blob: e51de52eeb94f71c9d6712a61d31e49f8e6f2f60 [file] [log] [blame]
NeilBrown32a76272005-06-21 17:17:14 -07001/*
2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
3 *
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
6 *
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
NeilBrown32a76272005-06-21 17:17:14 -070010 */
11
12/*
13 * Still to do:
14 *
15 * flush after percent set rather than just time based. (maybe both).
NeilBrown32a76272005-06-21 17:17:14 -070016 */
17
NeilBrownbff61972009-03-31 14:33:13 +110018#include <linux/blkdev.h>
NeilBrown32a76272005-06-21 17:17:14 -070019#include <linux/module.h>
NeilBrown32a76272005-06-21 17:17:14 -070020#include <linux/errno.h>
21#include <linux/slab.h>
22#include <linux/init.h>
NeilBrown32a76272005-06-21 17:17:14 -070023#include <linux/timer.h>
24#include <linux/sched.h>
25#include <linux/list.h>
26#include <linux/file.h>
27#include <linux/mount.h>
28#include <linux/buffer_head.h>
NeilBrown57148962012-03-19 12:46:40 +110029#include <linux/seq_file.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110030#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110031#include "bitmap.h"
NeilBrown32a76272005-06-21 17:17:14 -070032
NeilBrownac2f40b2010-06-01 19:37:31 +100033static inline char *bmname(struct bitmap *bitmap)
NeilBrown32a76272005-06-21 17:17:14 -070034{
35 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
36}
37
NeilBrown32a76272005-06-21 17:17:14 -070038/*
NeilBrown32a76272005-06-21 17:17:14 -070039 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
40 *
41 * 1) check to see if this page is allocated, if it's not then try to alloc
42 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
43 * page pointer directly as a counter
44 *
45 * if we find our page, we increment the page's refcount so that it stays
46 * allocated while we're using it
47 */
NeilBrown40cffcc2012-05-22 13:55:24 +100048static int bitmap_checkpage(struct bitmap_counts *bitmap,
NeilBrownac2f40b2010-06-01 19:37:31 +100049 unsigned long page, int create)
NeilBrownee305ac2009-09-23 18:06:44 +100050__releases(bitmap->lock)
51__acquires(bitmap->lock)
NeilBrown32a76272005-06-21 17:17:14 -070052{
53 unsigned char *mappage;
54
55 if (page >= bitmap->pages) {
NeilBrown1187cf02009-03-31 14:27:02 +110056 /* This can happen if bitmap_start_sync goes beyond
57 * End-of-device while looking for a whole page.
58 * It is harmless.
59 */
NeilBrown32a76272005-06-21 17:17:14 -070060 return -EINVAL;
61 }
62
NeilBrown32a76272005-06-21 17:17:14 -070063 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
64 return 0;
65
66 if (bitmap->bp[page].map) /* page is already allocated, just return */
67 return 0;
68
69 if (!create)
70 return -ENOENT;
71
NeilBrown32a76272005-06-21 17:17:14 -070072 /* this page has not been allocated yet */
73
NeilBrownac2f40b2010-06-01 19:37:31 +100074 spin_unlock_irq(&bitmap->lock);
NeilBrownd9590142015-02-02 17:08:03 +110075 /* It is possible that this is being called inside a
76 * prepare_to_wait/finish_wait loop from raid5c:make_request().
77 * In general it is not permitted to sleep in that context as it
78 * can cause the loop to spin freely.
79 * That doesn't apply here as we can only reach this point
80 * once with any loop.
81 * When this function completes, either bp[page].map or
82 * bp[page].hijacked. In either case, this function will
83 * abort before getting to this point again. So there is
84 * no risk of a free-spin, and so it is safe to assert
85 * that sleeping here is allowed.
86 */
87 sched_annotate_sleep();
NeilBrown792a1d42012-03-19 12:46:41 +110088 mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
NeilBrownac2f40b2010-06-01 19:37:31 +100089 spin_lock_irq(&bitmap->lock);
90
91 if (mappage == NULL) {
NeilBrown40cffcc2012-05-22 13:55:24 +100092 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
NeilBrown32a76272005-06-21 17:17:14 -070093 /* failed - set the hijacked flag so that we can use the
94 * pointer as a counter */
NeilBrown32a76272005-06-21 17:17:14 -070095 if (!bitmap->bp[page].map)
96 bitmap->bp[page].hijacked = 1;
NeilBrownac2f40b2010-06-01 19:37:31 +100097 } else if (bitmap->bp[page].map ||
98 bitmap->bp[page].hijacked) {
NeilBrown32a76272005-06-21 17:17:14 -070099 /* somebody beat us to getting the page */
NeilBrown792a1d42012-03-19 12:46:41 +1100100 kfree(mappage);
NeilBrown32a76272005-06-21 17:17:14 -0700101 return 0;
NeilBrownac2f40b2010-06-01 19:37:31 +1000102 } else {
103
104 /* no page was in place and we have one, so install it */
105
106 bitmap->bp[page].map = mappage;
107 bitmap->missing_pages--;
NeilBrown32a76272005-06-21 17:17:14 -0700108 }
NeilBrown32a76272005-06-21 17:17:14 -0700109 return 0;
110}
111
NeilBrown32a76272005-06-21 17:17:14 -0700112/* if page is completely empty, put it back on the free list, or dealloc it */
113/* if page was hijacked, unmark the flag so it might get alloced next time */
114/* Note: lock should be held when calling this */
NeilBrown40cffcc2012-05-22 13:55:24 +1000115static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
NeilBrown32a76272005-06-21 17:17:14 -0700116{
117 char *ptr;
118
119 if (bitmap->bp[page].count) /* page is still busy */
120 return;
121
122 /* page is no longer in use, it can be released */
123
124 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
125 bitmap->bp[page].hijacked = 0;
126 bitmap->bp[page].map = NULL;
NeilBrownac2f40b2010-06-01 19:37:31 +1000127 } else {
128 /* normal case, free the page */
129 ptr = bitmap->bp[page].map;
130 bitmap->bp[page].map = NULL;
131 bitmap->missing_pages++;
NeilBrown792a1d42012-03-19 12:46:41 +1100132 kfree(ptr);
NeilBrown32a76272005-06-21 17:17:14 -0700133 }
NeilBrown32a76272005-06-21 17:17:14 -0700134}
135
NeilBrown32a76272005-06-21 17:17:14 -0700136/*
137 * bitmap file handling - read and write the bitmap file and its superblock
138 */
139
NeilBrown32a76272005-06-21 17:17:14 -0700140/*
141 * basic page I/O operations
142 */
143
NeilBrowna654b9d82005-06-21 17:17:27 -0700144/* IO operations when bitmap is stored near all superblocks */
NeilBrown27581e52012-05-22 13:55:08 +1000145static int read_sb_page(struct mddev *mddev, loff_t offset,
146 struct page *page,
147 unsigned long index, int size)
NeilBrowna654b9d82005-06-21 17:17:27 -0700148{
149 /* choose a good rdev and read the page from there */
150
NeilBrown3cb03002011-10-11 16:45:26 +1100151 struct md_rdev *rdev;
NeilBrowna654b9d82005-06-21 17:17:27 -0700152 sector_t target;
NeilBrowna654b9d82005-06-21 17:17:27 -0700153
NeilBrowndafb20f2012-03-19 12:46:39 +1100154 rdev_for_each(rdev, mddev) {
NeilBrownb2d444d2005-11-08 21:39:31 -0800155 if (! test_bit(In_sync, &rdev->flags)
156 || test_bit(Faulty, &rdev->flags))
NeilBrownab904d62005-09-09 16:23:52 -0700157 continue;
158
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100159 target = offset + index * (PAGE_SIZE/512);
NeilBrowna654b9d82005-06-21 17:17:27 -0700160
NeilBrown2b193362010-10-27 15:16:40 +1100161 if (sync_page_io(rdev, target,
Martin K. Petersene1defc42009-05-22 17:17:49 -0400162 roundup(size, bdev_logical_block_size(rdev->bdev)),
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100163 page, READ, true)) {
NeilBrownab904d62005-09-09 16:23:52 -0700164 page->index = index;
NeilBrown27581e52012-05-22 13:55:08 +1000165 return 0;
NeilBrownab904d62005-09-09 16:23:52 -0700166 }
167 }
NeilBrown27581e52012-05-22 13:55:08 +1000168 return -EIO;
NeilBrowna654b9d82005-06-21 17:17:27 -0700169}
170
NeilBrownfd01b882011-10-11 16:47:53 +1100171static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000172{
173 /* Iterate the disks of an mddev, using rcu to protect access to the
174 * linked list, and raising the refcount of devices we return to ensure
175 * they don't disappear while in use.
176 * As devices are only added or removed when raid_disk is < 0 and
177 * nr_pending is 0 and In_sync is clear, the entries we return will
178 * still be in the same position on the list when we re-enter
Michael Wangfd177482012-10-11 13:43:21 +1100179 * list_for_each_entry_continue_rcu.
NeilBrown8532e342015-05-20 15:05:09 +1000180 *
181 * Note that if entered with 'rdev == NULL' to start at the
182 * beginning, we temporarily assign 'rdev' to an address which
183 * isn't really an rdev, but which can be used by
184 * list_for_each_entry_continue_rcu() to find the first entry.
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000185 */
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000186 rcu_read_lock();
187 if (rdev == NULL)
188 /* start at the beginning */
NeilBrown8532e342015-05-20 15:05:09 +1000189 rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000190 else {
191 /* release the previous rdev and start from there. */
192 rdev_dec_pending(rdev, mddev);
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000193 }
Michael Wangfd177482012-10-11 13:43:21 +1100194 list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000195 if (rdev->raid_disk >= 0 &&
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000196 !test_bit(Faulty, &rdev->flags)) {
197 /* this is a usable devices */
198 atomic_inc(&rdev->nr_pending);
199 rcu_read_unlock();
200 return rdev;
201 }
202 }
203 rcu_read_unlock();
204 return NULL;
205}
206
NeilBrownab6085c2007-05-23 13:58:10 -0700207static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
NeilBrowna654b9d82005-06-21 17:17:27 -0700208{
NeilBrown3cb03002011-10-11 16:45:26 +1100209 struct md_rdev *rdev = NULL;
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100210 struct block_device *bdev;
NeilBrownfd01b882011-10-11 16:47:53 +1100211 struct mddev *mddev = bitmap->mddev;
NeilBrown1ec885c2012-05-22 13:55:10 +1000212 struct bitmap_storage *store = &bitmap->storage;
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500213 int node_offset = 0;
214
215 if (mddev_is_clustered(bitmap->mddev))
216 node_offset = bitmap->cluster_slot * store->file_pages;
NeilBrowna654b9d82005-06-21 17:17:27 -0700217
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000218 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
NeilBrownac2f40b2010-06-01 19:37:31 +1000219 int size = PAGE_SIZE;
220 loff_t offset = mddev->bitmap_info.offset;
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100221
222 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
223
NeilBrown9b1215c2012-05-22 13:55:11 +1000224 if (page->index == store->file_pages-1) {
225 int last_page_size = store->bytes & (PAGE_SIZE-1);
226 if (last_page_size == 0)
227 last_page_size = PAGE_SIZE;
228 size = roundup(last_page_size,
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100229 bdev_logical_block_size(bdev));
NeilBrown9b1215c2012-05-22 13:55:11 +1000230 }
NeilBrownac2f40b2010-06-01 19:37:31 +1000231 /* Just make sure we aren't corrupting data or
232 * metadata
233 */
234 if (mddev->external) {
235 /* Bitmap could be anywhere. */
236 if (rdev->sb_start + offset + (page->index
237 * (PAGE_SIZE/512))
238 > rdev->data_offset
239 &&
240 rdev->sb_start + offset
241 < (rdev->data_offset + mddev->dev_sectors
242 + (PAGE_SIZE/512)))
243 goto bad_alignment;
244 } else if (offset < 0) {
245 /* DATA BITMAP METADATA */
246 if (offset
247 + (long)(page->index * (PAGE_SIZE/512))
248 + size/512 > 0)
249 /* bitmap runs in to metadata */
250 goto bad_alignment;
251 if (rdev->data_offset + mddev->dev_sectors
252 > rdev->sb_start + offset)
253 /* data runs in to bitmap */
254 goto bad_alignment;
255 } else if (rdev->sb_start < rdev->data_offset) {
256 /* METADATA BITMAP DATA */
257 if (rdev->sb_start
258 + offset
259 + page->index*(PAGE_SIZE/512) + size/512
260 > rdev->data_offset)
261 /* bitmap runs in to data */
262 goto bad_alignment;
263 } else {
264 /* DATA METADATA BITMAP - no problems */
265 }
266 md_super_write(mddev, rdev,
267 rdev->sb_start + offset
268 + page->index * (PAGE_SIZE/512),
269 size,
270 page);
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000271 }
NeilBrowna654b9d82005-06-21 17:17:27 -0700272
273 if (wait)
NeilBrowna9701a32005-11-08 21:39:34 -0800274 md_super_wait(mddev);
NeilBrowna654b9d82005-06-21 17:17:27 -0700275 return 0;
NeilBrown4b809912008-07-21 17:05:25 +1000276
277 bad_alignment:
NeilBrown4b809912008-07-21 17:05:25 +1000278 return -EINVAL;
NeilBrowna654b9d82005-06-21 17:17:27 -0700279}
280
NeilBrown4ad13662007-07-17 04:06:13 -0700281static void bitmap_file_kick(struct bitmap *bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700282/*
NeilBrowna654b9d82005-06-21 17:17:27 -0700283 * write out a page to a file
NeilBrown32a76272005-06-21 17:17:14 -0700284 */
NeilBrown4ad13662007-07-17 04:06:13 -0700285static void write_page(struct bitmap *bitmap, struct page *page, int wait)
NeilBrown32a76272005-06-21 17:17:14 -0700286{
NeilBrownd785a062006-06-26 00:27:48 -0700287 struct buffer_head *bh;
NeilBrown32a76272005-06-21 17:17:14 -0700288
NeilBrown1ec885c2012-05-22 13:55:10 +1000289 if (bitmap->storage.file == NULL) {
NeilBrownf0d76d72007-07-17 04:06:12 -0700290 switch (write_sb_page(bitmap, page, wait)) {
291 case -EINVAL:
NeilBrownb405fe92012-05-22 13:55:15 +1000292 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
NeilBrownf0d76d72007-07-17 04:06:12 -0700293 }
NeilBrown4ad13662007-07-17 04:06:13 -0700294 } else {
NeilBrowna654b9d82005-06-21 17:17:27 -0700295
NeilBrown4ad13662007-07-17 04:06:13 -0700296 bh = page_buffers(page);
NeilBrownc7084432006-01-06 00:20:45 -0800297
NeilBrown4ad13662007-07-17 04:06:13 -0700298 while (bh && bh->b_blocknr) {
299 atomic_inc(&bitmap->pending_writes);
300 set_buffer_locked(bh);
301 set_buffer_mapped(bh);
Jens Axboe721a9602011-03-09 11:56:30 +0100302 submit_bh(WRITE | REQ_SYNC, bh);
NeilBrown4ad13662007-07-17 04:06:13 -0700303 bh = bh->b_this_page;
304 }
NeilBrown32a76272005-06-21 17:17:14 -0700305
NeilBrownac2f40b2010-06-01 19:37:31 +1000306 if (wait)
NeilBrown4ad13662007-07-17 04:06:13 -0700307 wait_event(bitmap->write_wait,
308 atomic_read(&bitmap->pending_writes)==0);
NeilBrown32a76272005-06-21 17:17:14 -0700309 }
NeilBrownb405fe92012-05-22 13:55:15 +1000310 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
NeilBrown4ad13662007-07-17 04:06:13 -0700311 bitmap_file_kick(bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700312}
313
NeilBrownd785a062006-06-26 00:27:48 -0700314static void end_bitmap_write(struct buffer_head *bh, int uptodate)
NeilBrown32a76272005-06-21 17:17:14 -0700315{
NeilBrownd785a062006-06-26 00:27:48 -0700316 struct bitmap *bitmap = bh->b_private;
NeilBrownd785a062006-06-26 00:27:48 -0700317
NeilBrownb405fe92012-05-22 13:55:15 +1000318 if (!uptodate)
319 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
NeilBrownd785a062006-06-26 00:27:48 -0700320 if (atomic_dec_and_test(&bitmap->pending_writes))
321 wake_up(&bitmap->write_wait);
322}
323
324/* copied from buffer.c */
325static void
326__clear_page_buffers(struct page *page)
327{
328 ClearPagePrivate(page);
329 set_page_private(page, 0);
330 page_cache_release(page);
331}
332static void free_buffers(struct page *page)
333{
NeilBrown27581e52012-05-22 13:55:08 +1000334 struct buffer_head *bh;
NeilBrownd785a062006-06-26 00:27:48 -0700335
NeilBrown27581e52012-05-22 13:55:08 +1000336 if (!PagePrivate(page))
337 return;
338
339 bh = page_buffers(page);
NeilBrownd785a062006-06-26 00:27:48 -0700340 while (bh) {
341 struct buffer_head *next = bh->b_this_page;
342 free_buffer_head(bh);
343 bh = next;
344 }
345 __clear_page_buffers(page);
346 put_page(page);
347}
348
349/* read a page from a file.
