blob: b6d198f5181ed6d07f8f9c29782bd5a99e46f9ce [file] [log] [blame]
Arne Jansena2de7332011-03-08 14:14:00 +01001/*
Stefan Behrensb6bfebc2012-11-02 16:44:58 +01002 * Copyright (C) 2011, 2012 STRATO. All rights reserved.
Arne Jansena2de7332011-03-08 14:14:00 +01003 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
Arne Jansena2de7332011-03-08 14:14:00 +010019#include <linux/blkdev.h>
Jan Schmidt558540c2011-06-13 19:59:12 +020020#include <linux/ratelimit.h>
Arne Jansena2de7332011-03-08 14:14:00 +010021#include "ctree.h"
22#include "volumes.h"
23#include "disk-io.h"
24#include "ordered-data.h"
Jan Schmidt0ef8e452011-06-13 20:04:15 +020025#include "transaction.h"
Jan Schmidt558540c2011-06-13 19:59:12 +020026#include "backref.h"
Jan Schmidt5da6fcb2011-08-04 18:11:04 +020027#include "extent_io.h"
Stefan Behrensff023aa2012-11-06 11:43:11 +010028#include "dev-replace.h"
Stefan Behrens21adbd52011-11-09 13:44:05 +010029#include "check-integrity.h"
Josef Bacik606686e2012-06-04 14:03:51 -040030#include "rcu-string.h"
David Woodhouse53b381b2013-01-29 18:40:14 -050031#include "raid56.h"
Arne Jansena2de7332011-03-08 14:14:00 +010032
33/*
34 * This is only the first step towards a full-features scrub. It reads all
35 * extent and super block and verifies the checksums. In case a bad checksum
36 * is found or the extent cannot be read, good data will be written back if
37 * any can be found.
38 *
39 * Future enhancements:
Arne Jansena2de7332011-03-08 14:14:00 +010040 * - In case an unrepairable extent is encountered, track which files are
41 * affected and report them
Arne Jansena2de7332011-03-08 14:14:00 +010042 * - track and record media errors, throw out bad devices
Arne Jansena2de7332011-03-08 14:14:00 +010043 * - add a mode to also read unallocated space
Arne Jansena2de7332011-03-08 14:14:00 +010044 */
45
Stefan Behrensb5d67f62012-03-27 14:21:27 -040046struct scrub_block;
Stefan Behrensd9d181c2012-11-02 09:58:09 +010047struct scrub_ctx;
Arne Jansena2de7332011-03-08 14:14:00 +010048
Stefan Behrensff023aa2012-11-06 11:43:11 +010049/*
50 * the following three values only influence the performance.
51 * The last one configures the number of parallel and outstanding I/O
52 * operations. The first two values configure an upper limit for the number
53 * of (dynamically allocated) pages that are added to a bio.
54 */
55#define SCRUB_PAGES_PER_RD_BIO 32 /* 128k per bio */
56#define SCRUB_PAGES_PER_WR_BIO 32 /* 128k per bio */
57#define SCRUB_BIOS_PER_SCTX 64 /* 8MB per device in flight */
Stefan Behrens7a9e9982012-11-02 14:58:04 +010058
59/*
60 * the following value times PAGE_SIZE needs to be large enough to match the
61 * largest node/leaf/sector size that shall be supported.
62 * Values larger than BTRFS_STRIPE_LEN are not supported.
63 */
Stefan Behrensb5d67f62012-03-27 14:21:27 -040064#define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */
Arne Jansena2de7332011-03-08 14:14:00 +010065
66struct scrub_page {
Stefan Behrensb5d67f62012-03-27 14:21:27 -040067 struct scrub_block *sblock;
68 struct page *page;
Stefan Behrens442a4f62012-05-25 16:06:08 +020069 struct btrfs_device *dev;
Arne Jansena2de7332011-03-08 14:14:00 +010070 u64 flags; /* extent flags */
71 u64 generation;
Stefan Behrensb5d67f62012-03-27 14:21:27 -040072 u64 logical;
73 u64 physical;
Stefan Behrensff023aa2012-11-06 11:43:11 +010074 u64 physical_for_dev_replace;
Stefan Behrens7a9e9982012-11-02 14:58:04 +010075 atomic_t ref_count;
Stefan Behrensb5d67f62012-03-27 14:21:27 -040076 struct {
77 unsigned int mirror_num:8;
78 unsigned int have_csum:1;
79 unsigned int io_error:1;
80 };
Arne Jansena2de7332011-03-08 14:14:00 +010081 u8 csum[BTRFS_CSUM_SIZE];
82};
83
84struct scrub_bio {
85 int index;
Stefan Behrensd9d181c2012-11-02 09:58:09 +010086 struct scrub_ctx *sctx;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +010087 struct btrfs_device *dev;
Arne Jansena2de7332011-03-08 14:14:00 +010088 struct bio *bio;
89 int err;
90 u64 logical;
91 u64 physical;
Stefan Behrensff023aa2012-11-06 11:43:11 +010092#if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO
93 struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO];
94#else
95 struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO];
96#endif
Stefan Behrensb5d67f62012-03-27 14:21:27 -040097 int page_count;
Arne Jansena2de7332011-03-08 14:14:00 +010098 int next_free;
99 struct btrfs_work work;
100};
101
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400102struct scrub_block {
Stefan Behrens7a9e9982012-11-02 14:58:04 +0100103 struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK];
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400104 int page_count;
105 atomic_t outstanding_pages;
106 atomic_t ref_count; /* free mem on transition to zero */
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100107 struct scrub_ctx *sctx;
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400108 struct {
109 unsigned int header_error:1;
110 unsigned int checksum_error:1;
111 unsigned int no_io_error_seen:1;
Stefan Behrens442a4f62012-05-25 16:06:08 +0200112 unsigned int generation_error:1; /* also sets header_error */
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400113 };
114};
115
Stefan Behrensff023aa2012-11-06 11:43:11 +0100116struct scrub_wr_ctx {
117 struct scrub_bio *wr_curr_bio;
118 struct btrfs_device *tgtdev;
119 int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */
120 atomic_t flush_all_writes;
121 struct mutex wr_lock;
122};
123
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100124struct scrub_ctx {
Stefan Behrensff023aa2012-11-06 11:43:11 +0100125 struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX];
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100126 struct btrfs_root *dev_root;
Arne Jansena2de7332011-03-08 14:14:00 +0100127 int first_free;
128 int curr;
Stefan Behrensb6bfebc2012-11-02 16:44:58 +0100129 atomic_t bios_in_flight;
130 atomic_t workers_pending;
Arne Jansena2de7332011-03-08 14:14:00 +0100131 spinlock_t list_lock;
132 wait_queue_head_t list_wait;
133 u16 csum_size;
134 struct list_head csum_list;
135 atomic_t cancel_req;
Arne Jansen86287642011-03-23 16:34:19 +0100136 int readonly;
Stefan Behrensff023aa2012-11-06 11:43:11 +0100137 int pages_per_rd_bio;
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400138 u32 sectorsize;
139 u32 nodesize;
140 u32 leafsize;
Stefan Behrens63a212a2012-11-05 18:29:28 +0100141
142 int is_dev_replace;
Stefan Behrensff023aa2012-11-06 11:43:11 +0100143 struct scrub_wr_ctx wr_ctx;
Stefan Behrens63a212a2012-11-05 18:29:28 +0100144
Arne Jansena2de7332011-03-08 14:14:00 +0100145 /*
146 * statistics
147 */
148 struct btrfs_scrub_progress stat;
149 spinlock_t stat_lock;
150};
151
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200152struct scrub_fixup_nodatasum {
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100153 struct scrub_ctx *sctx;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100154 struct btrfs_device *dev;
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200155 u64 logical;
156 struct btrfs_root *root;
157 struct btrfs_work work;
158 int mirror_num;
159};
160
Josef Bacik652f25a2013-09-12 16:58:28 -0400161struct scrub_nocow_inode {
162 u64 inum;
163 u64 offset;
164 u64 root;
165 struct list_head list;
166};
167
Stefan Behrensff023aa2012-11-06 11:43:11 +0100168struct scrub_copy_nocow_ctx {
169 struct scrub_ctx *sctx;
170 u64 logical;
171 u64 len;
172 int mirror_num;
173 u64 physical_for_dev_replace;
Josef Bacik652f25a2013-09-12 16:58:28 -0400174 struct list_head inodes;
Stefan Behrensff023aa2012-11-06 11:43:11 +0100175 struct btrfs_work work;
176};
177
Jan Schmidt558540c2011-06-13 19:59:12 +0200178struct scrub_warning {
179 struct btrfs_path *path;
180 u64 extent_item_size;
181 char *scratch_buf;
182 char *msg_buf;
183 const char *errstr;
184 sector_t sector;
185 u64 logical;
186 struct btrfs_device *dev;
187 int msg_bufsize;
188 int scratch_bufsize;
189};
190
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400191
Stefan Behrensb6bfebc2012-11-02 16:44:58 +0100192static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
193static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
194static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
195static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400196static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100197static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
Stefan Behrens3ec706c2012-11-05 15:46:42 +0100198 struct btrfs_fs_info *fs_info,
Stefan Behrensff023aa2012-11-06 11:43:11 +0100199 struct scrub_block *original_sblock,
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400200 u64 length, u64 logical,
Stefan Behrensff023aa2012-11-06 11:43:11 +0100201 struct scrub_block *sblocks_for_recheck);
Stefan Behrens34f5c8e2012-11-02 16:16:26 +0100202static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
203 struct scrub_block *sblock, int is_metadata,
204 int have_csum, u8 *csum, u64 generation,
205 u16 csum_size);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400206static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
207 struct scrub_block *sblock,
208 int is_metadata, int have_csum,
209 const u8 *csum, u64 generation,
210 u16 csum_size);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400211static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
212 struct scrub_block *sblock_good,
213 int force_write);
214static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
215 struct scrub_block *sblock_good,
216 int page_num, int force_write);
Stefan Behrensff023aa2012-11-06 11:43:11 +0100217static void scrub_write_block_to_dev_replace(struct scrub_block *sblock);
218static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
219 int page_num);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400220static int scrub_checksum_data(struct scrub_block *sblock);
221static int scrub_checksum_tree_block(struct scrub_block *sblock);
222static int scrub_checksum_super(struct scrub_block *sblock);
223static void scrub_block_get(struct scrub_block *sblock);
224static void scrub_block_put(struct scrub_block *sblock);
Stefan Behrens7a9e9982012-11-02 14:58:04 +0100225static void scrub_page_get(struct scrub_page *spage);
226static void scrub_page_put(struct scrub_page *spage);
Stefan Behrensff023aa2012-11-06 11:43:11 +0100227static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
228 struct scrub_page *spage);
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100229static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100230 u64 physical, struct btrfs_device *dev, u64 flags,
Stefan Behrensff023aa2012-11-06 11:43:11 +0100231 u64 gen, int mirror_num, u8 *csum, int force,
232 u64 physical_for_dev_replace);
Stefan Behrens1623ede2012-03-27 14:21:26 -0400233static void scrub_bio_end_io(struct bio *bio, int err);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400234static void scrub_bio_end_io_worker(struct btrfs_work *work);
235static void scrub_block_complete(struct scrub_block *sblock);
Stefan Behrensff023aa2012-11-06 11:43:11 +0100236static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
237 u64 extent_logical, u64 extent_len,
238 u64 *extent_physical,
239 struct btrfs_device **extent_dev,
240 int *extent_mirror_num);
241static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
242 struct scrub_wr_ctx *wr_ctx,
243 struct btrfs_fs_info *fs_info,
244 struct btrfs_device *dev,
245 int is_dev_replace);
246static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
247static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
248 struct scrub_page *spage);
249static void scrub_wr_submit(struct scrub_ctx *sctx);
250static void scrub_wr_bio_end_io(struct bio *bio, int err);
251static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
252static int write_page_nocow(struct scrub_ctx *sctx,
253 u64 physical_for_dev_replace, struct page *page);
254static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
Josef Bacik652f25a2013-09-12 16:58:28 -0400255 struct scrub_copy_nocow_ctx *ctx);
Stefan Behrensff023aa2012-11-06 11:43:11 +0100256static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
257 int mirror_num, u64 physical_for_dev_replace);
258static void copy_nocow_pages_worker(struct btrfs_work *work);
Wang Shilongcb7ab022013-12-04 21:16:53 +0800259static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
Wang Shilong3cb09292013-12-04 21:15:19 +0800260static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
Stefan Behrens1623ede2012-03-27 14:21:26 -0400261
262
Stefan Behrensb6bfebc2012-11-02 16:44:58 +0100263static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
264{
265 atomic_inc(&sctx->bios_in_flight);
266}
267
268static void scrub_pending_bio_dec(struct scrub_ctx *sctx)
269{
270 atomic_dec(&sctx->bios_in_flight);
271 wake_up(&sctx->list_wait);
272}
273
Wang Shilongcb7ab022013-12-04 21:16:53 +0800274static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
Wang Shilong3cb09292013-12-04 21:15:19 +0800275{
276 while (atomic_read(&fs_info->scrub_pause_req)) {
277 mutex_unlock(&fs_info->scrub_lock);
278 wait_event(fs_info->scrub_pause_wait,
279 atomic_read(&fs_info->scrub_pause_req) == 0);
280 mutex_lock(&fs_info->scrub_lock);
281 }
282}
283
Wang Shilongcb7ab022013-12-04 21:16:53 +0800284static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
285{
286 atomic_inc(&fs_info->scrubs_paused);
287 wake_up(&fs_info->scrub_pause_wait);
288
289 mutex_lock(&fs_info->scrub_lock);
290 __scrub_blocked_if_needed(fs_info);
291 atomic_dec(&fs_info->scrubs_paused);
292 mutex_unlock(&fs_info->scrub_lock);
293
294 wake_up(&fs_info->scrub_pause_wait);
295}
296
Stefan Behrensb6bfebc2012-11-02 16:44:58 +0100297/*
298 * used for workers that require transaction commits (i.e., for the
299 * NOCOW case)
300 */
301static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
302{
303 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
304
305 /*
306 * increment scrubs_running to prevent cancel requests from
307 * completing as long as a worker is running. we must also
308 * increment scrubs_paused to prevent deadlocking on pause
309 * requests used for transactions commits (as the worker uses a
310 * transaction context). it is safe to regard the worker
311 * as paused for all matters practical. effectively, we only
312 * avoid cancellation requests from completing.
