blob: 2d826927a3bf9bd1da8c8f6850fa29e320011908 [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,
Guoqing Jiangc9d65032016-05-02 11:50:11 -040049 unsigned long page, int create, int no_hijack)
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");
Guoqing Jiangc9d65032016-05-02 11:50:11 -040093 /* We don't support hijack for cluster raid */
94 if (no_hijack)
95 return -ENOMEM;
NeilBrown32a76272005-06-21 17:17:14 -070096 /* failed - set the hijacked flag so that we can use the
97 * pointer as a counter */
NeilBrown32a76272005-06-21 17:17:14 -070098 if (!bitmap->bp[page].map)
99 bitmap->bp[page].hijacked = 1;
NeilBrownac2f40b2010-06-01 19:37:31 +1000100 } else if (bitmap->bp[page].map ||
101 bitmap->bp[page].hijacked) {
NeilBrown32a76272005-06-21 17:17:14 -0700102 /* somebody beat us to getting the page */
NeilBrown792a1d42012-03-19 12:46:41 +1100103 kfree(mappage);
NeilBrownac2f40b2010-06-01 19:37:31 +1000104 } else {
105
106 /* no page was in place and we have one, so install it */
107
108 bitmap->bp[page].map = mappage;
109 bitmap->missing_pages--;
NeilBrown32a76272005-06-21 17:17:14 -0700110 }
NeilBrown32a76272005-06-21 17:17:14 -0700111 return 0;
112}
113
NeilBrown32a76272005-06-21 17:17:14 -0700114/* if page is completely empty, put it back on the free list, or dealloc it */
115/* if page was hijacked, unmark the flag so it might get alloced next time */
116/* Note: lock should be held when calling this */
NeilBrown40cffcc2012-05-22 13:55:24 +1000117static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
NeilBrown32a76272005-06-21 17:17:14 -0700118{
119 char *ptr;
120
121 if (bitmap->bp[page].count) /* page is still busy */
122 return;
123
124 /* page is no longer in use, it can be released */
125
126 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
127 bitmap->bp[page].hijacked = 0;
128 bitmap->bp[page].map = NULL;
NeilBrownac2f40b2010-06-01 19:37:31 +1000129 } else {
130 /* normal case, free the page */
131 ptr = bitmap->bp[page].map;
132 bitmap->bp[page].map = NULL;
133 bitmap->missing_pages++;
NeilBrown792a1d42012-03-19 12:46:41 +1100134 kfree(ptr);
NeilBrown32a76272005-06-21 17:17:14 -0700135 }
NeilBrown32a76272005-06-21 17:17:14 -0700136}
137
NeilBrown32a76272005-06-21 17:17:14 -0700138/*
139 * bitmap file handling - read and write the bitmap file and its superblock
140 */
141
NeilBrown32a76272005-06-21 17:17:14 -0700142/*
143 * basic page I/O operations
144 */
145
NeilBrowna654b9d82005-06-21 17:17:27 -0700146/* IO operations when bitmap is stored near all superblocks */
NeilBrown27581e52012-05-22 13:55:08 +1000147static int read_sb_page(struct mddev *mddev, loff_t offset,
148 struct page *page,
149 unsigned long index, int size)
NeilBrowna654b9d82005-06-21 17:17:27 -0700150{
151 /* choose a good rdev and read the page from there */
152
NeilBrown3cb03002011-10-11 16:45:26 +1100153 struct md_rdev *rdev;
NeilBrowna654b9d82005-06-21 17:17:27 -0700154 sector_t target;
NeilBrowna654b9d82005-06-21 17:17:27 -0700155
NeilBrowndafb20f2012-03-19 12:46:39 +1100156 rdev_for_each(rdev, mddev) {
NeilBrownb2d444d2005-11-08 21:39:31 -0800157 if (! test_bit(In_sync, &rdev->flags)
158 || test_bit(Faulty, &rdev->flags))
NeilBrownab904d62005-09-09 16:23:52 -0700159 continue;
160
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100161 target = offset + index * (PAGE_SIZE/512);
NeilBrowna654b9d82005-06-21 17:17:27 -0700162
NeilBrown2b193362010-10-27 15:16:40 +1100163 if (sync_page_io(rdev, target,
Martin K. Petersene1defc42009-05-22 17:17:49 -0400164 roundup(size, bdev_logical_block_size(rdev->bdev)),
Mike Christie796a5cf2016-06-05 14:32:07 -0500165 page, REQ_OP_READ, 0, true)) {
NeilBrownab904d62005-09-09 16:23:52 -0700166 page->index = index;
NeilBrown27581e52012-05-22 13:55:08 +1000167 return 0;
NeilBrownab904d62005-09-09 16:23:52 -0700168 }
169 }
NeilBrown27581e52012-05-22 13:55:08 +1000170 return -EIO;
NeilBrowna654b9d82005-06-21 17:17:27 -0700171}
172
NeilBrownfd01b882011-10-11 16:47:53 +1100173static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000174{
175 /* Iterate the disks of an mddev, using rcu to protect access to the
176 * linked list, and raising the refcount of devices we return to ensure
177 * they don't disappear while in use.
178 * As devices are only added or removed when raid_disk is < 0 and
179 * nr_pending is 0 and In_sync is clear, the entries we return will
180 * still be in the same position on the list when we re-enter
Michael Wangfd177482012-10-11 13:43:21 +1100181 * list_for_each_entry_continue_rcu.
NeilBrown8532e342015-05-20 15:05:09 +1000182 *
183 * Note that if entered with 'rdev == NULL' to start at the
184 * beginning, we temporarily assign 'rdev' to an address which
185 * isn't really an rdev, but which can be used by
186 * list_for_each_entry_continue_rcu() to find the first entry.
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000187 */
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000188 rcu_read_lock();
189 if (rdev == NULL)
190 /* start at the beginning */
NeilBrown8532e342015-05-20 15:05:09 +1000191 rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000192 else {
193 /* release the previous rdev and start from there. */
194 rdev_dec_pending(rdev, mddev);
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000195 }
Michael Wangfd177482012-10-11 13:43:21 +1100196 list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000197 if (rdev->raid_disk >= 0 &&
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000198 !test_bit(Faulty, &rdev->flags)) {
199 /* this is a usable devices */
200 atomic_inc(&rdev->nr_pending);
201 rcu_read_unlock();
202 return rdev;
203 }
204 }
205 rcu_read_unlock();
206 return NULL;
207}
208
NeilBrownab6085c2007-05-23 13:58:10 -0700209static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
NeilBrowna654b9d82005-06-21 17:17:27 -0700210{
NeilBrown3cb03002011-10-11 16:45:26 +1100211 struct md_rdev *rdev = NULL;
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100212 struct block_device *bdev;
NeilBrownfd01b882011-10-11 16:47:53 +1100213 struct mddev *mddev = bitmap->mddev;
NeilBrown1ec885c2012-05-22 13:55:10 +1000214 struct bitmap_storage *store = &bitmap->storage;
NeilBrowna654b9d82005-06-21 17:17:27 -0700215
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000216 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
NeilBrownac2f40b2010-06-01 19:37:31 +1000217 int size = PAGE_SIZE;
218 loff_t offset = mddev->bitmap_info.offset;
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100219
220 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
221
NeilBrown9b1215c2012-05-22 13:55:11 +1000222 if (page->index == store->file_pages-1) {
223 int last_page_size = store->bytes & (PAGE_SIZE-1);
224 if (last_page_size == 0)
225 last_page_size = PAGE_SIZE;
226 size = roundup(last_page_size,
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100227 bdev_logical_block_size(bdev));
NeilBrown9b1215c2012-05-22 13:55:11 +1000228 }
NeilBrownac2f40b2010-06-01 19:37:31 +1000229 /* Just make sure we aren't corrupting data or
230 * metadata
231 */
232 if (mddev->external) {
233 /* Bitmap could be anywhere. */
234 if (rdev->sb_start + offset + (page->index
235 * (PAGE_SIZE/512))
236 > rdev->data_offset
237 &&
238 rdev->sb_start + offset
239 < (rdev->data_offset + mddev->dev_sectors
240 + (PAGE_SIZE/512)))
241 goto bad_alignment;
242 } else if (offset < 0) {
243 /* DATA BITMAP METADATA */
244 if (offset
245 + (long)(page->index * (PAGE_SIZE/512))
246 + size/512 > 0)
247 /* bitmap runs in to metadata */
248 goto bad_alignment;
249 if (rdev->data_offset + mddev->dev_sectors
250 > rdev->sb_start + offset)
251 /* data runs in to bitmap */
252 goto bad_alignment;
253 } else if (rdev->sb_start < rdev->data_offset) {
254 /* METADATA BITMAP DATA */
255 if (rdev->sb_start
256 + offset
257 + page->index*(PAGE_SIZE/512) + size/512
258 > rdev->data_offset)
259 /* bitmap runs in to data */
260 goto bad_alignment;
261 } else {
262 /* DATA METADATA BITMAP - no problems */
263 }
264 md_super_write(mddev, rdev,
265 rdev->sb_start + offset
266 + page->index * (PAGE_SIZE/512),
267 size,
268 page);
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000269 }
NeilBrowna654b9d82005-06-21 17:17:27 -0700270
271 if (wait)
NeilBrowna9701a32005-11-08 21:39:34 -0800272 md_super_wait(mddev);
NeilBrowna654b9d82005-06-21 17:17:27 -0700273 return 0;
NeilBrown4b809912008-07-21 17:05:25 +1000274
275 bad_alignment:
NeilBrown4b809912008-07-21 17:05:25 +1000276 return -EINVAL;
NeilBrowna654b9d82005-06-21 17:17:27 -0700277}
278
NeilBrown4ad13662007-07-17 04:06:13 -0700279static void bitmap_file_kick(struct bitmap *bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700280/*
NeilBrowna654b9d82005-06-21 17:17:27 -0700281 * write out a page to a file
NeilBrown32a76272005-06-21 17:17:14 -0700282 */
NeilBrown4ad13662007-07-17 04:06:13 -0700283static void write_page(struct bitmap *bitmap, struct page *page, int wait)
NeilBrown32a76272005-06-21 17:17:14 -0700284{
NeilBrownd785a062006-06-26 00:27:48 -0700285 struct buffer_head *bh;
NeilBrown32a76272005-06-21 17:17:14 -0700286
NeilBrown1ec885c2012-05-22 13:55:10 +1000287 if (bitmap->storage.file == NULL) {
NeilBrownf0d76d72007-07-17 04:06:12 -0700288 switch (write_sb_page(bitmap, page, wait)) {
289 case -EINVAL:
NeilBrownb405fe92012-05-22 13:55:15 +1000290 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
NeilBrownf0d76d72007-07-17 04:06:12 -0700291 }
NeilBrown4ad13662007-07-17 04:06:13 -0700292 } else {
NeilBrowna654b9d82005-06-21 17:17:27 -0700293
NeilBrown4ad13662007-07-17 04:06:13 -0700294 bh = page_buffers(page);
NeilBrownc7084432006-01-06 00:20:45 -0800295
NeilBrown4ad13662007-07-17 04:06:13 -0700296 while (bh && bh->b_blocknr) {
297 atomic_inc(&bitmap->pending_writes);
298 set_buffer_locked(bh);
299 set_buffer_mapped(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -0500300 submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
NeilBrown4ad13662007-07-17 04:06:13 -0700301 bh = bh->b_this_page;
302 }
NeilBrown32a76272005-06-21 17:17:14 -0700303
NeilBrownac2f40b2010-06-01 19:37:31 +1000304 if (wait)
NeilBrown4ad13662007-07-17 04:06:13 -0700305 wait_event(bitmap->write_wait,
306 atomic_read(&bitmap->pending_writes)==0);
NeilBrown32a76272005-06-21 17:17:14 -0700307 }
NeilBrownb405fe92012-05-22 13:55:15 +1000308 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
NeilBrown4ad13662007-07-17 04:06:13 -0700309 bitmap_file_kick(bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700310}
311
NeilBrownd785a062006-06-26 00:27:48 -0700312static void end_bitmap_write(struct buffer_head *bh, int uptodate)
NeilBrown32a76272005-06-21 17:17:14 -0700313{
NeilBrownd785a062006-06-26 00:27:48 -0700314 struct bitmap *bitmap = bh->b_private;
NeilBrownd785a062006-06-26 00:27:48 -0700315
NeilBrownb405fe92012-05-22 13:55:15 +1000316 if (!uptodate)
317 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
NeilBrownd785a062006-06-26 00:27:48 -0700318 if (atomic_dec_and_test(&bitmap->pending_writes))
319 wake_up(&bitmap->write_wait);
320}
321
322/* copied from buffer.c */
323static void
324__clear_page_buffers(struct page *page)
325{
326 ClearPagePrivate(page);
327 set_page_private(page, 0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300328 put_page(page);
NeilBrownd785a062006-06-26 00:27:48 -0700329}
330static void free_buffers(struct page *page)
331{
NeilBrown27581e52012-05-22 13:55:08 +1000332 struct buffer_head *bh;
NeilBrownd785a062006-06-26 00:27:48 -0700333
NeilBrown27581e52012-05-22 13:55:08 +1000334 if (!PagePrivate(page))
335 return;
336
337 bh = page_buffers(page);
NeilBrownd785a062006-06-26 00:27:48 -0700338 while (bh) {
339 struct buffer_head *next = bh->b_this_page;
340 free_buffer_head(bh);
341 bh = next;
342 }
343 __clear_page_buffers(page);
344 put_page(page);
345}
346
347/* read a page from a file.
348 * We both read the page, and attach buffers to the page to record the
349 * address of each block (using bmap). These addresses will be used
350 * to write the block later, completely bypassing the filesystem.
351 * This usage is similar to how swap files are handled, and allows us
352 * to write to a file with no concerns of memory allocation failing.
