blob: 663e6051bbad1b169203c6d533a7de96c57af336 [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>
NeilBrown43b2e5d2009-03-31 14:33:13 +110029#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110030#include "bitmap.h"
NeilBrown32a76272005-06-21 17:17:14 -070031
NeilBrowne384e582010-06-01 19:37:34 +100032#include <linux/dm-dirty-log.h>
NeilBrown32a76272005-06-21 17:17:14 -070033/* debug macros */
34
35#define DEBUG 0
36
37#if DEBUG
38/* these are for debugging purposes only! */
39
40/* define one and only one of these */
41#define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
42#define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
43#define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
44#define INJECT_FAULTS_4 0 /* undef */
45#define INJECT_FAULTS_5 0 /* undef */
46#define INJECT_FAULTS_6 0
47
48/* if these are defined, the driver will fail! debug only */
49#define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
50#define INJECT_FATAL_FAULT_2 0 /* undef */
51#define INJECT_FATAL_FAULT_3 0 /* undef */
52#endif
53
NeilBrown32a76272005-06-21 17:17:14 -070054#ifndef PRINTK
55# if DEBUG > 0
56# define PRINTK(x...) printk(KERN_DEBUG x)
57# else
58# define PRINTK(x...)
59# endif
60#endif
61
NeilBrownac2f40b2010-06-01 19:37:31 +100062static inline char *bmname(struct bitmap *bitmap)
NeilBrown32a76272005-06-21 17:17:14 -070063{
64 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
65}
66
NeilBrown32a76272005-06-21 17:17:14 -070067/*
68 * just a placeholder - calls kmalloc for bitmap pages
69 */
70static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
71{
72 unsigned char *page;
73
Olaf Hering44456d32005-07-27 11:45:17 -070074#ifdef INJECT_FAULTS_1
NeilBrown32a76272005-06-21 17:17:14 -070075 page = NULL;
76#else
NeilBrownac2f40b2010-06-01 19:37:31 +100077 page = kzalloc(PAGE_SIZE, GFP_NOIO);
NeilBrown32a76272005-06-21 17:17:14 -070078#endif
79 if (!page)
80 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
81 else
NeilBrowna654b9d82005-06-21 17:17:27 -070082 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
NeilBrown32a76272005-06-21 17:17:14 -070083 bmname(bitmap), page);
84 return page;
85}
86
87/*
88 * for now just a placeholder -- just calls kfree for bitmap pages
89 */
90static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
91{
92 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
93 kfree(page);
94}
95
96/*
97 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
98 *
99 * 1) check to see if this page is allocated, if it's not then try to alloc
100 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
101 * page pointer directly as a counter
102 *
103 * if we find our page, we increment the page's refcount so that it stays
104 * allocated while we're using it
105 */
NeilBrownac2f40b2010-06-01 19:37:31 +1000106static int bitmap_checkpage(struct bitmap *bitmap,
107 unsigned long page, int create)
NeilBrownee305ac2009-09-23 18:06:44 +1000108__releases(bitmap->lock)
109__acquires(bitmap->lock)
NeilBrown32a76272005-06-21 17:17:14 -0700110{
111 unsigned char *mappage;
112
113 if (page >= bitmap->pages) {
NeilBrown1187cf02009-03-31 14:27:02 +1100114 /* This can happen if bitmap_start_sync goes beyond
115 * End-of-device while looking for a whole page.
116 * It is harmless.
117 */
NeilBrown32a76272005-06-21 17:17:14 -0700118 return -EINVAL;
119 }
120
NeilBrown32a76272005-06-21 17:17:14 -0700121 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
122 return 0;
123
124 if (bitmap->bp[page].map) /* page is already allocated, just return */
125 return 0;
126
127 if (!create)
128 return -ENOENT;
129
NeilBrown32a76272005-06-21 17:17:14 -0700130 /* this page has not been allocated yet */
131
NeilBrownac2f40b2010-06-01 19:37:31 +1000132 spin_unlock_irq(&bitmap->lock);
133 mappage = bitmap_alloc_page(bitmap);
134 spin_lock_irq(&bitmap->lock);
135
136 if (mappage == NULL) {
NeilBrown32a76272005-06-21 17:17:14 -0700137 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
138 bmname(bitmap));
139 /* failed - set the hijacked flag so that we can use the
140 * pointer as a counter */
NeilBrown32a76272005-06-21 17:17:14 -0700141 if (!bitmap->bp[page].map)
142 bitmap->bp[page].hijacked = 1;
NeilBrownac2f40b2010-06-01 19:37:31 +1000143 } else if (bitmap->bp[page].map ||
144 bitmap->bp[page].hijacked) {
NeilBrown32a76272005-06-21 17:17:14 -0700145 /* somebody beat us to getting the page */
146 bitmap_free_page(bitmap, mappage);
147 return 0;
NeilBrownac2f40b2010-06-01 19:37:31 +1000148 } else {
149
150 /* no page was in place and we have one, so install it */
151
152 bitmap->bp[page].map = mappage;
153 bitmap->missing_pages--;
NeilBrown32a76272005-06-21 17:17:14 -0700154 }
NeilBrown32a76272005-06-21 17:17:14 -0700155 return 0;
156}
157
NeilBrown32a76272005-06-21 17:17:14 -0700158/* if page is completely empty, put it back on the free list, or dealloc it */
159/* if page was hijacked, unmark the flag so it might get alloced next time */
160/* Note: lock should be held when calling this */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800161static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
NeilBrown32a76272005-06-21 17:17:14 -0700162{
163 char *ptr;
164
165 if (bitmap->bp[page].count) /* page is still busy */
166 return;
167
168 /* page is no longer in use, it can be released */
169
170 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
171 bitmap->bp[page].hijacked = 0;
172 bitmap->bp[page].map = NULL;
NeilBrownac2f40b2010-06-01 19:37:31 +1000173 } else {
174 /* normal case, free the page */
175 ptr = bitmap->bp[page].map;
176 bitmap->bp[page].map = NULL;
177 bitmap->missing_pages++;
178 bitmap_free_page(bitmap, ptr);
NeilBrown32a76272005-06-21 17:17:14 -0700179 }
NeilBrown32a76272005-06-21 17:17:14 -0700180}
181
NeilBrown32a76272005-06-21 17:17:14 -0700182/*
183 * bitmap file handling - read and write the bitmap file and its superblock
184 */
185
NeilBrown32a76272005-06-21 17:17:14 -0700186/*
187 * basic page I/O operations
188 */
189
NeilBrowna654b9d82005-06-21 17:17:27 -0700190/* IO operations when bitmap is stored near all superblocks */
NeilBrownf6af9492009-12-14 12:49:54 +1100191static struct page *read_sb_page(mddev_t *mddev, loff_t offset,
NeilBrowna2ed9612008-12-19 16:25:01 +1100192 struct page *page,
193 unsigned long index, int size)
NeilBrowna654b9d82005-06-21 17:17:27 -0700194{
195 /* choose a good rdev and read the page from there */
196
197 mdk_rdev_t *rdev;
NeilBrowna654b9d82005-06-21 17:17:27 -0700198 sector_t target;
NeilBrownac2f40b2010-06-01 19:37:31 +1000199 int did_alloc = 0;
NeilBrowna654b9d82005-06-21 17:17:27 -0700200
NeilBrownac2f40b2010-06-01 19:37:31 +1000201 if (!page) {
NeilBrowna2ed9612008-12-19 16:25:01 +1100202 page = alloc_page(GFP_KERNEL);
NeilBrownac2f40b2010-06-01 19:37:31 +1000203 if (!page)
204 return ERR_PTR(-ENOMEM);
205 did_alloc = 1;
206 }
NeilBrowna654b9d82005-06-21 17:17:27 -0700207
Cheng Renquan159ec1f2009-01-09 08:31:08 +1100208 list_for_each_entry(rdev, &mddev->disks, same_set) {
NeilBrownb2d444d2005-11-08 21:39:31 -0800209 if (! test_bit(In_sync, &rdev->flags)
210 || test_bit(Faulty, &rdev->flags))
NeilBrownab904d62005-09-09 16:23:52 -0700211 continue;
212
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100213 target = offset + index * (PAGE_SIZE/512);
NeilBrowna654b9d82005-06-21 17:17:27 -0700214
NeilBrown2b193362010-10-27 15:16:40 +1100215 if (sync_page_io(rdev, target,
Martin K. Petersene1defc42009-05-22 17:17:49 -0400216 roundup(size, bdev_logical_block_size(rdev->bdev)),
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100217 page, READ, true)) {
NeilBrownab904d62005-09-09 16:23:52 -0700218 page->index = index;
NeilBrownce25c312006-06-26 00:27:49 -0700219 attach_page_buffers(page, NULL); /* so that free_buffer will
220 * quietly no-op */
NeilBrownab904d62005-09-09 16:23:52 -0700221 return page;
222 }
223 }
NeilBrownac2f40b2010-06-01 19:37:31 +1000224 if (did_alloc)
225 put_page(page);
NeilBrownab904d62005-09-09 16:23:52 -0700226 return ERR_PTR(-EIO);
NeilBrowna654b9d82005-06-21 17:17:27 -0700227
NeilBrowna654b9d82005-06-21 17:17:27 -0700228}
229
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000230static mdk_rdev_t *next_active_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
231{
232 /* Iterate the disks of an mddev, using rcu to protect access to the
233 * linked list, and raising the refcount of devices we return to ensure
234 * they don't disappear while in use.
235 * As devices are only added or removed when raid_disk is < 0 and
236 * nr_pending is 0 and In_sync is clear, the entries we return will
237 * still be in the same position on the list when we re-enter
238 * list_for_each_continue_rcu.
239 */
240 struct list_head *pos;
241 rcu_read_lock();
242 if (rdev == NULL)
243 /* start at the beginning */
244 pos = &mddev->disks;
245 else {
246 /* release the previous rdev and start from there. */
247 rdev_dec_pending(rdev, mddev);
248 pos = &rdev->same_set;
249 }
250 list_for_each_continue_rcu(pos, &mddev->disks) {
251 rdev = list_entry(pos, mdk_rdev_t, same_set);
252 if (rdev->raid_disk >= 0 &&
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000253 !test_bit(Faulty, &rdev->flags)) {
254 /* this is a usable devices */
255 atomic_inc(&rdev->nr_pending);
256 rcu_read_unlock();
257 return rdev;
258 }
259 }
260 rcu_read_unlock();
261 return NULL;
262}
263
NeilBrownab6085c2007-05-23 13:58:10 -0700264static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
NeilBrowna654b9d82005-06-21 17:17:27 -0700265{
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000266 mdk_rdev_t *rdev = NULL;
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100267 struct block_device *bdev;
NeilBrownab6085c2007-05-23 13:58:10 -0700268 mddev_t *mddev = bitmap->mddev;
NeilBrowna654b9d82005-06-21 17:17:27 -0700269
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000270 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
NeilBrownac2f40b2010-06-01 19:37:31 +1000271 int size = PAGE_SIZE;
272 loff_t offset = mddev->bitmap_info.offset;
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100273
274 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
275
NeilBrownac2f40b2010-06-01 19:37:31 +1000276 if (page->index == bitmap->file_pages-1)
277 size = roundup(bitmap->last_page_size,
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100278 bdev_logical_block_size(bdev));
NeilBrownac2f40b2010-06-01 19:37:31 +1000279 /* Just make sure we aren't corrupting data or
280 * metadata
281 */
282 if (mddev->external) {
283 /* Bitmap could be anywhere. */
284 if (rdev->sb_start + offset + (page->index
285 * (PAGE_SIZE/512))
286 > rdev->data_offset
287 &&
288 rdev->sb_start + offset
289 < (rdev->data_offset + mddev->dev_sectors
290 + (PAGE_SIZE/512)))
291 goto bad_alignment;
292 } else if (offset < 0) {
293 /* DATA BITMAP METADATA */
294 if (offset
295 + (long)(page->index * (PAGE_SIZE/512))
296 + size/512 > 0)
297 /* bitmap runs in to metadata */
298 goto bad_alignment;
299 if (rdev->data_offset + mddev->dev_sectors
300 > rdev->sb_start + offset)
301 /* data runs in to bitmap */
302 goto bad_alignment;
303 } else if (rdev->sb_start < rdev->data_offset) {
304 /* METADATA BITMAP DATA */
305 if (rdev->sb_start
306 + offset
307 + page->index*(PAGE_SIZE/512) + size/512
308 > rdev->data_offset)
309 /* bitmap runs in to data */
310 goto bad_alignment;
311 } else {
312 /* DATA METADATA BITMAP - no problems */
313 }
314 md_super_write(mddev, rdev,
315 rdev->sb_start + offset
316 + page->index * (PAGE_SIZE/512),
317 size,
318 page);
NeilBrownb2d2c4c2008-09-01 12:48:13 +1000319 }
NeilBrowna654b9d82005-06-21 17:17:27 -0700320
321 if (wait)
NeilBrowna9701a32005-11-08 21:39:34 -0800322 md_super_wait(mddev);
NeilBrowna654b9d82005-06-21 17:17:27 -0700323 return 0;
NeilBrown4b809912008-07-21 17:05:25 +1000324
325 bad_alignment:
NeilBrown4b809912008-07-21 17:05:25 +1000326 return -EINVAL;
NeilBrowna654b9d82005-06-21 17:17:27 -0700327}
328
NeilBrown4ad13662007-07-17 04:06:13 -0700329static void bitmap_file_kick(struct bitmap *bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700330/*
NeilBrowna654b9d82005-06-21 17:17:27 -0700331 * write out a page to a file
NeilBrown32a76272005-06-21 17:17:14 -0700332 */
NeilBrown4ad13662007-07-17 04:06:13 -0700333static void write_page(struct bitmap *bitmap, struct page *page, int wait)
NeilBrown32a76272005-06-21 17:17:14 -0700334{
NeilBrownd785a062006-06-26 00:27:48 -0700335 struct buffer_head *bh;
NeilBrown32a76272005-06-21 17:17:14 -0700336
NeilBrownf0d76d72007-07-17 04:06:12 -0700337 if (bitmap->file == NULL) {
338 switch (write_sb_page(bitmap, page, wait)) {
339 case -EINVAL:
340 bitmap->flags |= BITMAP_WRITE_ERROR;
NeilBrownf0d76d72007-07-17 04:06:12 -0700341 }
NeilBrown4ad13662007-07-17 04:06:13 -0700342 } else {
NeilBrowna654b9d82005-06-21 17:17:27 -0700343
NeilBrown4ad13662007-07-17 04:06:13 -0700344 bh = page_buffers(page);
NeilBrownc7084432006-01-06 00:20:45 -0800345
NeilBrown4ad13662007-07-17 04:06:13 -0700346 while (bh && bh->b_blocknr) {
347 atomic_inc(&bitmap->pending_writes);
348 set_buffer_locked(bh);
349 set_buffer_mapped(bh);
Jens Axboe721a9602011-03-09 11:56:30 +0100350 submit_bh(WRITE | REQ_SYNC, bh);
NeilBrown4ad13662007-07-17 04:06:13 -0700351 bh = bh->b_this_page;
352 }
NeilBrown32a76272005-06-21 17:17:14 -0700353
NeilBrownac2f40b2010-06-01 19:37:31 +1000354 if (wait)
NeilBrown4ad13662007-07-17 04:06:13 -0700355 wait_event(bitmap->write_wait,
356 atomic_read(&bitmap->pending_writes)==0);
NeilBrown32a76272005-06-21 17:17:14 -0700357 }
NeilBrown4ad13662007-07-17 04:06:13 -0700358 if (bitmap->flags & BITMAP_WRITE_ERROR)
359 bitmap_file_kick(bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700360}
361
NeilBrownd785a062006-06-26 00:27:48 -0700362static void end_bitmap_write(struct buffer_head *bh, int uptodate)
NeilBrown32a76272005-06-21 17:17:14 -0700363{
NeilBrownd785a062006-06-26 00:27:48 -0700364 struct bitmap *bitmap = bh->b_private;
365 unsigned long flags;
366
367 if (!uptodate) {
368 spin_lock_irqsave(&bitmap->lock, flags);
369 bitmap->flags |= BITMAP_WRITE_ERROR;
370 spin_unlock_irqrestore(&bitmap->lock, flags);
371 }
372 if (atomic_dec_and_test(&bitmap->pending_writes))
373 wake_up(&bitmap->write_wait);
374}
375
376/* copied from buffer.c */
377static void
378__clear_page_buffers(struct page *page)
379{
380 ClearPagePrivate(page);
381 set_page_private(page, 0);
382 page_cache_release(page);
383}
384static void free_buffers(struct page *page)
385{
386 struct buffer_head *bh = page_buffers(page);
387
388 while (bh) {
389 struct buffer_head *next = bh->b_this_page;
390 free_buffer_head(bh);
391 bh = next;
392 }
393 __clear_page_buffers(page);
394 put_page(page);
395}
396
397/* read a page from a file.