313 */
314 mutex_lock(&fs_info->scrub_lock);
315 atomic_inc(&fs_info->scrubs_running);
316 atomic_inc(&fs_info->scrubs_paused);
317 mutex_unlock(&fs_info->scrub_lock);
Wang Shilong32a44782014-02-19 19:24:19 +0800318
319 /*
320 * check if @scrubs_running=@scrubs_paused condition
321 * inside wait_event() is not an atomic operation.
322 * which means we may inc/dec @scrub_running/paused
323 * at any time. Let's wake up @scrub_pause_wait as
324 * much as we can to let commit transaction blocked less.
325 */
326 wake_up(&fs_info->scrub_pause_wait);
327
Stefan Behrensb6bfebc2012-11-02 16:44:58 +0100328 atomic_inc(&sctx->workers_pending);
329}
330
331/* used for workers that require transaction commits */
332static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
333{
334 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
335
336 /*
337 * see scrub_pending_trans_workers_inc() why we're pretending
338 * to be paused in the scrub counters
339 */
340 mutex_lock(&fs_info->scrub_lock);
341 atomic_dec(&fs_info->scrubs_running);
342 atomic_dec(&fs_info->scrubs_paused);
343 mutex_unlock(&fs_info->scrub_lock);
344 atomic_dec(&sctx->workers_pending);
345 wake_up(&fs_info->scrub_pause_wait);
346 wake_up(&sctx->list_wait);
347}
348
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100349static void scrub_free_csums(struct scrub_ctx *sctx)
Arne Jansena2de7332011-03-08 14:14:00 +0100350{
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100351 while (!list_empty(&sctx->csum_list)) {
Arne Jansena2de7332011-03-08 14:14:00 +0100352 struct btrfs_ordered_sum *sum;
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100353 sum = list_first_entry(&sctx->csum_list,
Arne Jansena2de7332011-03-08 14:14:00 +0100354 struct btrfs_ordered_sum, list);
355 list_del(&sum->list);
356 kfree(sum);
357 }
358}
359
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100360static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
Arne Jansena2de7332011-03-08 14:14:00 +0100361{
362 int i;
Arne Jansena2de7332011-03-08 14:14:00 +0100363
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100364 if (!sctx)
Arne Jansena2de7332011-03-08 14:14:00 +0100365 return;
366
Stefan Behrensff023aa2012-11-06 11:43:11 +0100367 scrub_free_wr_ctx(&sctx->wr_ctx);
368
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400369 /* this can happen when scrub is cancelled */
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100370 if (sctx->curr != -1) {
371 struct scrub_bio *sbio = sctx->bios[sctx->curr];
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400372
373 for (i = 0; i < sbio->page_count; i++) {
Stefan Behrensff023aa2012-11-06 11:43:11 +0100374 WARN_ON(!sbio->pagev[i]->page);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400375 scrub_block_put(sbio->pagev[i]->sblock);
376 }
377 bio_put(sbio->bio);
378 }
379
Stefan Behrensff023aa2012-11-06 11:43:11 +0100380 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100381 struct scrub_bio *sbio = sctx->bios[i];
Arne Jansena2de7332011-03-08 14:14:00 +0100382
383 if (!sbio)
384 break;
Arne Jansena2de7332011-03-08 14:14:00 +0100385 kfree(sbio);
386 }
387
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100388 scrub_free_csums(sctx);
389 kfree(sctx);
Arne Jansena2de7332011-03-08 14:14:00 +0100390}
391
392static noinline_for_stack
Stefan Behrens63a212a2012-11-05 18:29:28 +0100393struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
Arne Jansena2de7332011-03-08 14:14:00 +0100394{
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100395 struct scrub_ctx *sctx;
Arne Jansena2de7332011-03-08 14:14:00 +0100396 int i;
Arne Jansena2de7332011-03-08 14:14:00 +0100397 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
Stefan Behrensff023aa2012-11-06 11:43:11 +0100398 int pages_per_rd_bio;
399 int ret;
Arne Jansena2de7332011-03-08 14:14:00 +0100400
Stefan Behrensff023aa2012-11-06 11:43:11 +0100401 /*
402 * the setting of pages_per_rd_bio is correct for scrub but might
403 * be wrong for the dev_replace code where we might read from
404 * different devices in the initial huge bios. However, that
405 * code is able to correctly handle the case when adding a page
406 * to a bio fails.
407 */
408 if (dev->bdev)
409 pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO,
410 bio_get_nr_vecs(dev->bdev));
411 else
412 pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100413 sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
414 if (!sctx)
Arne Jansena2de7332011-03-08 14:14:00 +0100415 goto nomem;
Stefan Behrens63a212a2012-11-05 18:29:28 +0100416 sctx->is_dev_replace = is_dev_replace;
Stefan Behrensff023aa2012-11-06 11:43:11 +0100417 sctx->pages_per_rd_bio = pages_per_rd_bio;
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100418 sctx->curr = -1;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100419 sctx->dev_root = dev->dev_root;
Stefan Behrensff023aa2012-11-06 11:43:11 +0100420 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
Arne Jansena2de7332011-03-08 14:14:00 +0100421 struct scrub_bio *sbio;
422
423 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
424 if (!sbio)
425 goto nomem;
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100426 sctx->bios[i] = sbio;
Arne Jansena2de7332011-03-08 14:14:00 +0100427
Arne Jansena2de7332011-03-08 14:14:00 +0100428 sbio->index = i;
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100429 sbio->sctx = sctx;
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400430 sbio->page_count = 0;
Qu Wenruo0339ef22014-02-28 10:46:17 +0800431 btrfs_init_work(&sbio->work, scrub_bio_end_io_worker,
432 NULL, NULL);
Arne Jansena2de7332011-03-08 14:14:00 +0100433
Stefan Behrensff023aa2012-11-06 11:43:11 +0100434 if (i != SCRUB_BIOS_PER_SCTX - 1)
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100435 sctx->bios[i]->next_free = i + 1;
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200436 else
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100437 sctx->bios[i]->next_free = -1;
Arne Jansena2de7332011-03-08 14:14:00 +0100438 }
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100439 sctx->first_free = 0;
440 sctx->nodesize = dev->dev_root->nodesize;
441 sctx->leafsize = dev->dev_root->leafsize;
442 sctx->sectorsize = dev->dev_root->sectorsize;
Stefan Behrensb6bfebc2012-11-02 16:44:58 +0100443 atomic_set(&sctx->bios_in_flight, 0);
444 atomic_set(&sctx->workers_pending, 0);
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100445 atomic_set(&sctx->cancel_req, 0);
446 sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy);
447 INIT_LIST_HEAD(&sctx->csum_list);
Arne Jansena2de7332011-03-08 14:14:00 +0100448
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100449 spin_lock_init(&sctx->list_lock);
450 spin_lock_init(&sctx->stat_lock);
451 init_waitqueue_head(&sctx->list_wait);
Stefan Behrensff023aa2012-11-06 11:43:11 +0100452
453 ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
454 fs_info->dev_replace.tgtdev, is_dev_replace);
455 if (ret) {
456 scrub_free_ctx(sctx);
457 return ERR_PTR(ret);
458 }
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100459 return sctx;
Arne Jansena2de7332011-03-08 14:14:00 +0100460
461nomem:
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100462 scrub_free_ctx(sctx);
Arne Jansena2de7332011-03-08 14:14:00 +0100463 return ERR_PTR(-ENOMEM);
464}
465
Stefan Behrensff023aa2012-11-06 11:43:11 +0100466static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
467 void *warn_ctx)
Jan Schmidt558540c2011-06-13 19:59:12 +0200468{
469 u64 isize;
470 u32 nlink;
471 int ret;
472 int i;
473 struct extent_buffer *eb;
474 struct btrfs_inode_item *inode_item;
Stefan Behrensff023aa2012-11-06 11:43:11 +0100475 struct scrub_warning *swarn = warn_ctx;
Jan Schmidt558540c2011-06-13 19:59:12 +0200476 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
477 struct inode_fs_paths *ipath = NULL;
478 struct btrfs_root *local_root;
479 struct btrfs_key root_key;
480
481 root_key.objectid = root;
482 root_key.type = BTRFS_ROOT_ITEM_KEY;
483 root_key.offset = (u64)-1;
484 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
485 if (IS_ERR(local_root)) {
486 ret = PTR_ERR(local_root);
487 goto err;
488 }
489
490 ret = inode_item_info(inum, 0, local_root, swarn->path);
491 if (ret) {
492 btrfs_release_path(swarn->path);
493 goto err;
494 }
495
496 eb = swarn->path->nodes[0];
497 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
498 struct btrfs_inode_item);
499 isize = btrfs_inode_size(eb, inode_item);
500 nlink = btrfs_inode_nlink(eb, inode_item);
501 btrfs_release_path(swarn->path);
502
503 ipath = init_ipath(4096, local_root, swarn->path);
Dan Carpenter26bdef52011-11-16 11:28:01 +0300504 if (IS_ERR(ipath)) {
505 ret = PTR_ERR(ipath);
506 ipath = NULL;
507 goto err;
508 }
Jan Schmidt558540c2011-06-13 19:59:12 +0200509 ret = paths_from_inode(inum, ipath);
510
511 if (ret < 0)
512 goto err;
513
514 /*
515 * we deliberately ignore the bit ipath might have been too small to
516 * hold all of the paths here
517 */
518 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
Frank Holtonefe120a2013-12-20 11:37:06 -0500519 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
Jan Schmidt558540c2011-06-13 19:59:12 +0200520 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
521 "length %llu, links %u (path: %s)\n", swarn->errstr,
Josef Bacik606686e2012-06-04 14:03:51 -0400522 swarn->logical, rcu_str_deref(swarn->dev->name),
Jan Schmidt558540c2011-06-13 19:59:12 +0200523 (unsigned long long)swarn->sector, root, inum, offset,
524 min(isize - offset, (u64)PAGE_SIZE), nlink,
Jeff Mahoney745c4d82011-11-20 07:31:57 -0500525 (char *)(unsigned long)ipath->fspath->val[i]);
Jan Schmidt558540c2011-06-13 19:59:12 +0200526
527 free_ipath(ipath);
528 return 0;
529
530err:
Frank Holtonefe120a2013-12-20 11:37:06 -0500531 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
Jan Schmidt558540c2011-06-13 19:59:12 +0200532 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