353 */
NeilBrown27581e52012-05-22 13:55:08 +1000354static int read_page(struct file *file, unsigned long index,
355 struct bitmap *bitmap,
356 unsigned long count,
357 struct page *page)
NeilBrownd785a062006-06-26 00:27:48 -0700358{
NeilBrown27581e52012-05-22 13:55:08 +1000359 int ret = 0;
Al Viro496ad9a2013-01-23 17:07:38 -0500360 struct inode *inode = file_inode(file);
NeilBrownd785a062006-06-26 00:27:48 -0700361 struct buffer_head *bh;
362 sector_t block;
NeilBrown32a76272005-06-21 17:17:14 -0700363
NeilBrown36a4e1f2011-10-07 14:23:17 +1100364 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
365 (unsigned long long)index << PAGE_SHIFT);
NeilBrown32a76272005-06-21 17:17:14 -0700366
NeilBrownd785a062006-06-26 00:27:48 -0700367 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
368 if (!bh) {
NeilBrown27581e52012-05-22 13:55:08 +1000369 ret = -ENOMEM;
NeilBrownd785a062006-06-26 00:27:48 -0700370 goto out;
371 }
372 attach_page_buffers(page, bh);
373 block = index << (PAGE_SHIFT - inode->i_blkbits);
374 while (bh) {
375 if (count == 0)
376 bh->b_blocknr = 0;
377 else {
378 bh->b_blocknr = bmap(inode, block);
379 if (bh->b_blocknr == 0) {
380 /* Cannot use this file! */
NeilBrown27581e52012-05-22 13:55:08 +1000381 ret = -EINVAL;
NeilBrownd785a062006-06-26 00:27:48 -0700382 goto out;
383 }
384 bh->b_bdev = inode->i_sb->s_bdev;
385 if (count < (1<<inode->i_blkbits))
386 count = 0;
387 else
388 count -= (1<<inode->i_blkbits);
NeilBrown32a76272005-06-21 17:17:14 -0700389
NeilBrownd785a062006-06-26 00:27:48 -0700390 bh->b_end_io = end_bitmap_write;
391 bh->b_private = bitmap;
NeilBrownce25c312006-06-26 00:27:49 -0700392 atomic_inc(&bitmap->pending_writes);
393 set_buffer_locked(bh);
394 set_buffer_mapped(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -0500395 submit_bh(REQ_OP_READ, 0, bh);
NeilBrownd785a062006-06-26 00:27:48 -0700396 }
397 block++;
398 bh = bh->b_this_page;
399 }
NeilBrownd785a062006-06-26 00:27:48 -0700400 page->index = index;
NeilBrownce25c312006-06-26 00:27:49 -0700401
402 wait_event(bitmap->write_wait,
403 atomic_read(&bitmap->pending_writes)==0);
NeilBrownb405fe92012-05-22 13:55:15 +1000404 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
NeilBrown27581e52012-05-22 13:55:08 +1000405 ret = -EIO;
NeilBrown32a76272005-06-21 17:17:14 -0700406out:
NeilBrown27581e52012-05-22 13:55:08 +1000407 if (ret)
408 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
NeilBrown2d1f3b52006-01-06 00:20:31 -0800409 (int)PAGE_SIZE,
410 (unsigned long long)index << PAGE_SHIFT,
NeilBrown27581e52012-05-22 13:55:08 +1000411 ret);
412 return ret;
NeilBrown32a76272005-06-21 17:17:14 -0700413}
414
415/*
416 * bitmap file superblock operations
417 */
418
419/* update the event counter and sync the superblock to disk */
NeilBrown4ad13662007-07-17 04:06:13 -0700420void bitmap_update_sb(struct bitmap *bitmap)
NeilBrown32a76272005-06-21 17:17:14 -0700421{
422 bitmap_super_t *sb;
NeilBrown32a76272005-06-21 17:17:14 -0700423
424 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
NeilBrown4ad13662007-07-17 04:06:13 -0700425 return;
NeilBrownece5cff2009-12-14 12:49:56 +1100426 if (bitmap->mddev->bitmap_info.external)
427 return;
NeilBrown1ec885c2012-05-22 13:55:10 +1000428 if (!bitmap->storage.sb_page) /* no superblock */
NeilBrown4ad13662007-07-17 04:06:13 -0700429 return;
NeilBrown1ec885c2012-05-22 13:55:10 +1000430 sb = kmap_atomic(bitmap->storage.sb_page);
NeilBrown32a76272005-06-21 17:17:14 -0700431 sb->events = cpu_to_le64(bitmap->mddev->events);
NeilBrown8258c532011-05-11 14:26:30 +1000432 if (bitmap->mddev->events < bitmap->events_cleared)
Neil Browna0da84f2008-06-28 08:31:22 +1000433 /* rocking back to read-only */
434 bitmap->events_cleared = bitmap->mddev->events;
NeilBrown8258c532011-05-11 14:26:30 +1000435 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
436 sb->state = cpu_to_le32(bitmap->flags);
NeilBrown43a70502009-12-14 12:49:55 +1100437 /* Just in case these have been changed via sysfs: */
438 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
439 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
NeilBrownb81a0402012-05-22 13:55:26 +1000440 /* This might have been changed by a reshape */
441 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
442 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
Goldwyn Rodriguesc4ce8672014-03-29 10:20:02 -0500443 sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
NeilBrown1dff2b82012-05-22 13:55:34 +1000444 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
445 bitmap_info.space);
Cong Wangb2f46e62011-11-28 13:25:44 +0800446 kunmap_atomic(sb);
NeilBrown1ec885c2012-05-22 13:55:10 +1000447 write_page(bitmap, bitmap->storage.sb_page, 1);
NeilBrown32a76272005-06-21 17:17:14 -0700448}
449
450/* print out the bitmap file superblock */
451void bitmap_print_sb(struct bitmap *bitmap)
452{
453 bitmap_super_t *sb;
454
NeilBrown1ec885c2012-05-22 13:55:10 +1000455 if (!bitmap || !bitmap->storage.sb_page)
NeilBrown32a76272005-06-21 17:17:14 -0700456 return;
NeilBrown1ec885c2012-05-22 13:55:10 +1000457 sb = kmap_atomic(bitmap->storage.sb_page);
NeilBrown32a76272005-06-21 17:17:14 -0700458 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
NeilBrowna2cff262005-06-21 17:17:20 -0700459 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
460 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
461 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
NeilBrown32a76272005-06-21 17:17:14 -0700462 *(__u32 *)(sb->uuid+0),
463 *(__u32 *)(sb->uuid+4),
464 *(__u32 *)(sb->uuid+8),
465 *(__u32 *)(sb->uuid+12));
NeilBrowna2cff262005-06-21 17:17:20 -0700466 printk(KERN_DEBUG " events: %llu\n",
NeilBrown32a76272005-06-21 17:17:14 -0700467 (unsigned long long) le64_to_cpu(sb->events));
NeilBrowna2cff262005-06-21 17:17:20 -0700468 printk(KERN_DEBUG "events cleared: %llu\n",
NeilBrown32a76272005-06-21 17:17:14 -0700469 (unsigned long long) le64_to_cpu(sb->events_cleared));
NeilBrowna2cff262005-06-21 17:17:20 -0700470 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
471 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
472 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
473 printk(KERN_DEBUG " sync size: %llu KB\n",
474 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
NeilBrown4b6d2872005-09-09 16:23:47 -0700475 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
Cong Wangb2f46e62011-11-28 13:25:44 +0800476 kunmap_atomic(sb);
NeilBrown32a76272005-06-21 17:17:14 -0700477}
478
Jonathan Brassow9c810752011-06-08 17:59:30 -0500479/*
480 * bitmap_new_disk_sb
481 * @bitmap
482 *
483 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
484 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
485 * This function verifies 'bitmap_info' and populates the on-disk bitmap
486 * structure, which is to be written to disk.
487 *
488 * Returns: 0 on success, -Exxx on error
489 */
490static int bitmap_new_disk_sb(struct bitmap *bitmap)
491{
492 bitmap_super_t *sb;
493 unsigned long chunksize, daemon_sleep, write_behind;
Jonathan Brassow9c810752011-06-08 17:59:30 -0500494
Goldwyn Rodriguesd3b178a2015-07-22 12:09:17 -0500495 bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
Jianpeng Ma582e2e02012-10-11 13:45:36 +1100496 if (bitmap->storage.sb_page == NULL)
497 return -ENOMEM;
NeilBrown1ec885c2012-05-22 13:55:10 +1000498 bitmap->storage.sb_page->index = 0;
Jonathan Brassow9c810752011-06-08 17:59:30 -0500499
NeilBrown1ec885c2012-05-22 13:55:10 +1000500 sb = kmap_atomic(bitmap->storage.sb_page);
Jonathan Brassow9c810752011-06-08 17:59:30 -0500501
502 sb->magic = cpu_to_le32(BITMAP_MAGIC);
503 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
504
505 chunksize = bitmap->mddev->bitmap_info.chunksize;
506 BUG_ON(!chunksize);
507 if (!is_power_of_2(chunksize)) {
Cong Wangb2f46e62011-11-28 13:25:44 +0800508 kunmap_atomic(sb);
Jonathan Brassow9c810752011-06-08 17:59:30 -0500509 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
510 return -EINVAL;
511 }
512 sb->chunksize = cpu_to_le32(chunksize);
513
514 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
Eric Engestromc97e0602016-03-07 12:01:05 +0000515 if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
Jonathan Brassow9c810752011-06-08 17:59:30 -0500516 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
517 daemon_sleep = 5 * HZ;
518 }
519 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
520 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
521
522 /*
523 * FIXME: write_behind for RAID1. If not specified, what
524 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
525 */
526 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
527 if (write_behind > COUNTER_MAX)
528 write_behind = COUNTER_MAX / 2;
529 sb->write_behind = cpu_to_le32(write_behind);
530 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
531
532 /* keep the array size field of the bitmap superblock up to date */
533 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
534
535 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
536
NeilBrownb405fe92012-05-22 13:55:15 +1000537 set_bit(BITMAP_STALE, &bitmap->flags);
NeilBrown84e92342012-05-22 13:55:14 +1000538 sb->state = cpu_to_le32(bitmap->flags);
Jonathan Brassow9c810752011-06-08 17:59:30 -0500539 bitmap->events_cleared = bitmap->mddev->events;
540 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
Goldwyn Rodriguesd3b178a2015-07-22 12:09:17 -0500541 bitmap->mddev->bitmap_info.nodes = 0;
Jonathan Brassow9c810752011-06-08 17:59:30 -0500542
Cong Wangb2f46e62011-11-28 13:25:44 +0800543 kunmap_atomic(sb);
Jonathan Brassow9c810752011-06-08 17:59:30 -0500544
545 return 0;
546}
547
NeilBrown32a76272005-06-21 17:17:14 -0700548/* read the superblock from the bitmap file and initialize some bitmap fields */
549static int bitmap_read_sb(struct bitmap *bitmap)
550{
551 char *reason = NULL;
552 bitmap_super_t *sb;
NeilBrown4b6d2872005-09-09 16:23:47 -0700553 unsigned long chunksize, daemon_sleep, write_behind;
NeilBrown32a76272005-06-21 17:17:14 -0700554 unsigned long long events;
Goldwyn Rodriguesc4ce8672014-03-29 10:20:02 -0500555 int nodes = 0;
NeilBrown1dff2b82012-05-22 13:55:34 +1000556 unsigned long sectors_reserved = 0;
NeilBrown32a76272005-06-21 17:17:14 -0700557 int err = -EINVAL;
NeilBrown27581e52012-05-22 13:55:08 +1000558 struct page *sb_page;
Goldwyn Rodrigues33e38ac2015-07-01 12:19:56 +1000559 loff_t offset = bitmap->mddev->bitmap_info.offset;
NeilBrown32a76272005-06-21 17:17:14 -0700560
NeilBrown1ec885c2012-05-22 13:55:10 +1000561 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
NeilBrownef99bf42012-05-22 13:55:08 +1000562 chunksize = 128 * 1024 * 1024;
563 daemon_sleep = 5 * HZ;
564 write_behind = 0;
NeilBrownb405fe92012-05-22 13:55:15 +1000565 set_bit(BITMAP_STALE, &bitmap->flags);
NeilBrownef99bf42012-05-22 13:55:08 +1000566 err = 0;
567 goto out_no_sb;
568 }
NeilBrown32a76272005-06-21 17:17:14 -0700569 /* page 0 is the superblock, read it... */
NeilBrown27581e52012-05-22 13:55:08 +1000570 sb_page = alloc_page(GFP_KERNEL);
571 if (!sb_page)
572 return -ENOMEM;
NeilBrown1ec885c2012-05-22 13:55:10 +1000573 bitmap->storage.sb_page = sb_page;
NeilBrown27581e52012-05-22 13:55:08 +1000574
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500575re_read:
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -0500576 /* If cluster_slot is set, the cluster is setup */
577 if (bitmap->cluster_slot >= 0) {
Stephen Rothwell3b0e6aa2015-03-03 13:35:31 +1100578 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -0500579
Stephen Rothwell3b0e6aa2015-03-03 13:35:31 +1100580 sector_div(bm_blocks,
581 bitmap->mddev->bitmap_info.chunksize >> 9);
Goldwyn Rodrigues124eb762015-03-24 11:29:05 -0500582 /* bits to bytes */
583 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
584 /* to 4k blocks */
NeilBrown935f3d42015-03-02 17:02:29 +1100585 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
Goldwyn Rodrigues33e38ac2015-07-01 12:19:56 +1000586 offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3));
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -0500587 pr_info("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
Goldwyn Rodrigues33e38ac2015-07-01 12:19:56 +1000588 bitmap->cluster_slot, offset);
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -0500589 }
590
NeilBrown1ec885c2012-05-22 13:55:10 +1000591 if (bitmap->storage.file) {
592 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
NeilBrownf49d5e62007-01-26 00:57:03 -0800593 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
594
NeilBrown1ec885c2012-05-22 13:55:10 +1000595 err = read_page(bitmap->storage.file, 0,
NeilBrown27581e52012-05-22 13:55:08 +1000596 bitmap, bytes, sb_page);
NeilBrownf49d5e62007-01-26 00:57:03 -0800597 } else {
NeilBrown27581e52012-05-22 13:55:08 +1000598 err = read_sb_page(bitmap->mddev,
Goldwyn Rodrigues33e38ac2015-07-01 12:19:56 +1000599 offset,
NeilBrown27581e52012-05-22 13:55:08 +1000600 sb_page,
601 0, sizeof(bitmap_super_t));
NeilBrowna654b9d82005-06-21 17:17:27 -0700602 }
NeilBrown27581e52012-05-22 13:55:08 +1000603 if (err)
NeilBrown32a76272005-06-21 17:17:14 -0700604 return err;
NeilBrown32a76272005-06-21 17:17:14 -0700605
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500606 err = -EINVAL;
NeilBrown27581e52012-05-22 13:55:08 +1000607 sb = kmap_atomic(sb_page);
NeilBrown32a76272005-06-21 17:17:14 -0700608
NeilBrown32a76272005-06-21 17:17:14 -0700609 chunksize = le32_to_cpu(sb->chunksize);
NeilBrown1b04be92009-12-14 12:49:53 +1100610 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
NeilBrown4b6d2872005-09-09 16:23:47 -0700611 write_behind = le32_to_cpu(sb->write_behind);
NeilBrown1dff2b82012-05-22 13:55:34 +1000612 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
Goldwyn Rodrigues3c462c82015-08-19 07:35:54 +1000613 /* Setup nodes/clustername only if bitmap version is
614 * cluster-compatible
Goldwyn Rodriguesd3b178a2015-07-22 12:09:17 -0500615 */
Goldwyn Rodrigues3c462c82015-08-19 07:35:54 +1000616 if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
Goldwyn Rodriguesd3b178a2015-07-22 12:09:17 -0500617 nodes = le32_to_cpu(sb->nodes);
618 strlcpy(bitmap->mddev->bitmap_info.cluster_name,
619 sb->cluster_name, 64);
620 }
NeilBrown32a76272005-06-21 17:17:14 -0700621
622 /* verify that the bitmap-specific fields are valid */
623 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
624 reason = "bad magic";
NeilBrownbd926c62005-11-08 21:39:32 -0800625 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
Goldwyn Rodrigues3c462c82015-08-19 07:35:54 +1000626 le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
NeilBrown32a76272005-06-21 17:17:14 -0700627 reason = "unrecognized superblock version";
NeilBrown1187cf02009-03-31 14:27:02 +1100628 else if (chunksize < 512)
NeilBrown7dd5d342006-01-06 00:20:39 -0800629 reason = "bitmap chunksize too small";
Jonathan Brassowd7445402011-06-08 18:01:10 -0500630 else if (!is_power_of_2(chunksize))
NeilBrown32a76272005-06-21 17:17:14 -0700631 reason = "bitmap chunksize not a power of 2";
NeilBrown1b04be92009-12-14 12:49:53 +1100632 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
NeilBrown7dd5d342006-01-06 00:20:39 -0800633 reason = "daemon sleep period out of range";
NeilBrown4b6d2872005-09-09 16:23:47 -0700634 else if (write_behind > COUNTER_MAX)
635 reason = "write-behind limit out of range (0 - 16383)";
NeilBrown32a76272005-06-21 17:17:14 -0700636 if (reason) {
637 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
638 bmname(bitmap), reason);
639 goto out;
640 }
641
642 /* keep the array size field of the bitmap superblock up to date */
643 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
644
NeilBrown278c1ca2012-03-19 12:46:40 +1100645 if (bitmap->mddev->persistent) {
646 /*
647 * We have a persistent array superblock, so compare the
648 * bitmap's UUID and event counter to the mddev's
649 */
650 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
651 printk(KERN_INFO
652 "%s: bitmap superblock UUID mismatch\n",
653 bmname(bitmap));
654 goto out;
655 }
656 events = le64_to_cpu(sb->events);
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500657 if (!nodes && (events < bitmap->mddev->events)) {
NeilBrown278c1ca2012-03-19 12:46:40 +1100658 printk(KERN_INFO
659 "%s: bitmap file is out of date (%llu < %llu) "
660 "-- forcing full recovery\n",
661 bmname(bitmap), events,
662 (unsigned long long) bitmap->mddev->events);
NeilBrownb405fe92012-05-22 13:55:15 +1000663 set_bit(BITMAP_STALE, &bitmap->flags);
NeilBrown278c1ca2012-03-19 12:46:40 +1100664 }
665 }
NeilBrown32a76272005-06-21 17:17:14 -0700666
NeilBrown32a76272005-06-21 17:17:14 -0700667 /* assign fields using values from superblock */
NeilBrown4f2e6392006-10-21 10:24:09 -0700668 bitmap->flags |= le32_to_cpu(sb->state);
NeilBrownbd926c62005-11-08 21:39:32 -0800669 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
NeilBrownb405fe92012-05-22 13:55:15 +1000670 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
NeilBrown32a76272005-06-21 17:17:14 -0700671 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
Goldwyn Rodriguescf921cc2014-03-30 00:42:49 -0500672 strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
NeilBrown32a76272005-06-21 17:17:14 -0700673 err = 0;
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500674
NeilBrown32a76272005-06-21 17:17:14 -0700675out:
Cong Wangb2f46e62011-11-28 13:25:44 +0800676 kunmap_atomic(sb);
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -0500677 /* Assiging chunksize is required for "re_read" */
678 bitmap->mddev->bitmap_info.chunksize = chunksize;
Goldwyn Rodriguesf7357272015-07-22 12:09:16 -0500679 if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500680 err = md_setup_cluster(bitmap->mddev, nodes);
681 if (err) {
682 pr_err("%s: Could not setup cluster service (%d)\n",
683 bmname(bitmap), err);
684 goto out_no_sb;
685 }
686 bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500687 goto re_read;
688 }
689
690
NeilBrownef99bf42012-05-22 13:55:08 +1000691out_no_sb:
NeilBrownb405fe92012-05-22 13:55:15 +1000692 if (test_bit(BITMAP_STALE, &bitmap->flags))
NeilBrownef99bf42012-05-22 13:55:08 +1000693 bitmap->events_cleared = bitmap->mddev->events;
694 bitmap->mddev->bitmap_info.chunksize = chunksize;
695 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
696 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
Goldwyn Rodriguesc4ce8672014-03-29 10:20:02 -0500697 bitmap->mddev->bitmap_info.nodes = nodes;
NeilBrown1dff2b82012-05-22 13:55:34 +1000698 if (bitmap->mddev->bitmap_info.space == 0 ||
699 bitmap->mddev->bitmap_info.space > sectors_reserved)
700 bitmap->mddev->bitmap_info.space = sectors_reserved;
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500701 if (err) {
NeilBrown32a76272005-06-21 17:17:14 -0700702 bitmap_print_sb(bitmap);
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -0500703 if (bitmap->cluster_slot < 0)
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500704 md_cluster_stop(bitmap->mddev);
705 }
NeilBrown32a76272005-06-21 17:17:14 -0700706 return err;
707}
708
NeilBrown32a76272005-06-21 17:17:14 -0700709/*
710 * general bitmap file operations
711 */
712
NeilBrownece5cff2009-12-14 12:49:56 +1100713/*
714 * on-disk bitmap:
715 *
716 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
717 * file a page at a time. There's a superblock at the start of the file.
