| blob | 0d91644e80eb148ce65791e694b414585ddfe33e |
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 }
498 /*
499 * This routine returns the disk from which the requested read should
500 * be done. There is a per-array 'next expected sequential IO' sector
501 * number - if this matches on the next IO then we use the last disk.
502 * There is also a per-disk 'last know head position' sector that is
503 * maintained from IRQ contexts, both the normal and the resync IO
504 * completion handlers update this position correctly. If there is no
505 * perfect sequential match then we pick the disk whose head is closest.
506 *
507 * If there are 2 mirrors in the same 2 devices, performance degrades
508 * because position is mirror, not device based.
509 *
510 * The rdev for the device selected will have nr_pending incremented.
511 */
513 {
520 sector_t best_dist;
527 /*
528 * Check if we can balance. We can balance on the whole
529 * device if no resync is going on, or below the resync window.
530 * We take the first readable disk when above the resync window.
531 */
532 retry:
546 sector_t dist;
547 sector_t first_bad;
562 /* Don't balance among write-mostly, just
563 * use the first as a last resort */
568 /* Cannot use this */
575 }
577 }
578 /* This is a reasonable device to use. It might
579 * even be best.
580 */
584 /* already have a better device */
587 /* cannot read here. If this is the 'primary'
588 * device, then we must not read beyond
589 * bad_sectors from another device..
590 */
602 }
605 }
617 }
618 /* Don't change to another disk for sequential reads */
625 /*
626 * If buffered sequential IO size exceeds optimal
627 * iosize, check if there is idle disk. If yes, choose
628 * the idle disk. read_balance could already choose an
629 * idle disk before noticing it's a sequential IO in
630 * this disk. This doesn't matter because this disk
631 * will idle, next time it will be utilized after the
632 * first disk has IO size exceeds optimal iosize. In
633 * this way, iosize of the first disk will be optimal
634 * iosize at least. iosize of the second disk might be
635 * small, but not a big deal since when the second disk
636 * starts IO, the first disk is likely still busy.
637 */
645 }
647 }
648 /* If device is idle, use it */
652 }
660 }
665 }
666 }
668 /*
669 * If all disks are rotational, choose the closest disk. If any disk is
670 * non-rotational, choose the disk with less pending request even the
671 * disk is rotational, which might/might not be optimal for raids with
672 * mixed ratation/non-rotational disks depending on workload.
673 */
677 else
679 }
687 /* cannot risk returning a device that failed
688 * before we inc'ed nr_pending
689 */
692 }
699 }
704 }
709 {
730 }
731 }
732 }
734 }
737 }
740 {
756 /* Note the '|| 1' - when read_balance prefers
757 * non-congested targets, it can be removed
758 */
761 else
763 }
764 }
767 }
771 {
776 }
779 {
780 /* Any writes that have been queued but are awaiting
781 * bitmap updates get flushed here.
782 */
790 /* flush any pending bitmap writes to
791 * disk before proceeding w/ I/O */
800 /* Just ignore it */
802 else
805 }
808 }
810 /* Barriers....
811 * Sometimes we need to suspend IO while we do something else,
812 * either some resync/recovery, or reconfigure the array.
813 * To do this we raise a 'barrier'.
814 * The 'barrier' is a counter that can be raised multiple times
815 * to count how many activities are happening which preclude
816 * normal IO.
817 * We can only raise the barrier if there is no pending IO.
818 * i.e. if nr_pending == 0.
819 * We choose only to raise the barrier if no-one is waiting for the
820 * barrier to go down. This means that as soon as an IO request
821 * is ready, no other operations which require a barrier will start
822 * until the IO request has had a chance.
823 *
824 * So: regular IO calls 'wait_barrier'. When that returns there
825 * is no backgroup IO happening, It must arrange to call
826 * allow_barrier when it has finished its IO.
827 * backgroup IO calls must call raise_barrier. Once that returns
828 * there is no normal IO happeing. It must arrange to call
829 * lower_barrier when the particular background IO completes.
830 */
832 {
835 /* Wait until no block IO is waiting */
839 /* block any new IO from starting */
843 /* For these conditions we must wait:
844 * A: while the array is in frozen state
845 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
846 * the max count which allowed.
847 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
848 * next resync will reach to the window which normal bios are
849 * handling.