350 * We both read the page, and attach buffers to the page to record the
351 * address of each block (using bmap). These addresses will be used
352 * to write the block later, completely bypassing the filesystem.
353 * This usage is similar to how swap files are handled, and allows us
354 * to write to a file with no concerns of memory allocation failing.
355 */
NeilBrown27581e52012-05-22 13:55:08 +1000356static int read_page(struct file *file, unsigned long index,
357 struct bitmap *bitmap,
358 unsigned long count,
359 struct page *page)
NeilBrownd785a062006-06-26 00:27:48 -0700360{
NeilBrown27581e52012-05-22 13:55:08 +1000361 int ret = 0;
Al Viro496ad9a2013-01-23 17:07:38 -0500362 struct inode *inode = file_inode(file);
NeilBrownd785a062006-06-26 00:27:48 -0700363 struct buffer_head *bh;
364 sector_t block;
NeilBrown32a76272005-06-21 17:17:14 -0700365
NeilBrown36a4e1f2011-10-07 14:23:17 +1100366 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
367 (unsigned long long)index << PAGE_SHIFT);
NeilBrown32a76272005-06-21 17:17:14 -0700368
NeilBrownd785a062006-06-26 00:27:48 -0700369 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
370 if (!bh) {
NeilBrown27581e52012-05-22 13:55:08 +1000371 ret = -ENOMEM;
NeilBrownd785a062006-06-26 00:27:48 -0700372 goto out;
373 }
374 attach_page_buffers(page, bh);
375 block = index << (PAGE_SHIFT - inode->i_blkbits);
376 while (bh) {
377 if (count == 0)
378 bh->b_blocknr = 0;
379 else {
380 bh->b_blocknr = bmap(inode, block);
381 if (bh->b_blocknr == 0) {
382 /* Cannot use this file! */
NeilBrown27581e52012-05-22 13:55:08 +1000383 ret = -EINVAL;
NeilBrownd785a062006-06-26 00:27:48 -0700384 goto out;
385 }
386 bh->b_bdev = inode->i_sb->s_bdev;
387 if (count < (1<<inode->i_blkbits))
388 count = 0;
389 else
390 count -= (1<<inode->i_blkbits);
NeilBrown32a76272005-06-21 17:17:14 -0700391
NeilBrownd785a062006-06-26 00:27:48 -0700392 bh->b_end_io = end_bitmap_write;
393 bh->b_private = bitmap;
NeilBrownce25c312006-06-26 00:27:49 -0700394 atomic_inc(&bitmap->pending_writes);
395 set_buffer_locked(bh);
396 set_buffer_mapped(bh);
397 submit_bh(READ, bh);
NeilBrownd785a062006-06-26 00:27:48 -0700398 }
399 block++;
400 bh = bh->b_this_page;
401 }
NeilBrownd785a062006-06-26 00:27:48 -0700402 page->index = index;
NeilBrownce25c312006-06-26 00:27:49 -0700403
404 wait_event(bitmap->write_wait,
405 atomic_read(&bitmap->pending_writes)==0);
NeilBrownb405fe92012-05-22 13:55:15 +1000406 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
NeilBrown27581e52012-05-22 13:55:08 +1000407 ret = -EIO;
NeilBrown32a76272005-06-21 17:17:14 -0700408out:
NeilBrown27581e52012-05-22 13:55:08 +1000409 if (ret)
410 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
NeilBrown2d1f3b52006-01-06 00:20:31 -0800411 (int)PAGE_SIZE,
412 (unsigned long long)index << PAGE_SHIFT,
NeilBrown27581e52012-05-22 13:55:08 +1000413 ret);
414 return ret;
NeilBrown32a76272005-06-21 17:17:14 -0700415}
416
417/*
418 * bitmap file superblock operations
419 */
420
421/* update the event counter and sync the superblock to disk */
NeilBrown4ad13662007-07-17 04:06:13 -0700422void bitmap_update_sb(struct bitmap *bitmap)
NeilBrown32a76272005-06-21 17:17:14 -0700423{
424 bitmap_super_t *sb;
NeilBrown32a76272005-06-21 17:17:14 -0700425
426 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
NeilBrown4ad13662007-07-17 04:06:13 -0700427 return;
NeilBrownece5cff2009-12-14 12:49:56 +1100428 if (bitmap->mddev->bitmap_info.external)
429 return;
NeilBrown1ec885c2012-05-22 13:55:10 +1000430 if (!bitmap->storage.sb_page) /* no superblock */
NeilBrown4ad13662007-07-17 04:06:13 -0700431 return;
NeilBrown1ec885c2012-05-22 13:55:10 +1000432 sb = kmap_atomic(bitmap->storage.sb_page);
NeilBrown32a76272005-06-21 17:17:14 -0700433 sb->events = cpu_to_le64(bitmap->mddev->events);
NeilBrown8258c532011-05-11 14:26:30 +1000434 if (bitmap->mddev->events < bitmap->events_cleared)
Neil Browna0da84f2008-06-28 08:31:22 +1000435 /* rocking back to read-only */
436 bitmap->events_cleared = bitmap->mddev->events;
NeilBrown8258c532011-05-11 14:26:30 +1000437 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
438 sb->state = cpu_to_le32(bitmap->flags);
NeilBrown43a70502009-12-14 12:49:55 +1100439 /* Just in case these have been changed via sysfs: */
440 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
441 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
NeilBrownb81a0402012-05-22 13:55:26 +1000442 /* This might have been changed by a reshape */
443 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
444 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
Goldwyn Rodriguesc4ce8672014-03-29 10:20:02 -0500445 sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
NeilBrown1dff2b82012-05-22 13:55:34 +1000446 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
447 bitmap_info.space);
Cong Wangb2f46e62011-11-28 13:25:44 +0800448 kunmap_atomic(sb);
NeilBrown1ec885c2012-05-22 13:55:10 +1000449 write_page(bitmap, bitmap->storage.sb_page, 1);
NeilBrown32a76272005-06-21 17:17:14 -0700450}
451
452/* print out the bitmap file superblock */
453void bitmap_print_sb(struct bitmap *bitmap)
454{
455 bitmap_super_t *sb;
456
NeilBrown1ec885c2012-05-22 13:55:10 +1000457 if (!bitmap || !bitmap->storage.sb_page)
NeilBrown32a76272005-06-21 17:17:14 -0700458 return;
NeilBrown1ec885c2012-05-22 13:55:10 +1000459 sb = kmap_atomic(bitmap->storage.sb_page);
NeilBrown32a76272005-06-21 17:17:14 -0700460 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
NeilBrowna2cff262005-06-21 17:17:20 -0700461 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
462 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
463 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
NeilBrown32a76272005-06-21 17:17:14 -0700464 *(__u32 *)(sb->uuid+0),
465 *(__u32 *)(sb->uuid+4),
466 *(__u32 *)(sb->uuid+8),
467 *(__u32 *)(sb->uuid+12));
NeilBrowna2cff262005-06-21 17:17:20 -0700468 printk(KERN_DEBUG " events: %llu\n",
NeilBrown32a76272005-06-21 17:17:14 -0700469 (unsigned long long) le64_to_cpu(sb->events));
NeilBrowna2cff262005-06-21 17:17:20 -0700470 printk(KERN_DEBUG "events cleared: %llu\n",
NeilBrown32a76272005-06-21 17:17:14 -0700471 (unsigned long long) le64_to_cpu(sb->events_cleared));
NeilBrowna2cff262005-06-21 17:17:20 -0700472 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
473 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
474 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
475 printk(KERN_DEBUG " sync size: %llu KB\n",
476 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
NeilBrown4b6d2872005-09-09 16:23:47 -0700477 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
Cong Wangb2f46e62011-11-28 13:25:44 +0800478 kunmap_atomic(sb);
NeilBrown32a76272005-06-21 17:17:14 -0700479}
480
Jonathan Brassow9c810752011-06-08 17:59:30 -0500481/*
482 * bitmap_new_disk_sb
483 * @bitmap
484 *
485 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
486 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
487 * This function verifies 'bitmap_info' and populates the on-disk bitmap
488 * structure, which is to be written to disk.
489 *
490 * Returns: 0 on success, -Exxx on error
491 */
492static int bitmap_new_disk_sb(struct bitmap *bitmap)
493{
494 bitmap_super_t *sb;
495 unsigned long chunksize, daemon_sleep, write_behind;
Jonathan Brassow9c810752011-06-08 17:59:30 -0500496
Goldwyn Rodriguesd3b178a2015-07-22 12:09:17 -0500497 bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
Jianpeng Ma582e2e02012-10-11 13:45:36 +1100498 if (bitmap->storage.sb_page == NULL)
499 return -ENOMEM;
NeilBrown1ec885c2012-05-22 13:55:10 +1000500 bitmap->storage.sb_page->index = 0;
Jonathan Brassow9c810752011-06-08 17:59:30 -0500501
NeilBrown1ec885c2012-05-22 13:55:10 +1000502 sb = kmap_atomic(bitmap->storage.sb_page);
Jonathan Brassow9c810752011-06-08 17:59:30 -0500503
504 sb->magic = cpu_to_le32(BITMAP_MAGIC);
505 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
506
507 chunksize = bitmap->mddev->bitmap_info.chunksize;
508 BUG_ON(!chunksize);
509 if (!is_power_of_2(chunksize)) {
Cong Wangb2f46e62011-11-28 13:25:44 +0800510 kunmap_atomic(sb);
Jonathan Brassow9c810752011-06-08 17:59:30 -0500511 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
512 return -EINVAL;
513 }
514 sb->chunksize = cpu_to_le32(chunksize);
515
516 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
517 if (!daemon_sleep ||
518 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
519 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
520 daemon_sleep = 5 * HZ;
521 }
522 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
523 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
524
525 /*
526 * FIXME: write_behind for RAID1. If not specified, what
527 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
528 */
529 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
530 if (write_behind > COUNTER_MAX)
531 write_behind = COUNTER_MAX / 2;
532 sb->write_behind = cpu_to_le32(write_behind);
533 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
534
535 /* keep the array size field of the bitmap superblock up to date */
536 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
537
538 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
539
NeilBrownb405fe92012-05-22 13:55:15 +1000540 set_bit(BITMAP_STALE, &bitmap->flags);
NeilBrown84e92342012-05-22 13:55:14 +1000541 sb->state = cpu_to_le32(bitmap->flags);
Jonathan Brassow9c810752011-06-08 17:59:30 -0500542 bitmap->events_cleared = bitmap->mddev->events;
543 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
Goldwyn Rodriguesd3b178a2015-07-22 12:09:17 -0500544 bitmap->mddev->bitmap_info.nodes = 0;
Jonathan Brassow9c810752011-06-08 17:59:30 -0500545
Cong Wangb2f46e62011-11-28 13:25:44 +0800546 kunmap_atomic(sb);
Jonathan Brassow9c810752011-06-08 17:59:30 -0500547
548 return 0;
549}
550
NeilBrown32a76272005-06-21 17:17:14 -0700551/* read the superblock from the bitmap file and initialize some bitmap fields */
552static int bitmap_read_sb(struct bitmap *bitmap)
553{
554 char *reason = NULL;
555 bitmap_super_t *sb;
NeilBrown4b6d2872005-09-09 16:23:47 -0700556 unsigned long chunksize, daemon_sleep, write_behind;
NeilBrown32a76272005-06-21 17:17:14 -0700557 unsigned long long events;
Goldwyn Rodriguesc4ce8672014-03-29 10:20:02 -0500558 int nodes = 0;
NeilBrown1dff2b82012-05-22 13:55:34 +1000559 unsigned long sectors_reserved = 0;
NeilBrown32a76272005-06-21 17:17:14 -0700560 int err = -EINVAL;
NeilBrown27581e52012-05-22 13:55:08 +1000561 struct page *sb_page;
Goldwyn Rodrigues33e38ac2015-07-01 12:19:56 +1000562 loff_t offset = bitmap->mddev->bitmap_info.offset;
NeilBrown32a76272005-06-21 17:17:14 -0700563
NeilBrown1ec885c2012-05-22 13:55:10 +1000564 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
NeilBrownef99bf42012-05-22 13:55:08 +1000565 chunksize = 128 * 1024 * 1024;
566 daemon_sleep = 5 * HZ;
567 write_behind = 0;
NeilBrownb405fe92012-05-22 13:55:15 +1000568 set_bit(BITMAP_STALE, &bitmap->flags);
NeilBrownef99bf42012-05-22 13:55:08 +1000569 err = 0;
570 goto out_no_sb;
571 }
NeilBrown32a76272005-06-21 17:17:14 -0700572 /* page 0 is the superblock, read it... */
NeilBrown27581e52012-05-22 13:55:08 +1000573 sb_page = alloc_page(GFP_KERNEL);
574 if (!sb_page)
575 return -ENOMEM;
NeilBrown1ec885c2012-05-22 13:55:10 +1000576 bitmap->storage.sb_page = sb_page;
NeilBrown27581e52012-05-22 13:55:08 +1000577
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500578re_read:
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -0500579 /* If cluster_slot is set, the cluster is setup */
580 if (bitmap->cluster_slot >= 0) {
Stephen Rothwell3b0e6aa2015-03-03 13:35:31 +1100581 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -0500582
Stephen Rothwell3b0e6aa2015-03-03 13:35:31 +1100583 sector_div(bm_blocks,
584 bitmap->mddev->bitmap_info.chunksize >> 9);
Goldwyn Rodrigues124eb762015-03-24 11:29:05 -0500585 /* bits to bytes */
586 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
587 /* to 4k blocks */
NeilBrown935f3d42015-03-02 17:02:29 +1100588 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
Goldwyn Rodrigues33e38ac2015-07-01 12:19:56 +1000589 offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3));
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -0500590 pr_info("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
Goldwyn Rodrigues33e38ac2015-07-01 12:19:56 +1000591 bitmap->cluster_slot, offset);
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -0500592 }
593
NeilBrown1ec885c2012-05-22 13:55:10 +1000594 if (bitmap->storage.file) {
595 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
NeilBrownf49d5e62007-01-26 00:57:03 -0800596 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
597
NeilBrown1ec885c2012-05-22 13:55:10 +1000598 err = read_page(bitmap->storage.file, 0,
NeilBrown27581e52012-05-22 13:55:08 +1000599 bitmap, bytes, sb_page);
NeilBrownf49d5e62007-01-26 00:57:03 -0800600 } else {
NeilBrown27581e52012-05-22 13:55:08 +1000601 err = read_sb_page(bitmap->mddev,
Goldwyn Rodrigues33e38ac2015-07-01 12:19:56 +1000602 offset,
NeilBrown27581e52012-05-22 13:55:08 +1000603 sb_page,
604 0, sizeof(bitmap_super_t));
NeilBrowna654b9d82005-06-21 17:17:27 -0700605 }
NeilBrown27581e52012-05-22 13:55:08 +1000606 if (err)
NeilBrown32a76272005-06-21 17:17:14 -0700607 return err;
NeilBrown32a76272005-06-21 17:17:14 -0700608
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500609 err = -EINVAL;
NeilBrown27581e52012-05-22 13:55:08 +1000610 sb = kmap_atomic(sb_page);
NeilBrown32a76272005-06-21 17:17:14 -0700611
NeilBrown32a76272005-06-21 17:17:14 -0700612 chunksize = le32_to_cpu(sb->chunksize);
NeilBrown1b04be92009-12-14 12:49:53 +1100613 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
NeilBrown4b6d2872005-09-09 16:23:47 -0700614 write_behind = le32_to_cpu(sb->write_behind);
NeilBrown1dff2b82012-05-22 13:55:34 +1000615 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
Goldwyn Rodriguesd3b178a2015-07-22 12:09:17 -0500616 /* XXX: This is a hack to ensure that we don't use clustering
617 * in case:
618 * - dm-raid is in use and
619 * - the nodes written in bitmap_sb is erroneous.