718 */
NeilBrown32a76272005-06-21 17:17:14 -0700719/* calculate the index of the page that contains this bit */
NeilBrown1ec885c2012-05-22 13:55:10 +1000720static inline unsigned long file_page_index(struct bitmap_storage *store,
721 unsigned long chunk)
NeilBrown32a76272005-06-21 17:17:14 -0700722{
NeilBrown1ec885c2012-05-22 13:55:10 +1000723 if (store->sb_page)
NeilBrownece5cff2009-12-14 12:49:56 +1100724 chunk += sizeof(bitmap_super_t) << 3;
725 return chunk >> PAGE_BIT_SHIFT;
NeilBrown32a76272005-06-21 17:17:14 -0700726}
727
728/* calculate the (bit) offset of this bit within a page */
NeilBrown1ec885c2012-05-22 13:55:10 +1000729static inline unsigned long file_page_offset(struct bitmap_storage *store,
730 unsigned long chunk)
NeilBrown32a76272005-06-21 17:17:14 -0700731{
NeilBrown1ec885c2012-05-22 13:55:10 +1000732 if (store->sb_page)
NeilBrownece5cff2009-12-14 12:49:56 +1100733 chunk += sizeof(bitmap_super_t) << 3;
734 return chunk & (PAGE_BITS - 1);
NeilBrown32a76272005-06-21 17:17:14 -0700735}
736
737/*
738 * return a pointer to the page in the filemap that contains the given bit
739 *
NeilBrown32a76272005-06-21 17:17:14 -0700740 */
NeilBrown1ec885c2012-05-22 13:55:10 +1000741static inline struct page *filemap_get_page(struct bitmap_storage *store,
Jonathan Brassow3520fa42011-07-27 11:00:37 +1000742 unsigned long chunk)
NeilBrown32a76272005-06-21 17:17:14 -0700743{
NeilBrown1ec885c2012-05-22 13:55:10 +1000744 if (file_page_index(store, chunk) >= store->file_pages)
NeilBrownac2f40b2010-06-01 19:37:31 +1000745 return NULL;
NeilBrownf2e06c52014-05-28 13:39:23 +1000746 return store->filemap[file_page_index(store, chunk)];
NeilBrown32a76272005-06-21 17:17:14 -0700747}
748
NeilBrownd1244cb2012-05-22 13:55:12 +1000749static int bitmap_storage_alloc(struct bitmap_storage *store,
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500750 unsigned long chunks, int with_super,
751 int slot_number)
NeilBrownd1244cb2012-05-22 13:55:12 +1000752{
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500753 int pnum, offset = 0;
NeilBrownd1244cb2012-05-22 13:55:12 +1000754 unsigned long num_pages;
755 unsigned long bytes;
756
757 bytes = DIV_ROUND_UP(chunks, 8);
758 if (with_super)
759 bytes += sizeof(bitmap_super_t);
760
761 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
Guoqing Jiang7f86ffe2016-05-02 11:50:13 -0400762 offset = slot_number * num_pages;
NeilBrownd1244cb2012-05-22 13:55:12 +1000763
764 store->filemap = kmalloc(sizeof(struct page *)
765 * num_pages, GFP_KERNEL);
766 if (!store->filemap)
767 return -ENOMEM;
768
769 if (with_super && !store->sb_page) {
NeilBrownd60b4792012-05-22 13:55:25 +1000770 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
NeilBrownd1244cb2012-05-22 13:55:12 +1000771 if (store->sb_page == NULL)
772 return -ENOMEM;
NeilBrownd1244cb2012-05-22 13:55:12 +1000773 }
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500774
NeilBrownd1244cb2012-05-22 13:55:12 +1000775 pnum = 0;
776 if (store->sb_page) {
777 store->filemap[0] = store->sb_page;
778 pnum = 1;
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500779 store->sb_page->index = offset;
NeilBrownd1244cb2012-05-22 13:55:12 +1000780 }
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500781
NeilBrownd1244cb2012-05-22 13:55:12 +1000782 for ( ; pnum < num_pages; pnum++) {
NeilBrownd60b4792012-05-22 13:55:25 +1000783 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
NeilBrownd1244cb2012-05-22 13:55:12 +1000784 if (!store->filemap[pnum]) {
785 store->file_pages = pnum;
786 return -ENOMEM;
787 }
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500788 store->filemap[pnum]->index = pnum + offset;
NeilBrownd1244cb2012-05-22 13:55:12 +1000789 }
790 store->file_pages = pnum;
791
792 /* We need 4 bits per page, rounded up to a multiple
793 * of sizeof(unsigned long) */
794 store->filemap_attr = kzalloc(
795 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
796 GFP_KERNEL);
797 if (!store->filemap_attr)
798 return -ENOMEM;
799
800 store->bytes = bytes;
801
802 return 0;
803}
804
NeilBrownfae7d322012-05-22 13:55:21 +1000805static void bitmap_file_unmap(struct bitmap_storage *store)
NeilBrown32a76272005-06-21 17:17:14 -0700806{
807 struct page **map, *sb_page;
NeilBrown32a76272005-06-21 17:17:14 -0700808 int pages;
NeilBrownfae7d322012-05-22 13:55:21 +1000809 struct file *file;
NeilBrown32a76272005-06-21 17:17:14 -0700810
NeilBrownfae7d322012-05-22 13:55:21 +1000811 file = store->file;
NeilBrown1ec885c2012-05-22 13:55:10 +1000812 map = store->filemap;
NeilBrown1ec885c2012-05-22 13:55:10 +1000813 pages = store->file_pages;
NeilBrown1ec885c2012-05-22 13:55:10 +1000814 sb_page = store->sb_page;
NeilBrown32a76272005-06-21 17:17:14 -0700815
816 while (pages--)
NeilBrownece5cff2009-12-14 12:49:56 +1100817 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
NeilBrownd785a062006-06-26 00:27:48 -0700818 free_buffers(map[pages]);
NeilBrown32a76272005-06-21 17:17:14 -0700819 kfree(map);
NeilBrownfae7d322012-05-22 13:55:21 +1000820 kfree(store->filemap_attr);
NeilBrown32a76272005-06-21 17:17:14 -0700821
NeilBrownd785a062006-06-26 00:27:48 -0700822 if (sb_page)
823 free_buffers(sb_page);
NeilBrown32a76272005-06-21 17:17:14 -0700824
NeilBrownd785a062006-06-26 00:27:48 -0700825 if (file) {
Al Viro496ad9a2013-01-23 17:07:38 -0500826 struct inode *inode = file_inode(file);
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800827 invalidate_mapping_pages(inode->i_mapping, 0, -1);
NeilBrown32a76272005-06-21 17:17:14 -0700828 fput(file);
NeilBrownd785a062006-06-26 00:27:48 -0700829 }
NeilBrown32a76272005-06-21 17:17:14 -0700830}
831
NeilBrown32a76272005-06-21 17:17:14 -0700832/*
833 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
834 * then it is no longer reliable, so we stop using it and we mark the file
835 * as failed in the superblock
836 */
837static void bitmap_file_kick(struct bitmap *bitmap)
838{
839 char *path, *ptr = NULL;
840
NeilBrownb405fe92012-05-22 13:55:15 +1000841 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
NeilBrown4ad13662007-07-17 04:06:13 -0700842 bitmap_update_sb(bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700843
NeilBrown1ec885c2012-05-22 13:55:10 +1000844 if (bitmap->storage.file) {
NeilBrown4ad13662007-07-17 04:06:13 -0700845 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
846 if (path)
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +0200847 ptr = file_path(bitmap->storage.file,
NeilBrown1ec885c2012-05-22 13:55:10 +1000848 path, PAGE_SIZE);
Christoph Hellwig6bcfd602008-05-23 13:04:34 -0700849
NeilBrown4ad13662007-07-17 04:06:13 -0700850 printk(KERN_ALERT
851 "%s: kicking failed bitmap file %s from array!\n",
Christoph Hellwig6bcfd602008-05-23 13:04:34 -0700852 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
NeilBrown32a76272005-06-21 17:17:14 -0700853
NeilBrown4ad13662007-07-17 04:06:13 -0700854 kfree(path);
855 } else
856 printk(KERN_ALERT
857 "%s: disabling internal bitmap due to errors\n",
858 bmname(bitmap));
NeilBrowna654b9d82005-06-21 17:17:27 -0700859 }
NeilBrown32a76272005-06-21 17:17:14 -0700860}
861
862enum bitmap_page_attr {
NeilBrownac2f40b2010-06-01 19:37:31 +1000863 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
NeilBrown5a537df2011-09-21 15:37:46 +1000864 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
865 * i.e. counter is 1 or 2. */
NeilBrownac2f40b2010-06-01 19:37:31 +1000866 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
NeilBrown32a76272005-06-21 17:17:14 -0700867};
868
NeilBrownd1891222012-05-22 13:55:09 +1000869static inline void set_page_attr(struct bitmap *bitmap, int pnum,
870 enum bitmap_page_attr attr)
NeilBrown32a76272005-06-21 17:17:14 -0700871{
NeilBrownbdfd1142012-05-22 13:55:22 +1000872 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
NeilBrown32a76272005-06-21 17:17:14 -0700873}
874
NeilBrownd1891222012-05-22 13:55:09 +1000875static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
876 enum bitmap_page_attr attr)
NeilBrown32a76272005-06-21 17:17:14 -0700877{
NeilBrownbdfd1142012-05-22 13:55:22 +1000878 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
NeilBrown32a76272005-06-21 17:17:14 -0700879}
880
NeilBrownbdfd1142012-05-22 13:55:22 +1000881static inline int test_page_attr(struct bitmap *bitmap, int pnum,
882 enum bitmap_page_attr attr)
NeilBrown32a76272005-06-21 17:17:14 -0700883{
NeilBrown1ec885c2012-05-22 13:55:10 +1000884 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
NeilBrown32a76272005-06-21 17:17:14 -0700885}
886
NeilBrownbdfd1142012-05-22 13:55:22 +1000887static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
888 enum bitmap_page_attr attr)
889{
890 return test_and_clear_bit((pnum<<2) + attr,
891 bitmap->storage.filemap_attr);
892}
NeilBrown32a76272005-06-21 17:17:14 -0700893/*
894 * bitmap_file_set_bit -- called before performing a write to the md device
895 * to set (and eventually sync) a particular bit in the bitmap file
896 *
897 * we set the bit immediately, then we record the page number so that
898 * when an unplug occurs, we can flush the dirty pages out to disk
899 */
900static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
901{
902 unsigned long bit;
Jonathan Brassow3520fa42011-07-27 11:00:37 +1000903 struct page *page;
NeilBrown32a76272005-06-21 17:17:14 -0700904 void *kaddr;
NeilBrown40cffcc2012-05-22 13:55:24 +1000905 unsigned long chunk = block >> bitmap->counts.chunkshift;
Guoqing Jiang23cea66a2016-05-02 11:50:14 -0400906 struct bitmap_storage *store = &bitmap->storage;
907 unsigned long node_offset = 0;
908
909 if (mddev_is_clustered(bitmap->mddev))
910 node_offset = bitmap->cluster_slot * store->file_pages;
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 */
Guoqing Jiang23cea66a2016-05-02 11:50:14 -0400926 set_page_attr(bitmap, page->index - node_offset, 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;
Guoqing Jiang23cea66a2016-05-02 11:50:14 -0400935 struct bitmap_storage *store = &bitmap->storage;
936 unsigned long node_offset = 0;
937
938 if (mddev_is_clustered(bitmap->mddev))
939 node_offset = bitmap->cluster_slot * store->file_pages;
NeilBrownef99bf42012-05-22 13:55:08 +1000940
NeilBrown1ec885c2012-05-22 13:55:10 +1000941 page = filemap_get_page(&bitmap->storage, chunk);
NeilBrownef99bf42012-05-22 13:55:08 +1000942 if (!page)
943 return;
NeilBrown1ec885c2012-05-22 13:55:10 +1000944 bit = file_page_offset(&bitmap->storage, chunk);
NeilBrownef99bf42012-05-22 13:55:08 +1000945 paddr = kmap_atomic(page);
NeilBrownb405fe92012-05-22 13:55:15 +1000946 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
NeilBrownef99bf42012-05-22 13:55:08 +1000947 clear_bit(bit, paddr);
948 else
Akinobu Mita3f810b62013-04-24 11:42:41 +1000949 clear_bit_le(bit, paddr);
NeilBrownef99bf42012-05-22 13:55:08 +1000950 kunmap_atomic(paddr);
Guoqing Jiang23cea66a2016-05-02 11:50:14 -0400951 if (!test_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_NEEDWRITE)) {
952 set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_PENDING);
NeilBrownef99bf42012-05-22 13:55:08 +1000953 bitmap->allclean = 0;
954 }
955}
956
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -0500957static int bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
958{
959 unsigned long bit;
960 struct page *page;
961 void *paddr;
962 unsigned long chunk = block >> bitmap->counts.chunkshift;
963 int set = 0;
964
965 page = filemap_get_page(&bitmap->storage, chunk);
966 if (!page)
967 return -EINVAL;
968 bit = file_page_offset(&bitmap->storage, chunk);
969 paddr = kmap_atomic(page);
970 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
971 set = test_bit(bit, paddr);
972 else
973 set = test_bit_le(bit, paddr);
974 kunmap_atomic(paddr);
975 return set;
976}
977
978
NeilBrown32a76272005-06-21 17:17:14 -0700979/* this gets called when the md device is ready to unplug its underlying
980 * (slave) device queues -- before we let any writes go down, we need to
981 * sync the dirty pages of the bitmap file to disk */
NeilBrown4ad13662007-07-17 04:06:13 -0700982void bitmap_unplug(struct bitmap *bitmap)
NeilBrown32a76272005-06-21 17:17:14 -0700983{
NeilBrown74667122012-05-22 13:55:19 +1000984 unsigned long i;
NeilBrownec7a3192006-06-26 00:27:45 -0700985 int dirty, need_write;
NeilBrown32a76272005-06-21 17:17:14 -0700986
NeilBrown62f82fa2012-05-22 13:55:21 +1000987 if (!bitmap || !bitmap->storage.filemap ||
988 test_bit(BITMAP_STALE, &bitmap->flags))
NeilBrown4ad13662007-07-17 04:06:13 -0700989 return;
NeilBrown32a76272005-06-21 17:17:14 -0700990
991 /* look at each page to see if there are any set bits that need to be
992 * flushed out to disk */
NeilBrown1ec885c2012-05-22 13:55:10 +1000993 for (i = 0; i < bitmap->storage.file_pages; i++) {
NeilBrownbdfd1142012-05-22 13:55:22 +1000994 if (!bitmap->storage.filemap)
NeilBrown4ad13662007-07-17 04:06:13 -0700995 return;
NeilBrownbdfd1142012-05-22 13:55:22 +1000996 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
997 need_write = test_and_clear_page_attr(bitmap, i,
998 BITMAP_PAGE_NEEDWRITE);
999 if (dirty || need_write) {
NeilBrownd1891222012-05-22 13:55:09 +10001000 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
NeilBrownbdfd1142012-05-22 13:55:22 +10001001 write_page(bitmap, bitmap->storage.filemap[i], 0);
1002 }
NeilBrown32a76272005-06-21 17:17:14 -07001003 }
NeilBrown4b5060d2014-09-09 14:13:51 +10001004 if (bitmap->storage.file)
1005 wait_event(bitmap->write_wait,
1006 atomic_read(&bitmap->pending_writes)==0);
1007 else
1008 md_super_wait(bitmap->mddev);
1009
NeilBrownb405fe92012-05-22 13:55:15 +10001010 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
NeilBrownd785a062006-06-26 00:27:48 -07001011 bitmap_file_kick(bitmap);
NeilBrown32a76272005-06-21 17:17:14 -07001012}
NeilBrownac2f40b2010-06-01 19:37:31 +10001013EXPORT_SYMBOL(bitmap_unplug);
NeilBrown32a76272005-06-21 17:17:14 -07001014
NeilBrown6a079972005-09-09 16:23:44 -07001015static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
NeilBrown32a76272005-06-21 17:17:14 -07001016/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1017 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1018 * memory mapping of the bitmap file
1019 * Special cases:
1020 * if there's no bitmap file, or if the bitmap file had been
1021 * previously kicked from the array, we mark all the bits as
1022 * 1's in order to cause a full resync.