398 * We both read the page, and attach buffers to the page to record the
399 * address of each block (using bmap). These addresses will be used
400 * to write the block later, completely bypassing the filesystem.
401 * This usage is similar to how swap files are handled, and allows us
402 * to write to a file with no concerns of memory allocation failing.
403 */
404static struct page *read_page(struct file *file, unsigned long index,
405 struct bitmap *bitmap,
406 unsigned long count)
407{
NeilBrown32a76272005-06-21 17:17:14 -0700408 struct page *page = NULL;
Josef Sipekc649bb92006-12-08 02:37:19 -0800409 struct inode *inode = file->f_path.dentry->d_inode;
NeilBrownd785a062006-06-26 00:27:48 -0700410 struct buffer_head *bh;
411 sector_t block;
NeilBrown32a76272005-06-21 17:17:14 -0700412
NeilBrownac2f40b2010-06-01 19:37:31 +1000413 PRINTK("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
NeilBrown2d1f3b52006-01-06 00:20:31 -0800414 (unsigned long long)index << PAGE_SHIFT);
NeilBrown32a76272005-06-21 17:17:14 -0700415
NeilBrownd785a062006-06-26 00:27:48 -0700416 page = alloc_page(GFP_KERNEL);
417 if (!page)
418 page = ERR_PTR(-ENOMEM);
NeilBrown32a76272005-06-21 17:17:14 -0700419 if (IS_ERR(page))
420 goto out;
NeilBrownd785a062006-06-26 00:27:48 -0700421
NeilBrownd785a062006-06-26 00:27:48 -0700422 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
423 if (!bh) {
424 put_page(page);
425 page = ERR_PTR(-ENOMEM);
426 goto out;
427 }
428 attach_page_buffers(page, bh);
429 block = index << (PAGE_SHIFT - inode->i_blkbits);
430 while (bh) {
431 if (count == 0)
432 bh->b_blocknr = 0;
433 else {
434 bh->b_blocknr = bmap(inode, block);
435 if (bh->b_blocknr == 0) {
436 /* Cannot use this file! */
437 free_buffers(page);
438 page = ERR_PTR(-EINVAL);
439 goto out;
440 }
441 bh->b_bdev = inode->i_sb->s_bdev;
442 if (count < (1<<inode->i_blkbits))
443 count = 0;
444 else
445 count -= (1<<inode->i_blkbits);
NeilBrown32a76272005-06-21 17:17:14 -0700446
NeilBrownd785a062006-06-26 00:27:48 -0700447 bh->b_end_io = end_bitmap_write;
448 bh->b_private = bitmap;
NeilBrownce25c312006-06-26 00:27:49 -0700449 atomic_inc(&bitmap->pending_writes);
450 set_buffer_locked(bh);
451 set_buffer_mapped(bh);
452 submit_bh(READ, bh);
NeilBrownd785a062006-06-26 00:27:48 -0700453 }
454 block++;
455 bh = bh->b_this_page;
456 }
NeilBrownd785a062006-06-26 00:27:48 -0700457 page->index = index;
NeilBrownce25c312006-06-26 00:27:49 -0700458
459 wait_event(bitmap->write_wait,
460 atomic_read(&bitmap->pending_writes)==0);
461 if (bitmap->flags & BITMAP_WRITE_ERROR) {
462 free_buffers(page);
463 page = ERR_PTR(-EIO);
464 }
NeilBrown32a76272005-06-21 17:17:14 -0700465out:
466 if (IS_ERR(page))
NeilBrownac2f40b2010-06-01 19:37:31 +1000467 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %ld\n",
NeilBrown2d1f3b52006-01-06 00:20:31 -0800468 (int)PAGE_SIZE,
469 (unsigned long long)index << PAGE_SHIFT,
NeilBrown32a76272005-06-21 17:17:14 -0700470 PTR_ERR(page));
471 return page;
472}
473
474/*
475 * bitmap file superblock operations
476 */
477
478/* update the event counter and sync the superblock to disk */
NeilBrown4ad13662007-07-17 04:06:13 -0700479void bitmap_update_sb(struct bitmap *bitmap)
NeilBrown32a76272005-06-21 17:17:14 -0700480{
481 bitmap_super_t *sb;
482 unsigned long flags;
483
484 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
NeilBrown4ad13662007-07-17 04:06:13 -0700485 return;
NeilBrownece5cff2009-12-14 12:49:56 +1100486 if (bitmap->mddev->bitmap_info.external)
487 return;
NeilBrown32a76272005-06-21 17:17:14 -0700488 spin_lock_irqsave(&bitmap->lock, flags);
489 if (!bitmap->sb_page) { /* no superblock */
490 spin_unlock_irqrestore(&bitmap->lock, flags);
NeilBrown4ad13662007-07-17 04:06:13 -0700491 return;
NeilBrown32a76272005-06-21 17:17:14 -0700492 }
NeilBrown32a76272005-06-21 17:17:14 -0700493 spin_unlock_irqrestore(&bitmap->lock, flags);
H Hartley Sweeten7b928132010-03-08 16:02:40 +1100494 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
NeilBrown32a76272005-06-21 17:17:14 -0700495 sb->events = cpu_to_le64(bitmap->mddev->events);
NeilBrown8258c532011-05-11 14:26:30 +1000496 if (bitmap->mddev->events < bitmap->events_cleared)
Neil Browna0da84f2008-06-28 08:31:22 +1000497 /* rocking back to read-only */
498 bitmap->events_cleared = bitmap->mddev->events;
NeilBrown8258c532011-05-11 14:26:30 +1000499 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
500 sb->state = cpu_to_le32(bitmap->flags);
NeilBrown43a70502009-12-14 12:49:55 +1100501 /* Just in case these have been changed via sysfs: */
502 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
503 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
NeilBrownea03aff2006-01-06 00:20:34 -0800504 kunmap_atomic(sb, KM_USER0);
NeilBrown4ad13662007-07-17 04:06:13 -0700505 write_page(bitmap, bitmap->sb_page, 1);
NeilBrown32a76272005-06-21 17:17:14 -0700506}
507
508/* print out the bitmap file superblock */
509void bitmap_print_sb(struct bitmap *bitmap)
510{
511 bitmap_super_t *sb;
512
513 if (!bitmap || !bitmap->sb_page)
514 return;
H Hartley Sweeten7b928132010-03-08 16:02:40 +1100515 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
NeilBrown32a76272005-06-21 17:17:14 -0700516 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
NeilBrowna2cff262005-06-21 17:17:20 -0700517 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
518 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
519 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
NeilBrown32a76272005-06-21 17:17:14 -0700520 *(__u32 *)(sb->uuid+0),
521 *(__u32 *)(sb->uuid+4),
522 *(__u32 *)(sb->uuid+8),
523 *(__u32 *)(sb->uuid+12));
NeilBrowna2cff262005-06-21 17:17:20 -0700524 printk(KERN_DEBUG " events: %llu\n",
NeilBrown32a76272005-06-21 17:17:14 -0700525 (unsigned long long) le64_to_cpu(sb->events));
NeilBrowna2cff262005-06-21 17:17:20 -0700526 printk(KERN_DEBUG "events cleared: %llu\n",
NeilBrown32a76272005-06-21 17:17:14 -0700527 (unsigned long long) le64_to_cpu(sb->events_cleared));
NeilBrowna2cff262005-06-21 17:17:20 -0700528 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
529 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
530 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
531 printk(KERN_DEBUG " sync size: %llu KB\n",
532 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
NeilBrown4b6d2872005-09-09 16:23:47 -0700533 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
NeilBrownea03aff2006-01-06 00:20:34 -0800534 kunmap_atomic(sb, KM_USER0);
NeilBrown32a76272005-06-21 17:17:14 -0700535}
536
Jonathan Brassow9c810752011-06-08 17:59:30 -0500537/*
538 * bitmap_new_disk_sb
539 * @bitmap
540 *
541 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
542 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
543 * This function verifies 'bitmap_info' and populates the on-disk bitmap
544 * structure, which is to be written to disk.