533 "resolving failed with ret=%d\n", swarn->errstr,
Josef Bacik606686e2012-06-04 14:03:51 -0400534 swarn->logical, rcu_str_deref(swarn->dev->name),
Jan Schmidt558540c2011-06-13 19:59:12 +0200535 (unsigned long long)swarn->sector, root, inum, offset, ret);
536
537 free_ipath(ipath);
538 return 0;
539}
540
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400541static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
Jan Schmidt558540c2011-06-13 19:59:12 +0200542{
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100543 struct btrfs_device *dev;
544 struct btrfs_fs_info *fs_info;
Jan Schmidt558540c2011-06-13 19:59:12 +0200545 struct btrfs_path *path;
546 struct btrfs_key found_key;
547 struct extent_buffer *eb;
548 struct btrfs_extent_item *ei;
549 struct scrub_warning swarn;
Jan Schmidt558540c2011-06-13 19:59:12 +0200550 unsigned long ptr = 0;
Jan Schmidt4692cf52011-12-02 14:56:41 +0100551 u64 extent_item_pos;
Liu Bo69917e42012-09-07 20:01:28 -0600552 u64 flags = 0;
553 u64 ref_root;
554 u32 item_size;
555 u8 ref_level;
556 const int bufsize = 4096;
557 int ret;
Jan Schmidt558540c2011-06-13 19:59:12 +0200558
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100559 WARN_ON(sblock->page_count < 1);
Stefan Behrens7a9e9982012-11-02 14:58:04 +0100560 dev = sblock->pagev[0]->dev;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100561 fs_info = sblock->sctx->dev_root->fs_info;
562
Jan Schmidt558540c2011-06-13 19:59:12 +0200563 path = btrfs_alloc_path();
564
565 swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS);
566 swarn.msg_buf = kmalloc(bufsize, GFP_NOFS);
Stefan Behrens7a9e9982012-11-02 14:58:04 +0100567 swarn.sector = (sblock->pagev[0]->physical) >> 9;
568 swarn.logical = sblock->pagev[0]->logical;
Jan Schmidt558540c2011-06-13 19:59:12 +0200569 swarn.errstr = errstr;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100570 swarn.dev = NULL;
Jan Schmidt558540c2011-06-13 19:59:12 +0200571 swarn.msg_bufsize = bufsize;
572 swarn.scratch_bufsize = bufsize;
573
574 if (!path || !swarn.scratch_buf || !swarn.msg_buf)
575 goto out;
576
Liu Bo69917e42012-09-07 20:01:28 -0600577 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
578 &flags);
Jan Schmidt558540c2011-06-13 19:59:12 +0200579 if (ret < 0)
580 goto out;
581
Jan Schmidt4692cf52011-12-02 14:56:41 +0100582 extent_item_pos = swarn.logical - found_key.objectid;
Jan Schmidt558540c2011-06-13 19:59:12 +0200583 swarn.extent_item_size = found_key.offset;
584
585 eb = path->nodes[0];
586 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
587 item_size = btrfs_item_size_nr(eb, path->slots[0]);
588
Liu Bo69917e42012-09-07 20:01:28 -0600589 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
Jan Schmidt558540c2011-06-13 19:59:12 +0200590 do {
Liu Bo6eda71d2014-06-09 10:54:07 +0800591 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
592 item_size, &ref_root,
593 &ref_level);
Josef Bacik606686e2012-06-04 14:03:51 -0400594 printk_in_rcu(KERN_WARNING
Frank Holtonefe120a2013-12-20 11:37:06 -0500595 "BTRFS: %s at logical %llu on dev %s, "
Jan Schmidt558540c2011-06-13 19:59:12 +0200596 "sector %llu: metadata %s (level %d) in tree "
Josef Bacik606686e2012-06-04 14:03:51 -0400597 "%llu\n", errstr, swarn.logical,
598 rcu_str_deref(dev->name),
Jan Schmidt558540c2011-06-13 19:59:12 +0200599 (unsigned long long)swarn.sector,
600 ref_level ? "node" : "leaf",
601 ret < 0 ? -1 : ref_level,
602 ret < 0 ? -1 : ref_root);
603 } while (ret != 1);
Josef Bacikd8fe29e2013-03-29 08:09:34 -0600604 btrfs_release_path(path);
Jan Schmidt558540c2011-06-13 19:59:12 +0200605 } else {
Josef Bacikd8fe29e2013-03-29 08:09:34 -0600606 btrfs_release_path(path);
Jan Schmidt558540c2011-06-13 19:59:12 +0200607 swarn.path = path;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100608 swarn.dev = dev;
Jan Schmidt7a3ae2f2012-03-23 17:32:28 +0100609 iterate_extent_inodes(fs_info, found_key.objectid,
610 extent_item_pos, 1,
Jan Schmidt558540c2011-06-13 19:59:12 +0200611 scrub_print_warning_inode, &swarn);
612 }
613
614out:
615 btrfs_free_path(path);
616 kfree(swarn.scratch_buf);
617 kfree(swarn.msg_buf);
618}
619
Stefan Behrensff023aa2012-11-06 11:43:11 +0100620static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200621{
Jan Schmidt5da6fcb2011-08-04 18:11:04 +0200622 struct page *page = NULL;
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200623 unsigned long index;
Stefan Behrensff023aa2012-11-06 11:43:11 +0100624 struct scrub_fixup_nodatasum *fixup = fixup_ctx;
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200625 int ret;
Jan Schmidt5da6fcb2011-08-04 18:11:04 +0200626 int corrected = 0;
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200627 struct btrfs_key key;
Jan Schmidt5da6fcb2011-08-04 18:11:04 +0200628 struct inode *inode = NULL;
Liu Bo6f1c3602013-01-29 03:22:10 +0000629 struct btrfs_fs_info *fs_info;
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200630 u64 end = offset + PAGE_SIZE - 1;
631 struct btrfs_root *local_root;
Liu Bo6f1c3602013-01-29 03:22:10 +0000632 int srcu_index;
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200633
634 key.objectid = root;
635 key.type = BTRFS_ROOT_ITEM_KEY;
636 key.offset = (u64)-1;
Liu Bo6f1c3602013-01-29 03:22:10 +0000637
638 fs_info = fixup->root->fs_info;
639 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
640
641 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
642 if (IS_ERR(local_root)) {
643 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200644 return PTR_ERR(local_root);
Liu Bo6f1c3602013-01-29 03:22:10 +0000645 }
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200646
647 key.type = BTRFS_INODE_ITEM_KEY;
648 key.objectid = inum;
649 key.offset = 0;
Liu Bo6f1c3602013-01-29 03:22:10 +0000650 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
651 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200652 if (IS_ERR(inode))
653 return PTR_ERR(inode);
654
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200655 index = offset >> PAGE_CACHE_SHIFT;
656
657 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
Jan Schmidt5da6fcb2011-08-04 18:11:04 +0200658 if (!page) {
659 ret = -ENOMEM;
660 goto out;
661 }
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200662
Jan Schmidt5da6fcb2011-08-04 18:11:04 +0200663 if (PageUptodate(page)) {
Jan Schmidt5da6fcb2011-08-04 18:11:04 +0200664 if (PageDirty(page)) {
665 /*
666 * we need to write the data to the defect sector. the
667 * data that was in that sector is not in memory,
668 * because the page was modified. we must not write the
669 * modified page to that sector.
670 *
671 * TODO: what could be done here: wait for the delalloc
672 * runner to write out that page (might involve
673 * COW) and see whether the sector is still
674 * referenced afterwards.
675 *
676 * For the meantime, we'll treat this error
677 * incorrectable, although there is a chance that a
678 * later scrub will find the bad sector again and that
679 * there's no dirty page in memory, then.
680 */
681 ret = -EIO;
682 goto out;
683 }
Stefan Behrens3ec706c2012-11-05 15:46:42 +0100684 fs_info = BTRFS_I(inode)->root->fs_info;
685 ret = repair_io_failure(fs_info, offset, PAGE_SIZE,
Jan Schmidt5da6fcb2011-08-04 18:11:04 +0200686 fixup->logical, page,
687 fixup->mirror_num);
688 unlock_page(page);
689 corrected = !ret;
690 } else {
691 /*
692 * we need to get good data first. the general readpage path
693 * will call repair_io_failure for us, we just have to make
694 * sure we read the bad mirror.
695 */
696 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
697 EXTENT_DAMAGED, GFP_NOFS);
698 if (ret) {
699 /* set_extent_bits should give proper error */
700 WARN_ON(ret > 0);
701 if (ret > 0)
702 ret = -EFAULT;
703 goto out;
704 }
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200705
Jan Schmidt5da6fcb2011-08-04 18:11:04 +0200706 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
707 btrfs_get_extent,
708 fixup->mirror_num);
709 wait_on_page_locked(page);
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200710
Jan Schmidt5da6fcb2011-08-04 18:11:04 +0200711 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
712 end, EXTENT_DAMAGED, 0, NULL);
713 if (!corrected)
714 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
715 EXTENT_DAMAGED, GFP_NOFS);
716 }
717
718out:
719 if (page)
720 put_page(page);
Tobias Klauser7fb18a02014-04-25 14:58:05 +0200721
722 iput(inode);
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200723
724 if (ret < 0)
725 return ret;
726
727 if (ret == 0 && corrected) {
728 /*
729 * we only need to call readpage for one of the inodes belonging
730 * to this extent. so make iterate_extent_inodes stop
731 */
732 return 1;
733 }
734
735 return -EIO;
736}
737
738static void scrub_fixup_nodatasum(struct btrfs_work *work)
739{
740 int ret;
741 struct scrub_fixup_nodatasum *fixup;
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100742 struct scrub_ctx *sctx;
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200743 struct btrfs_trans_handle *trans = NULL;
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200744 struct btrfs_path *path;
745 int uncorrectable = 0;
746
747 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100748 sctx = fixup->sctx;
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200749
750 path = btrfs_alloc_path();
751 if (!path) {
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100752 spin_lock(&sctx->stat_lock);
753 ++sctx->stat.malloc_errors;
754 spin_unlock(&sctx->stat_lock);
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200755 uncorrectable = 1;
756 goto out;
757 }
758
759 trans = btrfs_join_transaction(fixup->root);
760 if (IS_ERR(trans)) {
761 uncorrectable = 1;
762 goto out;
763 }
764
765 /*
766 * the idea is to trigger a regular read through the standard path. we
767 * read a page from the (failed) logical address by specifying the
768 * corresponding copynum of the failed sector. thus, that readpage is
769 * expected to fail.