NeilBrown6a079972005-09-09 16:23:44 -07001023 *
1024 * We ignore all bits for sectors that end earlier than 'start'.
1025 * This is used when reading an out-of-date bitmap...
NeilBrown32a76272005-06-21 17:17:14 -07001026 */
NeilBrown6a079972005-09-09 16:23:44 -07001027static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
NeilBrown32a76272005-06-21 17:17:14 -07001028{
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -05001029 unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
NeilBrown27581e52012-05-22 13:55:08 +10001030 struct page *page = NULL;
NeilBrownd1244cb2012-05-22 13:55:12 +10001031 unsigned long bit_cnt = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001032 struct file *file;
NeilBrownd1244cb2012-05-22 13:55:12 +10001033 unsigned long offset;
NeilBrown32a76272005-06-21 17:17:14 -07001034 int outofdate;
1035 int ret = -ENOSPC;
NeilBrownea03aff2006-01-06 00:20:34 -08001036 void *paddr;
NeilBrown1ec885c2012-05-22 13:55:10 +10001037 struct bitmap_storage *store = &bitmap->storage;
NeilBrown32a76272005-06-21 17:17:14 -07001038
NeilBrown40cffcc2012-05-22 13:55:24 +10001039 chunks = bitmap->counts.chunks;
NeilBrown1ec885c2012-05-22 13:55:10 +10001040 file = store->file;
NeilBrown32a76272005-06-21 17:17:14 -07001041
NeilBrownef99bf42012-05-22 13:55:08 +10001042 if (!file && !bitmap->mddev->bitmap_info.offset) {
1043 /* No permanent bitmap - fill with '1s'. */
NeilBrown1ec885c2012-05-22 13:55:10 +10001044 store->filemap = NULL;
1045 store->file_pages = 0;
NeilBrownef99bf42012-05-22 13:55:08 +10001046 for (i = 0; i < chunks ; i++) {
1047 /* if the disk bit is set, set the memory bit */
NeilBrown40cffcc2012-05-22 13:55:24 +10001048 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
NeilBrownef99bf42012-05-22 13:55:08 +10001049 >= start);
1050 bitmap_set_memory_bits(bitmap,
NeilBrown40cffcc2012-05-22 13:55:24 +10001051 (sector_t)i << bitmap->counts.chunkshift,
NeilBrownef99bf42012-05-22 13:55:08 +10001052 needed);
1053 }
1054 return 0;
1055 }
NeilBrown32a76272005-06-21 17:17:14 -07001056
NeilBrownb405fe92012-05-22 13:55:15 +10001057 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
NeilBrown32a76272005-06-21 17:17:14 -07001058 if (outofdate)
1059 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
1060 "recovery\n", bmname(bitmap));
1061
NeilBrownd1244cb2012-05-22 13:55:12 +10001062 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
NeilBrown32a76272005-06-21 17:17:14 -07001063 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
NeilBrownd1244cb2012-05-22 13:55:12 +10001064 bmname(bitmap),
1065 (unsigned long) i_size_read(file->f_mapping->host),
1066 store->bytes);
NeilBrown4ad13662007-07-17 04:06:13 -07001067 goto err;
NeilBrown32a76272005-06-21 17:17:14 -07001068 }
NeilBrownbc7f77d2005-06-21 17:17:17 -07001069
NeilBrown32a76272005-06-21 17:17:14 -07001070 oldindex = ~0L;
NeilBrownd1244cb2012-05-22 13:55:12 +10001071 offset = 0;
1072 if (!bitmap->mddev->bitmap_info.external)
1073 offset = sizeof(bitmap_super_t);
NeilBrown32a76272005-06-21 17:17:14 -07001074
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -05001075 if (mddev_is_clustered(bitmap->mddev))
1076 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1077
NeilBrown32a76272005-06-21 17:17:14 -07001078 for (i = 0; i < chunks; i++) {
NeilBrownbd926c62005-11-08 21:39:32 -08001079 int b;
NeilBrown1ec885c2012-05-22 13:55:10 +10001080 index = file_page_index(&bitmap->storage, i);
1081 bit = file_page_offset(&bitmap->storage, i);
NeilBrown32a76272005-06-21 17:17:14 -07001082 if (index != oldindex) { /* this is a new page, read it in */
NeilBrownd785a062006-06-26 00:27:48 -07001083 int count;
NeilBrown32a76272005-06-21 17:17:14 -07001084 /* unmap the old page, we're done with it */
NeilBrownd1244cb2012-05-22 13:55:12 +10001085 if (index == store->file_pages-1)
1086 count = store->bytes - index * PAGE_SIZE;
NeilBrownd785a062006-06-26 00:27:48 -07001087 else
1088 count = PAGE_SIZE;
NeilBrown1ec885c2012-05-22 13:55:10 +10001089 page = store->filemap[index];
NeilBrown27581e52012-05-22 13:55:08 +10001090 if (file)
1091 ret = read_page(file, index, bitmap,
1092 count, page);
1093 else
1094 ret = read_sb_page(
1095 bitmap->mddev,
1096 bitmap->mddev->bitmap_info.offset,
1097 page,
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -05001098 index + node_offset, count);
NeilBrown27581e52012-05-22 13:55:08 +10001099
1100 if (ret)
NeilBrown4ad13662007-07-17 04:06:13 -07001101 goto err;
NeilBrowna654b9d82005-06-21 17:17:27 -07001102
NeilBrown32a76272005-06-21 17:17:14 -07001103 oldindex = index;
NeilBrown32a76272005-06-21 17:17:14 -07001104
1105 if (outofdate) {
1106 /*
1107 * if bitmap is out of date, dirty the
NeilBrownac2f40b2010-06-01 19:37:31 +10001108 * whole page and write it out
NeilBrown32a76272005-06-21 17:17:14 -07001109 */
Cong Wangb2f46e62011-11-28 13:25:44 +08001110 paddr = kmap_atomic(page);
NeilBrownea03aff2006-01-06 00:20:34 -08001111 memset(paddr + offset, 0xff,
NeilBrown6a079972005-09-09 16:23:44 -07001112 PAGE_SIZE - offset);
Cong Wangb2f46e62011-11-28 13:25:44 +08001113 kunmap_atomic(paddr);
NeilBrown4ad13662007-07-17 04:06:13 -07001114 write_page(bitmap, page, 1);
1115
1116 ret = -EIO;
NeilBrownb405fe92012-05-22 13:55:15 +10001117 if (test_bit(BITMAP_WRITE_ERROR,
1118 &bitmap->flags))
NeilBrown4ad13662007-07-17 04:06:13 -07001119 goto err;
NeilBrown32a76272005-06-21 17:17:14 -07001120 }
NeilBrown32a76272005-06-21 17:17:14 -07001121 }
Cong Wangb2f46e62011-11-28 13:25:44 +08001122 paddr = kmap_atomic(page);
NeilBrownb405fe92012-05-22 13:55:15 +10001123 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
NeilBrownea03aff2006-01-06 00:20:34 -08001124 b = test_bit(bit, paddr);
NeilBrownbd926c62005-11-08 21:39:32 -08001125 else
Akinobu Mita6b33aff2011-03-23 16:42:13 -07001126 b = test_bit_le(bit, paddr);
Cong Wangb2f46e62011-11-28 13:25:44 +08001127 kunmap_atomic(paddr);
NeilBrownbd926c62005-11-08 21:39:32 -08001128 if (b) {
NeilBrown32a76272005-06-21 17:17:14 -07001129 /* if the disk bit is set, set the memory bit */
NeilBrown40cffcc2012-05-22 13:55:24 +10001130 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
NeilBrowndb305e52009-05-07 12:49:06 +10001131 >= start);
1132 bitmap_set_memory_bits(bitmap,
NeilBrown40cffcc2012-05-22 13:55:24 +10001133 (sector_t)i << bitmap->counts.chunkshift,
NeilBrowndb305e52009-05-07 12:49:06 +10001134 needed);
NeilBrown32a76272005-06-21 17:17:14 -07001135 bit_cnt++;
1136 }
NeilBrown27581e52012-05-22 13:55:08 +10001137 offset = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001138 }
1139
NeilBrown32a76272005-06-21 17:17:14 -07001140 printk(KERN_INFO "%s: bitmap initialized from disk: "
NeilBrownd1244cb2012-05-22 13:55:12 +10001141 "read %lu pages, set %lu of %lu bits\n",
NeilBrown1ec885c2012-05-22 13:55:10 +10001142 bmname(bitmap), store->file_pages,
NeilBrownd1244cb2012-05-22 13:55:12 +10001143 bit_cnt, chunks);
NeilBrown32a76272005-06-21 17:17:14 -07001144
NeilBrown4ad13662007-07-17 04:06:13 -07001145 return 0;
1146
1147 err:
1148 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1149 bmname(bitmap), ret);
NeilBrown32a76272005-06-21 17:17:14 -07001150 return ret;
1151}
1152
NeilBrowna654b9d82005-06-21 17:17:27 -07001153void bitmap_write_all(struct bitmap *bitmap)
1154{
1155 /* We don't actually write all bitmap blocks here,
1156 * just flag them as needing to be written
1157 */
NeilBrownec7a3192006-06-26 00:27:45 -07001158 int i;
NeilBrowna654b9d82005-06-21 17:17:27 -07001159
NeilBrown1ec885c2012-05-22 13:55:10 +10001160 if (!bitmap || !bitmap->storage.filemap)
NeilBrownef99bf42012-05-22 13:55:08 +10001161 return;
NeilBrown1ec885c2012-05-22 13:55:10 +10001162 if (bitmap->storage.file)
NeilBrownef99bf42012-05-22 13:55:08 +10001163 /* Only one copy, so nothing needed */
1164 return;
1165
NeilBrown1ec885c2012-05-22 13:55:10 +10001166 for (i = 0; i < bitmap->storage.file_pages; i++)
NeilBrownd1891222012-05-22 13:55:09 +10001167 set_page_attr(bitmap, i,
NeilBrownec7a3192006-06-26 00:27:45 -07001168 BITMAP_PAGE_NEEDWRITE);
NeilBrown2585f3e2011-09-21 15:37:46 +10001169 bitmap->allclean = 0;
NeilBrowna654b9d82005-06-21 17:17:27 -07001170}
1171
NeilBrown40cffcc2012-05-22 13:55:24 +10001172static void bitmap_count_page(struct bitmap_counts *bitmap,
1173 sector_t offset, int inc)
NeilBrown32a76272005-06-21 17:17:14 -07001174{
NeilBrown61a0d802012-03-19 12:46:41 +11001175 sector_t chunk = offset >> bitmap->chunkshift;
NeilBrown32a76272005-06-21 17:17:14 -07001176 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1177 bitmap->bp[page].count += inc;
NeilBrown32a76272005-06-21 17:17:14 -07001178 bitmap_checkfree(bitmap, page);
1179}
NeilBrownbf07bb72012-05-22 13:55:06 +10001180
NeilBrown40cffcc2012-05-22 13:55:24 +10001181static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
NeilBrownbf07bb72012-05-22 13:55:06 +10001182{
1183 sector_t chunk = offset >> bitmap->chunkshift;
1184 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1185 struct bitmap_page *bp = &bitmap->bp[page];
1186
1187 if (!bp->pending)
1188 bp->pending = 1;
1189}
1190
NeilBrown40cffcc2012-05-22 13:55:24 +10001191static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
NeilBrown57dab0b2010-10-19 10:03:39 +11001192 sector_t offset, sector_t *blocks,
NeilBrown32a76272005-06-21 17:17:14 -07001193 int create);
1194
1195/*
1196 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1197 * out to disk
1198 */
1199
NeilBrownfd01b882011-10-11 16:47:53 +11001200void bitmap_daemon_work(struct mddev *mddev)
NeilBrown32a76272005-06-21 17:17:14 -07001201{
NeilBrownaa5cbd12009-12-14 12:49:46 +11001202 struct bitmap *bitmap;
NeilBrownaa3163f2005-06-21 17:17:22 -07001203 unsigned long j;
NeilBrownbf07bb72012-05-22 13:55:06 +10001204 unsigned long nextpage;
NeilBrown57dab0b2010-10-19 10:03:39 +11001205 sector_t blocks;
NeilBrown40cffcc2012-05-22 13:55:24 +10001206 struct bitmap_counts *counts;
NeilBrown32a76272005-06-21 17:17:14 -07001207
NeilBrownaa5cbd12009-12-14 12:49:46 +11001208 /* Use a mutex to guard daemon_work against
1209 * bitmap_destroy.
1210 */
NeilBrownc3d97142009-12-14 12:49:52 +11001211 mutex_lock(&mddev->bitmap_info.mutex);
NeilBrownaa5cbd12009-12-14 12:49:46 +11001212 bitmap = mddev->bitmap;
1213 if (bitmap == NULL) {
NeilBrownc3d97142009-12-14 12:49:52 +11001214 mutex_unlock(&mddev->bitmap_info.mutex);
NeilBrown4ad13662007-07-17 04:06:13 -07001215 return;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001216 }
NeilBrown42a04b52009-12-14 12:49:53 +11001217 if (time_before(jiffies, bitmap->daemon_lastrun
NeilBrown2e61ebb2011-12-23 10:17:50 +11001218 + mddev->bitmap_info.daemon_sleep))
NeilBrown7be3dfe2008-03-10 11:43:48 -07001219 goto done;
1220
NeilBrown32a76272005-06-21 17:17:14 -07001221 bitmap->daemon_lastrun = jiffies;
NeilBrown8311c292008-03-04 14:29:30 -08001222 if (bitmap->allclean) {
NeilBrown2e61ebb2011-12-23 10:17:50 +11001223 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001224 goto done;
NeilBrown8311c292008-03-04 14:29:30 -08001225 }
1226 bitmap->allclean = 1;
NeilBrown32a76272005-06-21 17:17:14 -07001227
NeilBrownbf07bb72012-05-22 13:55:06 +10001228 /* Any file-page which is PENDING now needs to be written.
1229 * So set NEEDWRITE now, then after we make any last-minute changes
1230 * we will write it.