620 */
621 if (!bitmap->mddev->sync_super) {
622 nodes = le32_to_cpu(sb->nodes);
623 strlcpy(bitmap->mddev->bitmap_info.cluster_name,
624 sb->cluster_name, 64);
625 }
NeilBrown32a76272005-06-21 17:17:14 -0700626
627 /* verify that the bitmap-specific fields are valid */
628 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
629 reason = "bad magic";
NeilBrownbd926c62005-11-08 21:39:32 -0800630 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
631 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
NeilBrown32a76272005-06-21 17:17:14 -0700632 reason = "unrecognized superblock version";
NeilBrown1187cf02009-03-31 14:27:02 +1100633 else if (chunksize < 512)
NeilBrown7dd5d342006-01-06 00:20:39 -0800634 reason = "bitmap chunksize too small";
Jonathan Brassowd7445402011-06-08 18:01:10 -0500635 else if (!is_power_of_2(chunksize))
NeilBrown32a76272005-06-21 17:17:14 -0700636 reason = "bitmap chunksize not a power of 2";
NeilBrown1b04be92009-12-14 12:49:53 +1100637 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
NeilBrown7dd5d342006-01-06 00:20:39 -0800638 reason = "daemon sleep period out of range";
NeilBrown4b6d2872005-09-09 16:23:47 -0700639 else if (write_behind > COUNTER_MAX)
640 reason = "write-behind limit out of range (0 - 16383)";
NeilBrown32a76272005-06-21 17:17:14 -0700641 if (reason) {
642 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
643 bmname(bitmap), reason);
644 goto out;
645 }
646
647 /* keep the array size field of the bitmap superblock up to date */
648 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
649
NeilBrown278c1ca2012-03-19 12:46:40 +1100650 if (bitmap->mddev->persistent) {
651 /*
652 * We have a persistent array superblock, so compare the
653 * bitmap's UUID and event counter to the mddev's
654 */
655 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
656 printk(KERN_INFO
657 "%s: bitmap superblock UUID mismatch\n",
658 bmname(bitmap));
659 goto out;
660 }
661 events = le64_to_cpu(sb->events);
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500662 if (!nodes && (events < bitmap->mddev->events)) {
NeilBrown278c1ca2012-03-19 12:46:40 +1100663 printk(KERN_INFO
664 "%s: bitmap file is out of date (%llu < %llu) "
665 "-- forcing full recovery\n",
666 bmname(bitmap), events,
667 (unsigned long long) bitmap->mddev->events);
NeilBrownb405fe92012-05-22 13:55:15 +1000668 set_bit(BITMAP_STALE, &bitmap->flags);
NeilBrown278c1ca2012-03-19 12:46:40 +1100669 }
670 }
NeilBrown32a76272005-06-21 17:17:14 -0700671
NeilBrown32a76272005-06-21 17:17:14 -0700672 /* assign fields using values from superblock */
NeilBrown4f2e6392006-10-21 10:24:09 -0700673 bitmap->flags |= le32_to_cpu(sb->state);
NeilBrownbd926c62005-11-08 21:39:32 -0800674 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
NeilBrownb405fe92012-05-22 13:55:15 +1000675 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
NeilBrown32a76272005-06-21 17:17:14 -0700676 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
Goldwyn Rodriguescf921cc2014-03-30 00:42:49 -0500677 strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
NeilBrown32a76272005-06-21 17:17:14 -0700678 err = 0;
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500679
NeilBrown32a76272005-06-21 17:17:14 -0700680out:
Cong Wangb2f46e62011-11-28 13:25:44 +0800681 kunmap_atomic(sb);
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -0500682 /* Assiging chunksize is required for "re_read" */
683 bitmap->mddev->bitmap_info.chunksize = chunksize;
Goldwyn Rodriguesf7357272015-07-22 12:09:16 -0500684 if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500685 err = md_setup_cluster(bitmap->mddev, nodes);
686 if (err) {
687 pr_err("%s: Could not setup cluster service (%d)\n",
688 bmname(bitmap), err);
689 goto out_no_sb;
690 }
691 bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500692 goto re_read;
693 }
694
695
NeilBrownef99bf42012-05-22 13:55:08 +1000696out_no_sb:
NeilBrownb405fe92012-05-22 13:55:15 +1000697 if (test_bit(BITMAP_STALE, &bitmap->flags))
NeilBrownef99bf42012-05-22 13:55:08 +1000698 bitmap->events_cleared = bitmap->mddev->events;
699 bitmap->mddev->bitmap_info.chunksize = chunksize;
700 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
701 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
Goldwyn Rodriguesc4ce8672014-03-29 10:20:02 -0500702 bitmap->mddev->bitmap_info.nodes = nodes;
NeilBrown1dff2b82012-05-22 13:55:34 +1000703 if (bitmap->mddev->bitmap_info.space == 0 ||
704 bitmap->mddev->bitmap_info.space > sectors_reserved)
705 bitmap->mddev->bitmap_info.space = sectors_reserved;
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500706 if (err) {
NeilBrown32a76272005-06-21 17:17:14 -0700707 bitmap_print_sb(bitmap);
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -0500708 if (bitmap->cluster_slot < 0)
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500709 md_cluster_stop(bitmap->mddev);
710 }
NeilBrown32a76272005-06-21 17:17:14 -0700711 return err;
712}
713
NeilBrown32a76272005-06-21 17:17:14 -0700714/*
715 * general bitmap file operations
716 */
717
NeilBrownece5cff2009-12-14 12:49:56 +1100718/*
719 * on-disk bitmap:
720 *
721 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
722 * file a page at a time. There's a superblock at the start of the file.
723 */
NeilBrown32a76272005-06-21 17:17:14 -0700724/* calculate the index of the page that contains this bit */
NeilBrown1ec885c2012-05-22 13:55:10 +1000725static inline unsigned long file_page_index(struct bitmap_storage *store,
726 unsigned long chunk)
NeilBrown32a76272005-06-21 17:17:14 -0700727{
NeilBrown1ec885c2012-05-22 13:55:10 +1000728 if (store->sb_page)
NeilBrownece5cff2009-12-14 12:49:56 +1100729 chunk += sizeof(bitmap_super_t) << 3;
730 return chunk >> PAGE_BIT_SHIFT;
NeilBrown32a76272005-06-21 17:17:14 -0700731}
732
733/* calculate the (bit) offset of this bit within a page */
NeilBrown1ec885c2012-05-22 13:55:10 +1000734static inline unsigned long file_page_offset(struct bitmap_storage *store,
735 unsigned long chunk)
NeilBrown32a76272005-06-21 17:17:14 -0700736{
NeilBrown1ec885c2012-05-22 13:55:10 +1000737 if (store->sb_page)
NeilBrownece5cff2009-12-14 12:49:56 +1100738 chunk += sizeof(bitmap_super_t) << 3;
739 return chunk & (PAGE_BITS - 1);
NeilBrown32a76272005-06-21 17:17:14 -0700740}
741
742/*
743 * return a pointer to the page in the filemap that contains the given bit
744 *
NeilBrown32a76272005-06-21 17:17:14 -0700745 */
NeilBrown1ec885c2012-05-22 13:55:10 +1000746static inline struct page *filemap_get_page(struct bitmap_storage *store,
Jonathan Brassow3520fa42011-07-27 11:00:37 +1000747 unsigned long chunk)
NeilBrown32a76272005-06-21 17:17:14 -0700748{
NeilBrown1ec885c2012-05-22 13:55:10 +1000749 if (file_page_index(store, chunk) >= store->file_pages)
NeilBrownac2f40b2010-06-01 19:37:31 +1000750 return NULL;
NeilBrownf2e06c52014-05-28 13:39:23 +1000751 return store->filemap[file_page_index(store, chunk)];
NeilBrown32a76272005-06-21 17:17:14 -0700752}
753
NeilBrownd1244cb2012-05-22 13:55:12 +1000754static int bitmap_storage_alloc(struct bitmap_storage *store,
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500755 unsigned long chunks, int with_super,
756 int slot_number)
NeilBrownd1244cb2012-05-22 13:55:12 +1000757{
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500758 int pnum, offset = 0;
NeilBrownd1244cb2012-05-22 13:55:12 +1000759 unsigned long num_pages;
760 unsigned long bytes;
761
762 bytes = DIV_ROUND_UP(chunks, 8);
763 if (with_super)
764 bytes += sizeof(bitmap_super_t);
765
766 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500767 offset = slot_number * (num_pages - 1);
NeilBrownd1244cb2012-05-22 13:55:12 +1000768
769 store->filemap = kmalloc(sizeof(struct page *)
770 * num_pages, GFP_KERNEL);
771 if (!store->filemap)
772 return -ENOMEM;
773
774 if (with_super && !store->sb_page) {
NeilBrownd60b4792012-05-22 13:55:25 +1000775 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
NeilBrownd1244cb2012-05-22 13:55:12 +1000776 if (store->sb_page == NULL)
777 return -ENOMEM;
NeilBrownd1244cb2012-05-22 13:55:12 +1000778 }
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500779
NeilBrownd1244cb2012-05-22 13:55:12 +1000780 pnum = 0;
781 if (store->sb_page) {
782 store->filemap[0] = store->sb_page;
783 pnum = 1;
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500784 store->sb_page->index = offset;
NeilBrownd1244cb2012-05-22 13:55:12 +1000785 }
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500786
NeilBrownd1244cb2012-05-22 13:55:12 +1000787 for ( ; pnum < num_pages; pnum++) {
NeilBrownd60b4792012-05-22 13:55:25 +1000788 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
NeilBrownd1244cb2012-05-22 13:55:12 +1000789 if (!store->filemap[pnum]) {
790 store->file_pages = pnum;
791 return -ENOMEM;
792 }
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500793 store->filemap[pnum]->index = pnum + offset;
NeilBrownd1244cb2012-05-22 13:55:12 +1000794 }
795 store->file_pages = pnum;
796
797 /* We need 4 bits per page, rounded up to a multiple
798 * of sizeof(unsigned long) */
799 store->filemap_attr = kzalloc(
800 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
801 GFP_KERNEL);
802 if (!store->filemap_attr)
803 return -ENOMEM;
804
805 store->bytes = bytes;
806
807 return 0;
808}
809
NeilBrownfae7d322012-05-22 13:55:21 +1000810static void bitmap_file_unmap(struct bitmap_storage *store)
NeilBrown32a76272005-06-21 17:17:14 -0700811{
812 struct page **map, *sb_page;
NeilBrown32a76272005-06-21 17:17:14 -0700813 int pages;
NeilBrownfae7d322012-05-22 13:55:21 +1000814 struct file *file;
NeilBrown32a76272005-06-21 17:17:14 -0700815
NeilBrownfae7d322012-05-22 13:55:21 +1000816 file = store->file;
NeilBrown1ec885c2012-05-22 13:55:10 +1000817 map = store->filemap;
NeilBrown1ec885c2012-05-22 13:55:10 +1000818 pages = store->file_pages;
NeilBrown1ec885c2012-05-22 13:55:10 +1000819 sb_page = store->sb_page;
NeilBrown32a76272005-06-21 17:17:14 -0700820
821 while (pages--)
NeilBrownece5cff2009-12-14 12:49:56 +1100822 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
NeilBrownd785a062006-06-26 00:27:48 -0700823 free_buffers(map[pages]);
NeilBrown32a76272005-06-21 17:17:14 -0700824 kfree(map);
NeilBrownfae7d322012-05-22 13:55:21 +1000825 kfree(store->filemap_attr);
NeilBrown32a76272005-06-21 17:17:14 -0700826
NeilBrownd785a062006-06-26 00:27:48 -0700827 if (sb_page)
828 free_buffers(sb_page);
NeilBrown32a76272005-06-21 17:17:14 -0700829
NeilBrownd785a062006-06-26 00:27:48 -0700830 if (file) {
Al Viro496ad9a2013-01-23 17:07:38 -0500831 struct inode *inode = file_inode(file);
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800832 invalidate_mapping_pages(inode->i_mapping, 0, -1);
NeilBrown32a76272005-06-21 17:17:14 -0700833 fput(file);
NeilBrownd785a062006-06-26 00:27:48 -0700834 }
NeilBrown32a76272005-06-21 17:17:14 -0700835}
836
NeilBrown32a76272005-06-21 17:17:14 -0700837/*
838 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
839 * then it is no longer reliable, so we stop using it and we mark the file
840 * as failed in the superblock
841 */
842static void bitmap_file_kick(struct bitmap *bitmap)
843{
844 char *path, *ptr = NULL;
845
NeilBrownb405fe92012-05-22 13:55:15 +1000846 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
NeilBrown4ad13662007-07-17 04:06:13 -0700847 bitmap_update_sb(bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700848
NeilBrown1ec885c2012-05-22 13:55:10 +1000849 if (bitmap->storage.file) {
NeilBrown4ad13662007-07-17 04:06:13 -0700850 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
851 if (path)
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +0200852 ptr = file_path(bitmap->storage.file,
NeilBrown1ec885c2012-05-22 13:55:10 +1000853 path, PAGE_SIZE);
Christoph Hellwig6bcfd602008-05-23 13:04:34 -0700854
NeilBrown4ad13662007-07-17 04:06:13 -0700855 printk(KERN_ALERT
856 "%s: kicking failed bitmap file %s from array!\n",
Christoph Hellwig6bcfd602008-05-23 13:04:34 -0700857 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
NeilBrown32a76272005-06-21 17:17:14 -0700858
NeilBrown4ad13662007-07-17 04:06:13 -0700859 kfree(path);
860 } else
861 printk(KERN_ALERT
862 "%s: disabling internal bitmap due to errors\n",
863 bmname(bitmap));
NeilBrowna654b9d82005-06-21 17:17:27 -0700864 }
NeilBrown32a76272005-06-21 17:17:14 -0700865}
866
867enum bitmap_page_attr {
NeilBrownac2f40b2010-06-01 19:37:31 +1000868 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
NeilBrown5a537df2011-09-21 15:37:46 +1000869 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
870 * i.e. counter is 1 or 2. */
NeilBrownac2f40b2010-06-01 19:37:31 +1000871 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
NeilBrown32a76272005-06-21 17:17:14 -0700872};
873
NeilBrownd1891222012-05-22 13:55:09 +1000874static inline void set_page_attr(struct bitmap *bitmap, int pnum,
875 enum bitmap_page_attr attr)
NeilBrown32a76272005-06-21 17:17:14 -0700876{
NeilBrownbdfd1142012-05-22 13:55:22 +1000877 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
NeilBrown32a76272005-06-21 17:17:14 -0700878}
879
NeilBrownd1891222012-05-22 13:55:09 +1000880static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
881 enum bitmap_page_attr attr)
NeilBrown32a76272005-06-21 17:17:14 -0700882{
NeilBrownbdfd1142012-05-22 13:55:22 +1000883 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
NeilBrown32a76272005-06-21 17:17:14 -0700884}
885
NeilBrownbdfd1142012-05-22 13:55:22 +1000886static inline int test_page_attr(struct bitmap *bitmap, int pnum,
887 enum bitmap_page_attr attr)
NeilBrown32a76272005-06-21 17:17:14 -0700888{
NeilBrown1ec885c2012-05-22 13:55:10 +1000889 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
NeilBrown32a76272005-06-21 17:17:14 -0700890}
891
NeilBrownbdfd1142012-05-22 13:55:22 +1000892static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
893 enum bitmap_page_attr attr)
894{
895 return test_and_clear_bit((pnum<<2) + attr,
896 bitmap->storage.filemap_attr);
897}
NeilBrown32a76272005-06-21 17:17:14 -0700898/*
899 * bitmap_file_set_bit -- called before performing a write to the md device
900 * to set (and eventually sync) a particular bit in the bitmap file
901 *
902 * we set the bit immediately, then we record the page number so that
903 * when an unplug occurs, we can flush the dirty pages out to disk
904 */
905static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
906{
907 unsigned long bit;
Jonathan Brassow3520fa42011-07-27 11:00:37 +1000908 struct page *page;
NeilBrown32a76272005-06-21 17:17:14 -0700909 void *kaddr;
NeilBrown40cffcc2012-05-22 13:55:24 +1000910 unsigned long chunk = block >> bitmap->counts.chunkshift;
NeilBrown32a76272005-06-21 17:17:14 -0700911
NeilBrown1ec885c2012-05-22 13:55:10 +1000912 page = filemap_get_page(&bitmap->storage, chunk);
Jonathan Brassow3520fa42011-07-27 11:00:37 +1000913 if (!page)
914 return;
NeilBrown1ec885c2012-05-22 13:55:10 +1000915 bit = file_page_offset(&bitmap->storage, chunk);
NeilBrown32a76272005-06-21 17:17:14 -0700916
Jonathan Brassow3520fa42011-07-27 11:00:37 +1000917 /* set the bit */
Cong Wangb2f46e62011-11-28 13:25:44 +0800918 kaddr = kmap_atomic(page);