545 *
546 * Returns: 0 on success, -Exxx on error
547 */
548static int bitmap_new_disk_sb(struct bitmap *bitmap)
549{
550 bitmap_super_t *sb;
551 unsigned long chunksize, daemon_sleep, write_behind;
552 int err = -EINVAL;
553
554 bitmap->sb_page = alloc_page(GFP_KERNEL);
555 if (IS_ERR(bitmap->sb_page)) {
556 err = PTR_ERR(bitmap->sb_page);
557 bitmap->sb_page = NULL;
558 return err;
559 }
560 bitmap->sb_page->index = 0;
561
562 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
563
564 sb->magic = cpu_to_le32(BITMAP_MAGIC);
565 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
566
567 chunksize = bitmap->mddev->bitmap_info.chunksize;
568 BUG_ON(!chunksize);
569 if (!is_power_of_2(chunksize)) {
570 kunmap_atomic(sb, KM_USER0);
571 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
572 return -EINVAL;
573 }
574 sb->chunksize = cpu_to_le32(chunksize);
575
576 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
577 if (!daemon_sleep ||
578 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
579 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
580 daemon_sleep = 5 * HZ;
581 }
582 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
583 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
584
585 /*
586 * FIXME: write_behind for RAID1. If not specified, what
587 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
588 */
589 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
590 if (write_behind > COUNTER_MAX)
591 write_behind = COUNTER_MAX / 2;
592 sb->write_behind = cpu_to_le32(write_behind);
593 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
594
595 /* keep the array size field of the bitmap superblock up to date */
596 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
597
598 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
599
600 bitmap->flags |= BITMAP_STALE;
601 sb->state |= cpu_to_le32(BITMAP_STALE);
602 bitmap->events_cleared = bitmap->mddev->events;
603 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
604
605 bitmap->flags |= BITMAP_HOSTENDIAN;
606 sb->version = cpu_to_le32(BITMAP_MAJOR_HOSTENDIAN);
607
608 kunmap_atomic(sb, KM_USER0);
609
610 return 0;
611}
612
NeilBrown32a76272005-06-21 17:17:14 -0700613/* read the superblock from the bitmap file and initialize some bitmap fields */
614static int bitmap_read_sb(struct bitmap *bitmap)
615{
616 char *reason = NULL;
617 bitmap_super_t *sb;
NeilBrown4b6d2872005-09-09 16:23:47 -0700618 unsigned long chunksize, daemon_sleep, write_behind;
NeilBrown32a76272005-06-21 17:17:14 -0700619 unsigned long long events;
620 int err = -EINVAL;
621
622 /* page 0 is the superblock, read it... */
NeilBrownf49d5e62007-01-26 00:57:03 -0800623 if (bitmap->file) {
624 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
625 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
626
627 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
628 } else {
NeilBrown42a04b52009-12-14 12:49:53 +1100629 bitmap->sb_page = read_sb_page(bitmap->mddev,
630 bitmap->mddev->bitmap_info.offset,
NeilBrowna2ed9612008-12-19 16:25:01 +1100631 NULL,
632 0, sizeof(bitmap_super_t));
NeilBrowna654b9d82005-06-21 17:17:27 -0700633 }
NeilBrown32a76272005-06-21 17:17:14 -0700634 if (IS_ERR(bitmap->sb_page)) {
635 err = PTR_ERR(bitmap->sb_page);
636 bitmap->sb_page = NULL;
637 return err;
638 }
639
H Hartley Sweeten7b928132010-03-08 16:02:40 +1100640 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
NeilBrown32a76272005-06-21 17:17:14 -0700641
NeilBrown32a76272005-06-21 17:17:14 -0700642 chunksize = le32_to_cpu(sb->chunksize);
NeilBrown1b04be92009-12-14 12:49:53 +1100643 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
NeilBrown4b6d2872005-09-09 16:23:47 -0700644 write_behind = le32_to_cpu(sb->write_behind);
NeilBrown32a76272005-06-21 17:17:14 -0700645
646 /* verify that the bitmap-specific fields are valid */
647 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
648 reason = "bad magic";
NeilBrownbd926c62005-11-08 21:39:32 -0800649 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
650 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
NeilBrown32a76272005-06-21 17:17:14 -0700651 reason = "unrecognized superblock version";
NeilBrown1187cf02009-03-31 14:27:02 +1100652 else if (chunksize < 512)
NeilBrown7dd5d342006-01-06 00:20:39 -0800653 reason = "bitmap chunksize too small";
Jonathan Brassowd7445402011-06-08 18:01:10 -0500654 else if (!is_power_of_2(chunksize))
NeilBrown32a76272005-06-21 17:17:14 -0700655 reason = "bitmap chunksize not a power of 2";
NeilBrown1b04be92009-12-14 12:49:53 +1100656 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
NeilBrown7dd5d342006-01-06 00:20:39 -0800657 reason = "daemon sleep period out of range";
NeilBrown4b6d2872005-09-09 16:23:47 -0700658 else if (write_behind > COUNTER_MAX)
659 reason = "write-behind limit out of range (0 - 16383)";
NeilBrown32a76272005-06-21 17:17:14 -0700660 if (reason) {
661 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
662 bmname(bitmap), reason);
663 goto out;
664 }
665
666 /* keep the array size field of the bitmap superblock up to date */
667 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
668
669 if (!bitmap->mddev->persistent)
670 goto success;
671
672 /*
673 * if we have a persistent array superblock, compare the
674 * bitmap's UUID and event counter to the mddev's
675 */
676 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
677 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
678 bmname(bitmap));
679 goto out;
680 }
681 events = le64_to_cpu(sb->events);
682 if (events < bitmap->mddev->events) {
683 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
684 "-- forcing full recovery\n", bmname(bitmap), events,
685 (unsigned long long) bitmap->mddev->events);
NeilBrown4f2e6392006-10-21 10:24:09 -0700686 sb->state |= cpu_to_le32(BITMAP_STALE);
NeilBrown32a76272005-06-21 17:17:14 -0700687 }
688success:
689 /* assign fields using values from superblock */
NeilBrown42a04b52009-12-14 12:49:53 +1100690 bitmap->mddev->bitmap_info.chunksize = chunksize;
691 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
NeilBrown42a04b52009-12-14 12:49:53 +1100692 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
NeilBrown4f2e6392006-10-21 10:24:09 -0700693 bitmap->flags |= le32_to_cpu(sb->state);
NeilBrownbd926c62005-11-08 21:39:32 -0800694 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
695 bitmap->flags |= BITMAP_HOSTENDIAN;
NeilBrown32a76272005-06-21 17:17:14 -0700696 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
NeilBrown8258c532011-05-11 14:26:30 +1000697 if (bitmap->flags & BITMAP_STALE)
NeilBrown6a079972005-09-09 16:23:44 -0700698 bitmap->events_cleared = bitmap->mddev->events;
NeilBrown32a76272005-06-21 17:17:14 -0700699 err = 0;
700out:
NeilBrownea03aff2006-01-06 00:20:34 -0800701 kunmap_atomic(sb, KM_USER0);
NeilBrown32a76272005-06-21 17:17:14 -0700702 if (err)
703 bitmap_print_sb(bitmap);
704 return err;
705}
706
707enum bitmap_mask_op {
708 MASK_SET,
709 MASK_UNSET
710};
711
NeilBrown4ad13662007-07-17 04:06:13 -0700712/* record the state of the bitmap in the superblock. Return the old value */
713static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
714 enum bitmap_mask_op op)
NeilBrown32a76272005-06-21 17:17:14 -0700715{
716 bitmap_super_t *sb;
717 unsigned long flags;
NeilBrown4ad13662007-07-17 04:06:13 -0700718 int old;
NeilBrown32a76272005-06-21 17:17:14 -0700719
720 spin_lock_irqsave(&bitmap->lock, flags);
Adrian Bunk7e317652006-03-25 03:07:51 -0800721 if (!bitmap->sb_page) { /* can't set the state */
NeilBrown32a76272005-06-21 17:17:14 -0700722 spin_unlock_irqrestore(&bitmap->lock, flags);
NeilBrown4ad13662007-07-17 04:06:13 -0700723 return 0;
NeilBrown32a76272005-06-21 17:17:14 -0700724 }
NeilBrown32a76272005-06-21 17:17:14 -0700725 spin_unlock_irqrestore(&bitmap->lock, flags);
H Hartley Sweeten7b928132010-03-08 16:02:40 +1100726 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
NeilBrown4ad13662007-07-17 04:06:13 -0700727 old = le32_to_cpu(sb->state) & bits;
NeilBrown32a76272005-06-21 17:17:14 -0700728 switch (op) {
NeilBrownac2f40b2010-06-01 19:37:31 +1000729 case MASK_SET:
730 sb->state |= cpu_to_le32(bits);
NeilBrown8258c532011-05-11 14:26:30 +1000731 bitmap->flags |= bits;
NeilBrownac2f40b2010-06-01 19:37:31 +1000732 break;
733 case MASK_UNSET:
734 sb->state &= cpu_to_le32(~bits);
NeilBrown8258c532011-05-11 14:26:30 +1000735 bitmap->flags &= ~bits;
NeilBrownac2f40b2010-06-01 19:37:31 +1000736 break;
737 default:
738 BUG();
NeilBrown32a76272005-06-21 17:17:14 -0700739 }
NeilBrownea03aff2006-01-06 00:20:34 -0800740 kunmap_atomic(sb, KM_USER0);
NeilBrown4ad13662007-07-17 04:06:13 -0700741 return old;
NeilBrown32a76272005-06-21 17:17:14 -0700742}
743
744/*
745 * general bitmap file operations
746 */
747
NeilBrownece5cff2009-12-14 12:49:56 +1100748/*
749 * on-disk bitmap:
750 *
751 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
752 * file a page at a time. There's a superblock at the start of the file.
753 */
NeilBrown32a76272005-06-21 17:17:14 -0700754/* calculate the index of the page that contains this bit */
NeilBrownece5cff2009-12-14 12:49:56 +1100755static inline unsigned long file_page_index(struct bitmap *bitmap, unsigned long chunk)
NeilBrown32a76272005-06-21 17:17:14 -0700756{
NeilBrownece5cff2009-12-14 12:49:56 +1100757 if (!bitmap->mddev->bitmap_info.external)
758 chunk += sizeof(bitmap_super_t) << 3;
759 return chunk >> PAGE_BIT_SHIFT;
NeilBrown32a76272005-06-21 17:17:14 -0700760}
761
762/* calculate the (bit) offset of this bit within a page */
NeilBrownece5cff2009-12-14 12:49:56 +1100763static inline unsigned long file_page_offset(struct bitmap *bitmap, unsigned long chunk)
NeilBrown32a76272005-06-21 17:17:14 -0700764{
NeilBrownece5cff2009-12-14 12:49:56 +1100765 if (!bitmap->mddev->bitmap_info.external)
766 chunk += sizeof(bitmap_super_t) << 3;
767 return chunk & (PAGE_BITS - 1);
NeilBrown32a76272005-06-21 17:17:14 -0700768}
769
770/*
771 * return a pointer to the page in the filemap that contains the given bit
772 *
773 * this lookup is complicated by the fact that the bitmap sb might be exactly
774 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
775 * 0 or page 1
776 */
777static inline struct page *filemap_get_page(struct bitmap *bitmap,
778 unsigned long chunk)
779{
NeilBrowne384e582010-06-01 19:37:34 +1000780 if (bitmap->filemap == NULL)
781 return NULL;
NeilBrownac2f40b2010-06-01 19:37:31 +1000782 if (file_page_index(bitmap, chunk) >= bitmap->file_pages)
783 return NULL;
NeilBrownece5cff2009-12-14 12:49:56 +1100784 return bitmap->filemap[file_page_index(bitmap, chunk)
785 - file_page_index(bitmap, 0)];
NeilBrown32a76272005-06-21 17:17:14 -0700786}
787
NeilBrown32a76272005-06-21 17:17:14 -0700788static void bitmap_file_unmap(struct bitmap *bitmap)
789{
790 struct page **map, *sb_page;
791 unsigned long *attr;
792 int pages;
793 unsigned long flags;
794
795 spin_lock_irqsave(&bitmap->lock, flags);
796 map = bitmap->filemap;
797 bitmap->filemap = NULL;
798 attr = bitmap->filemap_attr;
799 bitmap->filemap_attr = NULL;
800 pages = bitmap->file_pages;
801 bitmap->file_pages = 0;
802 sb_page = bitmap->sb_page;
803 bitmap->sb_page = NULL;
804 spin_unlock_irqrestore(&bitmap->lock, flags);
805
806 while (pages--)
NeilBrownece5cff2009-12-14 12:49:56 +1100807 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
NeilBrownd785a062006-06-26 00:27:48 -0700808 free_buffers(map[pages]);
NeilBrown32a76272005-06-21 17:17:14 -0700809 kfree(map);
810 kfree(attr);
811
NeilBrownd785a062006-06-26 00:27:48 -0700812 if (sb_page)
813 free_buffers(sb_page);
NeilBrown32a76272005-06-21 17:17:14 -0700814}
815
816static void bitmap_file_put(struct bitmap *bitmap)
817{
818 struct file *file;
NeilBrown32a76272005-06-21 17:17:14 -0700819 unsigned long flags;
820
821 spin_lock_irqsave(&bitmap->lock, flags);
822 file = bitmap->file;
823 bitmap->file = NULL;
824 spin_unlock_irqrestore(&bitmap->lock, flags);
825
NeilBrownd785a062006-06-26 00:27:48 -0700826 if (file)
827 wait_event(bitmap->write_wait,
828 atomic_read(&bitmap->pending_writes)==0);
NeilBrown32a76272005-06-21 17:17:14 -0700829 bitmap_file_unmap(bitmap);
830
NeilBrownd785a062006-06-26 00:27:48 -0700831 if (file) {
Josef Sipekc649bb92006-12-08 02:37:19 -0800832 struct inode *inode = file->f_path.dentry->d_inode;
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800833 invalidate_mapping_pages(inode->i_mapping, 0, -1);
NeilBrown32a76272005-06-21 17:17:14 -0700834 fput(file);
NeilBrownd785a062006-06-26 00:27:48 -0700835 }
NeilBrown32a76272005-06-21 17:17:14 -0700836}
837
NeilBrown32a76272005-06-21 17:17:14 -0700838/*
839 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