770 * that is the point where on-the-fly error correction will kick in
771 * (once it's finished) and rewrite the failed sector if a good copy
772 * can be found.
773 */
774 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
775 path, scrub_fixup_readpage,
776 fixup);
777 if (ret < 0) {
778 uncorrectable = 1;
779 goto out;
780 }
781 WARN_ON(ret != 1);
782
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100783 spin_lock(&sctx->stat_lock);
784 ++sctx->stat.corrected_errors;
785 spin_unlock(&sctx->stat_lock);
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200786
787out:
788 if (trans && !IS_ERR(trans))
789 btrfs_end_transaction(trans, fixup->root);
790 if (uncorrectable) {
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100791 spin_lock(&sctx->stat_lock);
792 ++sctx->stat.uncorrectable_errors;
793 spin_unlock(&sctx->stat_lock);
Stefan Behrensff023aa2012-11-06 11:43:11 +0100794 btrfs_dev_replace_stats_inc(
795 &sctx->dev_root->fs_info->dev_replace.
796 num_uncorrectable_read_errors);
Frank Holtonefe120a2013-12-20 11:37:06 -0500797 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
798 "unable to fixup (nodatasum) error at logical %llu on dev %s\n",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +0200799 fixup->logical, rcu_str_deref(fixup->dev->name));
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200800 }
801
802 btrfs_free_path(path);
803 kfree(fixup);
804
Stefan Behrensb6bfebc2012-11-02 16:44:58 +0100805 scrub_pending_trans_workers_dec(sctx);
Jan Schmidt0ef8e452011-06-13 20:04:15 +0200806}
807
Arne Jansena2de7332011-03-08 14:14:00 +0100808/*
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400809 * scrub_handle_errored_block gets called when either verification of the
810 * pages failed or the bio failed to read, e.g. with EIO. In the latter
811 * case, this function handles all pages in the bio, even though only one
812 * may be bad.
813 * The goal of this function is to repair the errored block by using the
814 * contents of one of the mirrors.
Arne Jansena2de7332011-03-08 14:14:00 +0100815 */
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400816static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
Arne Jansena2de7332011-03-08 14:14:00 +0100817{
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100818 struct scrub_ctx *sctx = sblock_to_check->sctx;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100819 struct btrfs_device *dev;
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400820 struct btrfs_fs_info *fs_info;
Arne Jansena2de7332011-03-08 14:14:00 +0100821 u64 length;
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400822 u64 logical;
823 u64 generation;
824 unsigned int failed_mirror_index;
825 unsigned int is_metadata;
826 unsigned int have_csum;
827 u8 *csum;
828 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
829 struct scrub_block *sblock_bad;
Arne Jansena2de7332011-03-08 14:14:00 +0100830 int ret;
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400831 int mirror_index;
832 int page_num;
833 int success;
834 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
835 DEFAULT_RATELIMIT_BURST);
Arne Jansena2de7332011-03-08 14:14:00 +0100836
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400837 BUG_ON(sblock_to_check->page_count < 1);
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100838 fs_info = sctx->dev_root->fs_info;
Stefan Behrens4ded4f62012-11-14 18:57:29 +0000839 if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) {
840 /*
841 * if we find an error in a super block, we just report it.
842 * They will get written with the next transaction commit
843 * anyway
844 */
845 spin_lock(&sctx->stat_lock);
846 ++sctx->stat.super_errors;
847 spin_unlock(&sctx->stat_lock);
848 return 0;
849 }
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400850 length = sblock_to_check->page_count * PAGE_SIZE;
Stefan Behrens7a9e9982012-11-02 14:58:04 +0100851 logical = sblock_to_check->pagev[0]->logical;
852 generation = sblock_to_check->pagev[0]->generation;
853 BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
854 failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
855 is_metadata = !(sblock_to_check->pagev[0]->flags &
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400856 BTRFS_EXTENT_FLAG_DATA);
Stefan Behrens7a9e9982012-11-02 14:58:04 +0100857 have_csum = sblock_to_check->pagev[0]->have_csum;
858 csum = sblock_to_check->pagev[0]->csum;
859 dev = sblock_to_check->pagev[0]->dev;
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400860
Stefan Behrensff023aa2012-11-06 11:43:11 +0100861 if (sctx->is_dev_replace && !is_metadata && !have_csum) {
862 sblocks_for_recheck = NULL;
863 goto nodatasum_case;
864 }
865
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400866 /*
867 * read all mirrors one after the other. This includes to
868 * re-read the extent or metadata block that failed (that was
869 * the cause that this fixup code is called) another time,
870 * page by page this time in order to know which pages
871 * caused I/O errors and which ones are good (for all mirrors).
872 * It is the goal to handle the situation when more than one
873 * mirror contains I/O errors, but the errors do not
874 * overlap, i.e. the data can be repaired by selecting the
875 * pages from those mirrors without I/O error on the
876 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
877 * would be that mirror #1 has an I/O error on the first page,
878 * the second page is good, and mirror #2 has an I/O error on
879 * the second page, but the first page is good.
880 * Then the first page of the first mirror can be repaired by
881 * taking the first page of the second mirror, and the
882 * second page of the second mirror can be repaired by
883 * copying the contents of the 2nd page of the 1st mirror.
884 * One more note: if the pages of one mirror contain I/O
885 * errors, the checksum cannot be verified. In order to get
886 * the best data for repairing, the first attempt is to find
887 * a mirror without I/O errors and with a validated checksum.
888 * Only if this is not possible, the pages are picked from
889 * mirrors with I/O errors without considering the checksum.
890 * If the latter is the case, at the end, the checksum of the
891 * repaired area is verified in order to correctly maintain
892 * the statistics.
893 */
894
895 sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS *
896 sizeof(*sblocks_for_recheck),
897 GFP_NOFS);
898 if (!sblocks_for_recheck) {
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100899 spin_lock(&sctx->stat_lock);
900 sctx->stat.malloc_errors++;
901 sctx->stat.read_errors++;
902 sctx->stat.uncorrectable_errors++;
903 spin_unlock(&sctx->stat_lock);
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100904 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400905 goto out;
906 }
907
908 /* setup the context, map the logical blocks and alloc the pages */
Stefan Behrensff023aa2012-11-06 11:43:11 +0100909 ret = scrub_setup_recheck_block(sctx, fs_info, sblock_to_check, length,
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400910 logical, sblocks_for_recheck);
911 if (ret) {
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100912 spin_lock(&sctx->stat_lock);
913 sctx->stat.read_errors++;
914 sctx->stat.uncorrectable_errors++;
915 spin_unlock(&sctx->stat_lock);
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100916 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400917 goto out;
918 }
919 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
920 sblock_bad = sblocks_for_recheck + failed_mirror_index;
921
922 /* build and submit the bios for the failed mirror, check checksums */
Stefan Behrens34f5c8e2012-11-02 16:16:26 +0100923 scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
924 csum, generation, sctx->csum_size);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400925
926 if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
927 sblock_bad->no_io_error_seen) {
928 /*
929 * the error disappeared after reading page by page, or
930 * the area was part of a huge bio and other parts of the
931 * bio caused I/O errors, or the block layer merged several
932 * read requests into one and the error is caused by a
933 * different bio (usually one of the two latter cases is
934 * the cause)
935 */
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100936 spin_lock(&sctx->stat_lock);
937 sctx->stat.unverified_errors++;
938 spin_unlock(&sctx->stat_lock);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400939
Stefan Behrensff023aa2012-11-06 11:43:11 +0100940 if (sctx->is_dev_replace)
941 scrub_write_block_to_dev_replace(sblock_bad);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400942 goto out;
943 }
944
945 if (!sblock_bad->no_io_error_seen) {
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100946 spin_lock(&sctx->stat_lock);
947 sctx->stat.read_errors++;
948 spin_unlock(&sctx->stat_lock);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400949 if (__ratelimit(&_rs))
950 scrub_print_warning("i/o error", sblock_to_check);
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100951 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400952 } else if (sblock_bad->checksum_error) {
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100953 spin_lock(&sctx->stat_lock);
954 sctx->stat.csum_errors++;
955 spin_unlock(&sctx->stat_lock);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400956 if (__ratelimit(&_rs))
957 scrub_print_warning("checksum error", sblock_to_check);
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100958 btrfs_dev_stat_inc_and_print(dev,
Stefan Behrens442a4f62012-05-25 16:06:08 +0200959 BTRFS_DEV_STAT_CORRUPTION_ERRS);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400960 } else if (sblock_bad->header_error) {
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100961 spin_lock(&sctx->stat_lock);
962 sctx->stat.verify_errors++;
963 spin_unlock(&sctx->stat_lock);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400964 if (__ratelimit(&_rs))
965 scrub_print_warning("checksum/header error",
966 sblock_to_check);
Stefan Behrens442a4f62012-05-25 16:06:08 +0200967 if (sblock_bad->generation_error)
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100968 btrfs_dev_stat_inc_and_print(dev,
Stefan Behrens442a4f62012-05-25 16:06:08 +0200969 BTRFS_DEV_STAT_GENERATION_ERRS);
970 else
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100971 btrfs_dev_stat_inc_and_print(dev,
Stefan Behrens442a4f62012-05-25 16:06:08 +0200972 BTRFS_DEV_STAT_CORRUPTION_ERRS);
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400973 }
974
Ilya Dryomov33ef30a2013-11-03 19:06:38 +0200975 if (sctx->readonly) {
976 ASSERT(!sctx->is_dev_replace);
977 goto out;
978 }
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400979
980 if (!is_metadata && !have_csum) {
981 struct scrub_fixup_nodatasum *fixup_nodatasum;
982
Stefan Behrensff023aa2012-11-06 11:43:11 +0100983nodatasum_case:
984 WARN_ON(sctx->is_dev_replace);
985
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400986 /*
987 * !is_metadata and !have_csum, this means that the data
988 * might not be COW'ed, that it might be modified
989 * concurrently. The general strategy to work on the
990 * commit root does not help in the case when COW is not
991 * used.
992 */
993 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
994 if (!fixup_nodatasum)
995 goto did_not_correct_error;
Stefan Behrensd9d181c2012-11-02 09:58:09 +0100996 fixup_nodatasum->sctx = sctx;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +0100997 fixup_nodatasum->dev = dev;
Stefan Behrensb5d67f62012-03-27 14:21:27 -0400998 fixup_nodatasum->logical = logical;
999 fixup_nodatasum->root = fs_info->extent_root;
1000 fixup_nodatasum->mirror_num = failed_mirror_index + 1;
Stefan Behrensb6bfebc2012-11-02 16:44:58 +01001001 scrub_pending_trans_workers_inc(sctx);
Qu Wenruo0339ef22014-02-28 10:46:17 +08001002 btrfs_init_work(&fixup_nodatasum->work, scrub_fixup_nodatasum,
1003 NULL, NULL);
1004 btrfs_queue_work(fs_info->scrub_workers,
1005 &fixup_nodatasum->work);
Arne Jansena2de7332011-03-08 14:14:00 +01001006 goto out;
1007 }
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001008
1009 /*
1010 * now build and submit the bios for the other mirrors, check
Stefan Behrenscb2ced72012-11-02 16:14:21 +01001011 * checksums.