850 * D: while there are any active requests in the current window.
851 */
862 }
865 {
873 }
876 {
887 else
889 }
892 }
895 {
901 /* Wait for the barrier to drop.
902 * However if there are already pending
903 * requests (preventing the barrier from
904 * rising completely), and the
905 * pre-process bio queue isn't empty,
906 * then don't wait, as we need to empty
907 * that queue to get the nr_pending
908 * count down.
909 */
919 }
927 NEXT_NORMALIO_DISTANCE;
932 else
935 }
936 }
941 }
944 sector_t bi_sector)
945 {
955 else
966 NEXT_NORMALIO_DISTANCE;
969 }
970 }
973 }
976 {
977 /* stop syncio and normal IO and wait for everything to
978 * go quite.
979 * We wait until nr_pending match nr_queued+extra
980 * This is called in the context of one normal IO request
981 * that has failed. Thus any sync request that might be pending
982 * will be blocked by nr_pending, and we need to wait for
983 * pending IO requests to complete or be queued for re-try.
984 * Thus the number queued (nr_queued) plus this request (extra)
985 * must match the number of pending IOs (nr_pending) before
986 * we continue.
987 */
995 }
997 {
998 /* reverse the effect of the freeze */
1003 }
1006 /* duplicate the data pages for behind I/O
1007 */
1009 {
1013 GFP_NOIO);
1026 }
1032 do_sync_io:
1039 }
1045 };
1048 {
1050 cb);
1064 }
1066 /* we aren't scheduling, so we can do the write-out directly. */
1076 /* Just ignore it */
1078 else
1081 }
1083 }
1086 {
1106 sector_t start_next_window;
1108 /*
1109 * Register the new request and wait if the reconstruction
1110 * thread has put up a bar for new requests.
1111 * Continue immediately if no resync is active currently.
1112 */
1119 /* As the suspend_* range is controlled by
1120 * userspace, we want an interruptible
1121 * wait.
1122 */
1132 }
1134 }
1140 /*
1141 * make_request() can abort the operation when READA is being
1142 * used and no empty request is available.
1143 *
1144 */
1153 /* We might need to issue multiple reads to different
1154 * devices if there are bad blocks around, so we keep
1155 * track of the number of reads in bio->bi_phys_segments.
1156 * If this is 0, there is only one r1_bio and no locking
1157 * will be needed when requests complete. If it is
1158 * non-zero, then it is the number of not-completed requests.
1159 */
1164 /*
1165 * read balancing logic:
1166 */
1169 read_again:
1173 /* couldn't find anywhere to read from */
1176 }
1180 bitmap) {
1181 /* Reading from a write-mostly device must
1182 * take care not to over-take any writes
1183 * that are 'behind'
1184 */
1187 }
1193 max_sectors);
1205 /* could not read all from this device, so we will
1206 * need another r1_bio.
1207 */
1215 else
1218 /* Cannot call generic_make_request directly
1219 * as that will be queued in __make_request
1220 * and subsequent mempool_alloc might block waiting
1221 * for it. So hand bio over to raid1d.
1222 */
1232 sectors_handled;
1237 }
1239 /*
1240 * WRITE:
1241 */
1246 }
1247 /* first select target devices under rcu_lock and
1248 * inc refcount on their rdev. Record them by setting
1249 * bios[x] to bio
1250 * If there are known/acknowledged bad blocks on any device on
1251 * which we have seen a write error, we want to avoid writing those
1252 * blocks.
1253 * This potentially requires several writes to write around
1254 * the bad blocks. Each set of writes gets it's own r1bio
1255 * with a set of bios attached.
1256 */
1259 retry_write:
1270 }
1277 }
1281 sector_t first_bad;
1286 max_sectors,
1289 /* mustn't write here until the bad block is
1290 * acknowledged*/
1294 }
1296 /* Cannot write here at all */
1299 /* mustn't write more than bad_sectors
1300 * to other devices yet
1301 */
1304 /* We don't set R1BIO_Degraded as that
1305 * only applies if the disk is
1306 * missing, so it might be re-added,
1307 * and we want to know to recover this
1308 * chunk.