NeilBrownb405fe92012-05-22 13:55:15 +1000919 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
Jonathan Brassow3520fa42011-07-27 11:00:37 +1000920 set_bit(bit, kaddr);
921 else
Akinobu Mita3f810b62013-04-24 11:42:41 +1000922 set_bit_le(bit, kaddr);
Cong Wangb2f46e62011-11-28 13:25:44 +0800923 kunmap_atomic(kaddr);
NeilBrown36a4e1f2011-10-07 14:23:17 +1100924 pr_debug("set file bit %lu page %lu\n", bit, page->index);
NeilBrown32a76272005-06-21 17:17:14 -0700925 /* record page number so it gets flushed to disk when unplug occurs */
NeilBrownd1891222012-05-22 13:55:09 +1000926 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
NeilBrown32a76272005-06-21 17:17:14 -0700927}
928
NeilBrownef99bf42012-05-22 13:55:08 +1000929static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
930{
931 unsigned long bit;
932 struct page *page;
933 void *paddr;
NeilBrown40cffcc2012-05-22 13:55:24 +1000934 unsigned long chunk = block >> bitmap->counts.chunkshift;
NeilBrownef99bf42012-05-22 13:55:08 +1000935
NeilBrown1ec885c2012-05-22 13:55:10 +1000936 page = filemap_get_page(&bitmap->storage, chunk);
NeilBrownef99bf42012-05-22 13:55:08 +1000937 if (!page)
938 return;
NeilBrown1ec885c2012-05-22 13:55:10 +1000939 bit = file_page_offset(&bitmap->storage, chunk);
NeilBrownef99bf42012-05-22 13:55:08 +1000940 paddr = kmap_atomic(page);
NeilBrownb405fe92012-05-22 13:55:15 +1000941 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
NeilBrownef99bf42012-05-22 13:55:08 +1000942 clear_bit(bit, paddr);
943 else
Akinobu Mita3f810b62013-04-24 11:42:41 +1000944 clear_bit_le(bit, paddr);
NeilBrownef99bf42012-05-22 13:55:08 +1000945 kunmap_atomic(paddr);
NeilBrownd1891222012-05-22 13:55:09 +1000946 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
947 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
NeilBrownef99bf42012-05-22 13:55:08 +1000948 bitmap->allclean = 0;
949 }
950}
951
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -0500952static int bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
953{
954 unsigned long bit;
955 struct page *page;
956 void *paddr;
957 unsigned long chunk = block >> bitmap->counts.chunkshift;
958 int set = 0;
959
960 page = filemap_get_page(&bitmap->storage, chunk);
961 if (!page)
962 return -EINVAL;
963 bit = file_page_offset(&bitmap->storage, chunk);
964 paddr = kmap_atomic(page);
965 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
966 set = test_bit(bit, paddr);
967 else
968 set = test_bit_le(bit, paddr);
969 kunmap_atomic(paddr);
970 return set;
971}
972
973
NeilBrown32a76272005-06-21 17:17:14 -0700974/* this gets called when the md device is ready to unplug its underlying
975 * (slave) device queues -- before we let any writes go down, we need to
976 * sync the dirty pages of the bitmap file to disk */
NeilBrown4ad13662007-07-17 04:06:13 -0700977void bitmap_unplug(struct bitmap *bitmap)
NeilBrown32a76272005-06-21 17:17:14 -0700978{
NeilBrown74667122012-05-22 13:55:19 +1000979 unsigned long i;
NeilBrownec7a3192006-06-26 00:27:45 -0700980 int dirty, need_write;
NeilBrown32a76272005-06-21 17:17:14 -0700981
NeilBrown62f82fa2012-05-22 13:55:21 +1000982 if (!bitmap || !bitmap->storage.filemap ||
983 test_bit(BITMAP_STALE, &bitmap->flags))
NeilBrown4ad13662007-07-17 04:06:13 -0700984 return;
NeilBrown32a76272005-06-21 17:17:14 -0700985
986 /* look at each page to see if there are any set bits that need to be
987 * flushed out to disk */
NeilBrown1ec885c2012-05-22 13:55:10 +1000988 for (i = 0; i < bitmap->storage.file_pages; i++) {
NeilBrownbdfd1142012-05-22 13:55:22 +1000989 if (!bitmap->storage.filemap)
NeilBrown4ad13662007-07-17 04:06:13 -0700990 return;
NeilBrownbdfd1142012-05-22 13:55:22 +1000991 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
992 need_write = test_and_clear_page_attr(bitmap, i,
993 BITMAP_PAGE_NEEDWRITE);
994 if (dirty || need_write) {
NeilBrownd1891222012-05-22 13:55:09 +1000995 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
NeilBrownbdfd1142012-05-22 13:55:22 +1000996 write_page(bitmap, bitmap->storage.filemap[i], 0);
997 }
NeilBrown32a76272005-06-21 17:17:14 -0700998 }
NeilBrown4b5060d2014-09-09 14:13:51 +1000999 if (bitmap->storage.file)
1000 wait_event(bitmap->write_wait,
1001 atomic_read(&bitmap->pending_writes)==0);
1002 else
1003 md_super_wait(bitmap->mddev);
1004
NeilBrownb405fe92012-05-22 13:55:15 +10001005 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
NeilBrownd785a062006-06-26 00:27:48 -07001006 bitmap_file_kick(bitmap);
NeilBrown32a76272005-06-21 17:17:14 -07001007}
NeilBrownac2f40b2010-06-01 19:37:31 +10001008EXPORT_SYMBOL(bitmap_unplug);
NeilBrown32a76272005-06-21 17:17:14 -07001009
NeilBrown6a079972005-09-09 16:23:44 -07001010static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
NeilBrown32a76272005-06-21 17:17:14 -07001011/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1012 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1013 * memory mapping of the bitmap file
1014 * Special cases:
1015 * if there's no bitmap file, or if the bitmap file had been
1016 * previously kicked from the array, we mark all the bits as
1017 * 1's in order to cause a full resync.
NeilBrown6a079972005-09-09 16:23:44 -07001018 *
1019 * We ignore all bits for sectors that end earlier than 'start'.
1020 * This is used when reading an out-of-date bitmap...
NeilBrown32a76272005-06-21 17:17:14 -07001021 */
NeilBrown6a079972005-09-09 16:23:44 -07001022static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
NeilBrown32a76272005-06-21 17:17:14 -07001023{
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -05001024 unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
NeilBrown27581e52012-05-22 13:55:08 +10001025 struct page *page = NULL;
NeilBrownd1244cb2012-05-22 13:55:12 +10001026 unsigned long bit_cnt = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001027 struct file *file;
NeilBrownd1244cb2012-05-22 13:55:12 +10001028 unsigned long offset;
NeilBrown32a76272005-06-21 17:17:14 -07001029 int outofdate;
1030 int ret = -ENOSPC;
NeilBrownea03aff2006-01-06 00:20:34 -08001031 void *paddr;
NeilBrown1ec885c2012-05-22 13:55:10 +10001032 struct bitmap_storage *store = &bitmap->storage;
NeilBrown32a76272005-06-21 17:17:14 -07001033
NeilBrown40cffcc2012-05-22 13:55:24 +10001034 chunks = bitmap->counts.chunks;
NeilBrown1ec885c2012-05-22 13:55:10 +10001035 file = store->file;
NeilBrown32a76272005-06-21 17:17:14 -07001036
NeilBrownef99bf42012-05-22 13:55:08 +10001037 if (!file && !bitmap->mddev->bitmap_info.offset) {
1038 /* No permanent bitmap - fill with '1s'. */
NeilBrown1ec885c2012-05-22 13:55:10 +10001039 store->filemap = NULL;
1040 store->file_pages = 0;
NeilBrownef99bf42012-05-22 13:55:08 +10001041 for (i = 0; i < chunks ; i++) {
1042 /* if the disk bit is set, set the memory bit */
NeilBrown40cffcc2012-05-22 13:55:24 +10001043 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
NeilBrownef99bf42012-05-22 13:55:08 +10001044 >= start);
1045 bitmap_set_memory_bits(bitmap,
NeilBrown40cffcc2012-05-22 13:55:24 +10001046 (sector_t)i << bitmap->counts.chunkshift,
NeilBrownef99bf42012-05-22 13:55:08 +10001047 needed);
1048 }
1049 return 0;
1050 }
NeilBrown32a76272005-06-21 17:17:14 -07001051
NeilBrownb405fe92012-05-22 13:55:15 +10001052 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
NeilBrown32a76272005-06-21 17:17:14 -07001053 if (outofdate)
1054 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
1055 "recovery\n", bmname(bitmap));
1056
NeilBrownd1244cb2012-05-22 13:55:12 +10001057 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
NeilBrown32a76272005-06-21 17:17:14 -07001058 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
NeilBrownd1244cb2012-05-22 13:55:12 +10001059 bmname(bitmap),
1060 (unsigned long) i_size_read(file->f_mapping->host),
1061 store->bytes);
NeilBrown4ad13662007-07-17 04:06:13 -07001062 goto err;
NeilBrown32a76272005-06-21 17:17:14 -07001063 }
NeilBrownbc7f77d2005-06-21 17:17:17 -07001064
NeilBrown32a76272005-06-21 17:17:14 -07001065 oldindex = ~0L;
NeilBrownd1244cb2012-05-22 13:55:12 +10001066 offset = 0;
1067 if (!bitmap->mddev->bitmap_info.external)
1068 offset = sizeof(bitmap_super_t);
NeilBrown32a76272005-06-21 17:17:14 -07001069
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -05001070 if (mddev_is_clustered(bitmap->mddev))
1071 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1072
NeilBrown32a76272005-06-21 17:17:14 -07001073 for (i = 0; i < chunks; i++) {
NeilBrownbd926c62005-11-08 21:39:32 -08001074 int b;
NeilBrown1ec885c2012-05-22 13:55:10 +10001075 index = file_page_index(&bitmap->storage, i);
1076 bit = file_page_offset(&bitmap->storage, i);
NeilBrown32a76272005-06-21 17:17:14 -07001077 if (index != oldindex) { /* this is a new page, read it in */
NeilBrownd785a062006-06-26 00:27:48 -07001078 int count;
NeilBrown32a76272005-06-21 17:17:14 -07001079 /* unmap the old page, we're done with it */
NeilBrownd1244cb2012-05-22 13:55:12 +10001080 if (index == store->file_pages-1)
1081 count = store->bytes - index * PAGE_SIZE;
NeilBrownd785a062006-06-26 00:27:48 -07001082 else
1083 count = PAGE_SIZE;
NeilBrown1ec885c2012-05-22 13:55:10 +10001084 page = store->filemap[index];
NeilBrown27581e52012-05-22 13:55:08 +10001085 if (file)
1086 ret = read_page(file, index, bitmap,
1087 count, page);
1088 else
1089 ret = read_sb_page(
1090 bitmap->mddev,
1091 bitmap->mddev->bitmap_info.offset,
1092 page,
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -05001093 index + node_offset, count);
NeilBrown27581e52012-05-22 13:55:08 +10001094
1095 if (ret)
NeilBrown4ad13662007-07-17 04:06:13 -07001096 goto err;
NeilBrowna654b9d82005-06-21 17:17:27 -07001097
NeilBrown32a76272005-06-21 17:17:14 -07001098 oldindex = index;
NeilBrown32a76272005-06-21 17:17:14 -07001099
1100 if (outofdate) {
1101 /*
1102 * if bitmap is out of date, dirty the
NeilBrownac2f40b2010-06-01 19:37:31 +10001103 * whole page and write it out
NeilBrown32a76272005-06-21 17:17:14 -07001104 */
Cong Wangb2f46e62011-11-28 13:25:44 +08001105 paddr = kmap_atomic(page);
NeilBrownea03aff2006-01-06 00:20:34 -08001106 memset(paddr + offset, 0xff,
NeilBrown6a079972005-09-09 16:23:44 -07001107 PAGE_SIZE - offset);
Cong Wangb2f46e62011-11-28 13:25:44 +08001108 kunmap_atomic(paddr);
NeilBrown4ad13662007-07-17 04:06:13 -07001109 write_page(bitmap, page, 1);
1110
1111 ret = -EIO;
NeilBrownb405fe92012-05-22 13:55:15 +10001112 if (test_bit(BITMAP_WRITE_ERROR,
1113 &bitmap->flags))
NeilBrown4ad13662007-07-17 04:06:13 -07001114 goto err;
NeilBrown32a76272005-06-21 17:17:14 -07001115 }
NeilBrown32a76272005-06-21 17:17:14 -07001116 }
Cong Wangb2f46e62011-11-28 13:25:44 +08001117 paddr = kmap_atomic(page);
NeilBrownb405fe92012-05-22 13:55:15 +10001118 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
NeilBrownea03aff2006-01-06 00:20:34 -08001119 b = test_bit(bit, paddr);
NeilBrownbd926c62005-11-08 21:39:32 -08001120 else
Akinobu Mita6b33aff2011-03-23 16:42:13 -07001121 b = test_bit_le(bit, paddr);
Cong Wangb2f46e62011-11-28 13:25:44 +08001122 kunmap_atomic(paddr);
NeilBrownbd926c62005-11-08 21:39:32 -08001123 if (b) {
NeilBrown32a76272005-06-21 17:17:14 -07001124 /* if the disk bit is set, set the memory bit */
NeilBrown40cffcc2012-05-22 13:55:24 +10001125 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
NeilBrowndb305e52009-05-07 12:49:06 +10001126 >= start);
1127 bitmap_set_memory_bits(bitmap,
NeilBrown40cffcc2012-05-22 13:55:24 +10001128 (sector_t)i << bitmap->counts.chunkshift,
NeilBrowndb305e52009-05-07 12:49:06 +10001129 needed);
NeilBrown32a76272005-06-21 17:17:14 -07001130 bit_cnt++;
1131 }
NeilBrown27581e52012-05-22 13:55:08 +10001132 offset = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001133 }
1134
NeilBrown32a76272005-06-21 17:17:14 -07001135 printk(KERN_INFO "%s: bitmap initialized from disk: "
NeilBrownd1244cb2012-05-22 13:55:12 +10001136 "read %lu pages, set %lu of %lu bits\n",
NeilBrown1ec885c2012-05-22 13:55:10 +10001137 bmname(bitmap), store->file_pages,
NeilBrownd1244cb2012-05-22 13:55:12 +10001138 bit_cnt, chunks);
NeilBrown32a76272005-06-21 17:17:14 -07001139
NeilBrown4ad13662007-07-17 04:06:13 -07001140 return 0;
1141
1142 err:
1143 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1144 bmname(bitmap), ret);
NeilBrown32a76272005-06-21 17:17:14 -07001145 return ret;
1146}
1147
NeilBrowna654b9d82005-06-21 17:17:27 -07001148void bitmap_write_all(struct bitmap *bitmap)
1149{
1150 /* We don't actually write all bitmap blocks here,
1151 * just flag them as needing to be written
1152 */
NeilBrownec7a3192006-06-26 00:27:45 -07001153 int i;
NeilBrowna654b9d82005-06-21 17:17:27 -07001154
NeilBrown1ec885c2012-05-22 13:55:10 +10001155 if (!bitmap || !bitmap->storage.filemap)
NeilBrownef99bf42012-05-22 13:55:08 +10001156 return;
NeilBrown1ec885c2012-05-22 13:55:10 +10001157 if (bitmap->storage.file)
NeilBrownef99bf42012-05-22 13:55:08 +10001158 /* Only one copy, so nothing needed */
1159 return;
1160
NeilBrown1ec885c2012-05-22 13:55:10 +10001161 for (i = 0; i < bitmap->storage.file_pages; i++)
NeilBrownd1891222012-05-22 13:55:09 +10001162 set_page_attr(bitmap, i,
NeilBrownec7a3192006-06-26 00:27:45 -07001163 BITMAP_PAGE_NEEDWRITE);
NeilBrown2585f3e2011-09-21 15:37:46 +10001164 bitmap->allclean = 0;
NeilBrowna654b9d82005-06-21 17:17:27 -07001165}
1166
NeilBrown40cffcc2012-05-22 13:55:24 +10001167static void bitmap_count_page(struct bitmap_counts *bitmap,
1168 sector_t offset, int inc)
NeilBrown32a76272005-06-21 17:17:14 -07001169{
NeilBrown61a0d802012-03-19 12:46:41 +11001170 sector_t chunk = offset >> bitmap->chunkshift;
NeilBrown32a76272005-06-21 17:17:14 -07001171 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1172 bitmap->bp[page].count += inc;
NeilBrown32a76272005-06-21 17:17:14 -07001173 bitmap_checkfree(bitmap, page);
1174}
NeilBrownbf07bb72012-05-22 13:55:06 +10001175
NeilBrown40cffcc2012-05-22 13:55:24 +10001176static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
NeilBrownbf07bb72012-05-22 13:55:06 +10001177{
1178 sector_t chunk = offset >> bitmap->chunkshift;
1179 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1180 struct bitmap_page *bp = &bitmap->bp[page];
1181
1182 if (!bp->pending)
1183 bp->pending = 1;
1184}
1185
NeilBrown40cffcc2012-05-22 13:55:24 +10001186static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
NeilBrown57dab0b2010-10-19 10:03:39 +11001187 sector_t offset, sector_t *blocks,
NeilBrown32a76272005-06-21 17:17:14 -07001188 int create);
1189
1190/*
1191 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1192 * out to disk
1193 */
1194
NeilBrownfd01b882011-10-11 16:47:53 +11001195void bitmap_daemon_work(struct mddev *mddev)
NeilBrown32a76272005-06-21 17:17:14 -07001196{
NeilBrownaa5cbd12009-12-14 12:49:46 +11001197 struct bitmap *bitmap;
NeilBrownaa3163f2005-06-21 17:17:22 -07001198 unsigned long j;
NeilBrownbf07bb72012-05-22 13:55:06 +10001199 unsigned long nextpage;
NeilBrown57dab0b2010-10-19 10:03:39 +11001200 sector_t blocks;
NeilBrown40cffcc2012-05-22 13:55:24 +10001201 struct bitmap_counts *counts;
NeilBrown32a76272005-06-21 17:17:14 -07001202
NeilBrownaa5cbd12009-12-14 12:49:46 +11001203 /* Use a mutex to guard daemon_work against
1204 * bitmap_destroy.