840 * then it is no longer reliable, so we stop using it and we mark the file
841 * as failed in the superblock
842 */
843static void bitmap_file_kick(struct bitmap *bitmap)
844{
845 char *path, *ptr = NULL;
846
NeilBrown4ad13662007-07-17 04:06:13 -0700847 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
848 bitmap_update_sb(bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700849
NeilBrown4ad13662007-07-17 04:06:13 -0700850 if (bitmap->file) {
851 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
852 if (path)
Christoph Hellwig6bcfd602008-05-23 13:04:34 -0700853 ptr = d_path(&bitmap->file->f_path, path,
854 PAGE_SIZE);
855
NeilBrown4ad13662007-07-17 04:06:13 -0700856 printk(KERN_ALERT
857 "%s: kicking failed bitmap file %s from array!\n",
Christoph Hellwig6bcfd602008-05-23 13:04:34 -0700858 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
NeilBrown32a76272005-06-21 17:17:14 -0700859
NeilBrown4ad13662007-07-17 04:06:13 -0700860 kfree(path);
861 } else
862 printk(KERN_ALERT
863 "%s: disabling internal bitmap due to errors\n",
864 bmname(bitmap));
NeilBrowna654b9d82005-06-21 17:17:27 -0700865 }
NeilBrown32a76272005-06-21 17:17:14 -0700866
867 bitmap_file_put(bitmap);
868
869 return;
870}
871
872enum bitmap_page_attr {
NeilBrownac2f40b2010-06-01 19:37:31 +1000873 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
874 BITMAP_PAGE_CLEAN = 1, /* there are bits that might need to be cleared */
875 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
NeilBrown32a76272005-06-21 17:17:14 -0700876};
877
878static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
879 enum bitmap_page_attr attr)
880{
NeilBrowne384e582010-06-01 19:37:34 +1000881 if (page)
882 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
883 else
884 __set_bit(attr, &bitmap->logattrs);
NeilBrown32a76272005-06-21 17:17:14 -0700885}
886
887static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
888 enum bitmap_page_attr attr)
889{
NeilBrowne384e582010-06-01 19:37:34 +1000890 if (page)
891 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
892 else
893 __clear_bit(attr, &bitmap->logattrs);
NeilBrown32a76272005-06-21 17:17:14 -0700894}
895
NeilBrownec7a3192006-06-26 00:27:45 -0700896static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
897 enum bitmap_page_attr attr)
NeilBrown32a76272005-06-21 17:17:14 -0700898{
NeilBrowne384e582010-06-01 19:37:34 +1000899 if (page)
900 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
901 else
902 return test_bit(attr, &bitmap->logattrs);
NeilBrown32a76272005-06-21 17:17:14 -0700903}
904
905/*
906 * bitmap_file_set_bit -- called before performing a write to the md device
907 * to set (and eventually sync) a particular bit in the bitmap file
908 *
909 * we set the bit immediately, then we record the page number so that
910 * when an unplug occurs, we can flush the dirty pages out to disk
911 */
912static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
913{
914 unsigned long bit;
NeilBrowne384e582010-06-01 19:37:34 +1000915 struct page *page = NULL;
NeilBrown32a76272005-06-21 17:17:14 -0700916 void *kaddr;
917 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
918
NeilBrowne384e582010-06-01 19:37:34 +1000919 if (!bitmap->filemap) {
920 struct dm_dirty_log *log = bitmap->mddev->bitmap_info.log;
921 if (log)
922 log->type->mark_region(log, chunk);
923 } else {
NeilBrown32a76272005-06-21 17:17:14 -0700924
NeilBrowne384e582010-06-01 19:37:34 +1000925 page = filemap_get_page(bitmap, chunk);
926 if (!page)
927 return;
928 bit = file_page_offset(bitmap, chunk);
NeilBrown32a76272005-06-21 17:17:14 -0700929
NeilBrowne384e582010-06-01 19:37:34 +1000930 /* set the bit */
931 kaddr = kmap_atomic(page, KM_USER0);
932 if (bitmap->flags & BITMAP_HOSTENDIAN)
933 set_bit(bit, kaddr);
934 else
Akinobu Mita6b33aff2011-03-23 16:42:13 -0700935 __test_and_set_bit_le(bit, kaddr);
NeilBrowne384e582010-06-01 19:37:34 +1000936 kunmap_atomic(kaddr, KM_USER0);
937 PRINTK("set file bit %lu page %lu\n", bit, page->index);
938 }
NeilBrown32a76272005-06-21 17:17:14 -0700939 /* record page number so it gets flushed to disk when unplug occurs */
940 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
NeilBrown32a76272005-06-21 17:17:14 -0700941}
942
943/* this gets called when the md device is ready to unplug its underlying
944 * (slave) device queues -- before we let any writes go down, we need to
945 * sync the dirty pages of the bitmap file to disk */
NeilBrown4ad13662007-07-17 04:06:13 -0700946void bitmap_unplug(struct bitmap *bitmap)
NeilBrown32a76272005-06-21 17:17:14 -0700947{
NeilBrownec7a3192006-06-26 00:27:45 -0700948 unsigned long i, flags;
949 int dirty, need_write;
NeilBrown32a76272005-06-21 17:17:14 -0700950 struct page *page;
951 int wait = 0;
952
953 if (!bitmap)
NeilBrown4ad13662007-07-17 04:06:13 -0700954 return;
NeilBrowne384e582010-06-01 19:37:34 +1000955 if (!bitmap->filemap) {
956 /* Must be using a dirty_log */
957 struct dm_dirty_log *log = bitmap->mddev->bitmap_info.log;
958 dirty = test_and_clear_bit(BITMAP_PAGE_DIRTY, &bitmap->logattrs);
959 need_write = test_and_clear_bit(BITMAP_PAGE_NEEDWRITE, &bitmap->logattrs);
960 if (dirty || need_write)
961 if (log->type->flush(log))
962 bitmap->flags |= BITMAP_WRITE_ERROR;
963 goto out;
964 }
NeilBrown32a76272005-06-21 17:17:14 -0700965
966 /* look at each page to see if there are any set bits that need to be
967 * flushed out to disk */
968 for (i = 0; i < bitmap->file_pages; i++) {
969 spin_lock_irqsave(&bitmap->lock, flags);
NeilBrowna654b9d82005-06-21 17:17:27 -0700970 if (!bitmap->filemap) {
NeilBrown32a76272005-06-21 17:17:14 -0700971 spin_unlock_irqrestore(&bitmap->lock, flags);
NeilBrown4ad13662007-07-17 04:06:13 -0700972 return;
NeilBrown32a76272005-06-21 17:17:14 -0700973 }
974 page = bitmap->filemap[i];
NeilBrownec7a3192006-06-26 00:27:45 -0700975 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
976 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
NeilBrown32a76272005-06-21 17:17:14 -0700977 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
978 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
NeilBrownec7a3192006-06-26 00:27:45 -0700979 if (dirty)
NeilBrown32a76272005-06-21 17:17:14 -0700980 wait = 1;
981 spin_unlock_irqrestore(&bitmap->lock, flags);
982
NeilBrownac2f40b2010-06-01 19:37:31 +1000983 if (dirty || need_write)
NeilBrown4ad13662007-07-17 04:06:13 -0700984 write_page(bitmap, page, 0);
NeilBrown32a76272005-06-21 17:17:14 -0700985 }
986 if (wait) { /* if any writes were performed, we need to wait on them */
NeilBrown0b79ccf2006-06-26 00:27:44 -0700987 if (bitmap->file)
NeilBrownd785a062006-06-26 00:27:48 -0700988 wait_event(bitmap->write_wait,
989 atomic_read(&bitmap->pending_writes)==0);
NeilBrown0b79ccf2006-06-26 00:27:44 -0700990 else
NeilBrowna9701a32005-11-08 21:39:34 -0800991 md_super_wait(bitmap->mddev);
NeilBrown32a76272005-06-21 17:17:14 -0700992 }
NeilBrowne384e582010-06-01 19:37:34 +1000993out:
NeilBrownd785a062006-06-26 00:27:48 -0700994 if (bitmap->flags & BITMAP_WRITE_ERROR)
995 bitmap_file_kick(bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700996}
NeilBrownac2f40b2010-06-01 19:37:31 +1000997EXPORT_SYMBOL(bitmap_unplug);
NeilBrown32a76272005-06-21 17:17:14 -0700998
NeilBrown6a079972005-09-09 16:23:44 -0700999static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
NeilBrown32a76272005-06-21 17:17:14 -07001000/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1001 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1002 * memory mapping of the bitmap file
1003 * Special cases:
1004 * if there's no bitmap file, or if the bitmap file had been
1005 * previously kicked from the array, we mark all the bits as
1006 * 1's in order to cause a full resync.
NeilBrown6a079972005-09-09 16:23:44 -07001007 *
1008 * We ignore all bits for sectors that end earlier than 'start'.
1009 * This is used when reading an out-of-date bitmap...
NeilBrown32a76272005-06-21 17:17:14 -07001010 */
NeilBrown6a079972005-09-09 16:23:44 -07001011static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
NeilBrown32a76272005-06-21 17:17:14 -07001012{
1013 unsigned long i, chunks, index, oldindex, bit;
1014 struct page *page = NULL, *oldpage = NULL;
1015 unsigned long num_pages, bit_cnt = 0;
1016 struct file *file;
NeilBrownd785a062006-06-26 00:27:48 -07001017 unsigned long bytes, offset;
NeilBrown32a76272005-06-21 17:17:14 -07001018 int outofdate;
1019 int ret = -ENOSPC;
NeilBrownea03aff2006-01-06 00:20:34 -08001020 void *paddr;
NeilBrown32a76272005-06-21 17:17:14 -07001021
1022 chunks = bitmap->chunks;
1023 file = bitmap->file;
1024
NeilBrown42a04b52009-12-14 12:49:53 +11001025 BUG_ON(!file && !bitmap->mddev->bitmap_info.offset);
NeilBrown32a76272005-06-21 17:17:14 -07001026
Olaf Hering44456d32005-07-27 11:45:17 -07001027#ifdef INJECT_FAULTS_3
NeilBrown32a76272005-06-21 17:17:14 -07001028 outofdate = 1;
1029#else
1030 outofdate = bitmap->flags & BITMAP_STALE;
1031#endif
1032 if (outofdate)
1033 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
1034 "recovery\n", bmname(bitmap));
1035
NeilBrowne384e582010-06-01 19:37:34 +10001036 bytes = DIV_ROUND_UP(bitmap->chunks, 8);
NeilBrownece5cff2009-12-14 12:49:56 +11001037 if (!bitmap->mddev->bitmap_info.external)
1038 bytes += sizeof(bitmap_super_t);
NeilBrownbc7f77d2005-06-21 17:17:17 -07001039
NeilBrowne384e582010-06-01 19:37:34 +10001040 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
NeilBrownbc7f77d2005-06-21 17:17:17 -07001041
NeilBrownece5cff2009-12-14 12:49:56 +11001042 if (file && i_size_read(file->f_mapping->host) < bytes) {
NeilBrown32a76272005-06-21 17:17:14 -07001043 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
1044 bmname(bitmap),
1045 (unsigned long) i_size_read(file->f_mapping->host),
NeilBrownece5cff2009-12-14 12:49:56 +11001046 bytes);
NeilBrown4ad13662007-07-17 04:06:13 -07001047 goto err;
NeilBrown32a76272005-06-21 17:17:14 -07001048 }
NeilBrownbc7f77d2005-06-21 17:17:17 -07001049
1050 ret = -ENOMEM;
1051
NeilBrown32a76272005-06-21 17:17:14 -07001052 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
NeilBrownbc7f77d2005-06-21 17:17:17 -07001053 if (!bitmap->filemap)
NeilBrown4ad13662007-07-17 04:06:13 -07001054 goto err;
NeilBrown32a76272005-06-21 17:17:14 -07001055
NeilBrowne16b68b2006-06-26 00:27:45 -07001056 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
1057 bitmap->filemap_attr = kzalloc(
NeilBrownac2f40b2010-06-01 19:37:31 +10001058 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
NeilBrowne16b68b2006-06-26 00:27:45 -07001059 GFP_KERNEL);
NeilBrownbc7f77d2005-06-21 17:17:17 -07001060 if (!bitmap->filemap_attr)
NeilBrown4ad13662007-07-17 04:06:13 -07001061 goto err;
NeilBrown32a76272005-06-21 17:17:14 -07001062
NeilBrown32a76272005-06-21 17:17:14 -07001063 oldindex = ~0L;
1064
1065 for (i = 0; i < chunks; i++) {
NeilBrownbd926c62005-11-08 21:39:32 -08001066 int b;
NeilBrownece5cff2009-12-14 12:49:56 +11001067 index = file_page_index(bitmap, i);
1068 bit = file_page_offset(bitmap, i);
NeilBrown32a76272005-06-21 17:17:14 -07001069 if (index != oldindex) { /* this is a new page, read it in */
NeilBrownd785a062006-06-26 00:27:48 -07001070 int count;
NeilBrown32a76272005-06-21 17:17:14 -07001071 /* unmap the old page, we're done with it */
NeilBrownd785a062006-06-26 00:27:48 -07001072 if (index == num_pages-1)
NeilBrownece5cff2009-12-14 12:49:56 +11001073 count = bytes - index * PAGE_SIZE;
NeilBrownd785a062006-06-26 00:27:48 -07001074 else
1075 count = PAGE_SIZE;
NeilBrownece5cff2009-12-14 12:49:56 +11001076 if (index == 0 && bitmap->sb_page) {
NeilBrown32a76272005-06-21 17:17:14 -07001077 /*
1078 * if we're here then the superblock page
1079 * contains some bits (PAGE_SIZE != sizeof sb)
1080 * we've already read it in, so just use it
1081 */
1082 page = bitmap->sb_page;
1083 offset = sizeof(bitmap_super_t);
NeilBrown53845272009-01-09 08:31:05 +11001084 if (!file)
Vasiliy Kulikov5c04f552010-10-01 14:18:12 -07001085 page = read_sb_page(
1086 bitmap->mddev,
1087 bitmap->mddev->bitmap_info.offset,
1088 page,
1089 index, count);
NeilBrowna654b9d82005-06-21 17:17:27 -07001090 } else if (file) {
NeilBrownd785a062006-06-26 00:27:48 -07001091 page = read_page(file, index, bitmap, count);
NeilBrowna654b9d82005-06-21 17:17:27 -07001092 offset = 0;
1093 } else {
NeilBrown42a04b52009-12-14 12:49:53 +11001094 page = read_sb_page(bitmap->mddev,
1095 bitmap->mddev->bitmap_info.offset,
NeilBrowna2ed9612008-12-19 16:25:01 +11001096 NULL,
1097 index, count);
NeilBrown32a76272005-06-21 17:17:14 -07001098 offset = 0;
1099 }
NeilBrowna654b9d82005-06-21 17:17:27 -07001100 if (IS_ERR(page)) { /* read error */
1101 ret = PTR_ERR(page);
NeilBrown4ad13662007-07-17 04:06:13 -07001102 goto err;