1012 * First try to pick the mirror which is completely without I/O
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001013 * errors and also does not have a checksum error.
1014 * If one is found, and if a checksum is present, the full block
1015 * that is known to contain an error is rewritten. Afterwards
1016 * the block is known to be corrected.
1017 * If a mirror is found which is completely correct, and no
1018 * checksum is present, only those pages are rewritten that had
1019 * an I/O error in the block to be repaired, since it cannot be
1020 * determined, which copy of the other pages is better (and it
1021 * could happen otherwise that a correct page would be
1022 * overwritten by a bad one).
1023 */
1024 for (mirror_index = 0;
1025 mirror_index < BTRFS_MAX_MIRRORS &&
1026 sblocks_for_recheck[mirror_index].page_count > 0;
1027 mirror_index++) {
Stefan Behrenscb2ced72012-11-02 16:14:21 +01001028 struct scrub_block *sblock_other;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001029
Stefan Behrenscb2ced72012-11-02 16:14:21 +01001030 if (mirror_index == failed_mirror_index)
1031 continue;
1032 sblock_other = sblocks_for_recheck + mirror_index;
1033
1034 /* build and submit the bios, check checksums */
Stefan Behrens34f5c8e2012-11-02 16:16:26 +01001035 scrub_recheck_block(fs_info, sblock_other, is_metadata,
1036 have_csum, csum, generation,
1037 sctx->csum_size);
1038
1039 if (!sblock_other->header_error &&
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001040 !sblock_other->checksum_error &&
1041 sblock_other->no_io_error_seen) {
Stefan Behrensff023aa2012-11-06 11:43:11 +01001042 if (sctx->is_dev_replace) {
1043 scrub_write_block_to_dev_replace(sblock_other);
1044 } else {
1045 int force_write = is_metadata || have_csum;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001046
Stefan Behrensff023aa2012-11-06 11:43:11 +01001047 ret = scrub_repair_block_from_good_copy(
1048 sblock_bad, sblock_other,
1049 force_write);
1050 }
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001051 if (0 == ret)
1052 goto corrected_error;
Arne Jansena2de7332011-03-08 14:14:00 +01001053 }
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001054 }
1055
1056 /*
Stefan Behrensff023aa2012-11-06 11:43:11 +01001057 * for dev_replace, pick good pages and write to the target device.
1058 */
1059 if (sctx->is_dev_replace) {
1060 success = 1;
1061 for (page_num = 0; page_num < sblock_bad->page_count;
1062 page_num++) {
1063 int sub_success;
1064
1065 sub_success = 0;
1066 for (mirror_index = 0;
1067 mirror_index < BTRFS_MAX_MIRRORS &&
1068 sblocks_for_recheck[mirror_index].page_count > 0;
1069 mirror_index++) {
1070 struct scrub_block *sblock_other =
1071 sblocks_for_recheck + mirror_index;
1072 struct scrub_page *page_other =
1073 sblock_other->pagev[page_num];
1074
1075 if (!page_other->io_error) {
1076 ret = scrub_write_page_to_dev_replace(
1077 sblock_other, page_num);
1078 if (ret == 0) {
1079 /* succeeded for this page */
1080 sub_success = 1;
1081 break;
1082 } else {
1083 btrfs_dev_replace_stats_inc(
1084 &sctx->dev_root->
1085 fs_info->dev_replace.
1086 num_write_errors);
1087 }
1088 }
1089 }
1090
1091 if (!sub_success) {
1092 /*
1093 * did not find a mirror to fetch the page
1094 * from. scrub_write_page_to_dev_replace()
1095 * handles this case (page->io_error), by
1096 * filling the block with zeros before
1097 * submitting the write request
1098 */
1099 success = 0;
1100 ret = scrub_write_page_to_dev_replace(
1101 sblock_bad, page_num);
1102 if (ret)
1103 btrfs_dev_replace_stats_inc(
1104 &sctx->dev_root->fs_info->
1105 dev_replace.num_write_errors);
1106 }
1107 }
1108
1109 goto out;
1110 }
1111
1112 /*
1113 * for regular scrub, repair those pages that are errored.
1114 * In case of I/O errors in the area that is supposed to be
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001115 * repaired, continue by picking good copies of those pages.
1116 * Select the good pages from mirrors to rewrite bad pages from
1117 * the area to fix. Afterwards verify the checksum of the block
1118 * that is supposed to be repaired. This verification step is
1119 * only done for the purpose of statistic counting and for the
1120 * final scrub report, whether errors remain.
1121 * A perfect algorithm could make use of the checksum and try
1122 * all possible combinations of pages from the different mirrors
1123 * until the checksum verification succeeds. For example, when
1124 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1125 * of mirror #2 is readable but the final checksum test fails,
1126 * then the 2nd page of mirror #3 could be tried, whether now
1127 * the final checksum succeedes. But this would be a rare
1128 * exception and is therefore not implemented. At least it is
1129 * avoided that the good copy is overwritten.
1130 * A more useful improvement would be to pick the sectors
1131 * without I/O error based on sector sizes (512 bytes on legacy
1132 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1133 * mirror could be repaired by taking 512 byte of a different
1134 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1135 * area are unreadable.
1136 */
1137
1138 /* can only fix I/O errors from here on */
1139 if (sblock_bad->no_io_error_seen)
1140 goto did_not_correct_error;
1141
1142 success = 1;
1143 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001144 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001145
1146 if (!page_bad->io_error)
1147 continue;
1148
1149 for (mirror_index = 0;
1150 mirror_index < BTRFS_MAX_MIRRORS &&
1151 sblocks_for_recheck[mirror_index].page_count > 0;
1152 mirror_index++) {
1153 struct scrub_block *sblock_other = sblocks_for_recheck +
1154 mirror_index;
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001155 struct scrub_page *page_other = sblock_other->pagev[
1156 page_num];
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001157
1158 if (!page_other->io_error) {
1159 ret = scrub_repair_page_from_good_copy(
1160 sblock_bad, sblock_other, page_num, 0);
1161 if (0 == ret) {
1162 page_bad->io_error = 0;
1163 break; /* succeeded for this page */
1164 }
Jan Schmidt13db62b2011-06-13 19:56:13 +02001165 }
1166 }
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001167
1168 if (page_bad->io_error) {
1169 /* did not find a mirror to copy the page from */
1170 success = 0;
1171 }
1172 }
1173
1174 if (success) {
1175 if (is_metadata || have_csum) {
1176 /*
1177 * need to verify the checksum now that all
1178 * sectors on disk are repaired (the write
1179 * request for data to be repaired is on its way).
1180 * Just be lazy and use scrub_recheck_block()
1181 * which re-reads the data before the checksum
1182 * is verified, but most likely the data comes out
1183 * of the page cache.
1184 */
Stefan Behrens34f5c8e2012-11-02 16:16:26 +01001185 scrub_recheck_block(fs_info, sblock_bad,
1186 is_metadata, have_csum, csum,
1187 generation, sctx->csum_size);
1188 if (!sblock_bad->header_error &&
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001189 !sblock_bad->checksum_error &&
1190 sblock_bad->no_io_error_seen)
1191 goto corrected_error;
1192 else
1193 goto did_not_correct_error;
1194 } else {
1195corrected_error:
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001196 spin_lock(&sctx->stat_lock);
1197 sctx->stat.corrected_errors++;
1198 spin_unlock(&sctx->stat_lock);
Josef Bacik606686e2012-06-04 14:03:51 -04001199 printk_ratelimited_in_rcu(KERN_ERR
Frank Holtonefe120a2013-12-20 11:37:06 -05001200 "BTRFS: fixed up error at logical %llu on dev %s\n",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02001201 logical, rcu_str_deref(dev->name));
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001202 }
1203 } else {
1204did_not_correct_error:
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001205 spin_lock(&sctx->stat_lock);
1206 sctx->stat.uncorrectable_errors++;
1207 spin_unlock(&sctx->stat_lock);
Josef Bacik606686e2012-06-04 14:03:51 -04001208 printk_ratelimited_in_rcu(KERN_ERR
Frank Holtonefe120a2013-12-20 11:37:06 -05001209 "BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02001210 logical, rcu_str_deref(dev->name));
Arne Jansena2de7332011-03-08 14:14:00 +01001211 }
1212
1213out:
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001214 if (sblocks_for_recheck) {
1215 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
1216 mirror_index++) {
1217 struct scrub_block *sblock = sblocks_for_recheck +
1218 mirror_index;
1219 int page_index;
1220
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001221 for (page_index = 0; page_index < sblock->page_count;
1222 page_index++) {
1223 sblock->pagev[page_index]->sblock = NULL;
1224 scrub_page_put(sblock->pagev[page_index]);
1225 }
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001226 }
1227 kfree(sblocks_for_recheck);
1228 }
1229
1230 return 0;
Arne Jansena2de7332011-03-08 14:14:00 +01001231}
1232
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001233static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
Stefan Behrens3ec706c2012-11-05 15:46:42 +01001234 struct btrfs_fs_info *fs_info,
Stefan Behrensff023aa2012-11-06 11:43:11 +01001235 struct scrub_block *original_sblock,
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001236 u64 length, u64 logical,
1237 struct scrub_block *sblocks_for_recheck)
Arne Jansena2de7332011-03-08 14:14:00 +01001238{
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001239 int page_index;
1240 int mirror_index;
1241 int ret;
1242
1243 /*
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001244 * note: the two members ref_count and outstanding_pages
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001245 * are not used (and not set) in the blocks that are used for
1246 * the recheck procedure
1247 */
1248
1249 page_index = 0;
1250 while (length > 0) {
1251 u64 sublen = min_t(u64, length, PAGE_SIZE);
1252 u64 mapped_length = sublen;
1253 struct btrfs_bio *bbio = NULL;
1254
1255 /*
1256 * with a length of PAGE_SIZE, each returned stripe
1257 * represents one mirror
1258 */
Stefan Behrens29a8d9a2012-11-06 14:16:24 +01001259 ret = btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, logical,
1260 &mapped_length, &bbio, 0);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001261 if (ret || !bbio || mapped_length < sublen) {
1262 kfree(bbio);
1263 return -EIO;
1264 }
1265
Stefan Behrensff023aa2012-11-06 11:43:11 +01001266 BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001267 for (mirror_index = 0; mirror_index < (int)bbio->num_stripes;
1268 mirror_index++) {
1269 struct scrub_block *sblock;
1270 struct scrub_page *page;
1271
1272 if (mirror_index >= BTRFS_MAX_MIRRORS)
1273 continue;
1274
1275 sblock = sblocks_for_recheck + mirror_index;
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001276 sblock->sctx = sctx;
1277 page = kzalloc(sizeof(*page), GFP_NOFS);
1278 if (!page) {
1279leave_nomem:
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001280 spin_lock(&sctx->stat_lock);
1281 sctx->stat.malloc_errors++;
1282 spin_unlock(&sctx->stat_lock);
Wei Yongjuncf93dcc2012-09-02 07:44:51 -06001283 kfree(bbio);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001284 return -ENOMEM;
1285 }
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001286 scrub_page_get(page);
1287 sblock->pagev[page_index] = page;
1288 page->logical = logical;
1289 page->physical = bbio->stripes[mirror_index].physical;
Stefan Behrensff023aa2012-11-06 11:43:11 +01001290 BUG_ON(page_index >= original_sblock->page_count);
1291 page->physical_for_dev_replace =
1292 original_sblock->pagev[page_index]->
1293 physical_for_dev_replace;
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001294 /* for missing devices, dev->bdev is NULL */
1295 page->dev = bbio->stripes[mirror_index].dev;
1296 page->mirror_num = mirror_index + 1;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001297 sblock->page_count++;
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001298 page->page = alloc_page(GFP_NOFS);
1299 if (!page->page)
1300 goto leave_nomem;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001301 }
1302 kfree(bbio);
1303 length -= sublen;
1304 logical += sublen;
1305 page_index++;
1306 }
1307
1308 return 0;
1309}
1310
1311/*
1312 * this function will check the on disk data for checksum errors, header
1313 * errors and read I/O errors. If any I/O errors happen, the exact pages
1314 * which are errored are marked as being bad. The goal is to enable scrub
1315 * to take those pages that are not errored from all the mirrors so that
1316 * the pages that are errored in the just handled mirror can be repaired.