1205 */
NeilBrownc3d97142009-12-14 12:49:52 +11001206 mutex_lock(&mddev->bitmap_info.mutex);
NeilBrownaa5cbd12009-12-14 12:49:46 +11001207 bitmap = mddev->bitmap;
1208 if (bitmap == NULL) {
NeilBrownc3d97142009-12-14 12:49:52 +11001209 mutex_unlock(&mddev->bitmap_info.mutex);
NeilBrown4ad13662007-07-17 04:06:13 -07001210 return;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001211 }
NeilBrown42a04b52009-12-14 12:49:53 +11001212 if (time_before(jiffies, bitmap->daemon_lastrun
NeilBrown2e61ebb2011-12-23 10:17:50 +11001213 + mddev->bitmap_info.daemon_sleep))
NeilBrown7be3dfe2008-03-10 11:43:48 -07001214 goto done;
1215
NeilBrown32a76272005-06-21 17:17:14 -07001216 bitmap->daemon_lastrun = jiffies;
NeilBrown8311c292008-03-04 14:29:30 -08001217 if (bitmap->allclean) {
NeilBrown2e61ebb2011-12-23 10:17:50 +11001218 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001219 goto done;
NeilBrown8311c292008-03-04 14:29:30 -08001220 }
1221 bitmap->allclean = 1;
NeilBrown32a76272005-06-21 17:17:14 -07001222
NeilBrownbf07bb72012-05-22 13:55:06 +10001223 /* Any file-page which is PENDING now needs to be written.
1224 * So set NEEDWRITE now, then after we make any last-minute changes
1225 * we will write it.
1226 */
NeilBrown1ec885c2012-05-22 13:55:10 +10001227 for (j = 0; j < bitmap->storage.file_pages; j++)
NeilBrownbdfd1142012-05-22 13:55:22 +10001228 if (test_and_clear_page_attr(bitmap, j,
1229 BITMAP_PAGE_PENDING))
NeilBrownd1891222012-05-22 13:55:09 +10001230 set_page_attr(bitmap, j,
NeilBrownbf07bb72012-05-22 13:55:06 +10001231 BITMAP_PAGE_NEEDWRITE);
NeilBrownbf07bb72012-05-22 13:55:06 +10001232
1233 if (bitmap->need_sync &&
1234 mddev->bitmap_info.external == 0) {
1235 /* Arrange for superblock update as well as
1236 * other changes */
1237 bitmap_super_t *sb;
1238 bitmap->need_sync = 0;
NeilBrown1ec885c2012-05-22 13:55:10 +10001239 if (bitmap->storage.filemap) {
1240 sb = kmap_atomic(bitmap->storage.sb_page);
NeilBrownef99bf42012-05-22 13:55:08 +10001241 sb->events_cleared =
1242 cpu_to_le64(bitmap->events_cleared);
1243 kunmap_atomic(sb);
NeilBrownd1891222012-05-22 13:55:09 +10001244 set_page_attr(bitmap, 0,
NeilBrownef99bf42012-05-22 13:55:08 +10001245 BITMAP_PAGE_NEEDWRITE);
1246 }
NeilBrownbf07bb72012-05-22 13:55:06 +10001247 }
1248 /* Now look at the bitmap counters and if any are '2' or '1',
1249 * decrement and handle accordingly.
1250 */
NeilBrown40cffcc2012-05-22 13:55:24 +10001251 counts = &bitmap->counts;
1252 spin_lock_irq(&counts->lock);
NeilBrownbf07bb72012-05-22 13:55:06 +10001253 nextpage = 0;
NeilBrown40cffcc2012-05-22 13:55:24 +10001254 for (j = 0; j < counts->chunks; j++) {
NeilBrown32a76272005-06-21 17:17:14 -07001255 bitmap_counter_t *bmc;
NeilBrown40cffcc2012-05-22 13:55:24 +10001256 sector_t block = (sector_t)j << counts->chunkshift;
Jonathan Brassow3520fa42011-07-27 11:00:37 +10001257
NeilBrownbf07bb72012-05-22 13:55:06 +10001258 if (j == nextpage) {
1259 nextpage += PAGE_COUNTER_RATIO;
NeilBrown40cffcc2012-05-22 13:55:24 +10001260 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
NeilBrownbf07bb72012-05-22 13:55:06 +10001261 j |= PAGE_COUNTER_MASK;
NeilBrownaa3163f2005-06-21 17:17:22 -07001262 continue;
1263 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001264 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001265 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001266 bmc = bitmap_get_counter(counts,
NeilBrownef99bf42012-05-22 13:55:08 +10001267 block,
NeilBrowndb305e52009-05-07 12:49:06 +10001268 &blocks, 0);
NeilBrown32a76272005-06-21 17:17:14 -07001269
NeilBrownbf07bb72012-05-22 13:55:06 +10001270 if (!bmc) {
1271 j |= PAGE_COUNTER_MASK;
1272 continue;
1273 }
1274 if (*bmc == 1 && !bitmap->need_sync) {
1275 /* We can clear the bit */
NeilBrownbf07bb72012-05-22 13:55:06 +10001276 *bmc = 0;
NeilBrown40cffcc2012-05-22 13:55:24 +10001277 bitmap_count_page(counts, block, -1);
NeilBrownef99bf42012-05-22 13:55:08 +10001278 bitmap_file_clear_bit(bitmap, block);
NeilBrownbf07bb72012-05-22 13:55:06 +10001279 } else if (*bmc && *bmc <= 2) {
1280 *bmc = 1;
NeilBrown40cffcc2012-05-22 13:55:24 +10001281 bitmap_set_pending(counts, block);
NeilBrown2585f3e2011-09-21 15:37:46 +10001282 bitmap->allclean = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001283 }
NeilBrown32a76272005-06-21 17:17:14 -07001284 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001285 spin_unlock_irq(&counts->lock);
NeilBrown32a76272005-06-21 17:17:14 -07001286
NeilBrownbf07bb72012-05-22 13:55:06 +10001287 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1288 * DIRTY pages need to be written by bitmap_unplug so it can wait
1289 * for them.
1290 * If we find any DIRTY page we stop there and let bitmap_unplug
1291 * handle all the rest. This is important in the case where
1292 * the first blocking holds the superblock and it has been updated.
1293 * We mustn't write any other blocks before the superblock.
1294 */
NeilBrown62f82fa2012-05-22 13:55:21 +10001295 for (j = 0;
1296 j < bitmap->storage.file_pages
1297 && !test_bit(BITMAP_STALE, &bitmap->flags);
1298 j++) {
NeilBrownd1891222012-05-22 13:55:09 +10001299 if (test_page_attr(bitmap, j,
NeilBrownbf07bb72012-05-22 13:55:06 +10001300 BITMAP_PAGE_DIRTY))
1301 /* bitmap_unplug will handle the rest */
1302 break;
NeilBrownbdfd1142012-05-22 13:55:22 +10001303 if (test_and_clear_page_attr(bitmap, j,
1304 BITMAP_PAGE_NEEDWRITE)) {
NeilBrown1ec885c2012-05-22 13:55:10 +10001305 write_page(bitmap, bitmap->storage.filemap[j], 0);
NeilBrownbf07bb72012-05-22 13:55:06 +10001306 }
1307 }
NeilBrownbf07bb72012-05-22 13:55:06 +10001308
NeilBrown7be3dfe2008-03-10 11:43:48 -07001309 done:
NeilBrown8311c292008-03-04 14:29:30 -08001310 if (bitmap->allclean == 0)
NeilBrown2e61ebb2011-12-23 10:17:50 +11001311 mddev->thread->timeout =
1312 mddev->bitmap_info.daemon_sleep;
NeilBrownc3d97142009-12-14 12:49:52 +11001313 mutex_unlock(&mddev->bitmap_info.mutex);
NeilBrown32a76272005-06-21 17:17:14 -07001314}
1315
NeilBrown40cffcc2012-05-22 13:55:24 +10001316static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
NeilBrown57dab0b2010-10-19 10:03:39 +11001317 sector_t offset, sector_t *blocks,
NeilBrown32a76272005-06-21 17:17:14 -07001318 int create)
NeilBrownee305ac2009-09-23 18:06:44 +10001319__releases(bitmap->lock)
1320__acquires(bitmap->lock)
NeilBrown32a76272005-06-21 17:17:14 -07001321{
1322 /* If 'create', we might release the lock and reclaim it.
1323 * The lock must have been taken with interrupts enabled.
1324 * If !create, we don't release the lock.
1325 */
NeilBrown61a0d802012-03-19 12:46:41 +11001326 sector_t chunk = offset >> bitmap->chunkshift;
NeilBrown32a76272005-06-21 17:17:14 -07001327 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1328 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1329 sector_t csize;
NeilBrownef425672010-06-01 19:37:33 +10001330 int err;
NeilBrown32a76272005-06-21 17:17:14 -07001331
NeilBrownef425672010-06-01 19:37:33 +10001332 err = bitmap_checkpage(bitmap, page, create);
1333
1334 if (bitmap->bp[page].hijacked ||
1335 bitmap->bp[page].map == NULL)
NeilBrown61a0d802012-03-19 12:46:41 +11001336 csize = ((sector_t)1) << (bitmap->chunkshift +
NeilBrownef425672010-06-01 19:37:33 +10001337 PAGE_COUNTER_SHIFT - 1);
1338 else
NeilBrown61a0d802012-03-19 12:46:41 +11001339 csize = ((sector_t)1) << bitmap->chunkshift;
NeilBrownef425672010-06-01 19:37:33 +10001340 *blocks = csize - (offset & (csize - 1));
1341
1342 if (err < 0)
NeilBrown32a76272005-06-21 17:17:14 -07001343 return NULL;
NeilBrownef425672010-06-01 19:37:33 +10001344
NeilBrown32a76272005-06-21 17:17:14 -07001345 /* now locked ... */
1346
1347 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1348 /* should we use the first or second counter field
1349 * of the hijacked pointer? */
1350 int hi = (pageoff > PAGE_COUNTER_MASK);
NeilBrown32a76272005-06-21 17:17:14 -07001351 return &((bitmap_counter_t *)
1352 &bitmap->bp[page].map)[hi];
NeilBrownef425672010-06-01 19:37:33 +10001353 } else /* page is allocated */
NeilBrown32a76272005-06-21 17:17:14 -07001354 return (bitmap_counter_t *)
1355 &(bitmap->bp[page].map[pageoff]);
NeilBrown32a76272005-06-21 17:17:14 -07001356}
1357
NeilBrown4b6d2872005-09-09 16:23:47 -07001358int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
NeilBrown32a76272005-06-21 17:17:14 -07001359{
NeilBrownac2f40b2010-06-01 19:37:31 +10001360 if (!bitmap)
1361 return 0;
NeilBrown4b6d2872005-09-09 16:23:47 -07001362
1363 if (behind) {
Paul Clements696fcd52010-03-08 16:02:37 +11001364 int bw;
NeilBrown4b6d2872005-09-09 16:23:47 -07001365 atomic_inc(&bitmap->behind_writes);
Paul Clements696fcd52010-03-08 16:02:37 +11001366 bw = atomic_read(&bitmap->behind_writes);
1367 if (bw > bitmap->behind_writes_used)
1368 bitmap->behind_writes_used = bw;
1369
NeilBrown36a4e1f2011-10-07 14:23:17 +11001370 pr_debug("inc write-behind count %d/%lu\n",
1371 bw, bitmap->mddev->bitmap_info.max_write_behind);
NeilBrown4b6d2872005-09-09 16:23:47 -07001372 }
1373
NeilBrown32a76272005-06-21 17:17:14 -07001374 while (sectors) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001375 sector_t blocks;
NeilBrown32a76272005-06-21 17:17:14 -07001376 bitmap_counter_t *bmc;
1377
NeilBrown40cffcc2012-05-22 13:55:24 +10001378 spin_lock_irq(&bitmap->counts.lock);
1379 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
NeilBrown32a76272005-06-21 17:17:14 -07001380 if (!bmc) {
NeilBrown40cffcc2012-05-22 13:55:24 +10001381 spin_unlock_irq(&bitmap->counts.lock);
NeilBrown32a76272005-06-21 17:17:14 -07001382 return 0;
1383 }
1384
Namhyung Kim27d5ea02011-06-09 11:42:57 +10001385 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
Neil Brownda6e1a32007-02-08 14:20:37 -08001386 DEFINE_WAIT(__wait);
1387 /* note that it is safe to do the prepare_to_wait
1388 * after the test as long as we do it before dropping
1389 * the spinlock.