NeilBrowna654b9d82005-06-21 17:17:27 -07001103 }
1104
NeilBrown32a76272005-06-21 17:17:14 -07001105 oldindex = index;
1106 oldpage = page;
NeilBrown32a76272005-06-21 17:17:14 -07001107
NeilBrownb74fd282009-05-07 12:47:19 +10001108 bitmap->filemap[bitmap->file_pages++] = page;
1109 bitmap->last_page_size = count;
1110
NeilBrown32a76272005-06-21 17:17:14 -07001111 if (outofdate) {
1112 /*
1113 * if bitmap is out of date, dirty the
NeilBrownac2f40b2010-06-01 19:37:31 +10001114 * whole page and write it out
NeilBrown32a76272005-06-21 17:17:14 -07001115 */
NeilBrownea03aff2006-01-06 00:20:34 -08001116 paddr = kmap_atomic(page, KM_USER0);
1117 memset(paddr + offset, 0xff,
NeilBrown6a079972005-09-09 16:23:44 -07001118 PAGE_SIZE - offset);
NeilBrownea03aff2006-01-06 00:20:34 -08001119 kunmap_atomic(paddr, KM_USER0);
NeilBrown4ad13662007-07-17 04:06:13 -07001120 write_page(bitmap, page, 1);
1121
1122 ret = -EIO;
NeilBrownb74fd282009-05-07 12:47:19 +10001123 if (bitmap->flags & BITMAP_WRITE_ERROR)
NeilBrown4ad13662007-07-17 04:06:13 -07001124 goto err;
NeilBrown32a76272005-06-21 17:17:14 -07001125 }
NeilBrown32a76272005-06-21 17:17:14 -07001126 }
NeilBrownea03aff2006-01-06 00:20:34 -08001127 paddr = kmap_atomic(page, KM_USER0);
NeilBrownbd926c62005-11-08 21:39:32 -08001128 if (bitmap->flags & BITMAP_HOSTENDIAN)
NeilBrownea03aff2006-01-06 00:20:34 -08001129 b = test_bit(bit, paddr);
NeilBrownbd926c62005-11-08 21:39:32 -08001130 else
Akinobu Mita6b33aff2011-03-23 16:42:13 -07001131 b = test_bit_le(bit, paddr);
NeilBrownea03aff2006-01-06 00:20:34 -08001132 kunmap_atomic(paddr, KM_USER0);
NeilBrownbd926c62005-11-08 21:39:32 -08001133 if (b) {
NeilBrown32a76272005-06-21 17:17:14 -07001134 /* if the disk bit is set, set the memory bit */
NeilBrowndb305e52009-05-07 12:49:06 +10001135 int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap))
1136 >= start);
1137 bitmap_set_memory_bits(bitmap,
1138 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1139 needed);
NeilBrown32a76272005-06-21 17:17:14 -07001140 bit_cnt++;
NeilBrown6a079972005-09-09 16:23:44 -07001141 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
NeilBrown32a76272005-06-21 17:17:14 -07001142 }
NeilBrown32a76272005-06-21 17:17:14 -07001143 }
1144
NeilBrownac2f40b2010-06-01 19:37:31 +10001145 /* everything went OK */
NeilBrown32a76272005-06-21 17:17:14 -07001146 ret = 0;
1147 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1148
NeilBrown32a76272005-06-21 17:17:14 -07001149 if (bit_cnt) { /* Kick recovery if any bits were set */
1150 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1151 md_wakeup_thread(bitmap->mddev->thread);
1152 }
1153
NeilBrown32a76272005-06-21 17:17:14 -07001154 printk(KERN_INFO "%s: bitmap initialized from disk: "
Jonathan Brassow9c810752011-06-08 17:59:30 -05001155 "read %lu/%lu pages, set %lu of %lu bits\n",
1156 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, chunks);
NeilBrown32a76272005-06-21 17:17:14 -07001157
NeilBrown4ad13662007-07-17 04:06:13 -07001158 return 0;
1159
1160 err:
1161 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1162 bmname(bitmap), ret);
NeilBrown32a76272005-06-21 17:17:14 -07001163 return ret;
1164}
1165
NeilBrowna654b9d82005-06-21 17:17:27 -07001166void bitmap_write_all(struct bitmap *bitmap)
1167{
1168 /* We don't actually write all bitmap blocks here,
1169 * just flag them as needing to be written
1170 */
NeilBrownec7a3192006-06-26 00:27:45 -07001171 int i;
NeilBrowna654b9d82005-06-21 17:17:27 -07001172
NeilBrownac2f40b2010-06-01 19:37:31 +10001173 for (i = 0; i < bitmap->file_pages; i++)
NeilBrownec7a3192006-06-26 00:27:45 -07001174 set_page_attr(bitmap, bitmap->filemap[i],
1175 BITMAP_PAGE_NEEDWRITE);
NeilBrowna654b9d82005-06-21 17:17:27 -07001176}
1177
NeilBrown32a76272005-06-21 17:17:14 -07001178static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1179{
1180 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1181 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1182 bitmap->bp[page].count += inc;
NeilBrown32a76272005-06-21 17:17:14 -07001183 bitmap_checkfree(bitmap, page);
1184}
1185static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
NeilBrown57dab0b2010-10-19 10:03:39 +11001186 sector_t offset, sector_t *blocks,
NeilBrown32a76272005-06-21 17:17:14 -07001187 int create);
1188
1189/*
1190 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1191 * out to disk
1192 */
1193
NeilBrownaa5cbd12009-12-14 12:49:46 +11001194void bitmap_daemon_work(mddev_t *mddev)
NeilBrown32a76272005-06-21 17:17:14 -07001195{
NeilBrownaa5cbd12009-12-14 12:49:46 +11001196 struct bitmap *bitmap;
NeilBrownaa3163f2005-06-21 17:17:22 -07001197 unsigned long j;
NeilBrown32a76272005-06-21 17:17:14 -07001198 unsigned long flags;
1199 struct page *page = NULL, *lastpage = NULL;
NeilBrown57dab0b2010-10-19 10:03:39 +11001200 sector_t blocks;
NeilBrownea03aff2006-01-06 00:20:34 -08001201 void *paddr;
NeilBrowne384e582010-06-01 19:37:34 +10001202 struct dm_dirty_log *log = mddev->bitmap_info.log;
NeilBrown32a76272005-06-21 17:17:14 -07001203
NeilBrownaa5cbd12009-12-14 12:49:46 +11001204 /* Use a mutex to guard daemon_work against
1205 * bitmap_destroy.
1206 */
NeilBrownc3d97142009-12-14 12:49:52 +11001207 mutex_lock(&mddev->bitmap_info.mutex);
NeilBrownaa5cbd12009-12-14 12:49:46 +11001208 bitmap = mddev->bitmap;
1209 if (bitmap == NULL) {
NeilBrownc3d97142009-12-14 12:49:52 +11001210 mutex_unlock(&mddev->bitmap_info.mutex);
NeilBrown4ad13662007-07-17 04:06:13 -07001211 return;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001212 }
NeilBrown42a04b52009-12-14 12:49:53 +11001213 if (time_before(jiffies, bitmap->daemon_lastrun
NeilBrown1b04be92009-12-14 12:49:53 +11001214 + bitmap->mddev->bitmap_info.daemon_sleep))
NeilBrown7be3dfe2008-03-10 11:43:48 -07001215 goto done;
1216
NeilBrown32a76272005-06-21 17:17:14 -07001217 bitmap->daemon_lastrun = jiffies;
NeilBrown8311c292008-03-04 14:29:30 -08001218 if (bitmap->allclean) {
1219 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001220 goto done;
NeilBrown8311c292008-03-04 14:29:30 -08001221 }
1222 bitmap->allclean = 1;
NeilBrown32a76272005-06-21 17:17:14 -07001223
NeilBrownbe512692009-05-26 09:41:17 +10001224 spin_lock_irqsave(&bitmap->lock, flags);
NeilBrown32a76272005-06-21 17:17:14 -07001225 for (j = 0; j < bitmap->chunks; j++) {
1226 bitmap_counter_t *bmc;
NeilBrowne384e582010-06-01 19:37:34 +10001227 if (!bitmap->filemap) {
1228 if (!log)
1229 /* error or shutdown */
1230 break;
1231 } else
1232 page = filemap_get_page(bitmap, j);
NeilBrown32a76272005-06-21 17:17:14 -07001233
1234 if (page != lastpage) {
NeilBrownaa3163f2005-06-21 17:17:22 -07001235 /* skip this page unless it's marked as needing cleaning */
NeilBrownec7a3192006-06-26 00:27:45 -07001236 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1237 int need_write = test_page_attr(bitmap, page,
1238 BITMAP_PAGE_NEEDWRITE);
NeilBrowna647e4b2006-06-26 00:27:46 -07001239 if (need_write)
NeilBrownaa3163f2005-06-21 17:17:22 -07001240 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
NeilBrowna647e4b2006-06-26 00:27:46 -07001241
NeilBrownaa3163f2005-06-21 17:17:22 -07001242 spin_unlock_irqrestore(&bitmap->lock, flags);
NeilBrown8311c292008-03-04 14:29:30 -08001243 if (need_write) {
NeilBrown4ad13662007-07-17 04:06:13 -07001244 write_page(bitmap, page, 0);
NeilBrown8311c292008-03-04 14:29:30 -08001245 bitmap->allclean = 0;
1246 }
NeilBrownbe512692009-05-26 09:41:17 +10001247 spin_lock_irqsave(&bitmap->lock, flags);
1248 j |= (PAGE_BITS - 1);
NeilBrownaa3163f2005-06-21 17:17:22 -07001249 continue;
1250 }
1251
NeilBrown32a76272005-06-21 17:17:14 -07001252 /* grab the new page, sync and release the old */
NeilBrown32a76272005-06-21 17:17:14 -07001253 if (lastpage != NULL) {
NeilBrownec7a3192006-06-26 00:27:45 -07001254 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
NeilBrown32a76272005-06-21 17:17:14 -07001255 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1256 spin_unlock_irqrestore(&bitmap->lock, flags);
NeilBrown4ad13662007-07-17 04:06:13 -07001257 write_page(bitmap, lastpage, 0);
NeilBrown32a76272005-06-21 17:17:14 -07001258 } else {
1259 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1260 spin_unlock_irqrestore(&bitmap->lock, flags);
1261 }
NeilBrown32a76272005-06-21 17:17:14 -07001262 } else
1263 spin_unlock_irqrestore(&bitmap->lock, flags);
1264 lastpage = page;
Neil Browna0da84f2008-06-28 08:31:22 +10001265
1266 /* We are possibly going to clear some bits, so make
1267 * sure that events_cleared is up-to-date.
1268 */
NeilBrownece5cff2009-12-14 12:49:56 +11001269 if (bitmap->need_sync &&
1270 bitmap->mddev->bitmap_info.external == 0) {
Neil Browna0da84f2008-06-28 08:31:22 +10001271 bitmap_super_t *sb;
1272 bitmap->need_sync = 0;
1273 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1274 sb->events_cleared =
1275 cpu_to_le64(bitmap->events_cleared);
1276 kunmap_atomic(sb, KM_USER0);
1277 write_page(bitmap, bitmap->sb_page, 1);
1278 }
NeilBrown32a76272005-06-21 17:17:14 -07001279 spin_lock_irqsave(&bitmap->lock, flags);
NeilBrownece5cff2009-12-14 12:49:56 +11001280 if (!bitmap->need_sync)
1281 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
NeilBrown32a76272005-06-21 17:17:14 -07001282 }
NeilBrowndb305e52009-05-07 12:49:06 +10001283 bmc = bitmap_get_counter(bitmap,
1284 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1285 &blocks, 0);
NeilBrown32a76272005-06-21 17:17:14 -07001286 if (bmc) {
NeilBrown8311c292008-03-04 14:29:30 -08001287 if (*bmc)
1288 bitmap->allclean = 0;
1289
NeilBrown32a76272005-06-21 17:17:14 -07001290 if (*bmc == 2) {
NeilBrownac2f40b2010-06-01 19:37:31 +10001291 *bmc = 1; /* maybe clear the bit next time */
NeilBrown32a76272005-06-21 17:17:14 -07001292 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
NeilBrownece5cff2009-12-14 12:49:56 +11001293 } else if (*bmc == 1 && !bitmap->need_sync) {
NeilBrown32a76272005-06-21 17:17:14 -07001294 /* we can clear the bit */
1295 *bmc = 0;
NeilBrowndb305e52009-05-07 12:49:06 +10001296 bitmap_count_page(bitmap,
1297 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
NeilBrown32a76272005-06-21 17:17:14 -07001298 -1);
1299
1300 /* clear the bit */
NeilBrowne384e582010-06-01 19:37:34 +10001301 if (page) {
1302 paddr = kmap_atomic(page, KM_USER0);
1303 if (bitmap->flags & BITMAP_HOSTENDIAN)
1304 clear_bit(file_page_offset(bitmap, j),
1305 paddr);
1306 else
Akinobu Mita6b33aff2011-03-23 16:42:13 -07001307 __test_and_clear_bit_le(file_page_offset(bitmap, j),
NeilBrowne384e582010-06-01 19:37:34 +10001308 paddr);
1309 kunmap_atomic(paddr, KM_USER0);
1310 } else
1311 log->type->clear_region(log, j);
NeilBrown32a76272005-06-21 17:17:14 -07001312 }
NeilBrownbe512692009-05-26 09:41:17 +10001313 } else
1314 j |= PAGE_COUNTER_MASK;
NeilBrown32a76272005-06-21 17:17:14 -07001315 }
NeilBrownbe512692009-05-26 09:41:17 +10001316 spin_unlock_irqrestore(&bitmap->lock, flags);
NeilBrown32a76272005-06-21 17:17:14 -07001317
1318 /* now sync the final page */
NeilBrowne384e582010-06-01 19:37:34 +10001319 if (lastpage != NULL || log != NULL) {
NeilBrown32a76272005-06-21 17:17:14 -07001320 spin_lock_irqsave(&bitmap->lock, flags);
NeilBrownec7a3192006-06-26 00:27:45 -07001321 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
NeilBrown32a76272005-06-21 17:17:14 -07001322 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1323 spin_unlock_irqrestore(&bitmap->lock, flags);
NeilBrowne384e582010-06-01 19:37:34 +10001324 if (lastpage)
1325 write_page(bitmap, lastpage, 0);
1326 else
1327 if (log->type->flush(log))
1328 bitmap->flags |= BITMAP_WRITE_ERROR;
NeilBrown32a76272005-06-21 17:17:14 -07001329 } else {
1330 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1331 spin_unlock_irqrestore(&bitmap->lock, flags);
1332 }
NeilBrown32a76272005-06-21 17:17:14 -07001333 }
1334
NeilBrown7be3dfe2008-03-10 11:43:48 -07001335 done:
NeilBrown8311c292008-03-04 14:29:30 -08001336 if (bitmap->allclean == 0)
NeilBrownac2f40b2010-06-01 19:37:31 +10001337 bitmap->mddev->thread->timeout =
NeilBrown1b04be92009-12-14 12:49:53 +11001338 bitmap->mddev->bitmap_info.daemon_sleep;
NeilBrownc3d97142009-12-14 12:49:52 +11001339 mutex_unlock(&mddev->bitmap_info.mutex);
NeilBrown32a76272005-06-21 17:17:14 -07001340}
1341
NeilBrown32a76272005-06-21 17:17:14 -07001342static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
NeilBrown57dab0b2010-10-19 10:03:39 +11001343 sector_t offset, sector_t *blocks,
NeilBrown32a76272005-06-21 17:17:14 -07001344 int create)
NeilBrownee305ac2009-09-23 18:06:44 +10001345__releases(bitmap->lock)
1346__acquires(bitmap->lock)
NeilBrown32a76272005-06-21 17:17:14 -07001347{
1348 /* If 'create', we might release the lock and reclaim it.