1231 */
NeilBrown1ec885c2012-05-22 13:55:10 +10001232 for (j = 0; j < bitmap->storage.file_pages; j++)
NeilBrownbdfd1142012-05-22 13:55:22 +10001233 if (test_and_clear_page_attr(bitmap, j,
1234 BITMAP_PAGE_PENDING))
NeilBrownd1891222012-05-22 13:55:09 +10001235 set_page_attr(bitmap, j,
NeilBrownbf07bb72012-05-22 13:55:06 +10001236 BITMAP_PAGE_NEEDWRITE);
NeilBrownbf07bb72012-05-22 13:55:06 +10001237
1238 if (bitmap->need_sync &&
1239 mddev->bitmap_info.external == 0) {
1240 /* Arrange for superblock update as well as
1241 * other changes */
1242 bitmap_super_t *sb;
1243 bitmap->need_sync = 0;
NeilBrown1ec885c2012-05-22 13:55:10 +10001244 if (bitmap->storage.filemap) {
1245 sb = kmap_atomic(bitmap->storage.sb_page);
NeilBrownef99bf42012-05-22 13:55:08 +10001246 sb->events_cleared =
1247 cpu_to_le64(bitmap->events_cleared);
1248 kunmap_atomic(sb);
NeilBrownd1891222012-05-22 13:55:09 +10001249 set_page_attr(bitmap, 0,
NeilBrownef99bf42012-05-22 13:55:08 +10001250 BITMAP_PAGE_NEEDWRITE);
1251 }
NeilBrownbf07bb72012-05-22 13:55:06 +10001252 }
1253 /* Now look at the bitmap counters and if any are '2' or '1',
1254 * decrement and handle accordingly.
1255 */
NeilBrown40cffcc2012-05-22 13:55:24 +10001256 counts = &bitmap->counts;
1257 spin_lock_irq(&counts->lock);
NeilBrownbf07bb72012-05-22 13:55:06 +10001258 nextpage = 0;
NeilBrown40cffcc2012-05-22 13:55:24 +10001259 for (j = 0; j < counts->chunks; j++) {
NeilBrown32a76272005-06-21 17:17:14 -07001260 bitmap_counter_t *bmc;
NeilBrown40cffcc2012-05-22 13:55:24 +10001261 sector_t block = (sector_t)j << counts->chunkshift;
Jonathan Brassow3520fa42011-07-27 11:00:37 +10001262
NeilBrownbf07bb72012-05-22 13:55:06 +10001263 if (j == nextpage) {
1264 nextpage += PAGE_COUNTER_RATIO;
NeilBrown40cffcc2012-05-22 13:55:24 +10001265 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
NeilBrownbf07bb72012-05-22 13:55:06 +10001266 j |= PAGE_COUNTER_MASK;
NeilBrownaa3163f2005-06-21 17:17:22 -07001267 continue;
1268 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001269 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001270 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001271 bmc = bitmap_get_counter(counts,
NeilBrownef99bf42012-05-22 13:55:08 +10001272 block,
NeilBrowndb305e52009-05-07 12:49:06 +10001273 &blocks, 0);
NeilBrown32a76272005-06-21 17:17:14 -07001274
NeilBrownbf07bb72012-05-22 13:55:06 +10001275 if (!bmc) {
1276 j |= PAGE_COUNTER_MASK;
1277 continue;
1278 }
1279 if (*bmc == 1 && !bitmap->need_sync) {
1280 /* We can clear the bit */
NeilBrownbf07bb72012-05-22 13:55:06 +10001281 *bmc = 0;
NeilBrown40cffcc2012-05-22 13:55:24 +10001282 bitmap_count_page(counts, block, -1);
NeilBrownef99bf42012-05-22 13:55:08 +10001283 bitmap_file_clear_bit(bitmap, block);
NeilBrownbf07bb72012-05-22 13:55:06 +10001284 } else if (*bmc && *bmc <= 2) {
1285 *bmc = 1;
NeilBrown40cffcc2012-05-22 13:55:24 +10001286 bitmap_set_pending(counts, block);
NeilBrown2585f3e2011-09-21 15:37:46 +10001287 bitmap->allclean = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001288 }
NeilBrown32a76272005-06-21 17:17:14 -07001289 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001290 spin_unlock_irq(&counts->lock);
NeilBrown32a76272005-06-21 17:17:14 -07001291
NeilBrownbf07bb72012-05-22 13:55:06 +10001292 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1293 * DIRTY pages need to be written by bitmap_unplug so it can wait
1294 * for them.
1295 * If we find any DIRTY page we stop there and let bitmap_unplug
1296 * handle all the rest. This is important in the case where
1297 * the first blocking holds the superblock and it has been updated.
1298 * We mustn't write any other blocks before the superblock.
1299 */
NeilBrown62f82fa2012-05-22 13:55:21 +10001300 for (j = 0;
1301 j < bitmap->storage.file_pages
1302 && !test_bit(BITMAP_STALE, &bitmap->flags);
1303 j++) {
NeilBrownd1891222012-05-22 13:55:09 +10001304 if (test_page_attr(bitmap, j,
NeilBrownbf07bb72012-05-22 13:55:06 +10001305 BITMAP_PAGE_DIRTY))
1306 /* bitmap_unplug will handle the rest */
1307 break;
NeilBrownbdfd1142012-05-22 13:55:22 +10001308 if (test_and_clear_page_attr(bitmap, j,
1309 BITMAP_PAGE_NEEDWRITE)) {
NeilBrown1ec885c2012-05-22 13:55:10 +10001310 write_page(bitmap, bitmap->storage.filemap[j], 0);
NeilBrownbf07bb72012-05-22 13:55:06 +10001311 }
1312 }
NeilBrownbf07bb72012-05-22 13:55:06 +10001313
NeilBrown7be3dfe2008-03-10 11:43:48 -07001314 done:
NeilBrown8311c292008-03-04 14:29:30 -08001315 if (bitmap->allclean == 0)
NeilBrown2e61ebb2011-12-23 10:17:50 +11001316 mddev->thread->timeout =
1317 mddev->bitmap_info.daemon_sleep;
NeilBrownc3d97142009-12-14 12:49:52 +11001318 mutex_unlock(&mddev->bitmap_info.mutex);
NeilBrown32a76272005-06-21 17:17:14 -07001319}
1320
NeilBrown40cffcc2012-05-22 13:55:24 +10001321static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
NeilBrown57dab0b2010-10-19 10:03:39 +11001322 sector_t offset, sector_t *blocks,
NeilBrown32a76272005-06-21 17:17:14 -07001323 int create)
NeilBrownee305ac2009-09-23 18:06:44 +10001324__releases(bitmap->lock)
1325__acquires(bitmap->lock)
NeilBrown32a76272005-06-21 17:17:14 -07001326{
1327 /* If 'create', we might release the lock and reclaim it.
1328 * The lock must have been taken with interrupts enabled.
1329 * If !create, we don't release the lock.
1330 */
NeilBrown61a0d802012-03-19 12:46:41 +11001331 sector_t chunk = offset >> bitmap->chunkshift;
NeilBrown32a76272005-06-21 17:17:14 -07001332 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1333 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1334 sector_t csize;
NeilBrownef425672010-06-01 19:37:33 +10001335 int err;
NeilBrown32a76272005-06-21 17:17:14 -07001336
Guoqing Jiangc9d65032016-05-02 11:50:11 -04001337 err = bitmap_checkpage(bitmap, page, create, 0);
NeilBrownef425672010-06-01 19:37:33 +10001338
1339 if (bitmap->bp[page].hijacked ||
1340 bitmap->bp[page].map == NULL)
NeilBrown61a0d802012-03-19 12:46:41 +11001341 csize = ((sector_t)1) << (bitmap->chunkshift +
NeilBrownef425672010-06-01 19:37:33 +10001342 PAGE_COUNTER_SHIFT - 1);
1343 else
NeilBrown61a0d802012-03-19 12:46:41 +11001344 csize = ((sector_t)1) << bitmap->chunkshift;
NeilBrownef425672010-06-01 19:37:33 +10001345 *blocks = csize - (offset & (csize - 1));
1346
1347 if (err < 0)
NeilBrown32a76272005-06-21 17:17:14 -07001348 return NULL;
NeilBrownef425672010-06-01 19:37:33 +10001349
NeilBrown32a76272005-06-21 17:17:14 -07001350 /* now locked ... */
1351
1352 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1353 /* should we use the first or second counter field
1354 * of the hijacked pointer? */
1355 int hi = (pageoff > PAGE_COUNTER_MASK);
NeilBrown32a76272005-06-21 17:17:14 -07001356 return &((bitmap_counter_t *)
1357 &bitmap->bp[page].map)[hi];
NeilBrownef425672010-06-01 19:37:33 +10001358 } else /* page is allocated */
NeilBrown32a76272005-06-21 17:17:14 -07001359 return (bitmap_counter_t *)
1360 &(bitmap->bp[page].map[pageoff]);
NeilBrown32a76272005-06-21 17:17:14 -07001361}
1362
NeilBrown4b6d2872005-09-09 16:23:47 -07001363int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
NeilBrown32a76272005-06-21 17:17:14 -07001364{
NeilBrownac2f40b2010-06-01 19:37:31 +10001365 if (!bitmap)
1366 return 0;
NeilBrown4b6d2872005-09-09 16:23:47 -07001367
1368 if (behind) {
Paul Clements696fcd52010-03-08 16:02:37 +11001369 int bw;
NeilBrown4b6d2872005-09-09 16:23:47 -07001370 atomic_inc(&bitmap->behind_writes);
Paul Clements696fcd52010-03-08 16:02:37 +11001371 bw = atomic_read(&bitmap->behind_writes);
1372 if (bw > bitmap->behind_writes_used)
1373 bitmap->behind_writes_used = bw;
1374
NeilBrown36a4e1f2011-10-07 14:23:17 +11001375 pr_debug("inc write-behind count %d/%lu\n",
1376 bw, bitmap->mddev->bitmap_info.max_write_behind);
NeilBrown4b6d2872005-09-09 16:23:47 -07001377 }
1378
NeilBrown32a76272005-06-21 17:17:14 -07001379 while (sectors) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001380 sector_t blocks;
NeilBrown32a76272005-06-21 17:17:14 -07001381 bitmap_counter_t *bmc;
1382
NeilBrown40cffcc2012-05-22 13:55:24 +10001383 spin_lock_irq(&bitmap->counts.lock);
1384 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
NeilBrown32a76272005-06-21 17:17:14 -07001385 if (!bmc) {
NeilBrown40cffcc2012-05-22 13:55:24 +10001386 spin_unlock_irq(&bitmap->counts.lock);
NeilBrown32a76272005-06-21 17:17:14 -07001387 return 0;
1388 }
1389
Namhyung Kim27d5ea02011-06-09 11:42:57 +10001390 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
Neil Brownda6e1a32007-02-08 14:20:37 -08001391 DEFINE_WAIT(__wait);
1392 /* note that it is safe to do the prepare_to_wait
1393 * after the test as long as we do it before dropping
1394 * the spinlock.
1395 */
1396 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1397 TASK_UNINTERRUPTIBLE);
NeilBrown40cffcc2012-05-22 13:55:24 +10001398 spin_unlock_irq(&bitmap->counts.lock);
NeilBrownf54a9d02012-08-02 08:33:20 +10001399 schedule();
Neil Brownda6e1a32007-02-08 14:20:37 -08001400 finish_wait(&bitmap->overflow_wait, &__wait);
1401 continue;
1402 }
1403
NeilBrownac2f40b2010-06-01 19:37:31 +10001404 switch (*bmc) {
NeilBrown32a76272005-06-21 17:17:14 -07001405 case 0:
1406 bitmap_file_set_bit(bitmap, offset);
NeilBrown40cffcc2012-05-22 13:55:24 +10001407 bitmap_count_page(&bitmap->counts, offset, 1);
NeilBrown32a76272005-06-21 17:17:14 -07001408 /* fall through */
1409 case 1:
1410 *bmc = 2;
1411 }
Neil Brownda6e1a32007-02-08 14:20:37 -08001412
NeilBrown32a76272005-06-21 17:17:14 -07001413 (*bmc)++;
1414
NeilBrown40cffcc2012-05-22 13:55:24 +10001415 spin_unlock_irq(&bitmap->counts.lock);
NeilBrown32a76272005-06-21 17:17:14 -07001416
1417 offset += blocks;
1418 if (sectors > blocks)
1419 sectors -= blocks;
NeilBrownac2f40b2010-06-01 19:37:31 +10001420 else
1421 sectors = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001422 }
1423 return 0;
1424}
NeilBrownac2f40b2010-06-01 19:37:31 +10001425EXPORT_SYMBOL(bitmap_startwrite);
NeilBrown32a76272005-06-21 17:17:14 -07001426
1427void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
NeilBrown4b6d2872005-09-09 16:23:47 -07001428 int success, int behind)
NeilBrown32a76272005-06-21 17:17:14 -07001429{
NeilBrownac2f40b2010-06-01 19:37:31 +10001430 if (!bitmap)
1431 return;
NeilBrown4b6d2872005-09-09 16:23:47 -07001432 if (behind) {
NeilBrowne5551902010-03-31 11:21:44 +11001433 if (atomic_dec_and_test(&bitmap->behind_writes))
1434 wake_up(&bitmap->behind_wait);
NeilBrown36a4e1f2011-10-07 14:23:17 +11001435 pr_debug("dec write-behind count %d/%lu\n",
1436 atomic_read(&bitmap->behind_writes),
1437 bitmap->mddev->bitmap_info.max_write_behind);
NeilBrown4b6d2872005-09-09 16:23:47 -07001438 }
1439
NeilBrown32a76272005-06-21 17:17:14 -07001440 while (sectors) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001441 sector_t blocks;
NeilBrown32a76272005-06-21 17:17:14 -07001442 unsigned long flags;
1443 bitmap_counter_t *bmc;
1444
NeilBrown40cffcc2012-05-22 13:55:24 +10001445 spin_lock_irqsave(&bitmap->counts.lock, flags);
1446 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
NeilBrown32a76272005-06-21 17:17:14 -07001447 if (!bmc) {
NeilBrown40cffcc2012-05-22 13:55:24 +10001448 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
NeilBrown32a76272005-06-21 17:17:14 -07001449 return;
1450 }
1451
NeilBrown961902c2011-12-23 09:57:48 +11001452 if (success && !bitmap->mddev->degraded &&
Neil Browna0da84f2008-06-28 08:31:22 +10001453 bitmap->events_cleared < bitmap->mddev->events) {
1454 bitmap->events_cleared = bitmap->mddev->events;
1455 bitmap->need_sync = 1;
NeilBrown5ff5aff2010-06-01 19:37:32 +10001456 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
Neil Browna0da84f2008-06-28 08:31:22 +10001457 }
1458
Namhyung Kim27d5ea02011-06-09 11:42:57 +10001459 if (!success && !NEEDED(*bmc))
NeilBrown32a76272005-06-21 17:17:14 -07001460 *bmc |= NEEDED_MASK;
1461
Namhyung Kim27d5ea02011-06-09 11:42:57 +10001462 if (COUNTER(*bmc) == COUNTER_MAX)
Neil Brownda6e1a32007-02-08 14:20:37 -08001463 wake_up(&bitmap->overflow_wait);
1464
NeilBrown32a76272005-06-21 17:17:14 -07001465 (*bmc)--;
NeilBrown2585f3e2011-09-21 15:37:46 +10001466 if (*bmc <= 2) {
NeilBrown40cffcc2012-05-22 13:55:24 +10001467 bitmap_set_pending(&bitmap->counts, offset);
NeilBrown2585f3e2011-09-21 15:37:46 +10001468 bitmap->allclean = 0;
1469 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001470 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
NeilBrown32a76272005-06-21 17:17:14 -07001471 offset += blocks;
1472 if (sectors > blocks)
1473 sectors -= blocks;
NeilBrownac2f40b2010-06-01 19:37:31 +10001474 else
1475 sectors = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001476 }
1477}
NeilBrownac2f40b2010-06-01 19:37:31 +10001478EXPORT_SYMBOL(bitmap_endwrite);
NeilBrown32a76272005-06-21 17:17:14 -07001479
NeilBrown57dab0b2010-10-19 10:03:39 +11001480static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
NeilBrown1187cf02009-03-31 14:27:02 +11001481 int degraded)
NeilBrown32a76272005-06-21 17:17:14 -07001482{
1483 bitmap_counter_t *bmc;
1484 int rv;
1485 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1486 *blocks = 1024;
1487 return 1; /* always resync if no bitmap */
1488 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001489 spin_lock_irq(&bitmap->counts.lock);
1490 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
NeilBrown32a76272005-06-21 17:17:14 -07001491 rv = 0;
1492 if (bmc) {
1493 /* locked */
1494 if (RESYNC(*bmc))
1495 rv = 1;
1496 else if (NEEDED(*bmc)) {
1497 rv = 1;
NeilBrown6a806c52005-07-15 03:56:35 -07001498 if (!degraded) { /* don't set/clear bits if degraded */
1499 *bmc |= RESYNC_MASK;
1500 *bmc &= ~NEEDED_MASK;
1501 }
NeilBrown32a76272005-06-21 17:17:14 -07001502 }
1503 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001504 spin_unlock_irq(&bitmap->counts.lock);
NeilBrown32a76272005-06-21 17:17:14 -07001505 return rv;
1506}
1507
NeilBrown57dab0b2010-10-19 10:03:39 +11001508int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
NeilBrown1187cf02009-03-31 14:27:02 +11001509 int degraded)
1510{
1511 /* bitmap_start_sync must always report on multiples of whole
1512 * pages, otherwise resync (which is very PAGE_SIZE based) will
1513 * get confused.