1309 * In this case the device is here,
1310 * and the fact that this chunk is not
1311 * in-sync is recorded in the bad
1312 * block log
1313 */
1315 }
1320 }
1321 }
1323 }
1327 /* Wait for this device to become unblocked */
1338 /*
1339 * We must make sure the multi r1bios of bio have
1340 * the same value of bi_phys_segments
1341 */
1344 /* Wait for the former r1bio(s) to complete */
1348 }
1351 /* We are splitting this write into multiple parts, so
1352 * we need to prepare for allocating another r1_bio.
1353 */
1358 else
1361 }
1377 /* do behind I/O ?
1378 * Not if there are too many, or cannot
1379 * allocate memory, or a reader on WriteMostly
1380 * is waiting for behind writes to flush */
1392 }
1397 /*
1398 * We trimmed the bio, so _all is legit
1399 */
1404 }
1421 else
1430 }
1434 }
1435 /* Mustn't call r1_bio_write_done before this next test,
1436 * as it could result in the bio being freed.
1437 */
1440 /* We need another r1_bio. It has already been counted
1441 * in bio->bi_phys_segments
1442 */
1450 }
1454 /* In case raid1d snuck in to freeze_array */
1456 }
1459 {
1470 }
1473 }
1477 {
1482 /*
1483 * If it is not operational, then we have already marked it as dead
1484 * else if it is the last working disks, ignore the error, let the
1485 * next level up know.
1486 * else mark the drive as failed
1487 */
1490 /*
1491 * Don't fail the drive, act as though we were just a
1492 * normal single drive.
1493 * However don't try a recovery from this drive as
1494 * it is very likely to fail.
1495 */
1498 }
1507 /*
1508 * if recovery is running, make sure it aborts.
1509 */
1517 }
1520 {
1527 }
1540 }
1542 }
1545 {
1559 }
1562 {
1568 /*
1569 * Find all failed disks within the RAID1 configuration
1570 * and mark them readable.
1571 * Called under mddev lock, so rcu protection not needed.
1572 * device_lock used to avoid races with raid1_end_read_request
1573 * which expects 'In_sync' flags and ->degraded to be consistent.
1574 */
1583 /* replacement has just become active */
1586 count++;
1588 /* Replaced device not technically
1589 * faulty, but we need to be sure
1590 * it gets removed and never re-added
1591 */
1595 }
1596 }
1601 count++;
1603 }
1604 }
1610 }
1614 {
1632 }
1645 /* As all devices are equivalent, we don't need a full recovery
1646 * if this was recently any drive of the array
1647 */
1652 }
1655 /* Add this device as a replacement */
1663 }
1664 }
1666 /* Some requests might not have seen this new
1667 * merge_bvec_fn. We must wait for them to complete
1668 * before merging the device fully.
1669 * First we make sure any code which has tested
1670 * our function has submitted the request, then
1671 * we wait for all outstanding requests to complete.
1672 */
1677 }
1683 }
1686 {
1701 }
1702 /* Only remove non-faulty devices if recovery
1703 * is not possible.
1704 */
1710 }
1714 /* lost the race, try later */
1719 /* We just removed a device that is being replaced.
1720 * Move down the replacement. We drain all IO before
1721 * doing this to avoid confusion.
1722 */
1734 }
1735 abort:
1739 }
1743 {
1748 /*
1749 * we have read a block, now it needs to be re-written,
1750 * or re-read if the read failed.
1751 * We don't do much here, just schedule handling by raid1d
1752 */
1758 }
1761 {
1767 sector_t first_bad;
1776 /* make sure these bits doesn't get cleared. */
1798 )
1809 }
1810 }
1811 }
1815 {
1817 /* success */
1825 }
1826 /* need to record an error - either for the block or the device */
1830 }
1833 {
1834 /* Try some synchronous reads of other devices to get
1835 * good data, much like with normal read errors. Only
1836 * read into the pages we already have so we don't
1837 * need to re-issue the read request.
1838 * We don't need to freeze the array, because being in an
1839 * active sync request, there is no normal IO, and
1840 * no overlapping syncs.
1841 * We don't need to check is_badblock() again as we
1842 * made sure that anything with a bad block in range
1843 * will have bi_end_io clear.
1844 */
1863 /* No rcu protection needed here devices
1864 * can only be removed when no resync is
1865 * active, and resync is currently active
1866 */
1873 }
1874 }
1875 d++;
1883 /* Cannot read from anywhere, this block is lost.