1317 */
Stefan Behrens34f5c8e2012-11-02 16:16:26 +01001318static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1319 struct scrub_block *sblock, int is_metadata,
1320 int have_csum, u8 *csum, u64 generation,
1321 u16 csum_size)
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001322{
1323 int page_num;
1324
1325 sblock->no_io_error_seen = 1;
1326 sblock->header_error = 0;
1327 sblock->checksum_error = 0;
1328
1329 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1330 struct bio *bio;
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001331 struct scrub_page *page = sblock->pagev[page_num];
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001332
Stefan Behrens442a4f62012-05-25 16:06:08 +02001333 if (page->dev->bdev == NULL) {
Stefan Behrensea9947b2012-05-04 15:16:07 -04001334 page->io_error = 1;
1335 sblock->no_io_error_seen = 0;
1336 continue;
1337 }
1338
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001339 WARN_ON(!page->page);
Chris Mason9be33952013-05-17 18:30:14 -04001340 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
Stefan Behrens34f5c8e2012-11-02 16:16:26 +01001341 if (!bio) {
1342 page->io_error = 1;
1343 sblock->no_io_error_seen = 0;
1344 continue;
1345 }
Stefan Behrens442a4f62012-05-25 16:06:08 +02001346 bio->bi_bdev = page->dev->bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001347 bio->bi_iter.bi_sector = page->physical >> 9;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001348
Stefan Behrens34f5c8e2012-11-02 16:16:26 +01001349 bio_add_page(bio, page->page, PAGE_SIZE, 0);
Kent Overstreet33879d42013-11-23 22:33:32 -08001350 if (btrfsic_submit_bio_wait(READ, bio))
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001351 sblock->no_io_error_seen = 0;
Kent Overstreet33879d42013-11-23 22:33:32 -08001352
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001353 bio_put(bio);
1354 }
1355
1356 if (sblock->no_io_error_seen)
1357 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
1358 have_csum, csum, generation,
1359 csum_size);
1360
Stefan Behrens34f5c8e2012-11-02 16:16:26 +01001361 return;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001362}
1363
1364static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
1365 struct scrub_block *sblock,
1366 int is_metadata, int have_csum,
1367 const u8 *csum, u64 generation,
1368 u16 csum_size)
1369{
1370 int page_num;
1371 u8 calculated_csum[BTRFS_CSUM_SIZE];
1372 u32 crc = ~(u32)0;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001373 void *mapped_buffer;
1374
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001375 WARN_ON(!sblock->pagev[0]->page);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001376 if (is_metadata) {
1377 struct btrfs_header *h;
1378
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001379 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001380 h = (struct btrfs_header *)mapped_buffer;
1381
Qu Wenruo3cae2102013-07-16 11:19:18 +08001382 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) ||
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001383 memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE) ||
1384 memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
Stefan Behrens442a4f62012-05-25 16:06:08 +02001385 BTRFS_UUID_SIZE)) {
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001386 sblock->header_error = 1;
Qu Wenruo3cae2102013-07-16 11:19:18 +08001387 } else if (generation != btrfs_stack_header_generation(h)) {
Stefan Behrens442a4f62012-05-25 16:06:08 +02001388 sblock->header_error = 1;
1389 sblock->generation_error = 1;
1390 }
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001391 csum = h->csum;
1392 } else {
1393 if (!have_csum)
1394 return;
1395
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001396 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001397 }
1398
1399 for (page_num = 0;;) {
1400 if (page_num == 0 && is_metadata)
Liu Bob0496682013-03-14 14:57:45 +00001401 crc = btrfs_csum_data(
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001402 ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE,
1403 crc, PAGE_SIZE - BTRFS_CSUM_SIZE);
1404 else
Liu Bob0496682013-03-14 14:57:45 +00001405 crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001406
Linus Torvalds9613beb2012-03-30 12:44:29 -07001407 kunmap_atomic(mapped_buffer);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001408 page_num++;
1409 if (page_num >= sblock->page_count)
1410 break;
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001411 WARN_ON(!sblock->pagev[page_num]->page);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001412
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001413 mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001414 }
1415
1416 btrfs_csum_final(crc, calculated_csum);
1417 if (memcmp(calculated_csum, csum, csum_size))
1418 sblock->checksum_error = 1;
1419}
1420
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001421static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1422 struct scrub_block *sblock_good,
1423 int force_write)
1424{
1425 int page_num;
1426 int ret = 0;
1427
1428 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1429 int ret_sub;
1430
1431 ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
1432 sblock_good,
1433 page_num,
1434 force_write);
1435 if (ret_sub)
1436 ret = ret_sub;
1437 }
1438
1439 return ret;
1440}
1441
1442static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
1443 struct scrub_block *sblock_good,
1444 int page_num, int force_write)
1445{
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001446 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1447 struct scrub_page *page_good = sblock_good->pagev[page_num];
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001448
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001449 BUG_ON(page_bad->page == NULL);
1450 BUG_ON(page_good->page == NULL);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001451 if (force_write || sblock_bad->header_error ||
1452 sblock_bad->checksum_error || page_bad->io_error) {
1453 struct bio *bio;
1454 int ret;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001455
Stefan Behrensff023aa2012-11-06 11:43:11 +01001456 if (!page_bad->dev->bdev) {
Frank Holtonefe120a2013-12-20 11:37:06 -05001457 printk_ratelimited(KERN_WARNING "BTRFS: "
1458 "scrub_repair_page_from_good_copy(bdev == NULL) "
1459 "is unexpected!\n");
Stefan Behrensff023aa2012-11-06 11:43:11 +01001460 return -EIO;
1461 }
1462
Chris Mason9be33952013-05-17 18:30:14 -04001463 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
Tsutomu Itohe627ee72012-04-12 16:03:56 -04001464 if (!bio)
1465 return -EIO;
Stefan Behrens442a4f62012-05-25 16:06:08 +02001466 bio->bi_bdev = page_bad->dev->bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001467 bio->bi_iter.bi_sector = page_bad->physical >> 9;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001468
1469 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
1470 if (PAGE_SIZE != ret) {
1471 bio_put(bio);
1472 return -EIO;
1473 }
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001474
Kent Overstreet33879d42013-11-23 22:33:32 -08001475 if (btrfsic_submit_bio_wait(WRITE, bio)) {
Stefan Behrens442a4f62012-05-25 16:06:08 +02001476 btrfs_dev_stat_inc_and_print(page_bad->dev,
1477 BTRFS_DEV_STAT_WRITE_ERRS);
Stefan Behrensff023aa2012-11-06 11:43:11 +01001478 btrfs_dev_replace_stats_inc(
1479 &sblock_bad->sctx->dev_root->fs_info->
1480 dev_replace.num_write_errors);
Stefan Behrens442a4f62012-05-25 16:06:08 +02001481 bio_put(bio);
1482 return -EIO;
1483 }
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001484 bio_put(bio);
1485 }
1486
1487 return 0;
1488}
1489
Stefan Behrensff023aa2012-11-06 11:43:11 +01001490static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
1491{
1492 int page_num;
1493
1494 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1495 int ret;
1496
1497 ret = scrub_write_page_to_dev_replace(sblock, page_num);
1498 if (ret)
1499 btrfs_dev_replace_stats_inc(
1500 &sblock->sctx->dev_root->fs_info->dev_replace.