1390 */
1391 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1392 TASK_UNINTERRUPTIBLE);
NeilBrown40cffcc2012-05-22 13:55:24 +10001393 spin_unlock_irq(&bitmap->counts.lock);
NeilBrownf54a9d02012-08-02 08:33:20 +10001394 schedule();
Neil Brownda6e1a32007-02-08 14:20:37 -08001395 finish_wait(&bitmap->overflow_wait, &__wait);
1396 continue;
1397 }
1398
NeilBrownac2f40b2010-06-01 19:37:31 +10001399 switch (*bmc) {
NeilBrown32a76272005-06-21 17:17:14 -07001400 case 0:
1401 bitmap_file_set_bit(bitmap, offset);
NeilBrown40cffcc2012-05-22 13:55:24 +10001402 bitmap_count_page(&bitmap->counts, offset, 1);
NeilBrown32a76272005-06-21 17:17:14 -07001403 /* fall through */
1404 case 1:
1405 *bmc = 2;
1406 }
Neil Brownda6e1a32007-02-08 14:20:37 -08001407
NeilBrown32a76272005-06-21 17:17:14 -07001408 (*bmc)++;
1409
NeilBrown40cffcc2012-05-22 13:55:24 +10001410 spin_unlock_irq(&bitmap->counts.lock);
NeilBrown32a76272005-06-21 17:17:14 -07001411
1412 offset += blocks;
1413 if (sectors > blocks)
1414 sectors -= blocks;
NeilBrownac2f40b2010-06-01 19:37:31 +10001415 else
1416 sectors = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001417 }
1418 return 0;
1419}
NeilBrownac2f40b2010-06-01 19:37:31 +10001420EXPORT_SYMBOL(bitmap_startwrite);
NeilBrown32a76272005-06-21 17:17:14 -07001421
1422void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
NeilBrown4b6d2872005-09-09 16:23:47 -07001423 int success, int behind)
NeilBrown32a76272005-06-21 17:17:14 -07001424{
NeilBrownac2f40b2010-06-01 19:37:31 +10001425 if (!bitmap)
1426 return;
NeilBrown4b6d2872005-09-09 16:23:47 -07001427 if (behind) {
NeilBrowne5551902010-03-31 11:21:44 +11001428 if (atomic_dec_and_test(&bitmap->behind_writes))
1429 wake_up(&bitmap->behind_wait);
NeilBrown36a4e1f2011-10-07 14:23:17 +11001430 pr_debug("dec write-behind count %d/%lu\n",
1431 atomic_read(&bitmap->behind_writes),
1432 bitmap->mddev->bitmap_info.max_write_behind);
NeilBrown4b6d2872005-09-09 16:23:47 -07001433 }
1434
NeilBrown32a76272005-06-21 17:17:14 -07001435 while (sectors) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001436 sector_t blocks;
NeilBrown32a76272005-06-21 17:17:14 -07001437 unsigned long flags;
1438 bitmap_counter_t *bmc;
1439
NeilBrown40cffcc2012-05-22 13:55:24 +10001440 spin_lock_irqsave(&bitmap->counts.lock, flags);
1441 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
NeilBrown32a76272005-06-21 17:17:14 -07001442 if (!bmc) {
NeilBrown40cffcc2012-05-22 13:55:24 +10001443 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
NeilBrown32a76272005-06-21 17:17:14 -07001444 return;
1445 }
1446
NeilBrown961902c2011-12-23 09:57:48 +11001447 if (success && !bitmap->mddev->degraded &&
Neil Browna0da84f2008-06-28 08:31:22 +10001448 bitmap->events_cleared < bitmap->mddev->events) {
1449 bitmap->events_cleared = bitmap->mddev->events;
1450 bitmap->need_sync = 1;
NeilBrown5ff5aff2010-06-01 19:37:32 +10001451 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
Neil Browna0da84f2008-06-28 08:31:22 +10001452 }
1453
Namhyung Kim27d5ea02011-06-09 11:42:57 +10001454 if (!success && !NEEDED(*bmc))
NeilBrown32a76272005-06-21 17:17:14 -07001455 *bmc |= NEEDED_MASK;
1456
Namhyung Kim27d5ea02011-06-09 11:42:57 +10001457 if (COUNTER(*bmc) == COUNTER_MAX)
Neil Brownda6e1a32007-02-08 14:20:37 -08001458 wake_up(&bitmap->overflow_wait);
1459
NeilBrown32a76272005-06-21 17:17:14 -07001460 (*bmc)--;
NeilBrown2585f3e2011-09-21 15:37:46 +10001461 if (*bmc <= 2) {
NeilBrown40cffcc2012-05-22 13:55:24 +10001462 bitmap_set_pending(&bitmap->counts, offset);
NeilBrown2585f3e2011-09-21 15:37:46 +10001463 bitmap->allclean = 0;
1464 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001465 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
NeilBrown32a76272005-06-21 17:17:14 -07001466 offset += blocks;
1467 if (sectors > blocks)
1468 sectors -= blocks;
NeilBrownac2f40b2010-06-01 19:37:31 +10001469 else
1470 sectors = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001471 }
1472}
NeilBrownac2f40b2010-06-01 19:37:31 +10001473EXPORT_SYMBOL(bitmap_endwrite);
NeilBrown32a76272005-06-21 17:17:14 -07001474
NeilBrown57dab0b2010-10-19 10:03:39 +11001475static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
NeilBrown1187cf02009-03-31 14:27:02 +11001476 int degraded)
NeilBrown32a76272005-06-21 17:17:14 -07001477{
1478 bitmap_counter_t *bmc;
1479 int rv;
1480 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1481 *blocks = 1024;
1482 return 1; /* always resync if no bitmap */
1483 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001484 spin_lock_irq(&bitmap->counts.lock);
1485 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
NeilBrown32a76272005-06-21 17:17:14 -07001486 rv = 0;
1487 if (bmc) {
1488 /* locked */
1489 if (RESYNC(*bmc))
1490 rv = 1;
1491 else if (NEEDED(*bmc)) {
1492 rv = 1;
NeilBrown6a806c52005-07-15 03:56:35 -07001493 if (!degraded) { /* don't set/clear bits if degraded */
1494 *bmc |= RESYNC_MASK;
1495 *bmc &= ~NEEDED_MASK;
1496 }
NeilBrown32a76272005-06-21 17:17:14 -07001497 }
1498 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001499 spin_unlock_irq(&bitmap->counts.lock);
NeilBrown32a76272005-06-21 17:17:14 -07001500 return rv;
1501}
1502
NeilBrown57dab0b2010-10-19 10:03:39 +11001503int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
NeilBrown1187cf02009-03-31 14:27:02 +11001504 int degraded)
1505{
1506 /* bitmap_start_sync must always report on multiples of whole
1507 * pages, otherwise resync (which is very PAGE_SIZE based) will
1508 * get confused.
1509 * So call __bitmap_start_sync repeatedly (if needed) until
1510 * At least PAGE_SIZE>>9 blocks are covered.
1511 * Return the 'or' of the result.
1512 */
1513 int rv = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11001514 sector_t blocks1;
NeilBrown1187cf02009-03-31 14:27:02 +11001515
1516 *blocks = 0;
1517 while (*blocks < (PAGE_SIZE>>9)) {
1518 rv |= __bitmap_start_sync(bitmap, offset,
1519 &blocks1, degraded);
1520 offset += blocks1;
1521 *blocks += blocks1;
1522 }
1523 return rv;
1524}
NeilBrownac2f40b2010-06-01 19:37:31 +10001525EXPORT_SYMBOL(bitmap_start_sync);
NeilBrown1187cf02009-03-31 14:27:02 +11001526
NeilBrown57dab0b2010-10-19 10:03:39 +11001527void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
NeilBrown32a76272005-06-21 17:17:14 -07001528{
1529 bitmap_counter_t *bmc;
1530 unsigned long flags;
NeilBrownac2f40b2010-06-01 19:37:31 +10001531
1532 if (bitmap == NULL) {
NeilBrown32a76272005-06-21 17:17:14 -07001533 *blocks = 1024;
1534 return;
1535 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001536 spin_lock_irqsave(&bitmap->counts.lock, flags);
1537 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
NeilBrown32a76272005-06-21 17:17:14 -07001538 if (bmc == NULL)
1539 goto unlock;
1540 /* locked */
NeilBrown32a76272005-06-21 17:17:14 -07001541 if (RESYNC(*bmc)) {
1542 *bmc &= ~RESYNC_MASK;
1543
1544 if (!NEEDED(*bmc) && aborted)
1545 *bmc |= NEEDED_MASK;
1546 else {
NeilBrown2585f3e2011-09-21 15:37:46 +10001547 if (*bmc <= 2) {
NeilBrown40cffcc2012-05-22 13:55:24 +10001548 bitmap_set_pending(&bitmap->counts, offset);
NeilBrown2585f3e2011-09-21 15:37:46 +10001549 bitmap->allclean = 0;
1550 }
NeilBrown32a76272005-06-21 17:17:14 -07001551 }
1552 }
1553 unlock:
NeilBrown40cffcc2012-05-22 13:55:24 +10001554 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
NeilBrown32a76272005-06-21 17:17:14 -07001555}
NeilBrownac2f40b2010-06-01 19:37:31 +10001556EXPORT_SYMBOL(bitmap_end_sync);
NeilBrown32a76272005-06-21 17:17:14 -07001557
1558void bitmap_close_sync(struct bitmap *bitmap)
1559{
1560 /* Sync has finished, and any bitmap chunks that weren't synced
1561 * properly have been aborted. It remains to us to clear the
1562 * RESYNC bit wherever it is still on
1563 */
1564 sector_t sector = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11001565 sector_t blocks;
NeilBrownb47490c2008-02-06 01:39:50 -08001566 if (!bitmap)
1567 return;
NeilBrown32a76272005-06-21 17:17:14 -07001568 while (sector < bitmap->mddev->resync_max_sectors) {
1569 bitmap_end_sync(bitmap, sector, &blocks, 0);
NeilBrownb47490c2008-02-06 01:39:50 -08001570 sector += blocks;
NeilBrown32a76272005-06-21 17:17:14 -07001571 }
1572}
NeilBrownac2f40b2010-06-01 19:37:31 +10001573EXPORT_SYMBOL(bitmap_close_sync);
NeilBrown32a76272005-06-21 17:17:14 -07001574
NeilBrownb47490c2008-02-06 01:39:50 -08001575void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1576{
1577 sector_t s = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11001578 sector_t blocks;
NeilBrownb47490c2008-02-06 01:39:50 -08001579
1580 if (!bitmap)
1581 return;
1582 if (sector == 0) {
1583 bitmap->last_end_sync = jiffies;
1584 return;
1585 }
1586 if (time_before(jiffies, (bitmap->last_end_sync
NeilBrown1b04be92009-12-14 12:49:53 +11001587 + bitmap->mddev->bitmap_info.daemon_sleep)))
NeilBrownb47490c2008-02-06 01:39:50 -08001588 return;
1589 wait_event(bitmap->mddev->recovery_wait,
1590 atomic_read(&bitmap->mddev->recovery_active) == 0);
1591
NeilBrown75d3da42011-01-14 09:14:34 +11001592 bitmap->mddev->curr_resync_completed = sector;
NeilBrown070dc6d2010-08-30 17:33:34 +10001593 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
NeilBrown40cffcc2012-05-22 13:55:24 +10001594 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
NeilBrownb47490c2008-02-06 01:39:50 -08001595 s = 0;
1596 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1597 bitmap_end_sync(bitmap, s, &blocks, 0);
1598 s += blocks;
1599 }
1600 bitmap->last_end_sync = jiffies;
NeilBrownacb180b2009-04-14 16:28:34 +10001601 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
NeilBrownb47490c2008-02-06 01:39:50 -08001602}
NeilBrownac2f40b2010-06-01 19:37:31 +10001603EXPORT_SYMBOL(bitmap_cond_end_sync);
NeilBrownb47490c2008-02-06 01:39:50 -08001604
NeilBrown6a079972005-09-09 16:23:44 -07001605static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
NeilBrown32a76272005-06-21 17:17:14 -07001606{
1607 /* For each chunk covered by any of these sectors, set the
NeilBrownef99bf42012-05-22 13:55:08 +10001608 * counter to 2 and possibly set resync_needed. They should all
NeilBrown32a76272005-06-21 17:17:14 -07001609 * be 0 at this point
1610 */
NeilBrown193f1c92005-08-04 12:53:33 -07001611
NeilBrown57dab0b2010-10-19 10:03:39 +11001612 sector_t secs;
NeilBrown193f1c92005-08-04 12:53:33 -07001613 bitmap_counter_t *bmc;
NeilBrown40cffcc2012-05-22 13:55:24 +10001614 spin_lock_irq(&bitmap->counts.lock);
1615 bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
NeilBrown193f1c92005-08-04 12:53:33 -07001616 if (!bmc) {
NeilBrown40cffcc2012-05-22 13:55:24 +10001617 spin_unlock_irq(&bitmap->counts.lock);
NeilBrown193f1c92005-08-04 12:53:33 -07001618 return;
NeilBrown32a76272005-06-21 17:17:14 -07001619 }
NeilBrownac2f40b2010-06-01 19:37:31 +10001620 if (!*bmc) {
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001621 *bmc = 2;
NeilBrown40cffcc2012-05-22 13:55:24 +10001622 bitmap_count_page(&bitmap->counts, offset, 1);
1623 bitmap_set_pending(&bitmap->counts, offset);
NeilBrown2585f3e2011-09-21 15:37:46 +10001624 bitmap->allclean = 0;
NeilBrown193f1c92005-08-04 12:53:33 -07001625 }
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001626 if (needed)
1627 *bmc |= NEEDED_MASK;
NeilBrown40cffcc2012-05-22 13:55:24 +10001628 spin_unlock_irq(&bitmap->counts.lock);
NeilBrown32a76272005-06-21 17:17:14 -07001629}
1630
Paul Clements9b1d1da2006-10-03 01:15:49 -07001631/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1632void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1633{
1634 unsigned long chunk;
1635
1636 for (chunk = s; chunk <= e; chunk++) {
NeilBrown40cffcc2012-05-22 13:55:24 +10001637 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
Paul Clements9b1d1da2006-10-03 01:15:49 -07001638 bitmap_set_memory_bits(bitmap, sec, 1);
1639 bitmap_file_set_bit(bitmap, sec);
NeilBrownffa23322009-12-14 12:49:56 +11001640 if (sec < bitmap->mddev->recovery_cp)
1641 /* We are asserting that the array is dirty,
1642 * so move the recovery_cp address back so
1643 * that it is obvious that it is dirty
1644 */
1645 bitmap->mddev->recovery_cp = sec;
Paul Clements9b1d1da2006-10-03 01:15:49 -07001646 }
1647}
1648
NeilBrown32a76272005-06-21 17:17:14 -07001649/*
NeilBrown6b8b3e82005-08-04 12:53:35 -07001650 * flush out any pending updates
1651 */
NeilBrownfd01b882011-10-11 16:47:53 +11001652void bitmap_flush(struct mddev *mddev)
NeilBrown6b8b3e82005-08-04 12:53:35 -07001653{
1654 struct bitmap *bitmap = mddev->bitmap;
NeilBrown42a04b52009-12-14 12:49:53 +11001655 long sleep;
NeilBrown6b8b3e82005-08-04 12:53:35 -07001656
1657 if (!bitmap) /* there was no bitmap */
1658 return;
1659
1660 /* run the daemon_work three time to ensure everything is flushed
1661 * that can be
1662 */
NeilBrown1b04be92009-12-14 12:49:53 +11001663 sleep = mddev->bitmap_info.daemon_sleep * 2;
NeilBrown42a04b52009-12-14 12:49:53 +11001664 bitmap->daemon_lastrun -= sleep;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001665 bitmap_daemon_work(mddev);
NeilBrown42a04b52009-12-14 12:49:53 +11001666 bitmap->daemon_lastrun -= sleep;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001667 bitmap_daemon_work(mddev);
NeilBrown42a04b52009-12-14 12:49:53 +11001668 bitmap->daemon_lastrun -= sleep;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001669 bitmap_daemon_work(mddev);
NeilBrown6b8b3e82005-08-04 12:53:35 -07001670 bitmap_update_sb(bitmap);
1671}
1672
1673/*
NeilBrown32a76272005-06-21 17:17:14 -07001674 * free memory that was allocated
1675 */
NeilBrown3178b0d2005-09-09 16:23:50 -07001676static void bitmap_free(struct bitmap *bitmap)
NeilBrown32a76272005-06-21 17:17:14 -07001677{
1678 unsigned long k, pages;
1679 struct bitmap_page *bp;
NeilBrown32a76272005-06-21 17:17:14 -07001680
1681 if (!bitmap) /* there was no bitmap */
1682 return;
1683
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001684 if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1685 bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -05001686 md_cluster_stop(bitmap->mddev);
1687
NeilBrownfae7d322012-05-22 13:55:21 +10001688 /* Shouldn't be needed - but just in case.... */
1689 wait_event(bitmap->write_wait,
1690 atomic_read(&bitmap->pending_writes) == 0);
1691
1692 /* release the bitmap file */
1693 bitmap_file_unmap(&bitmap->storage);
NeilBrown32a76272005-06-21 17:17:14 -07001694
NeilBrown40cffcc2012-05-22 13:55:24 +10001695 bp = bitmap->counts.bp;
1696 pages = bitmap->counts.pages;
NeilBrown32a76272005-06-21 17:17:14 -07001697
1698 /* free all allocated memory */
1699
NeilBrown32a76272005-06-21 17:17:14 -07001700 if (bp) /* deallocate the page memory */
1701 for (k = 0; k < pages; k++)
1702 if (bp[k].map && !bp[k].hijacked)
1703 kfree(bp[k].map);
1704 kfree(bp);
1705 kfree(bitmap);
1706}
NeilBrownaa5cbd12009-12-14 12:49:46 +11001707
NeilBrownfd01b882011-10-11 16:47:53 +11001708void bitmap_destroy(struct mddev *mddev)
NeilBrown3178b0d2005-09-09 16:23:50 -07001709{
1710 struct bitmap *bitmap = mddev->bitmap;
1711
1712 if (!bitmap) /* there was no bitmap */
1713 return;
1714
NeilBrownc3d97142009-12-14 12:49:52 +11001715 mutex_lock(&mddev->bitmap_info.mutex);
NeilBrown978a7a42014-12-15 12:56:58 +11001716 spin_lock(&mddev->lock);
NeilBrown3178b0d2005-09-09 16:23:50 -07001717 mddev->bitmap = NULL; /* disconnect from the md device */
NeilBrown978a7a42014-12-15 12:56:58 +11001718 spin_unlock(&mddev->lock);
NeilBrownc3d97142009-12-14 12:49:52 +11001719 mutex_unlock(&mddev->bitmap_info.mutex);
NeilBrownb15c2e52006-01-06 00:20:16 -08001720 if (mddev->thread)
1721 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
NeilBrown3178b0d2005-09-09 16:23:50 -07001722
NeilBrownece5cff2009-12-14 12:49:56 +11001723 if (bitmap->sysfs_can_clear)
1724 sysfs_put(bitmap->sysfs_can_clear);
1725
NeilBrown3178b0d2005-09-09 16:23:50 -07001726 bitmap_free(bitmap);
1727}
NeilBrown32a76272005-06-21 17:17:14 -07001728
1729/*
1730 * initialize the bitmap structure
1731 * if this returns an error, bitmap_destroy must be called to do clean up
1732 */
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001733struct bitmap *bitmap_create(struct mddev *mddev, int slot)
NeilBrown32a76272005-06-21 17:17:14 -07001734{
1735 struct bitmap *bitmap;
NeilBrown1f593902009-04-20 11:50:24 +10001736 sector_t blocks = mddev->resync_max_sectors;
NeilBrownc3d97142009-12-14 12:49:52 +11001737 struct file *file = mddev->bitmap_info.file;
NeilBrown32a76272005-06-21 17:17:14 -07001738 int err;
Tejun Heo324a56e2013-12-11 14:11:53 -05001739 struct kernfs_node *bm = NULL;
NeilBrown32a76272005-06-21 17:17:14 -07001740
Alexey Dobriyan5f6e3c832006-10-11 01:22:26 -07001741 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
NeilBrown32a76272005-06-21 17:17:14 -07001742
NeilBrownc3d97142009-12-14 12:49:52 +11001743 BUG_ON(file && mddev->bitmap_info.offset);
NeilBrowna654b9d82005-06-21 17:17:27 -07001744
NeilBrown9ffae0c2006-01-06 00:20:32 -08001745 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
NeilBrown32a76272005-06-21 17:17:14 -07001746 if (!bitmap)
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001747 return ERR_PTR(-ENOMEM);
NeilBrown32a76272005-06-21 17:17:14 -07001748
NeilBrown40cffcc2012-05-22 13:55:24 +10001749 spin_lock_init(&bitmap->counts.lock);
NeilBrownce25c312006-06-26 00:27:49 -07001750 atomic_set(&bitmap->pending_writes, 0);
1751 init_waitqueue_head(&bitmap->write_wait);
Neil Brownda6e1a32007-02-08 14:20:37 -08001752 init_waitqueue_head(&bitmap->overflow_wait);
NeilBrowne5551902010-03-31 11:21:44 +11001753 init_waitqueue_head(&bitmap->behind_wait);
NeilBrownce25c312006-06-26 00:27:49 -07001754
NeilBrown32a76272005-06-21 17:17:14 -07001755 bitmap->mddev = mddev;
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001756 bitmap->cluster_slot = slot;
NeilBrown32a76272005-06-21 17:17:14 -07001757
NeilBrown5ff5aff2010-06-01 19:37:32 +10001758 if (mddev->kobj.sd)
Tejun Heo388975c2013-09-11 23:19:13 -04001759 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
NeilBrownece5cff2009-12-14 12:49:56 +11001760 if (bm) {
Tejun Heo388975c2013-09-11 23:19:13 -04001761 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
NeilBrownece5cff2009-12-14 12:49:56 +11001762 sysfs_put(bm);
1763 } else
1764 bitmap->sysfs_can_clear = NULL;
1765
NeilBrown1ec885c2012-05-22 13:55:10 +10001766 bitmap->storage.file = file;
NeilBrownce25c312006-06-26 00:27:49 -07001767 if (file) {
1768 get_file(file);
NeilBrownae8fa282009-10-16 15:56:01 +11001769 /* As future accesses to this file will use bmap,
1770 * and bypass the page cache, we must sync the file
1771 * first.