1514 * So call __bitmap_start_sync repeatedly (if needed) until
1515 * At least PAGE_SIZE>>9 blocks are covered.
1516 * Return the 'or' of the result.
1517 */
1518 int rv = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11001519 sector_t blocks1;
NeilBrown1187cf02009-03-31 14:27:02 +11001520
1521 *blocks = 0;
1522 while (*blocks < (PAGE_SIZE>>9)) {
1523 rv |= __bitmap_start_sync(bitmap, offset,
1524 &blocks1, degraded);
1525 offset += blocks1;
1526 *blocks += blocks1;
1527 }
1528 return rv;
1529}
NeilBrownac2f40b2010-06-01 19:37:31 +10001530EXPORT_SYMBOL(bitmap_start_sync);
NeilBrown1187cf02009-03-31 14:27:02 +11001531
NeilBrown57dab0b2010-10-19 10:03:39 +11001532void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
NeilBrown32a76272005-06-21 17:17:14 -07001533{
1534 bitmap_counter_t *bmc;
1535 unsigned long flags;
NeilBrownac2f40b2010-06-01 19:37:31 +10001536
1537 if (bitmap == NULL) {
NeilBrown32a76272005-06-21 17:17:14 -07001538 *blocks = 1024;
1539 return;
1540 }
NeilBrown40cffcc2012-05-22 13:55:24 +10001541 spin_lock_irqsave(&bitmap->counts.lock, flags);
1542 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
NeilBrown32a76272005-06-21 17:17:14 -07001543 if (bmc == NULL)
1544 goto unlock;
1545 /* locked */
NeilBrown32a76272005-06-21 17:17:14 -07001546 if (RESYNC(*bmc)) {
1547 *bmc &= ~RESYNC_MASK;
1548
1549 if (!NEEDED(*bmc) && aborted)
1550 *bmc |= NEEDED_MASK;
1551 else {
NeilBrown2585f3e2011-09-21 15:37:46 +10001552 if (*bmc <= 2) {
NeilBrown40cffcc2012-05-22 13:55:24 +10001553 bitmap_set_pending(&bitmap->counts, offset);
NeilBrown2585f3e2011-09-21 15:37:46 +10001554 bitmap->allclean = 0;
1555 }
NeilBrown32a76272005-06-21 17:17:14 -07001556 }
1557 }
1558 unlock:
NeilBrown40cffcc2012-05-22 13:55:24 +10001559 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
NeilBrown32a76272005-06-21 17:17:14 -07001560}
NeilBrownac2f40b2010-06-01 19:37:31 +10001561EXPORT_SYMBOL(bitmap_end_sync);
NeilBrown32a76272005-06-21 17:17:14 -07001562
1563void bitmap_close_sync(struct bitmap *bitmap)
1564{
1565 /* Sync has finished, and any bitmap chunks that weren't synced
1566 * properly have been aborted. It remains to us to clear the
1567 * RESYNC bit wherever it is still on
1568 */
1569 sector_t sector = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11001570 sector_t blocks;
NeilBrownb47490c2008-02-06 01:39:50 -08001571 if (!bitmap)
1572 return;
NeilBrown32a76272005-06-21 17:17:14 -07001573 while (sector < bitmap->mddev->resync_max_sectors) {
1574 bitmap_end_sync(bitmap, sector, &blocks, 0);
NeilBrownb47490c2008-02-06 01:39:50 -08001575 sector += blocks;
NeilBrown32a76272005-06-21 17:17:14 -07001576 }
1577}
NeilBrownac2f40b2010-06-01 19:37:31 +10001578EXPORT_SYMBOL(bitmap_close_sync);
NeilBrown32a76272005-06-21 17:17:14 -07001579
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10001580void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force)
NeilBrownb47490c2008-02-06 01:39:50 -08001581{
1582 sector_t s = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11001583 sector_t blocks;
NeilBrownb47490c2008-02-06 01:39:50 -08001584
1585 if (!bitmap)
1586 return;
1587 if (sector == 0) {
1588 bitmap->last_end_sync = jiffies;
1589 return;
1590 }
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10001591 if (!force && time_before(jiffies, (bitmap->last_end_sync
NeilBrown1b04be92009-12-14 12:49:53 +11001592 + bitmap->mddev->bitmap_info.daemon_sleep)))
NeilBrownb47490c2008-02-06 01:39:50 -08001593 return;
1594 wait_event(bitmap->mddev->recovery_wait,
1595 atomic_read(&bitmap->mddev->recovery_active) == 0);
1596
NeilBrown75d3da42011-01-14 09:14:34 +11001597 bitmap->mddev->curr_resync_completed = sector;
NeilBrown070dc6d2010-08-30 17:33:34 +10001598 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
NeilBrown40cffcc2012-05-22 13:55:24 +10001599 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
NeilBrownb47490c2008-02-06 01:39:50 -08001600 s = 0;
1601 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1602 bitmap_end_sync(bitmap, s, &blocks, 0);
1603 s += blocks;
1604 }
1605 bitmap->last_end_sync = jiffies;
NeilBrownacb180b2009-04-14 16:28:34 +10001606 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
NeilBrownb47490c2008-02-06 01:39:50 -08001607}
NeilBrownac2f40b2010-06-01 19:37:31 +10001608EXPORT_SYMBOL(bitmap_cond_end_sync);
NeilBrownb47490c2008-02-06 01:39:50 -08001609
Guoqing Jiang18c9ff72016-05-02 11:50:12 -04001610void bitmap_sync_with_cluster(struct mddev *mddev,
1611 sector_t old_lo, sector_t old_hi,
1612 sector_t new_lo, sector_t new_hi)
1613{
1614 struct bitmap *bitmap = mddev->bitmap;
1615 sector_t sector, blocks = 0;
1616
1617 for (sector = old_lo; sector < new_lo; ) {
1618 bitmap_end_sync(bitmap, sector, &blocks, 0);
1619 sector += blocks;
1620 }
1621 WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n");
1622
1623 for (sector = old_hi; sector < new_hi; ) {
1624 bitmap_start_sync(bitmap, sector, &blocks, 0);
1625 sector += blocks;
1626 }
1627 WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n");
1628}
1629EXPORT_SYMBOL(bitmap_sync_with_cluster);
1630
NeilBrown6a079972005-09-09 16:23:44 -07001631static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
NeilBrown32a76272005-06-21 17:17:14 -07001632{
1633 /* For each chunk covered by any of these sectors, set the
NeilBrownef99bf42012-05-22 13:55:08 +10001634 * counter to 2 and possibly set resync_needed. They should all
NeilBrown32a76272005-06-21 17:17:14 -07001635 * be 0 at this point
1636 */
NeilBrown193f1c92005-08-04 12:53:33 -07001637
NeilBrown57dab0b2010-10-19 10:03:39 +11001638 sector_t secs;
NeilBrown193f1c92005-08-04 12:53:33 -07001639 bitmap_counter_t *bmc;
NeilBrown40cffcc2012-05-22 13:55:24 +10001640 spin_lock_irq(&bitmap->counts.lock);
1641 bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
NeilBrown193f1c92005-08-04 12:53:33 -07001642 if (!bmc) {
NeilBrown40cffcc2012-05-22 13:55:24 +10001643 spin_unlock_irq(&bitmap->counts.lock);
NeilBrown193f1c92005-08-04 12:53:33 -07001644 return;
NeilBrown32a76272005-06-21 17:17:14 -07001645 }
NeilBrownac2f40b2010-06-01 19:37:31 +10001646 if (!*bmc) {
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001647 *bmc = 2;
NeilBrown40cffcc2012-05-22 13:55:24 +10001648 bitmap_count_page(&bitmap->counts, offset, 1);
1649 bitmap_set_pending(&bitmap->counts, offset);
NeilBrown2585f3e2011-09-21 15:37:46 +10001650 bitmap->allclean = 0;
NeilBrown193f1c92005-08-04 12:53:33 -07001651 }
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001652 if (needed)
1653 *bmc |= NEEDED_MASK;
NeilBrown40cffcc2012-05-22 13:55:24 +10001654 spin_unlock_irq(&bitmap->counts.lock);
NeilBrown32a76272005-06-21 17:17:14 -07001655}
1656
Paul Clements9b1d1da2006-10-03 01:15:49 -07001657/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1658void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1659{
1660 unsigned long chunk;
1661
1662 for (chunk = s; chunk <= e; chunk++) {
NeilBrown40cffcc2012-05-22 13:55:24 +10001663 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
Paul Clements9b1d1da2006-10-03 01:15:49 -07001664 bitmap_set_memory_bits(bitmap, sec, 1);
1665 bitmap_file_set_bit(bitmap, sec);
NeilBrownffa23322009-12-14 12:49:56 +11001666 if (sec < bitmap->mddev->recovery_cp)
1667 /* We are asserting that the array is dirty,
1668 * so move the recovery_cp address back so
1669 * that it is obvious that it is dirty
1670 */
1671 bitmap->mddev->recovery_cp = sec;
Paul Clements9b1d1da2006-10-03 01:15:49 -07001672 }
1673}
1674
NeilBrown32a76272005-06-21 17:17:14 -07001675/*
NeilBrown6b8b3e82005-08-04 12:53:35 -07001676 * flush out any pending updates
1677 */
NeilBrownfd01b882011-10-11 16:47:53 +11001678void bitmap_flush(struct mddev *mddev)
NeilBrown6b8b3e82005-08-04 12:53:35 -07001679{
1680 struct bitmap *bitmap = mddev->bitmap;
NeilBrown42a04b52009-12-14 12:49:53 +11001681 long sleep;
NeilBrown6b8b3e82005-08-04 12:53:35 -07001682
1683 if (!bitmap) /* there was no bitmap */
1684 return;
1685
1686 /* run the daemon_work three time to ensure everything is flushed
1687 * that can be
1688 */
NeilBrown1b04be92009-12-14 12:49:53 +11001689 sleep = mddev->bitmap_info.daemon_sleep * 2;
NeilBrown42a04b52009-12-14 12:49:53 +11001690 bitmap->daemon_lastrun -= sleep;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001691 bitmap_daemon_work(mddev);
NeilBrown42a04b52009-12-14 12:49:53 +11001692 bitmap->daemon_lastrun -= sleep;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001693 bitmap_daemon_work(mddev);
NeilBrown42a04b52009-12-14 12:49:53 +11001694 bitmap->daemon_lastrun -= sleep;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001695 bitmap_daemon_work(mddev);
NeilBrown6b8b3e82005-08-04 12:53:35 -07001696 bitmap_update_sb(bitmap);
1697}
1698
1699/*
NeilBrown32a76272005-06-21 17:17:14 -07001700 * free memory that was allocated
1701 */
NeilBrown3178b0d2005-09-09 16:23:50 -07001702static void bitmap_free(struct bitmap *bitmap)
NeilBrown32a76272005-06-21 17:17:14 -07001703{
1704 unsigned long k, pages;
1705 struct bitmap_page *bp;
NeilBrown32a76272005-06-21 17:17:14 -07001706
1707 if (!bitmap) /* there was no bitmap */
1708 return;
1709
Guoqing Jiangf9a67b12016-04-01 17:08:49 +08001710 if (bitmap->sysfs_can_clear)
1711 sysfs_put(bitmap->sysfs_can_clear);
1712
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001713 if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1714 bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -05001715 md_cluster_stop(bitmap->mddev);
1716
NeilBrownfae7d322012-05-22 13:55:21 +10001717 /* Shouldn't be needed - but just in case.... */
1718 wait_event(bitmap->write_wait,
1719 atomic_read(&bitmap->pending_writes) == 0);
1720
1721 /* release the bitmap file */
1722 bitmap_file_unmap(&bitmap->storage);
NeilBrown32a76272005-06-21 17:17:14 -07001723
NeilBrown40cffcc2012-05-22 13:55:24 +10001724 bp = bitmap->counts.bp;
1725 pages = bitmap->counts.pages;
NeilBrown32a76272005-06-21 17:17:14 -07001726
1727 /* free all allocated memory */
1728
NeilBrown32a76272005-06-21 17:17:14 -07001729 if (bp) /* deallocate the page memory */
1730 for (k = 0; k < pages; k++)
1731 if (bp[k].map && !bp[k].hijacked)
1732 kfree(bp[k].map);
1733 kfree(bp);
1734 kfree(bitmap);
1735}
NeilBrownaa5cbd12009-12-14 12:49:46 +11001736
NeilBrownfd01b882011-10-11 16:47:53 +11001737void bitmap_destroy(struct mddev *mddev)
NeilBrown3178b0d2005-09-09 16:23:50 -07001738{
1739 struct bitmap *bitmap = mddev->bitmap;
1740
1741 if (!bitmap) /* there was no bitmap */
1742 return;
1743
NeilBrownc3d97142009-12-14 12:49:52 +11001744 mutex_lock(&mddev->bitmap_info.mutex);
NeilBrown978a7a42014-12-15 12:56:58 +11001745 spin_lock(&mddev->lock);
NeilBrown3178b0d2005-09-09 16:23:50 -07001746 mddev->bitmap = NULL; /* disconnect from the md device */
NeilBrown978a7a42014-12-15 12:56:58 +11001747 spin_unlock(&mddev->lock);
NeilBrownc3d97142009-12-14 12:49:52 +11001748 mutex_unlock(&mddev->bitmap_info.mutex);
NeilBrownb15c2e52006-01-06 00:20:16 -08001749 if (mddev->thread)
1750 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
NeilBrown3178b0d2005-09-09 16:23:50 -07001751
1752 bitmap_free(bitmap);
1753}
NeilBrown32a76272005-06-21 17:17:14 -07001754
1755/*
1756 * initialize the bitmap structure
1757 * if this returns an error, bitmap_destroy must be called to do clean up
Guoqing Jiangf9a67b12016-04-01 17:08:49 +08001758 * once mddev->bitmap is set
NeilBrown32a76272005-06-21 17:17:14 -07001759 */
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001760struct bitmap *bitmap_create(struct mddev *mddev, int slot)
NeilBrown32a76272005-06-21 17:17:14 -07001761{
1762 struct bitmap *bitmap;
NeilBrown1f593902009-04-20 11:50:24 +10001763 sector_t blocks = mddev->resync_max_sectors;
NeilBrownc3d97142009-12-14 12:49:52 +11001764 struct file *file = mddev->bitmap_info.file;
NeilBrown32a76272005-06-21 17:17:14 -07001765 int err;
Tejun Heo324a56e2013-12-11 14:11:53 -05001766 struct kernfs_node *bm = NULL;
NeilBrown32a76272005-06-21 17:17:14 -07001767
Alexey Dobriyan5f6e3c832006-10-11 01:22:26 -07001768 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
NeilBrown32a76272005-06-21 17:17:14 -07001769
NeilBrownc3d97142009-12-14 12:49:52 +11001770 BUG_ON(file && mddev->bitmap_info.offset);
NeilBrowna654b9d82005-06-21 17:17:27 -07001771
NeilBrown9ffae0c2006-01-06 00:20:32 -08001772 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
NeilBrown32a76272005-06-21 17:17:14 -07001773 if (!bitmap)
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001774 return ERR_PTR(-ENOMEM);
NeilBrown32a76272005-06-21 17:17:14 -07001775
NeilBrown40cffcc2012-05-22 13:55:24 +10001776 spin_lock_init(&bitmap->counts.lock);
NeilBrownce25c312006-06-26 00:27:49 -07001777 atomic_set(&bitmap->pending_writes, 0);
1778 init_waitqueue_head(&bitmap->write_wait);
Neil Brownda6e1a32007-02-08 14:20:37 -08001779 init_waitqueue_head(&bitmap->overflow_wait);
NeilBrowne5551902010-03-31 11:21:44 +11001780 init_waitqueue_head(&bitmap->behind_wait);
NeilBrownce25c312006-06-26 00:27:49 -07001781
NeilBrown32a76272005-06-21 17:17:14 -07001782 bitmap->mddev = mddev;
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001783 bitmap->cluster_slot = slot;
NeilBrown32a76272005-06-21 17:17:14 -07001784
NeilBrown5ff5aff2010-06-01 19:37:32 +10001785 if (mddev->kobj.sd)
Tejun Heo388975c2013-09-11 23:19:13 -04001786 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
NeilBrownece5cff2009-12-14 12:49:56 +11001787 if (bm) {
Tejun Heo388975c2013-09-11 23:19:13 -04001788 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
NeilBrownece5cff2009-12-14 12:49:56 +11001789 sysfs_put(bm);
1790 } else
1791 bitmap->sysfs_can_clear = NULL;
1792
NeilBrown1ec885c2012-05-22 13:55:10 +10001793 bitmap->storage.file = file;
NeilBrownce25c312006-06-26 00:27:49 -07001794 if (file) {
1795 get_file(file);
NeilBrownae8fa282009-10-16 15:56:01 +11001796 /* As future accesses to this file will use bmap,
1797 * and bypass the page cache, we must sync the file
1798 * first.