1349 * The lock must have been taken with interrupts enabled.
1350 * If !create, we don't release the lock.
1351 */
1352 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1353 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1354 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1355 sector_t csize;
NeilBrownef425672010-06-01 19:37:33 +10001356 int err;
NeilBrown32a76272005-06-21 17:17:14 -07001357
NeilBrownef425672010-06-01 19:37:33 +10001358 err = bitmap_checkpage(bitmap, page, create);
1359
1360 if (bitmap->bp[page].hijacked ||
1361 bitmap->bp[page].map == NULL)
1362 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1363 PAGE_COUNTER_SHIFT - 1);
1364 else
NeilBrown32a76272005-06-21 17:17:14 -07001365 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
NeilBrownef425672010-06-01 19:37:33 +10001366 *blocks = csize - (offset & (csize - 1));
1367
1368 if (err < 0)
NeilBrown32a76272005-06-21 17:17:14 -07001369 return NULL;
NeilBrownef425672010-06-01 19:37:33 +10001370
NeilBrown32a76272005-06-21 17:17:14 -07001371 /* now locked ... */
1372
1373 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1374 /* should we use the first or second counter field
1375 * of the hijacked pointer? */
1376 int hi = (pageoff > PAGE_COUNTER_MASK);
NeilBrown32a76272005-06-21 17:17:14 -07001377 return &((bitmap_counter_t *)
1378 &bitmap->bp[page].map)[hi];
NeilBrownef425672010-06-01 19:37:33 +10001379 } else /* page is allocated */
NeilBrown32a76272005-06-21 17:17:14 -07001380 return (bitmap_counter_t *)
1381 &(bitmap->bp[page].map[pageoff]);
NeilBrown32a76272005-06-21 17:17:14 -07001382}
1383
NeilBrown4b6d2872005-09-09 16:23:47 -07001384int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
NeilBrown32a76272005-06-21 17:17:14 -07001385{
NeilBrownac2f40b2010-06-01 19:37:31 +10001386 if (!bitmap)
1387 return 0;
NeilBrown4b6d2872005-09-09 16:23:47 -07001388
1389 if (behind) {
Paul Clements696fcd52010-03-08 16:02:37 +11001390 int bw;
NeilBrown4b6d2872005-09-09 16:23:47 -07001391 atomic_inc(&bitmap->behind_writes);
Paul Clements696fcd52010-03-08 16:02:37 +11001392 bw = atomic_read(&bitmap->behind_writes);
1393 if (bw > bitmap->behind_writes_used)
1394 bitmap->behind_writes_used = bw;
1395
NeilBrown4b6d2872005-09-09 16:23:47 -07001396 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
Paul Clements696fcd52010-03-08 16:02:37 +11001397 bw, bitmap->max_write_behind);
NeilBrown4b6d2872005-09-09 16:23:47 -07001398 }
1399
NeilBrown32a76272005-06-21 17:17:14 -07001400 while (sectors) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001401 sector_t blocks;
NeilBrown32a76272005-06-21 17:17:14 -07001402 bitmap_counter_t *bmc;
1403
1404 spin_lock_irq(&bitmap->lock);
1405 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1406 if (!bmc) {
1407 spin_unlock_irq(&bitmap->lock);
1408 return 0;
1409 }
1410
Namhyung Kim27d5ea02011-06-09 11:42:57 +10001411 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
Neil Brownda6e1a32007-02-08 14:20:37 -08001412 DEFINE_WAIT(__wait);
1413 /* note that it is safe to do the prepare_to_wait
1414 * after the test as long as we do it before dropping
1415 * the spinlock.
1416 */
1417 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1418 TASK_UNINTERRUPTIBLE);
1419 spin_unlock_irq(&bitmap->lock);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001420 io_schedule();
Neil Brownda6e1a32007-02-08 14:20:37 -08001421 finish_wait(&bitmap->overflow_wait, &__wait);
1422 continue;
1423 }
1424
NeilBrownac2f40b2010-06-01 19:37:31 +10001425 switch (*bmc) {
NeilBrown32a76272005-06-21 17:17:14 -07001426 case 0:
1427 bitmap_file_set_bit(bitmap, offset);
NeilBrownac2f40b2010-06-01 19:37:31 +10001428 bitmap_count_page(bitmap, offset, 1);
NeilBrown32a76272005-06-21 17:17:14 -07001429 /* fall through */
1430 case 1:
1431 *bmc = 2;
1432 }
Neil Brownda6e1a32007-02-08 14:20:37 -08001433
NeilBrown32a76272005-06-21 17:17:14 -07001434 (*bmc)++;
1435
1436 spin_unlock_irq(&bitmap->lock);
1437
1438 offset += blocks;
1439 if (sectors > blocks)
1440 sectors -= blocks;
NeilBrownac2f40b2010-06-01 19:37:31 +10001441 else
1442 sectors = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001443 }
NeilBrown8311c292008-03-04 14:29:30 -08001444 bitmap->allclean = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001445 return 0;
1446}
NeilBrownac2f40b2010-06-01 19:37:31 +10001447EXPORT_SYMBOL(bitmap_startwrite);
NeilBrown32a76272005-06-21 17:17:14 -07001448
1449void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
NeilBrown4b6d2872005-09-09 16:23:47 -07001450 int success, int behind)
NeilBrown32a76272005-06-21 17:17:14 -07001451{
NeilBrownac2f40b2010-06-01 19:37:31 +10001452 if (!bitmap)
1453 return;
NeilBrown4b6d2872005-09-09 16:23:47 -07001454 if (behind) {
NeilBrowne5551902010-03-31 11:21:44 +11001455 if (atomic_dec_and_test(&bitmap->behind_writes))
1456 wake_up(&bitmap->behind_wait);
NeilBrown4b6d2872005-09-09 16:23:47 -07001457 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1458 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1459 }
NeilBrownd0a4bb42009-03-31 14:27:02 +11001460 if (bitmap->mddev->degraded)
1461 /* Never clear bits or update events_cleared when degraded */
1462 success = 0;
NeilBrown4b6d2872005-09-09 16:23:47 -07001463
NeilBrown32a76272005-06-21 17:17:14 -07001464 while (sectors) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001465 sector_t blocks;
NeilBrown32a76272005-06-21 17:17:14 -07001466 unsigned long flags;
1467 bitmap_counter_t *bmc;
1468
1469 spin_lock_irqsave(&bitmap->lock, flags);
1470 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1471 if (!bmc) {
1472 spin_unlock_irqrestore(&bitmap->lock, flags);
1473 return;
1474 }
1475
Neil Browna0da84f2008-06-28 08:31:22 +10001476 if (success &&
1477 bitmap->events_cleared < bitmap->mddev->events) {
1478 bitmap->events_cleared = bitmap->mddev->events;
1479 bitmap->need_sync = 1;
NeilBrown5ff5aff2010-06-01 19:37:32 +10001480 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
Neil Browna0da84f2008-06-28 08:31:22 +10001481 }
1482
Namhyung Kim27d5ea02011-06-09 11:42:57 +10001483 if (!success && !NEEDED(*bmc))
NeilBrown32a76272005-06-21 17:17:14 -07001484 *bmc |= NEEDED_MASK;
1485
Namhyung Kim27d5ea02011-06-09 11:42:57 +10001486 if (COUNTER(*bmc) == COUNTER_MAX)
Neil Brownda6e1a32007-02-08 14:20:37 -08001487 wake_up(&bitmap->overflow_wait);
1488
NeilBrown32a76272005-06-21 17:17:14 -07001489 (*bmc)--;
NeilBrownac2f40b2010-06-01 19:37:31 +10001490 if (*bmc <= 2)
NeilBrown32a76272005-06-21 17:17:14 -07001491 set_page_attr(bitmap,
NeilBrowne384e582010-06-01 19:37:34 +10001492 filemap_get_page(
1493 bitmap,
1494 offset >> CHUNK_BLOCK_SHIFT(bitmap)),
NeilBrown32a76272005-06-21 17:17:14 -07001495 BITMAP_PAGE_CLEAN);
NeilBrownac2f40b2010-06-01 19:37:31 +10001496
NeilBrown32a76272005-06-21 17:17:14 -07001497 spin_unlock_irqrestore(&bitmap->lock, flags);
1498 offset += blocks;
1499 if (sectors > blocks)
1500 sectors -= blocks;
NeilBrownac2f40b2010-06-01 19:37:31 +10001501 else
1502 sectors = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001503 }
1504}
NeilBrownac2f40b2010-06-01 19:37:31 +10001505EXPORT_SYMBOL(bitmap_endwrite);
NeilBrown32a76272005-06-21 17:17:14 -07001506
NeilBrown57dab0b2010-10-19 10:03:39 +11001507static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
NeilBrown1187cf02009-03-31 14:27:02 +11001508 int degraded)
NeilBrown32a76272005-06-21 17:17:14 -07001509{
1510 bitmap_counter_t *bmc;
1511 int rv;
1512 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1513 *blocks = 1024;
1514 return 1; /* always resync if no bitmap */
1515 }
1516 spin_lock_irq(&bitmap->lock);
1517 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1518 rv = 0;
1519 if (bmc) {
1520 /* locked */
1521 if (RESYNC(*bmc))
1522 rv = 1;
1523 else if (NEEDED(*bmc)) {
1524 rv = 1;
NeilBrown6a806c52005-07-15 03:56:35 -07001525 if (!degraded) { /* don't set/clear bits if degraded */
1526 *bmc |= RESYNC_MASK;
1527 *bmc &= ~NEEDED_MASK;
1528 }
NeilBrown32a76272005-06-21 17:17:14 -07001529 }
1530 }
1531 spin_unlock_irq(&bitmap->lock);
NeilBrown8311c292008-03-04 14:29:30 -08001532 bitmap->allclean = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001533 return rv;
1534}
1535
NeilBrown57dab0b2010-10-19 10:03:39 +11001536int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
NeilBrown1187cf02009-03-31 14:27:02 +11001537 int degraded)
1538{
1539 /* bitmap_start_sync must always report on multiples of whole
1540 * pages, otherwise resync (which is very PAGE_SIZE based) will
1541 * get confused.
1542 * So call __bitmap_start_sync repeatedly (if needed) until
1543 * At least PAGE_SIZE>>9 blocks are covered.
1544 * Return the 'or' of the result.