1501 num_write_errors);
1502 }
1503}
1504
1505static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
1506 int page_num)
1507{
1508 struct scrub_page *spage = sblock->pagev[page_num];
1509
1510 BUG_ON(spage->page == NULL);
1511 if (spage->io_error) {
1512 void *mapped_buffer = kmap_atomic(spage->page);
1513
1514 memset(mapped_buffer, 0, PAGE_CACHE_SIZE);
1515 flush_dcache_page(spage->page);
1516 kunmap_atomic(mapped_buffer);
1517 }
1518 return scrub_add_page_to_wr_bio(sblock->sctx, spage);
1519}
1520
1521static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
1522 struct scrub_page *spage)
1523{
1524 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1525 struct scrub_bio *sbio;
1526 int ret;
1527
1528 mutex_lock(&wr_ctx->wr_lock);
1529again:
1530 if (!wr_ctx->wr_curr_bio) {
1531 wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio),
1532 GFP_NOFS);
1533 if (!wr_ctx->wr_curr_bio) {
1534 mutex_unlock(&wr_ctx->wr_lock);
1535 return -ENOMEM;
1536 }
1537 wr_ctx->wr_curr_bio->sctx = sctx;
1538 wr_ctx->wr_curr_bio->page_count = 0;
1539 }
1540 sbio = wr_ctx->wr_curr_bio;
1541 if (sbio->page_count == 0) {
1542 struct bio *bio;
1543
1544 sbio->physical = spage->physical_for_dev_replace;
1545 sbio->logical = spage->logical;
1546 sbio->dev = wr_ctx->tgtdev;
1547 bio = sbio->bio;
1548 if (!bio) {
Chris Mason9be33952013-05-17 18:30:14 -04001549 bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
Stefan Behrensff023aa2012-11-06 11:43:11 +01001550 if (!bio) {
1551 mutex_unlock(&wr_ctx->wr_lock);
1552 return -ENOMEM;
1553 }
1554 sbio->bio = bio;
1555 }
1556
1557 bio->bi_private = sbio;
1558 bio->bi_end_io = scrub_wr_bio_end_io;
1559 bio->bi_bdev = sbio->dev->bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001560 bio->bi_iter.bi_sector = sbio->physical >> 9;
Stefan Behrensff023aa2012-11-06 11:43:11 +01001561 sbio->err = 0;
1562 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1563 spage->physical_for_dev_replace ||
1564 sbio->logical + sbio->page_count * PAGE_SIZE !=
1565 spage->logical) {
1566 scrub_wr_submit(sctx);
1567 goto again;
1568 }
1569
1570 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1571 if (ret != PAGE_SIZE) {
1572 if (sbio->page_count < 1) {
1573 bio_put(sbio->bio);
1574 sbio->bio = NULL;
1575 mutex_unlock(&wr_ctx->wr_lock);
1576 return -EIO;
1577 }
1578 scrub_wr_submit(sctx);
1579 goto again;
1580 }
1581
1582 sbio->pagev[sbio->page_count] = spage;
1583 scrub_page_get(spage);
1584 sbio->page_count++;
1585 if (sbio->page_count == wr_ctx->pages_per_wr_bio)
1586 scrub_wr_submit(sctx);
1587 mutex_unlock(&wr_ctx->wr_lock);
1588
1589 return 0;
1590}
1591
1592static void scrub_wr_submit(struct scrub_ctx *sctx)
1593{
1594 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1595 struct scrub_bio *sbio;
1596
1597 if (!wr_ctx->wr_curr_bio)
1598 return;
1599
1600 sbio = wr_ctx->wr_curr_bio;
1601 wr_ctx->wr_curr_bio = NULL;
1602 WARN_ON(!sbio->bio->bi_bdev);
1603 scrub_pending_bio_inc(sctx);
1604 /* process all writes in a single worker thread. Then the block layer
1605 * orders the requests before sending them to the driver which
1606 * doubled the write performance on spinning disks when measured
1607 * with Linux 3.5 */
1608 btrfsic_submit_bio(WRITE, sbio->bio);
1609}
1610
1611static void scrub_wr_bio_end_io(struct bio *bio, int err)
1612{
1613 struct scrub_bio *sbio = bio->bi_private;
1614 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
1615
1616 sbio->err = err;
1617 sbio->bio = bio;
1618
Qu Wenruo0339ef22014-02-28 10:46:17 +08001619 btrfs_init_work(&sbio->work, scrub_wr_bio_end_io_worker, NULL, NULL);
1620 btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
Stefan Behrensff023aa2012-11-06 11:43:11 +01001621}
1622
1623static void scrub_wr_bio_end_io_worker(struct btrfs_work *work)
1624{
1625 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
1626 struct scrub_ctx *sctx = sbio->sctx;
1627 int i;
1628
1629 WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO);
1630 if (sbio->err) {
1631 struct btrfs_dev_replace *dev_replace =
1632 &sbio->sctx->dev_root->fs_info->dev_replace;
1633
1634 for (i = 0; i < sbio->page_count; i++) {
1635 struct scrub_page *spage = sbio->pagev[i];
1636
1637 spage->io_error = 1;
1638 btrfs_dev_replace_stats_inc(&dev_replace->
1639 num_write_errors);
1640 }
1641 }
1642
1643 for (i = 0; i < sbio->page_count; i++)
1644 scrub_page_put(sbio->pagev[i]);
1645
1646 bio_put(sbio->bio);
1647 kfree(sbio);
1648 scrub_pending_bio_dec(sctx);
1649}
1650
1651static int scrub_checksum(struct scrub_block *sblock)
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001652{
1653 u64 flags;
1654 int ret;
1655
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001656 WARN_ON(sblock->page_count < 1);
1657 flags = sblock->pagev[0]->flags;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001658 ret = 0;
1659 if (flags & BTRFS_EXTENT_FLAG_DATA)
1660 ret = scrub_checksum_data(sblock);
1661 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1662 ret = scrub_checksum_tree_block(sblock);
1663 else if (flags & BTRFS_EXTENT_FLAG_SUPER)
1664 (void)scrub_checksum_super(sblock);
1665 else
1666 WARN_ON(1);
1667 if (ret)
1668 scrub_handle_errored_block(sblock);
Stefan Behrensff023aa2012-11-06 11:43:11 +01001669
1670 return ret;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001671}
1672
1673static int scrub_checksum_data(struct scrub_block *sblock)
1674{
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001675 struct scrub_ctx *sctx = sblock->sctx;
Arne Jansena2de7332011-03-08 14:14:00 +01001676 u8 csum[BTRFS_CSUM_SIZE];
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001677 u8 *on_disk_csum;
1678 struct page *page;
1679 void *buffer;
Arne Jansena2de7332011-03-08 14:14:00 +01001680 u32 crc = ~(u32)0;
1681 int fail = 0;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001682 u64 len;
1683 int index;
Arne Jansena2de7332011-03-08 14:14:00 +01001684
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001685 BUG_ON(sblock->page_count < 1);
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001686 if (!sblock->pagev[0]->have_csum)
Arne Jansena2de7332011-03-08 14:14:00 +01001687 return 0;
1688
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001689 on_disk_csum = sblock->pagev[0]->csum;
1690 page = sblock->pagev[0]->page;
Linus Torvalds9613beb2012-03-30 12:44:29 -07001691 buffer = kmap_atomic(page);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001692
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001693 len = sctx->sectorsize;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001694 index = 0;
1695 for (;;) {
1696 u64 l = min_t(u64, len, PAGE_SIZE);
1697
Liu Bob0496682013-03-14 14:57:45 +00001698 crc = btrfs_csum_data(buffer, crc, l);
Linus Torvalds9613beb2012-03-30 12:44:29 -07001699 kunmap_atomic(buffer);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001700 len -= l;
1701 if (len == 0)
1702 break;
1703 index++;
1704 BUG_ON(index >= sblock->page_count);
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001705 BUG_ON(!sblock->pagev[index]->page);
1706 page = sblock->pagev[index]->page;
Linus Torvalds9613beb2012-03-30 12:44:29 -07001707 buffer = kmap_atomic(page);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001708 }
1709
Arne Jansena2de7332011-03-08 14:14:00 +01001710 btrfs_csum_final(crc, csum);
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001711 if (memcmp(csum, on_disk_csum, sctx->csum_size))
Arne Jansena2de7332011-03-08 14:14:00 +01001712 fail = 1;
1713
Arne Jansena2de7332011-03-08 14:14:00 +01001714 return fail;
1715}
1716
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001717static int scrub_checksum_tree_block(struct scrub_block *sblock)
Arne Jansena2de7332011-03-08 14:14:00 +01001718{
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001719 struct scrub_ctx *sctx = sblock->sctx;
Arne Jansena2de7332011-03-08 14:14:00 +01001720 struct btrfs_header *h;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01001721 struct btrfs_root *root = sctx->dev_root;
Arne Jansena2de7332011-03-08 14:14:00 +01001722 struct btrfs_fs_info *fs_info = root->fs_info;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001723 u8 calculated_csum[BTRFS_CSUM_SIZE];
1724 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1725 struct page *page;
1726 void *mapped_buffer;
1727 u64 mapped_size;
1728 void *p;
Arne Jansena2de7332011-03-08 14:14:00 +01001729 u32 crc = ~(u32)0;
1730 int fail = 0;
1731 int crc_fail = 0;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001732 u64 len;
1733 int index;
1734
1735 BUG_ON(sblock->page_count < 1);
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001736 page = sblock->pagev[0]->page;
Linus Torvalds9613beb2012-03-30 12:44:29 -07001737 mapped_buffer = kmap_atomic(page);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001738 h = (struct btrfs_header *)mapped_buffer;
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001739 memcpy(on_disk_csum, h->csum, sctx->csum_size);
Arne Jansena2de7332011-03-08 14:14:00 +01001740
1741 /*
1742 * we don't use the getter functions here, as we
1743 * a) don't have an extent buffer and
1744 * b) the page is already kmapped
1745 */
Arne Jansena2de7332011-03-08 14:14:00 +01001746
Qu Wenruo3cae2102013-07-16 11:19:18 +08001747 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
Arne Jansena2de7332011-03-08 14:14:00 +01001748 ++fail;
1749
Qu Wenruo3cae2102013-07-16 11:19:18 +08001750 if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h))
Arne Jansena2de7332011-03-08 14:14:00 +01001751 ++fail;
1752
1753 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
1754 ++fail;
1755
1756 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1757 BTRFS_UUID_SIZE))
1758 ++fail;
1759
Stefan Behrensff023aa2012-11-06 11:43:11 +01001760 WARN_ON(sctx->nodesize != sctx->leafsize);
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001761 len = sctx->nodesize - BTRFS_CSUM_SIZE;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001762 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1763 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1764 index = 0;
1765 for (;;) {
1766 u64 l = min_t(u64, len, mapped_size);
1767
Liu Bob0496682013-03-14 14:57:45 +00001768 crc = btrfs_csum_data(p, crc, l);
Linus Torvalds9613beb2012-03-30 12:44:29 -07001769 kunmap_atomic(mapped_buffer);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001770 len -= l;
1771 if (len == 0)
1772 break;
1773 index++;
1774 BUG_ON(index >= sblock->page_count);
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001775 BUG_ON(!sblock->pagev[index]->page);
1776 page = sblock->pagev[index]->page;
Linus Torvalds9613beb2012-03-30 12:44:29 -07001777 mapped_buffer = kmap_atomic(page);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001778 mapped_size = PAGE_SIZE;
1779 p = mapped_buffer;
1780 }
1781
1782 btrfs_csum_final(crc, calculated_csum);
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001783 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
Arne Jansena2de7332011-03-08 14:14:00 +01001784 ++crc_fail;
1785
Arne Jansena2de7332011-03-08 14:14:00 +01001786 return fail || crc_fail;
1787}
1788
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001789static int scrub_checksum_super(struct scrub_block *sblock)
Arne Jansena2de7332011-03-08 14:14:00 +01001790{
1791 struct btrfs_super_block *s;
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001792 struct scrub_ctx *sctx = sblock->sctx;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01001793 struct btrfs_root *root = sctx->dev_root;
Arne Jansena2de7332011-03-08 14:14:00 +01001794 struct btrfs_fs_info *fs_info = root->fs_info;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001795 u8 calculated_csum[BTRFS_CSUM_SIZE];
1796 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1797 struct page *page;
1798 void *mapped_buffer;
1799 u64 mapped_size;
1800 void *p;
Arne Jansena2de7332011-03-08 14:14:00 +01001801 u32 crc = ~(u32)0;
Stefan Behrens442a4f62012-05-25 16:06:08 +02001802 int fail_gen = 0;
1803 int fail_cor = 0;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001804 u64 len;
1805 int index;
Arne Jansena2de7332011-03-08 14:14:00 +01001806
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001807 BUG_ON(sblock->page_count < 1);
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001808 page = sblock->pagev[0]->page;
Linus Torvalds9613beb2012-03-30 12:44:29 -07001809 mapped_buffer = kmap_atomic(page);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001810 s = (struct btrfs_super_block *)mapped_buffer;
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001811 memcpy(on_disk_csum, s->csum, sctx->csum_size);
Arne Jansena2de7332011-03-08 14:14:00 +01001812
Qu Wenruo3cae2102013-07-16 11:19:18 +08001813 if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
Stefan Behrens442a4f62012-05-25 16:06:08 +02001814 ++fail_cor;
Arne Jansena2de7332011-03-08 14:14:00 +01001815
Qu Wenruo3cae2102013-07-16 11:19:18 +08001816 if (sblock->pagev[0]->generation != btrfs_super_generation(s))
Stefan Behrens442a4f62012-05-25 16:06:08 +02001817 ++fail_gen;
Arne Jansena2de7332011-03-08 14:14:00 +01001818
1819 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
Stefan Behrens442a4f62012-05-25 16:06:08 +02001820 ++fail_cor;
Arne Jansena2de7332011-03-08 14:14:00 +01001821
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001822 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
1823 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1824 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1825 index = 0;
1826 for (;;) {
1827 u64 l = min_t(u64, len, mapped_size);
1828
Liu Bob0496682013-03-14 14:57:45 +00001829 crc = btrfs_csum_data(p, crc, l);
Linus Torvalds9613beb2012-03-30 12:44:29 -07001830 kunmap_atomic(mapped_buffer);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001831 len -= l;
1832 if (len == 0)
1833 break;
1834 index++;
1835 BUG_ON(index >= sblock->page_count);
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001836 BUG_ON(!sblock->pagev[index]->page);
1837 page = sblock->pagev[index]->page;
Linus Torvalds9613beb2012-03-30 12:44:29 -07001838 mapped_buffer = kmap_atomic(page);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001839 mapped_size = PAGE_SIZE;
1840 p = mapped_buffer;
1841 }
1842
1843 btrfs_csum_final(crc, calculated_csum);
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001844 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
Stefan Behrens442a4f62012-05-25 16:06:08 +02001845 ++fail_cor;
Arne Jansena2de7332011-03-08 14:14:00 +01001846
Stefan Behrens442a4f62012-05-25 16:06:08 +02001847 if (fail_cor + fail_gen) {
Arne Jansena2de7332011-03-08 14:14:00 +01001848 /*
1849 * if we find an error in a super block, we just report it.