1772 */
Christoph Hellwig8018ab02010-03-22 17:32:25 +01001773 vfs_fsync(file, 1);
NeilBrownce25c312006-06-26 00:27:49 -07001774 }
NeilBrown42a04b52009-12-14 12:49:53 +11001775 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
Jonathan Brassow9c810752011-06-08 17:59:30 -05001776 if (!mddev->bitmap_info.external) {
1777 /*
1778 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1779 * instructing us to create a new on-disk bitmap instance.
1780 */
1781 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1782 err = bitmap_new_disk_sb(bitmap);
1783 else
1784 err = bitmap_read_sb(bitmap);
1785 } else {
NeilBrownece5cff2009-12-14 12:49:56 +11001786 err = 0;
1787 if (mddev->bitmap_info.chunksize == 0 ||
1788 mddev->bitmap_info.daemon_sleep == 0)
1789 /* chunksize and time_base need to be
1790 * set first. */
1791 err = -EINVAL;
1792 }
NeilBrown32a76272005-06-21 17:17:14 -07001793 if (err)
NeilBrown3178b0d2005-09-09 16:23:50 -07001794 goto error;
NeilBrown32a76272005-06-21 17:17:14 -07001795
NeilBrown624ce4f2009-12-14 12:49:56 +11001796 bitmap->daemon_lastrun = jiffies;
NeilBrownd60b4792012-05-22 13:55:25 +10001797 err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1798 if (err)
NeilBrown3178b0d2005-09-09 16:23:50 -07001799 goto error;
NeilBrown32a76272005-06-21 17:17:14 -07001800
NeilBrown69e51b42010-06-01 19:37:35 +10001801 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
NeilBrownd60b4792012-05-22 13:55:25 +10001802 bitmap->counts.pages, bmname(bitmap));
NeilBrown69e51b42010-06-01 19:37:35 +10001803
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001804 err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1805 if (err)
1806 goto error;
NeilBrown69e51b42010-06-01 19:37:35 +10001807
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001808 return bitmap;
NeilBrown69e51b42010-06-01 19:37:35 +10001809 error:
1810 bitmap_free(bitmap);
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001811 return ERR_PTR(err);
NeilBrown69e51b42010-06-01 19:37:35 +10001812}
1813
NeilBrownfd01b882011-10-11 16:47:53 +11001814int bitmap_load(struct mddev *mddev)
NeilBrown69e51b42010-06-01 19:37:35 +10001815{
1816 int err = 0;
Jonathan Brassow3520fa42011-07-27 11:00:37 +10001817 sector_t start = 0;
NeilBrown69e51b42010-06-01 19:37:35 +10001818 sector_t sector = 0;
1819 struct bitmap *bitmap = mddev->bitmap;
1820
1821 if (!bitmap)
1822 goto out;
1823
1824 /* Clear out old bitmap info first: Either there is none, or we
1825 * are resuming after someone else has possibly changed things,
1826 * so we should forget old cached info.
1827 * All chunks should be clean, but some might need_sync.
1828 */
1829 while (sector < mddev->resync_max_sectors) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001830 sector_t blocks;
NeilBrown69e51b42010-06-01 19:37:35 +10001831 bitmap_start_sync(bitmap, sector, &blocks, 0);
1832 sector += blocks;
1833 }
1834 bitmap_close_sync(bitmap);
1835
Jonathan Brassow3520fa42011-07-27 11:00:37 +10001836 if (mddev->degraded == 0
1837 || bitmap->events_cleared == mddev->events)
1838 /* no need to keep dirty bits to optimise a
1839 * re-add of a missing device */
1840 start = mddev->recovery_cp;
NeilBrown69e51b42010-06-01 19:37:35 +10001841
NeilBrownafbaa902012-04-12 16:05:06 +10001842 mutex_lock(&mddev->bitmap_info.mutex);
Jonathan Brassow3520fa42011-07-27 11:00:37 +10001843 err = bitmap_init_from_disk(bitmap, start);
NeilBrownafbaa902012-04-12 16:05:06 +10001844 mutex_unlock(&mddev->bitmap_info.mutex);
Jonathan Brassow3520fa42011-07-27 11:00:37 +10001845
NeilBrown32a76272005-06-21 17:17:14 -07001846 if (err)
NeilBrown69e51b42010-06-01 19:37:35 +10001847 goto out;
NeilBrownb405fe92012-05-22 13:55:15 +10001848 clear_bit(BITMAP_STALE, &bitmap->flags);
NeilBrownef99bf42012-05-22 13:55:08 +10001849
1850 /* Kick recovery in case any bits were set */
1851 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
NeilBrown3178b0d2005-09-09 16:23:50 -07001852
NeilBrown1b04be92009-12-14 12:49:53 +11001853 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
NeilBrown9cd30fd2009-12-14 12:49:54 +11001854 md_wakeup_thread(mddev->thread);
NeilBrownb15c2e52006-01-06 00:20:16 -08001855
NeilBrown4ad13662007-07-17 04:06:13 -07001856 bitmap_update_sb(bitmap);
1857
NeilBrownb405fe92012-05-22 13:55:15 +10001858 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
NeilBrown69e51b42010-06-01 19:37:35 +10001859 err = -EIO;
1860out:
NeilBrown3178b0d2005-09-09 16:23:50 -07001861 return err;
NeilBrown32a76272005-06-21 17:17:14 -07001862}
NeilBrown69e51b42010-06-01 19:37:35 +10001863EXPORT_SYMBOL_GPL(bitmap_load);
NeilBrown32a76272005-06-21 17:17:14 -07001864
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001865/* Loads the bitmap associated with slot and copies the resync information
1866 * to our bitmap
1867 */
1868int bitmap_copy_from_slot(struct mddev *mddev, int slot,
Goldwyn Rodrigues97f6cd32015-04-14 10:45:42 -05001869 sector_t *low, sector_t *high, bool clear_bits)
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001870{
1871 int rv = 0, i, j;
1872 sector_t block, lo = 0, hi = 0;
1873 struct bitmap_counts *counts;
1874 struct bitmap *bitmap = bitmap_create(mddev, slot);
1875
1876 if (IS_ERR(bitmap))
1877 return PTR_ERR(bitmap);
1878
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001879 rv = bitmap_init_from_disk(bitmap, 0);
1880 if (rv)
1881 goto err;
1882
1883 counts = &bitmap->counts;
1884 for (j = 0; j < counts->chunks; j++) {
1885 block = (sector_t)j << counts->chunkshift;
1886 if (bitmap_file_test_bit(bitmap, block)) {
1887 if (!lo)
1888 lo = block;
1889 hi = block;
1890 bitmap_file_clear_bit(bitmap, block);
1891 bitmap_set_memory_bits(mddev->bitmap, block, 1);
1892 bitmap_file_set_bit(mddev->bitmap, block);
1893 }
1894 }
1895
Goldwyn Rodrigues97f6cd32015-04-14 10:45:42 -05001896 if (clear_bits) {
1897 bitmap_update_sb(bitmap);
1898 /* Setting this for the ev_page should be enough.
1899 * And we do not require both write_all and PAGE_DIRT either
1900 */
1901 for (i = 0; i < bitmap->storage.file_pages; i++)
1902 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1903 bitmap_write_all(bitmap);
1904 bitmap_unplug(bitmap);
1905 }
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001906 *low = lo;
1907 *high = hi;
1908err:
1909 bitmap_free(bitmap);
1910 return rv;
1911}
1912EXPORT_SYMBOL_GPL(bitmap_copy_from_slot);
1913
1914
NeilBrown57148962012-03-19 12:46:40 +11001915void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1916{
1917 unsigned long chunk_kb;
NeilBrown40cffcc2012-05-22 13:55:24 +10001918 struct bitmap_counts *counts;
NeilBrown57148962012-03-19 12:46:40 +11001919
1920 if (!bitmap)
1921 return;
1922
NeilBrown40cffcc2012-05-22 13:55:24 +10001923 counts = &bitmap->counts;
1924
NeilBrown57148962012-03-19 12:46:40 +11001925 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1926 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1927 "%lu%s chunk",
NeilBrown40cffcc2012-05-22 13:55:24 +10001928 counts->pages - counts->missing_pages,
1929 counts->pages,
1930 (counts->pages - counts->missing_pages)
NeilBrown57148962012-03-19 12:46:40 +11001931 << (PAGE_SHIFT - 10),
1932 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1933 chunk_kb ? "KB" : "B");
NeilBrown1ec885c2012-05-22 13:55:10 +10001934 if (bitmap->storage.file) {
NeilBrown57148962012-03-19 12:46:40 +11001935 seq_printf(seq, ", file: ");
Miklos Szeredi2726d562015-06-19 10:30:28 +02001936 seq_file_path(seq, bitmap->storage.file, " \t\n");
NeilBrown57148962012-03-19 12:46:40 +11001937 }
1938
1939 seq_printf(seq, "\n");
NeilBrown57148962012-03-19 12:46:40 +11001940}
1941
NeilBrownd60b4792012-05-22 13:55:25 +10001942int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
1943 int chunksize, int init)
1944{
1945 /* If chunk_size is 0, choose an appropriate chunk size.
1946 * Then possibly allocate new storage space.
1947 * Then quiesce, copy bits, replace bitmap, and re-start
1948 *
1949 * This function is called both to set up the initial bitmap
1950 * and to resize the bitmap while the array is active.
1951 * If this happens as a result of the array being resized,
1952 * chunksize will be zero, and we need to choose a suitable
1953 * chunksize, otherwise we use what we are given.
1954 */
1955 struct bitmap_storage store;
1956 struct bitmap_counts old_counts;
1957 unsigned long chunks;
1958 sector_t block;
1959 sector_t old_blocks, new_blocks;
1960 int chunkshift;
1961 int ret = 0;
1962 long pages;
1963 struct bitmap_page *new_bp;
1964
1965 if (chunksize == 0) {
1966 /* If there is enough space, leave the chunk size unchanged,
1967 * else increase by factor of two until there is enough space.
1968 */
1969 long bytes;
1970 long space = bitmap->mddev->bitmap_info.space;
1971
1972 if (space == 0) {
1973 /* We don't know how much space there is, so limit
1974 * to current size - in sectors.