1884 * Record a bad block on each device. If that doesn't
1885 * work just disable and interrupt the recovery.
1886 * Don't fail devices as that won't really help.
1887 */
1899 }
1907 }
1908 /* Try next page */
1911 idx++;
1913 }
1916 /* write it back and re-read */
1920 d--;
1929 }
1930 }
1935 d--;
1943 }
1946 idx ++;
1947 }
1951 }
1954 {
1955 /* We have read all readable devices. If we haven't
1956 * got the block, then there is no hope left.
1957 * If we have, then we want to do a comparison
1958 * and skip the write if everything is the same.
1959 * If any blocks failed to read, then we need to
1960 * attempt an over-write
1961 */
1968 /* Fix variable parts of all bios */
1977 /* fixup the bio for reuse, but preserve BIO_UPTODATE */
1997 else
2000 }
2001 }
2008 }
2018 /* Now we can 'fixup' the BIO_UPTODATE flag */
2030 }
2037 /* No need to write to this device. */
2041 }
2044 }
2046 }
2049 {
2058 /* ouch - failed to read all of that. */
2065 /*
2066 * schedule writes
2067 */
2083 }
2086 /* if we're here, all write(s) have completed, so clean up */
2094 }
2095 }
2096 }
2098 /*
2099 * This is a kernel thread which:
2100 *
2101 * 1. Retries failed read operations on working mirrors.
2102 * 2. Updates the raid superblock when problems encounter.
2103 * 3. Performs writes following reads for array synchronising.
2104 */
2108 {
2121 /* Note: no rcu protection needed here
2122 * as this is synchronous in the raid1d thread
2123 * which is the thread that might remove
2124 * a device. If raid1d ever becomes multi-threaded....
2125 */
2126 sector_t first_bad;
2140 d++;
2143 }
2147 /* Cannot read from anywhere - mark it bad */
2152 }
2153 /* write it back and re-read */
2158 d--;
2164 }
2170 d--;
2178 "md/raid1:%s: read error corrected "
2184 }
2185 }
2186 }
2189 }
2190 }
2193 {
2198 /* bio has the data to be written to device 'i' where
2199 * we just recently had a write error.
2200 * We repeatedly clone the bio and trim down to one block,
2201 * then try the write. Where the write fails we record
2202 * a bad block.
2203 * It is conceivable that the bio doesn't exactly align with
2204 * blocks. We must handle this somehow.
2205 *
2206 * We currently own a reference on the rdev.
2207 */
2210 sector_t sector;
2228 /* Write at 'sector' for 'sectors'*/
2235 vec++;
2236 vcnt--;
2237 }
2245 }
2255 /* failure! */
2264 }
2266 }
2269 {
2280 }
2285 }
2286 }
2289 }
2292 {
2302 /* This drive got a write error. We need to
2303 * narrow down and record precise write
2304 * errors.
2305 */
2309 /* an I/O failed, we can't clear the bitmap */
2311 }
2314 }
2318 }
2321 {
2330 /* we got a read error. Maybe the drive is bad. Maybe just
2331 * the block and we can fix it.
2332 * We freeze all other IO, and try reading the block from
2333 * other devices. When we find one, we re-write
2334 * and check it that fixes the read error.
2335 * This is all done synchronously while the array is
2336 * frozen
2337 */
2349 read_more:
2357 const unsigned long do_sync
2363 }
2367 max_sectors);
2371 "md/raid1:%s: redirecting sector %llu"
2382 /* Drat - have to split this up more */
2390 else
2404 sectors_handled;
2409 }
2410 }
2413 {
2432 }
2444 else
2451 else
2452 /* just a partial read to be scheduled from separate
2453 * context
2454 */
2460 }
2462 }
2466 {
2477 }
2479 /*
2480 * perform a "sync" on one "block"
2481 *
2482 * We need to make sure that no normal I/O request - particularly write
2483 * requests - conflict with active sync requests.
2484 *
2485 * This is achieved by tracking pending requests and a 'barrier' concept
2486 * that can be installed to exclude normal IO requests.
2487 */
2489 static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
2490 {
2499 sector_t sync_blocks;
2510 /* If we aborted, we need to abort the
2511 * sync on the 'current' bitmap chunk (there will
2512 * only be one in raid1 resync.