1545 */
1546 int rv = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11001547 sector_t blocks1;
NeilBrown1187cf02009-03-31 14:27:02 +11001548
1549 *blocks = 0;
1550 while (*blocks < (PAGE_SIZE>>9)) {
1551 rv |= __bitmap_start_sync(bitmap, offset,
1552 &blocks1, degraded);
1553 offset += blocks1;
1554 *blocks += blocks1;
1555 }
1556 return rv;
1557}
NeilBrownac2f40b2010-06-01 19:37:31 +10001558EXPORT_SYMBOL(bitmap_start_sync);
NeilBrown1187cf02009-03-31 14:27:02 +11001559
NeilBrown57dab0b2010-10-19 10:03:39 +11001560void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
NeilBrown32a76272005-06-21 17:17:14 -07001561{
1562 bitmap_counter_t *bmc;
1563 unsigned long flags;
NeilBrownac2f40b2010-06-01 19:37:31 +10001564
1565 if (bitmap == NULL) {
NeilBrown32a76272005-06-21 17:17:14 -07001566 *blocks = 1024;
1567 return;
1568 }
1569 spin_lock_irqsave(&bitmap->lock, flags);
1570 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1571 if (bmc == NULL)
1572 goto unlock;
1573 /* locked */
NeilBrown32a76272005-06-21 17:17:14 -07001574 if (RESYNC(*bmc)) {
1575 *bmc &= ~RESYNC_MASK;
1576
1577 if (!NEEDED(*bmc) && aborted)
1578 *bmc |= NEEDED_MASK;
1579 else {
NeilBrownac2f40b2010-06-01 19:37:31 +10001580 if (*bmc <= 2)
NeilBrown32a76272005-06-21 17:17:14 -07001581 set_page_attr(bitmap,
1582 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1583 BITMAP_PAGE_CLEAN);
NeilBrown32a76272005-06-21 17:17:14 -07001584 }
1585 }
1586 unlock:
1587 spin_unlock_irqrestore(&bitmap->lock, flags);
NeilBrown8311c292008-03-04 14:29:30 -08001588 bitmap->allclean = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001589}
NeilBrownac2f40b2010-06-01 19:37:31 +10001590EXPORT_SYMBOL(bitmap_end_sync);
NeilBrown32a76272005-06-21 17:17:14 -07001591
1592void bitmap_close_sync(struct bitmap *bitmap)
1593{
1594 /* Sync has finished, and any bitmap chunks that weren't synced
1595 * properly have been aborted. It remains to us to clear the
1596 * RESYNC bit wherever it is still on
1597 */
1598 sector_t sector = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11001599 sector_t blocks;
NeilBrownb47490c2008-02-06 01:39:50 -08001600 if (!bitmap)
1601 return;
NeilBrown32a76272005-06-21 17:17:14 -07001602 while (sector < bitmap->mddev->resync_max_sectors) {
1603 bitmap_end_sync(bitmap, sector, &blocks, 0);
NeilBrownb47490c2008-02-06 01:39:50 -08001604 sector += blocks;
NeilBrown32a76272005-06-21 17:17:14 -07001605 }
1606}
NeilBrownac2f40b2010-06-01 19:37:31 +10001607EXPORT_SYMBOL(bitmap_close_sync);
NeilBrown32a76272005-06-21 17:17:14 -07001608
NeilBrownb47490c2008-02-06 01:39:50 -08001609void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1610{
1611 sector_t s = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11001612 sector_t blocks;
NeilBrownb47490c2008-02-06 01:39:50 -08001613
1614 if (!bitmap)
1615 return;
1616 if (sector == 0) {
1617 bitmap->last_end_sync = jiffies;
1618 return;
1619 }
1620 if (time_before(jiffies, (bitmap->last_end_sync
NeilBrown1b04be92009-12-14 12:49:53 +11001621 + bitmap->mddev->bitmap_info.daemon_sleep)))
NeilBrownb47490c2008-02-06 01:39:50 -08001622 return;
1623 wait_event(bitmap->mddev->recovery_wait,
1624 atomic_read(&bitmap->mddev->recovery_active) == 0);
1625
NeilBrown75d3da42011-01-14 09:14:34 +11001626 bitmap->mddev->curr_resync_completed = sector;
NeilBrown070dc6d2010-08-30 17:33:34 +10001627 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
NeilBrownb47490c2008-02-06 01:39:50 -08001628 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1629 s = 0;
1630 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1631 bitmap_end_sync(bitmap, s, &blocks, 0);
1632 s += blocks;
1633 }
1634 bitmap->last_end_sync = jiffies;
NeilBrownacb180b2009-04-14 16:28:34 +10001635 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
NeilBrownb47490c2008-02-06 01:39:50 -08001636}
NeilBrownac2f40b2010-06-01 19:37:31 +10001637EXPORT_SYMBOL(bitmap_cond_end_sync);
NeilBrownb47490c2008-02-06 01:39:50 -08001638
NeilBrown6a079972005-09-09 16:23:44 -07001639static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
NeilBrown32a76272005-06-21 17:17:14 -07001640{
1641 /* For each chunk covered by any of these sectors, set the
NeilBrown193f1c92005-08-04 12:53:33 -07001642 * counter to 1 and set resync_needed. They should all
NeilBrown32a76272005-06-21 17:17:14 -07001643 * be 0 at this point
1644 */
NeilBrown193f1c92005-08-04 12:53:33 -07001645
NeilBrown57dab0b2010-10-19 10:03:39 +11001646 sector_t secs;
NeilBrown193f1c92005-08-04 12:53:33 -07001647 bitmap_counter_t *bmc;
1648 spin_lock_irq(&bitmap->lock);
1649 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1650 if (!bmc) {
NeilBrown32a76272005-06-21 17:17:14 -07001651 spin_unlock_irq(&bitmap->lock);
NeilBrown193f1c92005-08-04 12:53:33 -07001652 return;
NeilBrown32a76272005-06-21 17:17:14 -07001653 }
NeilBrownac2f40b2010-06-01 19:37:31 +10001654 if (!*bmc) {
NeilBrown193f1c92005-08-04 12:53:33 -07001655 struct page *page;
NeilBrownac2f40b2010-06-01 19:37:31 +10001656 *bmc = 1 | (needed ? NEEDED_MASK : 0);
NeilBrown193f1c92005-08-04 12:53:33 -07001657 bitmap_count_page(bitmap, offset, 1);
1658 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1659 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1660 }
1661 spin_unlock_irq(&bitmap->lock);
NeilBrown8311c292008-03-04 14:29:30 -08001662 bitmap->allclean = 0;
NeilBrown32a76272005-06-21 17:17:14 -07001663}
1664
Paul Clements9b1d1da2006-10-03 01:15:49 -07001665/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1666void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1667{
1668 unsigned long chunk;
1669
1670 for (chunk = s; chunk <= e; chunk++) {
NeilBrowndb305e52009-05-07 12:49:06 +10001671 sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap);
Paul Clements9b1d1da2006-10-03 01:15:49 -07001672 bitmap_set_memory_bits(bitmap, sec, 1);
1673 bitmap_file_set_bit(bitmap, sec);
NeilBrownffa23322009-12-14 12:49:56 +11001674 if (sec < bitmap->mddev->recovery_cp)
1675 /* We are asserting that the array is dirty,
1676 * so move the recovery_cp address back so
1677 * that it is obvious that it is dirty
1678 */
1679 bitmap->mddev->recovery_cp = sec;
Paul Clements9b1d1da2006-10-03 01:15:49 -07001680 }
1681}
1682
NeilBrown32a76272005-06-21 17:17:14 -07001683/*
NeilBrown6b8b3e82005-08-04 12:53:35 -07001684 * flush out any pending updates
1685 */
1686void bitmap_flush(mddev_t *mddev)
1687{
1688 struct bitmap *bitmap = mddev->bitmap;
NeilBrown42a04b52009-12-14 12:49:53 +11001689 long sleep;
NeilBrown6b8b3e82005-08-04 12:53:35 -07001690
1691 if (!bitmap) /* there was no bitmap */
1692 return;
1693
1694 /* run the daemon_work three time to ensure everything is flushed
1695 * that can be
1696 */
NeilBrown1b04be92009-12-14 12:49:53 +11001697 sleep = mddev->bitmap_info.daemon_sleep * 2;
NeilBrown42a04b52009-12-14 12:49:53 +11001698 bitmap->daemon_lastrun -= sleep;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001699 bitmap_daemon_work(mddev);
NeilBrown42a04b52009-12-14 12:49:53 +11001700 bitmap->daemon_lastrun -= sleep;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001701 bitmap_daemon_work(mddev);
NeilBrown42a04b52009-12-14 12:49:53 +11001702 bitmap->daemon_lastrun -= sleep;
NeilBrownaa5cbd12009-12-14 12:49:46 +11001703 bitmap_daemon_work(mddev);
NeilBrown6b8b3e82005-08-04 12:53:35 -07001704 bitmap_update_sb(bitmap);
1705}
1706
1707/*
NeilBrown32a76272005-06-21 17:17:14 -07001708 * free memory that was allocated
1709 */
NeilBrown3178b0d2005-09-09 16:23:50 -07001710static void bitmap_free(struct bitmap *bitmap)
NeilBrown32a76272005-06-21 17:17:14 -07001711{
1712 unsigned long k, pages;
1713 struct bitmap_page *bp;
NeilBrown32a76272005-06-21 17:17:14 -07001714
1715 if (!bitmap) /* there was no bitmap */
1716 return;
1717
NeilBrown32a76272005-06-21 17:17:14 -07001718 /* release the bitmap file and kill the daemon */
1719 bitmap_file_put(bitmap);
1720
1721 bp = bitmap->bp;
1722 pages = bitmap->pages;
1723
1724 /* free all allocated memory */
1725
NeilBrown32a76272005-06-21 17:17:14 -07001726 if (bp) /* deallocate the page memory */
1727 for (k = 0; k < pages; k++)
1728 if (bp[k].map && !bp[k].hijacked)
1729 kfree(bp[k].map);
1730 kfree(bp);
1731 kfree(bitmap);
1732}
NeilBrownaa5cbd12009-12-14 12:49:46 +11001733
NeilBrown3178b0d2005-09-09 16:23:50 -07001734void bitmap_destroy(mddev_t *mddev)
1735{
1736 struct bitmap *bitmap = mddev->bitmap;
1737
1738 if (!bitmap) /* there was no bitmap */
1739 return;
1740
NeilBrownc3d97142009-12-14 12:49:52 +11001741 mutex_lock(&mddev->bitmap_info.mutex);
NeilBrown3178b0d2005-09-09 16:23:50 -07001742 mddev->bitmap = NULL; /* disconnect from the md device */
NeilBrownc3d97142009-12-14 12:49:52 +11001743 mutex_unlock(&mddev->bitmap_info.mutex);
NeilBrownb15c2e52006-01-06 00:20:16 -08001744 if (mddev->thread)
1745 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
NeilBrown3178b0d2005-09-09 16:23:50 -07001746
NeilBrownece5cff2009-12-14 12:49:56 +11001747 if (bitmap->sysfs_can_clear)
1748 sysfs_put(bitmap->sysfs_can_clear);
1749
NeilBrown3178b0d2005-09-09 16:23:50 -07001750 bitmap_free(bitmap);
1751}
NeilBrown32a76272005-06-21 17:17:14 -07001752
1753/*
1754 * initialize the bitmap structure
1755 * if this returns an error, bitmap_destroy must be called to do clean up
1756 */
1757int bitmap_create(mddev_t *mddev)
1758{
1759 struct bitmap *bitmap;
NeilBrown1f593902009-04-20 11:50:24 +10001760 sector_t blocks = mddev->resync_max_sectors;
NeilBrown32a76272005-06-21 17:17:14 -07001761 unsigned long chunks;
1762 unsigned long pages;
NeilBrownc3d97142009-12-14 12:49:52 +11001763 struct file *file = mddev->bitmap_info.file;
NeilBrown32a76272005-06-21 17:17:14 -07001764 int err;
NeilBrown5ff5aff2010-06-01 19:37:32 +10001765 struct sysfs_dirent *bm = NULL;
NeilBrown32a76272005-06-21 17:17:14 -07001766
Alexey Dobriyan5f6e3c832006-10-11 01:22:26 -07001767 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
NeilBrown32a76272005-06-21 17:17:14 -07001768
NeilBrowne384e582010-06-01 19:37:34 +10001769 if (!file
1770 && !mddev->bitmap_info.offset
1771 && !mddev->bitmap_info.log) /* bitmap disabled, nothing to do */
NeilBrown32a76272005-06-21 17:17:14 -07001772 return 0;
1773
NeilBrownc3d97142009-12-14 12:49:52 +11001774 BUG_ON(file && mddev->bitmap_info.offset);
NeilBrowne384e582010-06-01 19:37:34 +10001775 BUG_ON(mddev->bitmap_info.offset && mddev->bitmap_info.log);
NeilBrowna654b9d82005-06-21 17:17:27 -07001776
NeilBrown9ffae0c2006-01-06 00:20:32 -08001777 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
NeilBrown32a76272005-06-21 17:17:14 -07001778 if (!bitmap)
1779 return -ENOMEM;
1780
NeilBrown32a76272005-06-21 17:17:14 -07001781 spin_lock_init(&bitmap->lock);
NeilBrownce25c312006-06-26 00:27:49 -07001782 atomic_set(&bitmap->pending_writes, 0);
1783 init_waitqueue_head(&bitmap->write_wait);
Neil Brownda6e1a32007-02-08 14:20:37 -08001784 init_waitqueue_head(&bitmap->overflow_wait);
NeilBrowne5551902010-03-31 11:21:44 +11001785 init_waitqueue_head(&bitmap->behind_wait);
NeilBrownce25c312006-06-26 00:27:49 -07001786
NeilBrown32a76272005-06-21 17:17:14 -07001787 bitmap->mddev = mddev;
NeilBrown32a76272005-06-21 17:17:14 -07001788
NeilBrown5ff5aff2010-06-01 19:37:32 +10001789 if (mddev->kobj.sd)
1790 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
NeilBrownece5cff2009-12-14 12:49:56 +11001791 if (bm) {
Eric W. Biederman3ff195b2010-03-30 11:31:26 -07001792 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
NeilBrownece5cff2009-12-14 12:49:56 +11001793 sysfs_put(bm);
1794 } else
1795 bitmap->sysfs_can_clear = NULL;
1796
NeilBrown32a76272005-06-21 17:17:14 -07001797 bitmap->file = file;
NeilBrownce25c312006-06-26 00:27:49 -07001798 if (file) {
1799 get_file(file);
NeilBrownae8fa282009-10-16 15:56:01 +11001800 /* As future accesses to this file will use bmap,
1801 * and bypass the page cache, we must sync the file
1802 * first.
1803 */
Christoph Hellwig8018ab02010-03-22 17:32:25 +01001804 vfs_fsync(file, 1);
NeilBrownce25c312006-06-26 00:27:49 -07001805 }
NeilBrown42a04b52009-12-14 12:49:53 +11001806 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
Jonathan Brassow9c810752011-06-08 17:59:30 -05001807 if (!mddev->bitmap_info.external) {
1808 /*
1809 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1810 * instructing us to create a new on-disk bitmap instance.