1975 */
1976 bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
1977 if (!bitmap->mddev->bitmap_info.external)
1978 bytes += sizeof(bitmap_super_t);
1979 space = DIV_ROUND_UP(bytes, 512);
1980 bitmap->mddev->bitmap_info.space = space;
1981 }
1982 chunkshift = bitmap->counts.chunkshift;
1983 chunkshift--;
1984 do {
1985 /* 'chunkshift' is shift from block size to chunk size */
1986 chunkshift++;
1987 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1988 bytes = DIV_ROUND_UP(chunks, 8);
1989 if (!bitmap->mddev->bitmap_info.external)
1990 bytes += sizeof(bitmap_super_t);
1991 } while (bytes > (space << 9));
1992 } else
1993 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
1994
1995 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1996 memset(&store, 0, sizeof(store));
1997 if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
1998 ret = bitmap_storage_alloc(&store, chunks,
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -05001999 !bitmap->mddev->bitmap_info.external,
2000 bitmap->cluster_slot);
NeilBrownd60b4792012-05-22 13:55:25 +10002001 if (ret)
2002 goto err;
2003
2004 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2005
2006 new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL);
2007 ret = -ENOMEM;
2008 if (!new_bp) {
2009 bitmap_file_unmap(&store);
2010 goto err;
2011 }
2012
2013 if (!init)
2014 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2015
2016 store.file = bitmap->storage.file;
2017 bitmap->storage.file = NULL;
2018
2019 if (store.sb_page && bitmap->storage.sb_page)
2020 memcpy(page_address(store.sb_page),
2021 page_address(bitmap->storage.sb_page),
2022 sizeof(bitmap_super_t));
2023 bitmap_file_unmap(&bitmap->storage);
2024 bitmap->storage = store;
2025
2026 old_counts = bitmap->counts;
2027 bitmap->counts.bp = new_bp;
2028 bitmap->counts.pages = pages;
2029 bitmap->counts.missing_pages = pages;
2030 bitmap->counts.chunkshift = chunkshift;
2031 bitmap->counts.chunks = chunks;
2032 bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
2033 BITMAP_BLOCK_SHIFT);
2034
2035 blocks = min(old_counts.chunks << old_counts.chunkshift,
2036 chunks << chunkshift);
2037
2038 spin_lock_irq(&bitmap->counts.lock);
2039 for (block = 0; block < blocks; ) {
2040 bitmap_counter_t *bmc_old, *bmc_new;
2041 int set;
2042
2043 bmc_old = bitmap_get_counter(&old_counts, block,
2044 &old_blocks, 0);
2045 set = bmc_old && NEEDED(*bmc_old);
2046
2047 if (set) {
2048 bmc_new = bitmap_get_counter(&bitmap->counts, block,
2049 &new_blocks, 1);
2050 if (*bmc_new == 0) {
2051 /* need to set on-disk bits too. */
2052 sector_t end = block + new_blocks;
2053 sector_t start = block >> chunkshift;
2054 start <<= chunkshift;
2055 while (start < end) {
2056 bitmap_file_set_bit(bitmap, block);
2057 start += 1 << chunkshift;
2058 }
2059 *bmc_new = 2;
2060 bitmap_count_page(&bitmap->counts,
2061 block, 1);
2062 bitmap_set_pending(&bitmap->counts,
2063 block);
2064 }
2065 *bmc_new |= NEEDED_MASK;
2066 if (new_blocks < old_blocks)
2067 old_blocks = new_blocks;
2068 }
2069 block += old_blocks;
2070 }
2071
2072 if (!init) {
2073 int i;
2074 while (block < (chunks << chunkshift)) {
2075 bitmap_counter_t *bmc;
2076 bmc = bitmap_get_counter(&bitmap->counts, block,
2077 &new_blocks, 1);
2078 if (bmc) {
2079 /* new space. It needs to be resynced, so
2080 * we set NEEDED_MASK.
2081 */
2082 if (*bmc == 0) {
2083 *bmc = NEEDED_MASK | 2;
2084 bitmap_count_page(&bitmap->counts,
2085 block, 1);
2086 bitmap_set_pending(&bitmap->counts,
2087 block);
2088 }
2089 }
2090 block += new_blocks;
2091 }
2092 for (i = 0; i < bitmap->storage.file_pages; i++)
2093 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2094 }
2095 spin_unlock_irq(&bitmap->counts.lock);
2096
2097 if (!init) {
2098 bitmap_unplug(bitmap);
2099 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2100 }
2101 ret = 0;
2102err:
2103 return ret;
2104}
2105EXPORT_SYMBOL_GPL(bitmap_resize);
2106
NeilBrown43a70502009-12-14 12:49:55 +11002107static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002108location_show(struct mddev *mddev, char *page)
NeilBrown43a70502009-12-14 12:49:55 +11002109{
2110 ssize_t len;
NeilBrownac2f40b2010-06-01 19:37:31 +10002111 if (mddev->bitmap_info.file)
NeilBrown43a70502009-12-14 12:49:55 +11002112 len = sprintf(page, "file");
NeilBrownac2f40b2010-06-01 19:37:31 +10002113 else if (mddev->bitmap_info.offset)
NeilBrown43a70502009-12-14 12:49:55 +11002114 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
NeilBrownac2f40b2010-06-01 19:37:31 +10002115 else
NeilBrown43a70502009-12-14 12:49:55 +11002116 len = sprintf(page, "none");
2117 len += sprintf(page+len, "\n");
2118 return len;
2119}
2120
2121static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002122location_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown43a70502009-12-14 12:49:55 +11002123{
2124
2125 if (mddev->pers) {
2126 if (!mddev->pers->quiesce)
2127 return -EBUSY;
2128 if (mddev->recovery || mddev->sync_thread)
2129 return -EBUSY;
2130 }
2131
2132 if (mddev->bitmap || mddev->bitmap_info.file ||
2133 mddev->bitmap_info.offset) {
2134 /* bitmap already configured. Only option is to clear it */
2135 if (strncmp(buf, "none", 4) != 0)
2136 return -EBUSY;
2137 if (mddev->pers) {
2138 mddev->pers->quiesce(mddev, 1);
2139 bitmap_destroy(mddev);
2140 mddev->pers->quiesce(mddev, 0);
2141 }
2142 mddev->bitmap_info.offset = 0;
2143 if (mddev->bitmap_info.file) {
2144 struct file *f = mddev->bitmap_info.file;
2145 mddev->bitmap_info.file = NULL;
NeilBrown43a70502009-12-14 12:49:55 +11002146 fput(f);
2147 }
2148 } else {
2149 /* No bitmap, OK to set a location */
2150 long long offset;
2151 if (strncmp(buf, "none", 4) == 0)
2152 /* nothing to be done */;
2153 else if (strncmp(buf, "file:", 5) == 0) {
2154 /* Not supported yet */
2155 return -EINVAL;
2156 } else {
2157 int rv;
2158 if (buf[0] == '+')
Jingoo Hanb29bebd2013-06-01 16:15:16 +09002159 rv = kstrtoll(buf+1, 10, &offset);
NeilBrown43a70502009-12-14 12:49:55 +11002160 else
Jingoo Hanb29bebd2013-06-01 16:15:16 +09002161 rv = kstrtoll(buf, 10, &offset);
NeilBrown43a70502009-12-14 12:49:55 +11002162 if (rv)
2163 return rv;
2164 if (offset == 0)
2165 return -EINVAL;
NeilBrownece5cff2009-12-14 12:49:56 +11002166 if (mddev->bitmap_info.external == 0 &&
2167 mddev->major_version == 0 &&
NeilBrown43a70502009-12-14 12:49:55 +11002168 offset != mddev->bitmap_info.default_offset)
2169 return -EINVAL;
2170 mddev->bitmap_info.offset = offset;
2171 if (mddev->pers) {
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05002172 struct bitmap *bitmap;
NeilBrown43a70502009-12-14 12:49:55 +11002173 mddev->pers->quiesce(mddev, 1);
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05002174 bitmap = bitmap_create(mddev, -1);
2175 if (IS_ERR(bitmap))
2176 rv = PTR_ERR(bitmap);
2177 else {
2178 mddev->bitmap = bitmap;
NeilBrown4474ca42012-03-19 12:46:37 +11002179 rv = bitmap_load(mddev);
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05002180 if (rv) {
2181 bitmap_destroy(mddev);
2182 mddev->bitmap_info.offset = 0;
2183 }
NeilBrown43a70502009-12-14 12:49:55 +11002184 }
2185 mddev->pers->quiesce(mddev, 0);
2186 if (rv)
2187 return rv;
2188 }
2189 }
2190 }
2191 if (!mddev->external) {
2192 /* Ensure new bitmap info is stored in
2193 * metadata promptly.
2194 */
2195 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2196 md_wakeup_thread(mddev->thread);
2197 }
2198 return len;
2199}
2200
2201static struct md_sysfs_entry bitmap_location =
2202__ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2203
NeilBrown6409bb02012-05-22 13:55:07 +10002204/* 'bitmap/space' is the space available at 'location' for the
2205 * bitmap. This allows the kernel to know when it is safe to
2206 * resize the bitmap to match a resized array.
2207 */
2208static ssize_t
2209space_show(struct mddev *mddev, char *page)
2210{
2211 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2212}
2213
2214static ssize_t
2215space_store(struct mddev *mddev, const char *buf, size_t len)
2216{
2217 unsigned long sectors;
2218 int rv;
2219
2220 rv = kstrtoul(buf, 10, &sectors);
2221 if (rv)
2222 return rv;
2223
2224 if (sectors == 0)
2225 return -EINVAL;
2226
2227 if (mddev->bitmap &&
NeilBrown9b1215c2012-05-22 13:55:11 +10002228 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
NeilBrown6409bb02012-05-22 13:55:07 +10002229 return -EFBIG; /* Bitmap is too big for this small space */
2230
2231 /* could make sure it isn't too big, but that isn't really
2232 * needed - user-space should be careful.
2233 */
2234 mddev->bitmap_info.space = sectors;
2235 return len;
2236}
2237
2238static struct md_sysfs_entry bitmap_space =
2239__ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2240
NeilBrown43a70502009-12-14 12:49:55 +11002241static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002242timeout_show(struct mddev *mddev, char *page)
NeilBrown43a70502009-12-14 12:49:55 +11002243{
2244 ssize_t len;
2245 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2246 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
NeilBrownac2f40b2010-06-01 19:37:31 +10002247
NeilBrown43a70502009-12-14 12:49:55 +11002248 len = sprintf(page, "%lu", secs);
2249 if (jifs)
2250 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2251 len += sprintf(page+len, "\n");
2252 return len;
2253}
2254
2255static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002256timeout_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown43a70502009-12-14 12:49:55 +11002257{
2258 /* timeout can be set at any time */
2259 unsigned long timeout;
2260 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2261 if (rv)
2262 return rv;
2263
2264 /* just to make sure we don't overflow... */
2265 if (timeout >= LONG_MAX / HZ)
2266 return -EINVAL;
2267
2268 timeout = timeout * HZ / 10000;
2269
2270 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2271 timeout = MAX_SCHEDULE_TIMEOUT-1;
2272 if (timeout < 1)
2273 timeout = 1;
2274 mddev->bitmap_info.daemon_sleep = timeout;
2275 if (mddev->thread) {
2276 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2277 * the bitmap is all clean and we don't need to
2278 * adjust the timeout right now
2279 */
2280 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2281 mddev->thread->timeout = timeout;
2282 md_wakeup_thread(mddev->thread);
2283 }
2284 }
2285 return len;
2286}
2287
2288static struct md_sysfs_entry bitmap_timeout =
2289__ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2290
2291static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002292backlog_show(struct mddev *mddev, char *page)
NeilBrown43a70502009-12-14 12:49:55 +11002293{
2294 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2295}
2296
2297static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002298backlog_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown43a70502009-12-14 12:49:55 +11002299{
2300 unsigned long backlog;
Jingoo Hanb29bebd2013-06-01 16:15:16 +09002301 int rv = kstrtoul(buf, 10, &backlog);
NeilBrown43a70502009-12-14 12:49:55 +11002302 if (rv)
2303 return rv;
2304 if (backlog > COUNTER_MAX)
2305 return -EINVAL;
2306 mddev->bitmap_info.max_write_behind = backlog;
2307 return len;
2308}
2309
2310static struct md_sysfs_entry bitmap_backlog =
2311__ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2312
2313static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002314chunksize_show(struct mddev *mddev, char *page)
NeilBrown43a70502009-12-14 12:49:55 +11002315{
2316 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2317}
2318
2319static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002320chunksize_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown43a70502009-12-14 12:49:55 +11002321{
2322 /* Can only be changed when no bitmap is active */
2323 int rv;
2324 unsigned long csize;
2325 if (mddev->bitmap)
2326 return -EBUSY;
Jingoo Hanb29bebd2013-06-01 16:15:16 +09002327 rv = kstrtoul(buf, 10, &csize);
NeilBrown43a70502009-12-14 12:49:55 +11002328 if (rv)
2329 return rv;
2330 if (csize < 512 ||
2331 !is_power_of_2(csize))
2332 return -EINVAL;
2333 mddev->bitmap_info.chunksize = csize;
2334 return len;
2335}
2336
2337static struct md_sysfs_entry bitmap_chunksize =
2338__ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2339
NeilBrownfd01b882011-10-11 16:47:53 +11002340static ssize_t metadata_show(struct mddev *mddev, char *page)
NeilBrownece5cff2009-12-14 12:49:56 +11002341{
Goldwyn Rodriguesc4ce8672014-03-29 10:20:02 -05002342 if (mddev_is_clustered(mddev))
2343 return sprintf(page, "clustered\n");
NeilBrownece5cff2009-12-14 12:49:56 +11002344 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2345 ? "external" : "internal"));
2346}
2347
NeilBrownfd01b882011-10-11 16:47:53 +11002348static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrownece5cff2009-12-14 12:49:56 +11002349{
2350 if (mddev->bitmap ||
2351 mddev->bitmap_info.file ||
2352 mddev->bitmap_info.offset)
2353 return -EBUSY;
2354 if (strncmp(buf, "external", 8) == 0)
2355 mddev->bitmap_info.external = 1;
Goldwyn Rodriguesc4ce8672014-03-29 10:20:02 -05002356 else if ((strncmp(buf, "internal", 8) == 0) ||
2357 (strncmp(buf, "clustered", 9) == 0))
NeilBrownece5cff2009-12-14 12:49:56 +11002358 mddev->bitmap_info.external = 0;
2359 else
2360 return -EINVAL;
2361 return len;
2362}
2363
2364static struct md_sysfs_entry bitmap_metadata =
2365__ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2366
NeilBrownfd01b882011-10-11 16:47:53 +11002367static ssize_t can_clear_show(struct mddev *mddev, char *page)
NeilBrownece5cff2009-12-14 12:49:56 +11002368{
2369 int len;
NeilBrownb7b17c92014-12-15 12:56:59 +11002370 spin_lock(&mddev->lock);
NeilBrownece5cff2009-12-14 12:49:56 +11002371 if (mddev->bitmap)
2372 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2373 "false" : "true"));
2374 else
2375 len = sprintf(page, "\n");
NeilBrownb7b17c92014-12-15 12:56:59 +11002376 spin_unlock(&mddev->lock);
NeilBrownece5cff2009-12-14 12:49:56 +11002377 return len;
2378}
2379
NeilBrownfd01b882011-10-11 16:47:53 +11002380static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrownece5cff2009-12-14 12:49:56 +11002381{
2382 if (mddev->bitmap == NULL)
2383 return -ENOENT;
2384 if (strncmp(buf, "false", 5) == 0)
2385 mddev->bitmap->need_sync = 1;
2386 else if (strncmp(buf, "true", 4) == 0) {
2387 if (mddev->degraded)
2388 return -EBUSY;
2389 mddev->bitmap->need_sync = 0;
2390 } else
2391 return -EINVAL;
2392 return len;
2393}
2394
2395static struct md_sysfs_entry bitmap_can_clear =
2396__ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2397
Paul Clements696fcd52010-03-08 16:02:37 +11002398static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002399behind_writes_used_show(struct mddev *mddev, char *page)
Paul Clements696fcd52010-03-08 16:02:37 +11002400{
NeilBrownb7b17c92014-12-15 12:56:59 +11002401 ssize_t ret;
2402 spin_lock(&mddev->lock);
Paul Clements696fcd52010-03-08 16:02:37 +11002403 if (mddev->bitmap == NULL)
NeilBrownb7b17c92014-12-15 12:56:59 +11002404 ret = sprintf(page, "0\n");
2405 else
2406 ret = sprintf(page, "%lu\n",
2407 mddev->bitmap->behind_writes_used);
2408 spin_unlock(&mddev->lock);
2409 return ret;
Paul Clements696fcd52010-03-08 16:02:37 +11002410}
2411
2412static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002413behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
Paul Clements696fcd52010-03-08 16:02:37 +11002414{
2415 if (mddev->bitmap)
2416 mddev->bitmap->behind_writes_used = 0;
2417 return len;
2418}
2419
2420static struct md_sysfs_entry max_backlog_used =
2421__ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2422 behind_writes_used_show, behind_writes_used_reset);
2423
NeilBrown43a70502009-12-14 12:49:55 +11002424static struct attribute *md_bitmap_attrs[] = {
2425 &bitmap_location.attr,
NeilBrown6409bb02012-05-22 13:55:07 +10002426 &bitmap_space.attr,
NeilBrown43a70502009-12-14 12:49:55 +11002427 &bitmap_timeout.attr,
2428 &bitmap_backlog.attr,
2429 &bitmap_chunksize.attr,
NeilBrownece5cff2009-12-14 12:49:56 +11002430 &bitmap_metadata.attr,
2431 &bitmap_can_clear.attr,
Paul Clements696fcd52010-03-08 16:02:37 +11002432 &max_backlog_used.attr,
NeilBrown43a70502009-12-14 12:49:55 +11002433 NULL
2434};
2435struct attribute_group md_bitmap_group = {
2436 .name = "bitmap",
2437 .attrs = md_bitmap_attrs,
2438};
2439