2106 * which is the thread that might remove
2107 * a device. If raid1d ever becomes multi-threaded....
2108 */
2109 sector_t first_bad;
2123 d++;
2126 }
2130 /* Cannot read from anywhere - mark it bad */
2135 }
2136 /* write it back and re-read */
2141 d--;
2147 }
2153 d--;
2161 "md/raid1:%s: read error corrected "
2167 }
2168 }
2169 }
2172 }
2173 }
2176 {
2181 /* bio has the data to be written to device 'i' where
2182 * we just recently had a write error.
2183 * We repeatedly clone the bio and trim down to one block,
2184 * then try the write. Where the write fails we record
2185 * a bad block.
2186 * It is conceivable that the bio doesn't exactly align with
2187 * blocks. We must handle this somehow.
2188 *
2189 * We currently own a reference on the rdev.
2190 */
2193 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 }
2449 {
2460 }
2462 /*
2463 * perform a "sync" on one "block"
2464 *
2465 * We need to make sure that no normal I/O request - particularly write
2466 * requests - conflict with active sync requests.
2467 *
2468 * This is achieved by tracking pending requests and a 'barrier' concept
2469 * that can be installed to exclude normal IO requests.
2470 */
2472 static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
2473 {
2482 sector_t sync_blocks;
2493 /* If we aborted, we need to abort the
2494 * sync on the 'current' bitmap chunk (there will
2495 * only be one in raid1 resync.
2496 * We can find the current addess in mddev->curr_resync
2497 */
2507 }
2515 }
2516 /* before building a request, check if we can skip these blocks..
2517 * This call the bitmap_start_sync doesn't actually record anything
2518 */
2521 /* We can skip this block, and probably several more */
2524 }
2525 /*
2526 * If there is non-resync activity waiting for a turn,
2527 * and resync is going fast enough,
2528 * then let it though before starting on this new sync request.
2529 */
2540 /*
2541 * If we get a correctably read error during resync or recovery,
2542 * we might want to read from a different device. So we
2543 * flag all drives that could conceivably be read from for READ,
2544 * and any others (which will be non-In_sync devices) for WRITE.
2545 * If a read fails, we try reading from something else for which READ
2546 * is OK.
2547 */
2567 write_targets ++;
2569 /* may need to read from here */
2582 }
2583 }
2591 }
2594 read_targets++;
2598 /*
2599 * The device is suitable for reading (InSync),
2600 * but has bad block(s) here. Let's try to correct them,
2601 * if we are doing resync or repair. Otherwise, leave
2602 * this device alone for this sync request.
2603 */
2606 write_targets++;
2607 }
2608 }
2614 }
2615 }
2622 /* These sectors are bad on all InSync devices, so we
2623 * need to mark them bad on all write targets
2624 */
2632 }
2638 /* Cannot record the badblocks, so need to
2639 * abort the resync.
2640 * If there are multiple read targets, could just
2641 * fail the really bad ones ???
2642 */
2649 }
2651 /* only resync enough to reach the next bad->good
2652 * transition */
2654 }
2657 /* extra read targets are also write targets */
2661 /* There is nowhere to write, so all non-sync
2662 * drives must be failed - so we are finished
2663 */
2664 sector_t rv;
2671 }
2695 }
2702 /* stop here */
2705 i--;
2709 /* remove last page from this bio */
2713 }
2715 }
2716 }
2717 }
2722 bio_full:
2725 /* For a user-requested sync, we read all readable devices and do a
2726 * compare
2727 */
2733 read_targets--;
2736 }
2737 }
2744 }
2746 }
2749 {
2754 }
2757 {
2770 GFP_KERNEL);
2783 r1bio_pool_free,
2800 else
2812 }
2834 /* This slot has a replacement. */
2836 /* No original, just make the replacement
2837 * a recovering spare
2838 */
2843 /* Original is not in_sync - bad */
2845 }
2853 }
2854 }
2863 }
2867 abort:
2875 }
2877 }
2881 {
2892 }
2897 }
2898 /*
2899 * copy the already verified devices into our private RAID1
2900 * bookkeeping area. [whatever we allocate in run(),
2901 * should be freed in stop()]
2902 */
2905 else
2921 }
2942 /*
2943 * Ok, everything is just fine now
2944 */
2959 else
2962 }
2968 }
2971 {
2975 /* wait for behind writes to complete */
2979 /* need to kick something here to make sure I/O goes? */
2982 }
2996 }
2999 {
3000 /* no resync is happening, and there is enough space
3001 * on all devices, so we can resize.
3002 * We need to make sure resync covers any new space.
3003 * If the array is shrinking we should possibly wait until
3004 * any io in the removed space completes, but it hardly seems
3005 * worth it.
3006 */
3015 }
3023 }
3027 }
3030 {
3031 /* We need to:
3032 * 1/ resize the r1bio_pool
3033 * 2/ resize conf->mirrors
3034 *
3035 * We allocate a new r1bio_pool if we can.
3036 * Then raise a device barrier and wait until all IO stops.
3037 * Then resize conf->mirrors and swap in the new r1bio pool.
3038 *
3039 * At the same time, we "pack" the devices so that all the missing
3040 * devices have the higher raid_disk numbers.
3041 */
3050 /* Cannot change chunk_size, layout, or level */
3058 }
3070 cnt++;
3073 }
3086 }
3088 GFP_KERNEL);
3093 }
3097 /* ok, everything is stopped */
3111 }
3114 }
3133 }
3136 {
3149 }
3150 }
3153 {
3154 /* raid1 can take over:
3155 * raid5 with 2 devices, any layout or chunk size
3156 */
3164 /* Array must appear to be quiesced */
3167 }
3169 }
3172 {
3190 };
3193 {
3195 }
3198 {
3200 }