1811 */
1812 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1813 err = bitmap_new_disk_sb(bitmap);
1814 else
1815 err = bitmap_read_sb(bitmap);
1816 } else {
NeilBrownece5cff2009-12-14 12:49:56 +11001817 err = 0;
1818 if (mddev->bitmap_info.chunksize == 0 ||
1819 mddev->bitmap_info.daemon_sleep == 0)
1820 /* chunksize and time_base need to be
1821 * set first. */
1822 err = -EINVAL;
1823 }
NeilBrown32a76272005-06-21 17:17:14 -07001824 if (err)
NeilBrown3178b0d2005-09-09 16:23:50 -07001825 goto error;
NeilBrown32a76272005-06-21 17:17:14 -07001826
NeilBrown624ce4f2009-12-14 12:49:56 +11001827 bitmap->daemon_lastrun = jiffies;
NeilBrown42a04b52009-12-14 12:49:53 +11001828 bitmap->chunkshift = ffz(~mddev->bitmap_info.chunksize);
NeilBrown32a76272005-06-21 17:17:14 -07001829
1830 /* now that chunksize and chunkshift are set, we can use these macros */
NeilBrownac2f40b2010-06-01 19:37:31 +10001831 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >>
NeilBrown1f593902009-04-20 11:50:24 +10001832 CHUNK_BLOCK_SHIFT(bitmap);
NeilBrownac2f40b2010-06-01 19:37:31 +10001833 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
NeilBrown32a76272005-06-21 17:17:14 -07001834
1835 BUG_ON(!pages);
1836
1837 bitmap->chunks = chunks;
1838 bitmap->pages = pages;
1839 bitmap->missing_pages = pages;
1840 bitmap->counter_bits = COUNTER_BITS;
1841
1842 bitmap->syncchunk = ~0UL;
1843
Olaf Hering44456d32005-07-27 11:45:17 -07001844#ifdef INJECT_FATAL_FAULT_1
NeilBrown32a76272005-06-21 17:17:14 -07001845 bitmap->bp = NULL;
1846#else
NeilBrown9ffae0c2006-01-06 00:20:32 -08001847 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
NeilBrown32a76272005-06-21 17:17:14 -07001848#endif
NeilBrown3178b0d2005-09-09 16:23:50 -07001849 err = -ENOMEM;
NeilBrown32a76272005-06-21 17:17:14 -07001850 if (!bitmap->bp)
NeilBrown3178b0d2005-09-09 16:23:50 -07001851 goto error;
NeilBrown32a76272005-06-21 17:17:14 -07001852
NeilBrown69e51b42010-06-01 19:37:35 +10001853 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1854 pages, bmname(bitmap));
1855
1856 mddev->bitmap = bitmap;
1857
1858
1859 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1860
1861 error:
1862 bitmap_free(bitmap);
1863 return err;
1864}
1865
1866int bitmap_load(mddev_t *mddev)
1867{
1868 int err = 0;
1869 sector_t sector = 0;
1870 struct bitmap *bitmap = mddev->bitmap;
1871
1872 if (!bitmap)
1873 goto out;
1874
1875 /* Clear out old bitmap info first: Either there is none, or we
1876 * are resuming after someone else has possibly changed things,
1877 * so we should forget old cached info.
1878 * All chunks should be clean, but some might need_sync.
1879 */
1880 while (sector < mddev->resync_max_sectors) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001881 sector_t blocks;
NeilBrown69e51b42010-06-01 19:37:35 +10001882 bitmap_start_sync(bitmap, sector, &blocks, 0);
1883 sector += blocks;
1884 }
1885 bitmap_close_sync(bitmap);
1886
NeilBrowne384e582010-06-01 19:37:34 +10001887 if (mddev->bitmap_info.log) {
1888 unsigned long i;
1889 struct dm_dirty_log *log = mddev->bitmap_info.log;
1890 for (i = 0; i < bitmap->chunks; i++)
1891 if (!log->type->in_sync(log, i, 1))
1892 bitmap_set_memory_bits(bitmap,
1893 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1894 1);
NeilBrown69e51b42010-06-01 19:37:35 +10001895 } else {
1896 sector_t start = 0;
1897 if (mddev->degraded == 0
1898 || bitmap->events_cleared == mddev->events)
1899 /* no need to keep dirty bits to optimise a
1900 * re-add of a missing device */
1901 start = mddev->recovery_cp;
1902
NeilBrowne384e582010-06-01 19:37:34 +10001903 err = bitmap_init_from_disk(bitmap, start);
NeilBrown69e51b42010-06-01 19:37:35 +10001904 }
NeilBrown32a76272005-06-21 17:17:14 -07001905 if (err)
NeilBrown69e51b42010-06-01 19:37:35 +10001906 goto out;
NeilBrown3178b0d2005-09-09 16:23:50 -07001907
NeilBrown1b04be92009-12-14 12:49:53 +11001908 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
NeilBrown9cd30fd2009-12-14 12:49:54 +11001909 md_wakeup_thread(mddev->thread);
NeilBrownb15c2e52006-01-06 00:20:16 -08001910
NeilBrown4ad13662007-07-17 04:06:13 -07001911 bitmap_update_sb(bitmap);
1912
NeilBrown69e51b42010-06-01 19:37:35 +10001913 if (bitmap->flags & BITMAP_WRITE_ERROR)
1914 err = -EIO;
1915out:
NeilBrown3178b0d2005-09-09 16:23:50 -07001916 return err;
NeilBrown32a76272005-06-21 17:17:14 -07001917}
NeilBrown69e51b42010-06-01 19:37:35 +10001918EXPORT_SYMBOL_GPL(bitmap_load);
NeilBrown32a76272005-06-21 17:17:14 -07001919
NeilBrown43a70502009-12-14 12:49:55 +11001920static ssize_t
1921location_show(mddev_t *mddev, char *page)
1922{
1923 ssize_t len;
NeilBrownac2f40b2010-06-01 19:37:31 +10001924 if (mddev->bitmap_info.file)
NeilBrown43a70502009-12-14 12:49:55 +11001925 len = sprintf(page, "file");
NeilBrownac2f40b2010-06-01 19:37:31 +10001926 else if (mddev->bitmap_info.offset)
NeilBrown43a70502009-12-14 12:49:55 +11001927 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
NeilBrownac2f40b2010-06-01 19:37:31 +10001928 else
NeilBrown43a70502009-12-14 12:49:55 +11001929 len = sprintf(page, "none");
1930 len += sprintf(page+len, "\n");
1931 return len;
1932}
1933
1934static ssize_t
1935location_store(mddev_t *mddev, const char *buf, size_t len)
1936{
1937
1938 if (mddev->pers) {
1939 if (!mddev->pers->quiesce)
1940 return -EBUSY;
1941 if (mddev->recovery || mddev->sync_thread)
1942 return -EBUSY;
1943 }
1944
1945 if (mddev->bitmap || mddev->bitmap_info.file ||
1946 mddev->bitmap_info.offset) {
1947 /* bitmap already configured. Only option is to clear it */
1948 if (strncmp(buf, "none", 4) != 0)
1949 return -EBUSY;
1950 if (mddev->pers) {
1951 mddev->pers->quiesce(mddev, 1);
1952 bitmap_destroy(mddev);
1953 mddev->pers->quiesce(mddev, 0);
1954 }
1955 mddev->bitmap_info.offset = 0;
1956 if (mddev->bitmap_info.file) {
1957 struct file *f = mddev->bitmap_info.file;
1958 mddev->bitmap_info.file = NULL;
1959 restore_bitmap_write_access(f);
1960 fput(f);
1961 }
1962 } else {
1963 /* No bitmap, OK to set a location */
1964 long long offset;
1965 if (strncmp(buf, "none", 4) == 0)
1966 /* nothing to be done */;
1967 else if (strncmp(buf, "file:", 5) == 0) {
1968 /* Not supported yet */
1969 return -EINVAL;
1970 } else {
1971 int rv;
1972 if (buf[0] == '+')
1973 rv = strict_strtoll(buf+1, 10, &offset);
1974 else
1975 rv = strict_strtoll(buf, 10, &offset);
1976 if (rv)
1977 return rv;
1978 if (offset == 0)
1979 return -EINVAL;
NeilBrownece5cff2009-12-14 12:49:56 +11001980 if (mddev->bitmap_info.external == 0 &&
1981 mddev->major_version == 0 &&
NeilBrown43a70502009-12-14 12:49:55 +11001982 offset != mddev->bitmap_info.default_offset)
1983 return -EINVAL;
1984 mddev->bitmap_info.offset = offset;
1985 if (mddev->pers) {
1986 mddev->pers->quiesce(mddev, 1);
1987 rv = bitmap_create(mddev);
1988 if (rv) {
1989 bitmap_destroy(mddev);
1990 mddev->bitmap_info.offset = 0;
1991 }
1992 mddev->pers->quiesce(mddev, 0);
1993 if (rv)
1994 return rv;
1995 }
1996 }
1997 }
1998 if (!mddev->external) {
1999 /* Ensure new bitmap info is stored in
2000 * metadata promptly.
2001 */
2002 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2003 md_wakeup_thread(mddev->thread);
2004 }
2005 return len;
2006}
2007
2008static struct md_sysfs_entry bitmap_location =
2009__ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2010
2011static ssize_t
2012timeout_show(mddev_t *mddev, char *page)
2013{
2014 ssize_t len;
2015 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2016 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
NeilBrownac2f40b2010-06-01 19:37:31 +10002017
NeilBrown43a70502009-12-14 12:49:55 +11002018 len = sprintf(page, "%lu", secs);
2019 if (jifs)
2020 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2021 len += sprintf(page+len, "\n");
2022 return len;
2023}
2024
2025static ssize_t
2026timeout_store(mddev_t *mddev, const char *buf, size_t len)
2027{
2028 /* timeout can be set at any time */
2029 unsigned long timeout;
2030 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2031 if (rv)
2032 return rv;
2033
2034 /* just to make sure we don't overflow... */
2035 if (timeout >= LONG_MAX / HZ)
2036 return -EINVAL;
2037
2038 timeout = timeout * HZ / 10000;
2039
2040 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2041 timeout = MAX_SCHEDULE_TIMEOUT-1;
2042 if (timeout < 1)
2043 timeout = 1;
2044 mddev->bitmap_info.daemon_sleep = timeout;
2045 if (mddev->thread) {
2046 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2047 * the bitmap is all clean and we don't need to
2048 * adjust the timeout right now
2049 */
2050 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2051 mddev->thread->timeout = timeout;
2052 md_wakeup_thread(mddev->thread);
2053 }
2054 }
2055 return len;
2056}
2057
2058static struct md_sysfs_entry bitmap_timeout =
2059__ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2060
2061static ssize_t
2062backlog_show(mddev_t *mddev, char *page)
2063{
2064 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2065}
2066
2067static ssize_t
2068backlog_store(mddev_t *mddev, const char *buf, size_t len)
2069{
2070 unsigned long backlog;
2071 int rv = strict_strtoul(buf, 10, &backlog);
2072 if (rv)
2073 return rv;
2074 if (backlog > COUNTER_MAX)
2075 return -EINVAL;
2076 mddev->bitmap_info.max_write_behind = backlog;
2077 return len;
2078}
2079
2080static struct md_sysfs_entry bitmap_backlog =
2081__ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2082
2083static ssize_t
2084chunksize_show(mddev_t *mddev, char *page)
2085{
2086 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2087}
2088
2089static ssize_t
2090chunksize_store(mddev_t *mddev, const char *buf, size_t len)
2091{
2092 /* Can only be changed when no bitmap is active */
2093 int rv;
2094 unsigned long csize;
2095 if (mddev->bitmap)
2096 return -EBUSY;
2097 rv = strict_strtoul(buf, 10, &csize);
2098 if (rv)
2099 return rv;
2100 if (csize < 512 ||
2101 !is_power_of_2(csize))
2102 return -EINVAL;
2103 mddev->bitmap_info.chunksize = csize;
2104 return len;
2105}
2106
2107static struct md_sysfs_entry bitmap_chunksize =
2108__ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2109
NeilBrownece5cff2009-12-14 12:49:56 +11002110static ssize_t metadata_show(mddev_t *mddev, char *page)
2111{
2112 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2113 ? "external" : "internal"));
2114}
2115
2116static ssize_t metadata_store(mddev_t *mddev, const char *buf, size_t len)
2117{
2118 if (mddev->bitmap ||
2119 mddev->bitmap_info.file ||
2120 mddev->bitmap_info.offset)
2121 return -EBUSY;
2122 if (strncmp(buf, "external", 8) == 0)
2123 mddev->bitmap_info.external = 1;
2124 else if (strncmp(buf, "internal", 8) == 0)
2125 mddev->bitmap_info.external = 0;
2126 else
2127 return -EINVAL;
2128 return len;
2129}
2130
2131static struct md_sysfs_entry bitmap_metadata =
2132__ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2133
2134static ssize_t can_clear_show(mddev_t *mddev, char *page)
2135{
2136 int len;
2137 if (mddev->bitmap)
2138 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2139 "false" : "true"));
2140 else
2141 len = sprintf(page, "\n");
2142 return len;
2143}
2144
2145static ssize_t can_clear_store(mddev_t *mddev, const char *buf, size_t len)
2146{
2147 if (mddev->bitmap == NULL)
2148 return -ENOENT;
2149 if (strncmp(buf, "false", 5) == 0)
2150 mddev->bitmap->need_sync = 1;
2151 else if (strncmp(buf, "true", 4) == 0) {
2152 if (mddev->degraded)
2153 return -EBUSY;
2154 mddev->bitmap->need_sync = 0;
2155 } else
2156 return -EINVAL;
2157 return len;
2158}
2159
2160static struct md_sysfs_entry bitmap_can_clear =
2161__ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2162
Paul Clements696fcd52010-03-08 16:02:37 +11002163static ssize_t
2164behind_writes_used_show(mddev_t *mddev, char *page)
2165{
2166 if (mddev->bitmap == NULL)
2167 return sprintf(page, "0\n");
2168 return sprintf(page, "%lu\n",
2169 mddev->bitmap->behind_writes_used);
2170}
2171
2172static ssize_t
2173behind_writes_used_reset(mddev_t *mddev, const char *buf, size_t len)
2174{
2175 if (mddev->bitmap)
2176 mddev->bitmap->behind_writes_used = 0;
2177 return len;
2178}
2179
2180static struct md_sysfs_entry max_backlog_used =
2181__ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2182 behind_writes_used_show, behind_writes_used_reset);
2183
NeilBrown43a70502009-12-14 12:49:55 +11002184static struct attribute *md_bitmap_attrs[] = {
2185 &bitmap_location.attr,
2186 &bitmap_timeout.attr,
2187 &bitmap_backlog.attr,
2188 &bitmap_chunksize.attr,
NeilBrownece5cff2009-12-14 12:49:56 +11002189 &bitmap_metadata.attr,
2190 &bitmap_can_clear.attr,
Paul Clements696fcd52010-03-08 16:02:37 +11002191 &max_backlog_used.attr,
NeilBrown43a70502009-12-14 12:49:55 +11002192 NULL
2193};
2194struct attribute_group md_bitmap_group = {
2195 .name = "bitmap",
2196 .attrs = md_bitmap_attrs,
2197};
2198