1799 */
Christoph Hellwig8018ab02010-03-22 17:32:25 +01001800 vfs_fsync(file, 1);
NeilBrownce25c312006-06-26 00:27:49 -07001801 }
NeilBrown42a04b52009-12-14 12:49:53 +11001802 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
Jonathan Brassow9c810752011-06-08 17:59:30 -05001803 if (!mddev->bitmap_info.external) {
1804 /*
1805 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1806 * instructing us to create a new on-disk bitmap instance.
1807 */
1808 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1809 err = bitmap_new_disk_sb(bitmap);
1810 else
1811 err = bitmap_read_sb(bitmap);
1812 } else {
NeilBrownece5cff2009-12-14 12:49:56 +11001813 err = 0;
1814 if (mddev->bitmap_info.chunksize == 0 ||
1815 mddev->bitmap_info.daemon_sleep == 0)
1816 /* chunksize and time_base need to be
1817 * set first. */
1818 err = -EINVAL;
1819 }
NeilBrown32a76272005-06-21 17:17:14 -07001820 if (err)
NeilBrown3178b0d2005-09-09 16:23:50 -07001821 goto error;
NeilBrown32a76272005-06-21 17:17:14 -07001822
NeilBrown624ce4f2009-12-14 12:49:56 +11001823 bitmap->daemon_lastrun = jiffies;
NeilBrownd60b4792012-05-22 13:55:25 +10001824 err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1825 if (err)
NeilBrown3178b0d2005-09-09 16:23:50 -07001826 goto error;
NeilBrown32a76272005-06-21 17:17:14 -07001827
NeilBrown69e51b42010-06-01 19:37:35 +10001828 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
NeilBrownd60b4792012-05-22 13:55:25 +10001829 bitmap->counts.pages, bmname(bitmap));
NeilBrown69e51b42010-06-01 19:37:35 +10001830
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001831 err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1832 if (err)
1833 goto error;
NeilBrown69e51b42010-06-01 19:37:35 +10001834
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001835 return bitmap;
NeilBrown69e51b42010-06-01 19:37:35 +10001836 error:
1837 bitmap_free(bitmap);
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05001838 return ERR_PTR(err);
NeilBrown69e51b42010-06-01 19:37:35 +10001839}
1840
NeilBrownfd01b882011-10-11 16:47:53 +11001841int bitmap_load(struct mddev *mddev)
NeilBrown69e51b42010-06-01 19:37:35 +10001842{
1843 int err = 0;
Jonathan Brassow3520fa42011-07-27 11:00:37 +10001844 sector_t start = 0;
NeilBrown69e51b42010-06-01 19:37:35 +10001845 sector_t sector = 0;
1846 struct bitmap *bitmap = mddev->bitmap;
1847
1848 if (!bitmap)
1849 goto out;
1850
Guoqing Jiang51e453a2016-05-04 02:17:09 -04001851 if (mddev_is_clustered(mddev))
1852 md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes);
1853
NeilBrown69e51b42010-06-01 19:37:35 +10001854 /* Clear out old bitmap info first: Either there is none, or we
1855 * are resuming after someone else has possibly changed things,
1856 * so we should forget old cached info.
1857 * All chunks should be clean, but some might need_sync.
1858 */
1859 while (sector < mddev->resync_max_sectors) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001860 sector_t blocks;
NeilBrown69e51b42010-06-01 19:37:35 +10001861 bitmap_start_sync(bitmap, sector, &blocks, 0);
1862 sector += blocks;
1863 }
1864 bitmap_close_sync(bitmap);
1865
Jonathan Brassow3520fa42011-07-27 11:00:37 +10001866 if (mddev->degraded == 0
1867 || bitmap->events_cleared == mddev->events)
1868 /* no need to keep dirty bits to optimise a
1869 * re-add of a missing device */
1870 start = mddev->recovery_cp;
NeilBrown69e51b42010-06-01 19:37:35 +10001871
NeilBrownafbaa902012-04-12 16:05:06 +10001872 mutex_lock(&mddev->bitmap_info.mutex);
Jonathan Brassow3520fa42011-07-27 11:00:37 +10001873 err = bitmap_init_from_disk(bitmap, start);
NeilBrownafbaa902012-04-12 16:05:06 +10001874 mutex_unlock(&mddev->bitmap_info.mutex);
Jonathan Brassow3520fa42011-07-27 11:00:37 +10001875
NeilBrown32a76272005-06-21 17:17:14 -07001876 if (err)
NeilBrown69e51b42010-06-01 19:37:35 +10001877 goto out;
NeilBrownb405fe92012-05-22 13:55:15 +10001878 clear_bit(BITMAP_STALE, &bitmap->flags);
NeilBrownef99bf42012-05-22 13:55:08 +10001879
1880 /* Kick recovery in case any bits were set */
1881 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
NeilBrown3178b0d2005-09-09 16:23:50 -07001882
NeilBrown1b04be92009-12-14 12:49:53 +11001883 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
NeilBrown9cd30fd2009-12-14 12:49:54 +11001884 md_wakeup_thread(mddev->thread);
NeilBrownb15c2e52006-01-06 00:20:16 -08001885
NeilBrown4ad13662007-07-17 04:06:13 -07001886 bitmap_update_sb(bitmap);
1887
NeilBrownb405fe92012-05-22 13:55:15 +10001888 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
NeilBrown69e51b42010-06-01 19:37:35 +10001889 err = -EIO;
1890out:
NeilBrown3178b0d2005-09-09 16:23:50 -07001891 return err;
NeilBrown32a76272005-06-21 17:17:14 -07001892}
NeilBrown69e51b42010-06-01 19:37:35 +10001893EXPORT_SYMBOL_GPL(bitmap_load);
NeilBrown32a76272005-06-21 17:17:14 -07001894
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001895/* Loads the bitmap associated with slot and copies the resync information
1896 * to our bitmap
1897 */
1898int bitmap_copy_from_slot(struct mddev *mddev, int slot,
Goldwyn Rodrigues97f6cd32015-04-14 10:45:42 -05001899 sector_t *low, sector_t *high, bool clear_bits)
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001900{
1901 int rv = 0, i, j;
1902 sector_t block, lo = 0, hi = 0;
1903 struct bitmap_counts *counts;
1904 struct bitmap *bitmap = bitmap_create(mddev, slot);
1905
Shaohua Lif71f1cf92016-09-13 10:28:00 -07001906 if (IS_ERR(bitmap))
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001907 return PTR_ERR(bitmap);
1908
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001909 rv = bitmap_init_from_disk(bitmap, 0);
1910 if (rv)
1911 goto err;
1912
1913 counts = &bitmap->counts;
1914 for (j = 0; j < counts->chunks; j++) {
1915 block = (sector_t)j << counts->chunkshift;
1916 if (bitmap_file_test_bit(bitmap, block)) {
1917 if (!lo)
1918 lo = block;
1919 hi = block;
1920 bitmap_file_clear_bit(bitmap, block);
1921 bitmap_set_memory_bits(mddev->bitmap, block, 1);
1922 bitmap_file_set_bit(mddev->bitmap, block);
1923 }
1924 }
1925
Goldwyn Rodrigues97f6cd32015-04-14 10:45:42 -05001926 if (clear_bits) {
1927 bitmap_update_sb(bitmap);
Guoqing Jiangc84400c2016-05-02 11:50:15 -04001928 /* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs
1929 * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */
Goldwyn Rodrigues97f6cd32015-04-14 10:45:42 -05001930 for (i = 0; i < bitmap->storage.file_pages; i++)
Guoqing Jiangc84400c2016-05-02 11:50:15 -04001931 if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING))
1932 set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
Goldwyn Rodrigues97f6cd32015-04-14 10:45:42 -05001933 bitmap_unplug(bitmap);
1934 }
Guoqing Jiangc84400c2016-05-02 11:50:15 -04001935 bitmap_unplug(mddev->bitmap);
Goldwyn Rodrigues11dd35d2014-06-07 00:36:26 -05001936 *low = lo;
1937 *high = hi;
1938err:
1939 bitmap_free(bitmap);
1940 return rv;
1941}
1942EXPORT_SYMBOL_GPL(bitmap_copy_from_slot);
1943
1944
NeilBrown57148962012-03-19 12:46:40 +11001945void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1946{
1947 unsigned long chunk_kb;
NeilBrown40cffcc2012-05-22 13:55:24 +10001948 struct bitmap_counts *counts;
NeilBrown57148962012-03-19 12:46:40 +11001949
1950 if (!bitmap)
1951 return;
1952
NeilBrown40cffcc2012-05-22 13:55:24 +10001953 counts = &bitmap->counts;
1954
NeilBrown57148962012-03-19 12:46:40 +11001955 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1956 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1957 "%lu%s chunk",
NeilBrown40cffcc2012-05-22 13:55:24 +10001958 counts->pages - counts->missing_pages,
1959 counts->pages,
1960 (counts->pages - counts->missing_pages)
NeilBrown57148962012-03-19 12:46:40 +11001961 << (PAGE_SHIFT - 10),
1962 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1963 chunk_kb ? "KB" : "B");
NeilBrown1ec885c2012-05-22 13:55:10 +10001964 if (bitmap->storage.file) {
NeilBrown57148962012-03-19 12:46:40 +11001965 seq_printf(seq, ", file: ");
Miklos Szeredi2726d562015-06-19 10:30:28 +02001966 seq_file_path(seq, bitmap->storage.file, " \t\n");
NeilBrown57148962012-03-19 12:46:40 +11001967 }
1968
1969 seq_printf(seq, "\n");
NeilBrown57148962012-03-19 12:46:40 +11001970}
1971
NeilBrownd60b4792012-05-22 13:55:25 +10001972int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
1973 int chunksize, int init)
1974{
1975 /* If chunk_size is 0, choose an appropriate chunk size.
1976 * Then possibly allocate new storage space.
1977 * Then quiesce, copy bits, replace bitmap, and re-start
1978 *
1979 * This function is called both to set up the initial bitmap
1980 * and to resize the bitmap while the array is active.
1981 * If this happens as a result of the array being resized,
1982 * chunksize will be zero, and we need to choose a suitable
1983 * chunksize, otherwise we use what we are given.
1984 */
1985 struct bitmap_storage store;
1986 struct bitmap_counts old_counts;
1987 unsigned long chunks;
1988 sector_t block;
1989 sector_t old_blocks, new_blocks;
1990 int chunkshift;
1991 int ret = 0;
1992 long pages;
1993 struct bitmap_page *new_bp;
1994
1995 if (chunksize == 0) {
1996 /* If there is enough space, leave the chunk size unchanged,
1997 * else increase by factor of two until there is enough space.
1998 */
1999 long bytes;
2000 long space = bitmap->mddev->bitmap_info.space;
2001
2002 if (space == 0) {
2003 /* We don't know how much space there is, so limit
2004 * to current size - in sectors.
2005 */
2006 bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
2007 if (!bitmap->mddev->bitmap_info.external)
2008 bytes += sizeof(bitmap_super_t);
2009 space = DIV_ROUND_UP(bytes, 512);
2010 bitmap->mddev->bitmap_info.space = space;
2011 }
2012 chunkshift = bitmap->counts.chunkshift;
2013 chunkshift--;
2014 do {
2015 /* 'chunkshift' is shift from block size to chunk size */
2016 chunkshift++;
2017 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2018 bytes = DIV_ROUND_UP(chunks, 8);
2019 if (!bitmap->mddev->bitmap_info.external)
2020 bytes += sizeof(bitmap_super_t);
2021 } while (bytes > (space << 9));
2022 } else
2023 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
2024
2025 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2026 memset(&store, 0, sizeof(store));
2027 if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
2028 ret = bitmap_storage_alloc(&store, chunks,
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -05002029 !bitmap->mddev->bitmap_info.external,
NeilBrownda6fb7a2015-10-01 16:03:38 +10002030 mddev_is_clustered(bitmap->mddev)
2031 ? bitmap->cluster_slot : 0);
NeilBrownd60b4792012-05-22 13:55:25 +10002032 if (ret)
2033 goto err;
2034
2035 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2036
2037 new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL);
2038 ret = -ENOMEM;
2039 if (!new_bp) {
2040 bitmap_file_unmap(&store);
2041 goto err;
2042 }
2043
2044 if (!init)
2045 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2046
2047 store.file = bitmap->storage.file;
2048 bitmap->storage.file = NULL;
2049
2050 if (store.sb_page && bitmap->storage.sb_page)
2051 memcpy(page_address(store.sb_page),
2052 page_address(bitmap->storage.sb_page),
2053 sizeof(bitmap_super_t));
2054 bitmap_file_unmap(&bitmap->storage);
2055 bitmap->storage = store;
2056
2057 old_counts = bitmap->counts;
2058 bitmap->counts.bp = new_bp;
2059 bitmap->counts.pages = pages;
2060 bitmap->counts.missing_pages = pages;
2061 bitmap->counts.chunkshift = chunkshift;
2062 bitmap->counts.chunks = chunks;
2063 bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
2064 BITMAP_BLOCK_SHIFT);
2065
2066 blocks = min(old_counts.chunks << old_counts.chunkshift,
2067 chunks << chunkshift);
2068
2069 spin_lock_irq(&bitmap->counts.lock);
Guoqing Jiangc9d65032016-05-02 11:50:11 -04002070 /* For cluster raid, need to pre-allocate bitmap */
2071 if (mddev_is_clustered(bitmap->mddev)) {
2072 unsigned long page;
2073 for (page = 0; page < pages; page++) {
2074 ret = bitmap_checkpage(&bitmap->counts, page, 1, 1);
2075 if (ret) {
2076 unsigned long k;
2077
2078 /* deallocate the page memory */
2079 for (k = 0; k < page; k++) {
kbuild test robotbc47e842016-05-02 11:50:16 -04002080 kfree(new_bp[k].map);
Guoqing Jiangc9d65032016-05-02 11:50:11 -04002081 }
2082
2083 /* restore some fields from old_counts */
2084 bitmap->counts.bp = old_counts.bp;
2085 bitmap->counts.pages = old_counts.pages;
2086 bitmap->counts.missing_pages = old_counts.pages;
2087 bitmap->counts.chunkshift = old_counts.chunkshift;
2088 bitmap->counts.chunks = old_counts.chunks;
2089 bitmap->mddev->bitmap_info.chunksize = 1 << (old_counts.chunkshift +
2090 BITMAP_BLOCK_SHIFT);
2091 blocks = old_counts.chunks << old_counts.chunkshift;
2092 pr_err("Could not pre-allocate in-memory bitmap for cluster raid\n");
2093 break;
2094 } else
2095 bitmap->counts.bp[page].count += 1;
2096 }
2097 }
2098
NeilBrownd60b4792012-05-22 13:55:25 +10002099 for (block = 0; block < blocks; ) {
2100 bitmap_counter_t *bmc_old, *bmc_new;
2101 int set;
2102
2103 bmc_old = bitmap_get_counter(&old_counts, block,
2104 &old_blocks, 0);
2105 set = bmc_old && NEEDED(*bmc_old);
2106
2107 if (set) {
2108 bmc_new = bitmap_get_counter(&bitmap->counts, block,
2109 &new_blocks, 1);
2110 if (*bmc_new == 0) {
2111 /* need to set on-disk bits too. */
2112 sector_t end = block + new_blocks;
2113 sector_t start = block >> chunkshift;
2114 start <<= chunkshift;
2115 while (start < end) {
2116 bitmap_file_set_bit(bitmap, block);
2117 start += 1 << chunkshift;
2118 }
2119 *bmc_new = 2;
2120 bitmap_count_page(&bitmap->counts,
2121 block, 1);
2122 bitmap_set_pending(&bitmap->counts,
2123 block);
2124 }
2125 *bmc_new |= NEEDED_MASK;
2126 if (new_blocks < old_blocks)
2127 old_blocks = new_blocks;
2128 }
2129 block += old_blocks;
2130 }
2131
2132 if (!init) {
2133 int i;
2134 while (block < (chunks << chunkshift)) {
2135 bitmap_counter_t *bmc;
2136 bmc = bitmap_get_counter(&bitmap->counts, block,
2137 &new_blocks, 1);
2138 if (bmc) {
2139 /* new space. It needs to be resynced, so
2140 * we set NEEDED_MASK.