2513 * We can find the current addess in mddev->curr_resync
2514 */
2524 }
2532 }
2533 /* before building a request, check if we can skip these blocks..
2534 * This call the bitmap_start_sync doesn't actually record anything
2535 */
2538 /* We can skip this block, and probably several more */
2541 }
2542 /*
2543 * If there is non-resync activity waiting for a turn,
2544 * and resync is going fast enough,
2545 * then let it though before starting on this new sync request.
2546 */
2556 /*
2557 * If we get a correctably read error during resync or recovery,
2558 * we might want to read from a different device. So we
2559 * flag all drives that could conceivably be read from for READ,
2560 * and any others (which will be non-In_sync devices) for WRITE.
2561 * If a read fails, we try reading from something else for which READ
2562 * is OK.
2563 */
2583 write_targets ++;
2585 /* may need to read from here */
2598 }
2599 }
2607 }
2610 read_targets++;
2614 /*
2615 * The device is suitable for reading (InSync),
2616 * but has bad block(s) here. Let's try to correct them,
2617 * if we are doing resync or repair. Otherwise, leave
2618 * this device alone for this sync request.
2619 */
2622 write_targets++;
2623 }
2624 }
2630 }
2631 }
2638 /* These sectors are bad on all InSync devices, so we
2639 * need to mark them bad on all write targets
2640 */
2648 }
2654 /* Cannot record the badblocks, so need to
2655 * abort the resync.
2656 * If there are multiple read targets, could just
2657 * fail the really bad ones ???
2658 */
2665 }
2667 /* only resync enough to reach the next bad->good
2668 * transition */
2670 }
2673 /* extra read targets are also write targets */
2677 /* There is nowhere to write, so all non-sync
2678 * drives must be failed - so we are finished
2679 */
2680 sector_t rv;
2687 }
2711 }
2718 /* stop here */
2721 i--;
2725 /* remove last page from this bio */
2729 }
2731 }
2732 }
2733 }
2738 bio_full:
2741 /* For a user-requested sync, we read all readable devices and do a
2742 * compare
2743 */
2749 read_targets--;
2752 }
2753 }
2760 }
2762 }
2765 {
2770 }
2773 {
2786 GFP_KERNEL);
2799 r1bio_pool_free,
2816 else
2828 }
2850 /* This slot has a replacement. */
2852 /* No original, just make the replacement
2853 * a recovering spare
2854 */
2859 /* Original is not in_sync - bad */
2861 }
2869 }
2870 }
2879 }
2883 abort:
2891 }
2893 }
2897 {
2908 }
2913 }
2914 /*
2915 * copy the already verified devices into our private RAID1
2916 * bookkeeping area. [whatever we allocate in run(),
2917 * should be freed in stop()]
2918 */
2921 else
2937 }
2958 /*
2959 * Ok, everything is just fine now
2960 */
2975 else
2978 }
2984 }
2987 {
2991 /* wait for behind writes to complete */
2995 /* need to kick something here to make sure I/O goes? */
2998 }
3012 }
3015 {
3016 /* no resync is happening, and there is enough space
3017 * on all devices, so we can resize.
3018 * We need to make sure resync covers any new space.
3019 * If the array is shrinking we should possibly wait until
3020 * any io in the removed space completes, but it hardly seems
3021 * worth it.
3022 */
3031 }
3039 }
3043 }
3046 {
3047 /* We need to:
3048 * 1/ resize the r1bio_pool
3049 * 2/ resize conf->mirrors
3050 *
3051 * We allocate a new r1bio_pool if we can.
3052 * Then raise a device barrier and wait until all IO stops.
3053 * Then resize conf->mirrors and swap in the new r1bio pool.
3054 *
3055 * At the same time, we "pack" the devices so that all the missing
3056 * devices have the higher raid_disk numbers.
3057 */
3066 /* Cannot change chunk_size, layout, or level */
3074 }
3086 cnt++;
3089 }
3102 }
3104 GFP_KERNEL);
3109 }
3113 /* ok, everything is stopped */
3127 }
3130 }
3149 }
3152 {
3165 }
3166 }
3169 {
3170 /* raid1 can take over:
3171 * raid5 with 2 devices, any layout or chunk size
3172 */
3180 /* Array must appear to be quiesced */
3183 }
3185 }
3188 {
3206 };
3209 {
3211 }
3214 {
3216 }