2141 */
2142 if (*bmc == 0) {
2143 *bmc = NEEDED_MASK | 2;
2144 bitmap_count_page(&bitmap->counts,
2145 block, 1);
2146 bitmap_set_pending(&bitmap->counts,
2147 block);
2148 }
2149 }
2150 block += new_blocks;
2151 }
2152 for (i = 0; i < bitmap->storage.file_pages; i++)
2153 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2154 }
2155 spin_unlock_irq(&bitmap->counts.lock);
2156
2157 if (!init) {
2158 bitmap_unplug(bitmap);
2159 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2160 }
2161 ret = 0;
2162err:
2163 return ret;
2164}
2165EXPORT_SYMBOL_GPL(bitmap_resize);
2166
NeilBrown43a70502009-12-14 12:49:55 +11002167static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002168location_show(struct mddev *mddev, char *page)
NeilBrown43a70502009-12-14 12:49:55 +11002169{
2170 ssize_t len;
NeilBrownac2f40b2010-06-01 19:37:31 +10002171 if (mddev->bitmap_info.file)
NeilBrown43a70502009-12-14 12:49:55 +11002172 len = sprintf(page, "file");
NeilBrownac2f40b2010-06-01 19:37:31 +10002173 else if (mddev->bitmap_info.offset)
NeilBrown43a70502009-12-14 12:49:55 +11002174 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
NeilBrownac2f40b2010-06-01 19:37:31 +10002175 else
NeilBrown43a70502009-12-14 12:49:55 +11002176 len = sprintf(page, "none");
2177 len += sprintf(page+len, "\n");
2178 return len;
2179}
2180
2181static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002182location_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown43a70502009-12-14 12:49:55 +11002183{
Shaohua Lid9dd26b2016-07-30 10:05:31 -07002184 int rv;
NeilBrown43a70502009-12-14 12:49:55 +11002185
Shaohua Lid9dd26b2016-07-30 10:05:31 -07002186 rv = mddev_lock(mddev);
2187 if (rv)
2188 return rv;
NeilBrown43a70502009-12-14 12:49:55 +11002189 if (mddev->pers) {
Shaohua Lid9dd26b2016-07-30 10:05:31 -07002190 if (!mddev->pers->quiesce) {
2191 rv = -EBUSY;
2192 goto out;
2193 }
2194 if (mddev->recovery || mddev->sync_thread) {
2195 rv = -EBUSY;
2196 goto out;
2197 }
NeilBrown43a70502009-12-14 12:49:55 +11002198 }
2199
2200 if (mddev->bitmap || mddev->bitmap_info.file ||
2201 mddev->bitmap_info.offset) {
2202 /* bitmap already configured. Only option is to clear it */
Shaohua Lid9dd26b2016-07-30 10:05:31 -07002203 if (strncmp(buf, "none", 4) != 0) {
2204 rv = -EBUSY;
2205 goto out;
2206 }
NeilBrown43a70502009-12-14 12:49:55 +11002207 if (mddev->pers) {
2208 mddev->pers->quiesce(mddev, 1);
2209 bitmap_destroy(mddev);
2210 mddev->pers->quiesce(mddev, 0);
2211 }
2212 mddev->bitmap_info.offset = 0;
2213 if (mddev->bitmap_info.file) {
2214 struct file *f = mddev->bitmap_info.file;
2215 mddev->bitmap_info.file = NULL;
NeilBrown43a70502009-12-14 12:49:55 +11002216 fput(f);
2217 }
2218 } else {
2219 /* No bitmap, OK to set a location */
2220 long long offset;
2221 if (strncmp(buf, "none", 4) == 0)
2222 /* nothing to be done */;
2223 else if (strncmp(buf, "file:", 5) == 0) {
2224 /* Not supported yet */
Shaohua Lid9dd26b2016-07-30 10:05:31 -07002225 rv = -EINVAL;
2226 goto out;
NeilBrown43a70502009-12-14 12:49:55 +11002227 } else {
NeilBrown43a70502009-12-14 12:49:55 +11002228 if (buf[0] == '+')
Jingoo Hanb29bebd2013-06-01 16:15:16 +09002229 rv = kstrtoll(buf+1, 10, &offset);
NeilBrown43a70502009-12-14 12:49:55 +11002230 else
Jingoo Hanb29bebd2013-06-01 16:15:16 +09002231 rv = kstrtoll(buf, 10, &offset);
NeilBrown43a70502009-12-14 12:49:55 +11002232 if (rv)
Shaohua Lid9dd26b2016-07-30 10:05:31 -07002233 goto out;
2234 if (offset == 0) {
2235 rv = -EINVAL;
2236 goto out;
2237 }
NeilBrownece5cff2009-12-14 12:49:56 +11002238 if (mddev->bitmap_info.external == 0 &&
2239 mddev->major_version == 0 &&
Shaohua Lid9dd26b2016-07-30 10:05:31 -07002240 offset != mddev->bitmap_info.default_offset) {
2241 rv = -EINVAL;
2242 goto out;
2243 }
NeilBrown43a70502009-12-14 12:49:55 +11002244 mddev->bitmap_info.offset = offset;
2245 if (mddev->pers) {
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05002246 struct bitmap *bitmap;
NeilBrown43a70502009-12-14 12:49:55 +11002247 mddev->pers->quiesce(mddev, 1);
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05002248 bitmap = bitmap_create(mddev, -1);
2249 if (IS_ERR(bitmap))
2250 rv = PTR_ERR(bitmap);
2251 else {
2252 mddev->bitmap = bitmap;
NeilBrown4474ca42012-03-19 12:46:37 +11002253 rv = bitmap_load(mddev);
Guoqing Jiangf9a67b12016-04-01 17:08:49 +08002254 if (rv)
Goldwyn Rodriguesf9209a32014-06-06 12:43:49 -05002255 mddev->bitmap_info.offset = 0;
NeilBrown43a70502009-12-14 12:49:55 +11002256 }
2257 mddev->pers->quiesce(mddev, 0);
Guoqing Jiangf9a67b12016-04-01 17:08:49 +08002258 if (rv) {
2259 bitmap_destroy(mddev);
Shaohua Lid9dd26b2016-07-30 10:05:31 -07002260 goto out;
Guoqing Jiangf9a67b12016-04-01 17:08:49 +08002261 }
NeilBrown43a70502009-12-14 12:49:55 +11002262 }
2263 }
2264 }
2265 if (!mddev->external) {
2266 /* Ensure new bitmap info is stored in
2267 * metadata promptly.
2268 */
2269 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2270 md_wakeup_thread(mddev->thread);
2271 }
Shaohua Lid9dd26b2016-07-30 10:05:31 -07002272 rv = 0;
2273out:
2274 mddev_unlock(mddev);
2275 if (rv)
2276 return rv;
NeilBrown43a70502009-12-14 12:49:55 +11002277 return len;
2278}
2279
2280static struct md_sysfs_entry bitmap_location =
2281__ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2282
NeilBrown6409bb02012-05-22 13:55:07 +10002283/* 'bitmap/space' is the space available at 'location' for the
2284 * bitmap. This allows the kernel to know when it is safe to
2285 * resize the bitmap to match a resized array.
2286 */
2287static ssize_t
2288space_show(struct mddev *mddev, char *page)
2289{
2290 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2291}
2292
2293static ssize_t
2294space_store(struct mddev *mddev, const char *buf, size_t len)
2295{
2296 unsigned long sectors;
2297 int rv;
2298
2299 rv = kstrtoul(buf, 10, &sectors);
2300 if (rv)
2301 return rv;
2302
2303 if (sectors == 0)
2304 return -EINVAL;
2305
2306 if (mddev->bitmap &&
NeilBrown9b1215c2012-05-22 13:55:11 +10002307 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
NeilBrown6409bb02012-05-22 13:55:07 +10002308 return -EFBIG; /* Bitmap is too big for this small space */
2309
2310 /* could make sure it isn't too big, but that isn't really
2311 * needed - user-space should be careful.
2312 */
2313 mddev->bitmap_info.space = sectors;
2314 return len;
2315}
2316
2317static struct md_sysfs_entry bitmap_space =
2318__ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2319
NeilBrown43a70502009-12-14 12:49:55 +11002320static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002321timeout_show(struct mddev *mddev, char *page)
NeilBrown43a70502009-12-14 12:49:55 +11002322{
2323 ssize_t len;
2324 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2325 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
NeilBrownac2f40b2010-06-01 19:37:31 +10002326
NeilBrown43a70502009-12-14 12:49:55 +11002327 len = sprintf(page, "%lu", secs);
2328 if (jifs)
2329 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2330 len += sprintf(page+len, "\n");
2331 return len;
2332}
2333
2334static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002335timeout_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown43a70502009-12-14 12:49:55 +11002336{
2337 /* timeout can be set at any time */
2338 unsigned long timeout;
2339 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2340 if (rv)
2341 return rv;
2342
2343 /* just to make sure we don't overflow... */
2344 if (timeout >= LONG_MAX / HZ)
2345 return -EINVAL;
2346
2347 timeout = timeout * HZ / 10000;
2348
2349 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2350 timeout = MAX_SCHEDULE_TIMEOUT-1;
2351 if (timeout < 1)
2352 timeout = 1;
2353 mddev->bitmap_info.daemon_sleep = timeout;
2354 if (mddev->thread) {
2355 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2356 * the bitmap is all clean and we don't need to
2357 * adjust the timeout right now
2358 */
2359 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2360 mddev->thread->timeout = timeout;
2361 md_wakeup_thread(mddev->thread);
2362 }
2363 }
2364 return len;
2365}
2366
2367static struct md_sysfs_entry bitmap_timeout =
2368__ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2369
2370static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002371backlog_show(struct mddev *mddev, char *page)
NeilBrown43a70502009-12-14 12:49:55 +11002372{
2373 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2374}
2375
2376static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002377backlog_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown43a70502009-12-14 12:49:55 +11002378{
2379 unsigned long backlog;
Jingoo Hanb29bebd2013-06-01 16:15:16 +09002380 int rv = kstrtoul(buf, 10, &backlog);
NeilBrown43a70502009-12-14 12:49:55 +11002381 if (rv)
2382 return rv;
2383 if (backlog > COUNTER_MAX)
2384 return -EINVAL;
2385 mddev->bitmap_info.max_write_behind = backlog;
2386 return len;
2387}
2388
2389static struct md_sysfs_entry bitmap_backlog =
2390__ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2391
2392static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002393chunksize_show(struct mddev *mddev, char *page)
NeilBrown43a70502009-12-14 12:49:55 +11002394{
2395 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2396}
2397
2398static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002399chunksize_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown43a70502009-12-14 12:49:55 +11002400{
2401 /* Can only be changed when no bitmap is active */
2402 int rv;
2403 unsigned long csize;
2404 if (mddev->bitmap)
2405 return -EBUSY;
Jingoo Hanb29bebd2013-06-01 16:15:16 +09002406 rv = kstrtoul(buf, 10, &csize);
NeilBrown43a70502009-12-14 12:49:55 +11002407 if (rv)
2408 return rv;
2409 if (csize < 512 ||
2410 !is_power_of_2(csize))
2411 return -EINVAL;
2412 mddev->bitmap_info.chunksize = csize;
2413 return len;
2414}
2415
2416static struct md_sysfs_entry bitmap_chunksize =
2417__ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2418
NeilBrownfd01b882011-10-11 16:47:53 +11002419static ssize_t metadata_show(struct mddev *mddev, char *page)
NeilBrownece5cff2009-12-14 12:49:56 +11002420{
Goldwyn Rodriguesc4ce8672014-03-29 10:20:02 -05002421 if (mddev_is_clustered(mddev))
2422 return sprintf(page, "clustered\n");
NeilBrownece5cff2009-12-14 12:49:56 +11002423 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2424 ? "external" : "internal"));
2425}
2426
NeilBrownfd01b882011-10-11 16:47:53 +11002427static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrownece5cff2009-12-14 12:49:56 +11002428{
2429 if (mddev->bitmap ||
2430 mddev->bitmap_info.file ||
2431 mddev->bitmap_info.offset)
2432 return -EBUSY;
2433 if (strncmp(buf, "external", 8) == 0)
2434 mddev->bitmap_info.external = 1;
Goldwyn Rodriguesc4ce8672014-03-29 10:20:02 -05002435 else if ((strncmp(buf, "internal", 8) == 0) ||
2436 (strncmp(buf, "clustered", 9) == 0))
NeilBrownece5cff2009-12-14 12:49:56 +11002437 mddev->bitmap_info.external = 0;
2438 else
2439 return -EINVAL;
2440 return len;
2441}
2442
2443static struct md_sysfs_entry bitmap_metadata =
2444__ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2445
NeilBrownfd01b882011-10-11 16:47:53 +11002446static ssize_t can_clear_show(struct mddev *mddev, char *page)
NeilBrownece5cff2009-12-14 12:49:56 +11002447{
2448 int len;
NeilBrownb7b17c92014-12-15 12:56:59 +11002449 spin_lock(&mddev->lock);
NeilBrownece5cff2009-12-14 12:49:56 +11002450 if (mddev->bitmap)
2451 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2452 "false" : "true"));
2453 else
2454 len = sprintf(page, "\n");
NeilBrownb7b17c92014-12-15 12:56:59 +11002455 spin_unlock(&mddev->lock);
NeilBrownece5cff2009-12-14 12:49:56 +11002456 return len;
2457}
2458
NeilBrownfd01b882011-10-11 16:47:53 +11002459static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrownece5cff2009-12-14 12:49:56 +11002460{
2461 if (mddev->bitmap == NULL)
2462 return -ENOENT;
2463 if (strncmp(buf, "false", 5) == 0)
2464 mddev->bitmap->need_sync = 1;
2465 else if (strncmp(buf, "true", 4) == 0) {
2466 if (mddev->degraded)
2467 return -EBUSY;
2468 mddev->bitmap->need_sync = 0;
2469 } else
2470 return -EINVAL;
2471 return len;
2472}
2473
2474static struct md_sysfs_entry bitmap_can_clear =
2475__ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2476
Paul Clements696fcd52010-03-08 16:02:37 +11002477static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002478behind_writes_used_show(struct mddev *mddev, char *page)
Paul Clements696fcd52010-03-08 16:02:37 +11002479{
NeilBrownb7b17c92014-12-15 12:56:59 +11002480 ssize_t ret;
2481 spin_lock(&mddev->lock);
Paul Clements696fcd52010-03-08 16:02:37 +11002482 if (mddev->bitmap == NULL)
NeilBrownb7b17c92014-12-15 12:56:59 +11002483 ret = sprintf(page, "0\n");
2484 else
2485 ret = sprintf(page, "%lu\n",
2486 mddev->bitmap->behind_writes_used);
2487 spin_unlock(&mddev->lock);
2488 return ret;
Paul Clements696fcd52010-03-08 16:02:37 +11002489}
2490
2491static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11002492behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
Paul Clements696fcd52010-03-08 16:02:37 +11002493{
2494 if (mddev->bitmap)
2495 mddev->bitmap->behind_writes_used = 0;
2496 return len;
2497}
2498
2499static struct md_sysfs_entry max_backlog_used =
2500__ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2501 behind_writes_used_show, behind_writes_used_reset);
2502
NeilBrown43a70502009-12-14 12:49:55 +11002503static struct attribute *md_bitmap_attrs[] = {
2504 &bitmap_location.attr,
NeilBrown6409bb02012-05-22 13:55:07 +10002505 &bitmap_space.attr,
NeilBrown43a70502009-12-14 12:49:55 +11002506 &bitmap_timeout.attr,
2507 &bitmap_backlog.attr,
2508 &bitmap_chunksize.attr,
NeilBrownece5cff2009-12-14 12:49:56 +11002509 &bitmap_metadata.attr,
2510 &bitmap_can_clear.attr,
Paul Clements696fcd52010-03-08 16:02:37 +11002511 &max_backlog_used.attr,
NeilBrown43a70502009-12-14 12:49:55 +11002512 NULL
2513};
2514struct attribute_group md_bitmap_group = {
2515 .name = "bitmap",
2516 .attrs = md_bitmap_attrs,
2517};
2518