1850 * They will get written with the next transaction commit
1851 * anyway
1852 */
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001853 spin_lock(&sctx->stat_lock);
1854 ++sctx->stat.super_errors;
1855 spin_unlock(&sctx->stat_lock);
Stefan Behrens442a4f62012-05-25 16:06:08 +02001856 if (fail_cor)
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001857 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
Stefan Behrens442a4f62012-05-25 16:06:08 +02001858 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1859 else
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001860 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
Stefan Behrens442a4f62012-05-25 16:06:08 +02001861 BTRFS_DEV_STAT_GENERATION_ERRS);
Arne Jansena2de7332011-03-08 14:14:00 +01001862 }
1863
Stefan Behrens442a4f62012-05-25 16:06:08 +02001864 return fail_cor + fail_gen;
Arne Jansena2de7332011-03-08 14:14:00 +01001865}
1866
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001867static void scrub_block_get(struct scrub_block *sblock)
1868{
1869 atomic_inc(&sblock->ref_count);
1870}
1871
1872static void scrub_block_put(struct scrub_block *sblock)
1873{
1874 if (atomic_dec_and_test(&sblock->ref_count)) {
1875 int i;
1876
1877 for (i = 0; i < sblock->page_count; i++)
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001878 scrub_page_put(sblock->pagev[i]);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001879 kfree(sblock);
1880 }
1881}
1882
Stefan Behrens7a9e9982012-11-02 14:58:04 +01001883static void scrub_page_get(struct scrub_page *spage)
1884{
1885 atomic_inc(&spage->ref_count);
1886}
1887
1888static void scrub_page_put(struct scrub_page *spage)
1889{
1890 if (atomic_dec_and_test(&spage->ref_count)) {
1891 if (spage->page)
1892 __free_page(spage->page);
1893 kfree(spage);
1894 }
1895}
1896
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001897static void scrub_submit(struct scrub_ctx *sctx)
Arne Jansena2de7332011-03-08 14:14:00 +01001898{
1899 struct scrub_bio *sbio;
1900
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001901 if (sctx->curr == -1)
Stefan Behrens1623ede2012-03-27 14:21:26 -04001902 return;
Arne Jansena2de7332011-03-08 14:14:00 +01001903
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001904 sbio = sctx->bios[sctx->curr];
1905 sctx->curr = -1;
Stefan Behrensb6bfebc2012-11-02 16:44:58 +01001906 scrub_pending_bio_inc(sctx);
Arne Jansena2de7332011-03-08 14:14:00 +01001907
Stefan Behrensff023aa2012-11-06 11:43:11 +01001908 if (!sbio->bio->bi_bdev) {
1909 /*
1910 * this case should not happen. If btrfs_map_block() is
1911 * wrong, it could happen for dev-replace operations on
1912 * missing devices when no mirrors are available, but in
1913 * this case it should already fail the mount.
1914 * This case is handled correctly (but _very_ slowly).
1915 */
1916 printk_ratelimited(KERN_WARNING
Frank Holtonefe120a2013-12-20 11:37:06 -05001917 "BTRFS: scrub_submit(bio bdev == NULL) is unexpected!\n");
Stefan Behrensff023aa2012-11-06 11:43:11 +01001918 bio_endio(sbio->bio, -EIO);
1919 } else {
1920 btrfsic_submit_bio(READ, sbio->bio);
1921 }
Arne Jansena2de7332011-03-08 14:14:00 +01001922}
1923
Stefan Behrensff023aa2012-11-06 11:43:11 +01001924static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
1925 struct scrub_page *spage)
Arne Jansena2de7332011-03-08 14:14:00 +01001926{
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001927 struct scrub_block *sblock = spage->sblock;
Arne Jansena2de7332011-03-08 14:14:00 +01001928 struct scrub_bio *sbio;
Arne Jansen69f4cb52011-11-11 08:17:10 -05001929 int ret;
Arne Jansena2de7332011-03-08 14:14:00 +01001930
1931again:
1932 /*
1933 * grab a fresh bio or wait for one to become available
1934 */
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001935 while (sctx->curr == -1) {
1936 spin_lock(&sctx->list_lock);
1937 sctx->curr = sctx->first_free;
1938 if (sctx->curr != -1) {
1939 sctx->first_free = sctx->bios[sctx->curr]->next_free;
1940 sctx->bios[sctx->curr]->next_free = -1;
1941 sctx->bios[sctx->curr]->page_count = 0;
1942 spin_unlock(&sctx->list_lock);
Arne Jansena2de7332011-03-08 14:14:00 +01001943 } else {
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001944 spin_unlock(&sctx->list_lock);
1945 wait_event(sctx->list_wait, sctx->first_free != -1);
Arne Jansena2de7332011-03-08 14:14:00 +01001946 }
1947 }
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001948 sbio = sctx->bios[sctx->curr];
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001949 if (sbio->page_count == 0) {
Arne Jansen69f4cb52011-11-11 08:17:10 -05001950 struct bio *bio;
1951
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001952 sbio->physical = spage->physical;
1953 sbio->logical = spage->logical;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01001954 sbio->dev = spage->dev;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001955 bio = sbio->bio;
1956 if (!bio) {
Chris Mason9be33952013-05-17 18:30:14 -04001957 bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001958 if (!bio)
1959 return -ENOMEM;
1960 sbio->bio = bio;
1961 }
Arne Jansen69f4cb52011-11-11 08:17:10 -05001962
1963 bio->bi_private = sbio;
1964 bio->bi_end_io = scrub_bio_end_io;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01001965 bio->bi_bdev = sbio->dev->bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001966 bio->bi_iter.bi_sector = sbio->physical >> 9;
Arne Jansen69f4cb52011-11-11 08:17:10 -05001967 sbio->err = 0;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001968 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1969 spage->physical ||
1970 sbio->logical + sbio->page_count * PAGE_SIZE !=
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01001971 spage->logical ||
1972 sbio->dev != spage->dev) {
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001973 scrub_submit(sctx);
Arne Jansen69f4cb52011-11-11 08:17:10 -05001974 goto again;
1975 }
1976
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001977 sbio->pagev[sbio->page_count] = spage;
1978 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1979 if (ret != PAGE_SIZE) {
1980 if (sbio->page_count < 1) {
1981 bio_put(sbio->bio);
1982 sbio->bio = NULL;
1983 return -EIO;
1984 }
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001985 scrub_submit(sctx);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001986 goto again;
Arne Jansena2de7332011-03-08 14:14:00 +01001987 }
Arne Jansen1bc87792011-05-28 21:57:55 +02001988
Stefan Behrensff023aa2012-11-06 11:43:11 +01001989 scrub_block_get(sblock); /* one for the page added to the bio */
Stefan Behrensb5d67f62012-03-27 14:21:27 -04001990 atomic_inc(&sblock->outstanding_pages);
1991 sbio->page_count++;
Stefan Behrensff023aa2012-11-06 11:43:11 +01001992 if (sbio->page_count == sctx->pages_per_rd_bio)
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001993 scrub_submit(sctx);
Arne Jansena2de7332011-03-08 14:14:00 +01001994
1995 return 0;
1996}
1997
Stefan Behrensd9d181c2012-11-02 09:58:09 +01001998static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01001999 u64 physical, struct btrfs_device *dev, u64 flags,
Stefan Behrensff023aa2012-11-06 11:43:11 +01002000 u64 gen, int mirror_num, u8 *csum, int force,
2001 u64 physical_for_dev_replace)
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002002{
2003 struct scrub_block *sblock;
2004 int index;
2005
2006 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2007 if (!sblock) {
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002008 spin_lock(&sctx->stat_lock);
2009 sctx->stat.malloc_errors++;
2010 spin_unlock(&sctx->stat_lock);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002011 return -ENOMEM;
2012 }
2013
Stefan Behrens7a9e9982012-11-02 14:58:04 +01002014 /* one ref inside this function, plus one for each page added to
2015 * a bio later on */
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002016 atomic_set(&sblock->ref_count, 1);
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002017 sblock->sctx = sctx;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002018 sblock->no_io_error_seen = 1;
2019
2020 for (index = 0; len > 0; index++) {
Stefan Behrens7a9e9982012-11-02 14:58:04 +01002021 struct scrub_page *spage;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002022 u64 l = min_t(u64, len, PAGE_SIZE);
2023
Stefan Behrens7a9e9982012-11-02 14:58:04 +01002024 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2025 if (!spage) {
2026leave_nomem:
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002027 spin_lock(&sctx->stat_lock);
2028 sctx->stat.malloc_errors++;
2029 spin_unlock(&sctx->stat_lock);
Stefan Behrens7a9e9982012-11-02 14:58:04 +01002030 scrub_block_put(sblock);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002031 return -ENOMEM;
2032 }
Stefan Behrens7a9e9982012-11-02 14:58:04 +01002033 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2034 scrub_page_get(spage);
2035 sblock->pagev[index] = spage;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002036 spage->sblock = sblock;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002037 spage->dev = dev;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002038 spage->flags = flags;
2039 spage->generation = gen;
2040 spage->logical = logical;
2041 spage->physical = physical;
Stefan Behrensff023aa2012-11-06 11:43:11 +01002042 spage->physical_for_dev_replace = physical_for_dev_replace;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002043 spage->mirror_num = mirror_num;
2044 if (csum) {
2045 spage->have_csum = 1;
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002046 memcpy(spage->csum, csum, sctx->csum_size);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002047 } else {
2048 spage->have_csum = 0;
2049 }
2050 sblock->page_count++;
Stefan Behrens7a9e9982012-11-02 14:58:04 +01002051 spage->page = alloc_page(GFP_NOFS);
2052 if (!spage->page)
2053 goto leave_nomem;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002054 len -= l;
2055 logical += l;
2056 physical += l;
Stefan Behrensff023aa2012-11-06 11:43:11 +01002057 physical_for_dev_replace += l;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002058 }
2059
Stefan Behrens7a9e9982012-11-02 14:58:04 +01002060 WARN_ON(sblock->page_count == 0);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002061 for (index = 0; index < sblock->page_count; index++) {
Stefan Behrens7a9e9982012-11-02 14:58:04 +01002062 struct scrub_page *spage = sblock->pagev[index];
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002063 int ret;
2064
Stefan Behrensff023aa2012-11-06 11:43:11 +01002065 ret = scrub_add_page_to_rd_bio(sctx, spage);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002066 if (ret) {
2067 scrub_block_put(sblock);
2068 return ret;
2069 }
2070 }
2071
2072 if (force)
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002073 scrub_submit(sctx);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002074
2075 /* last one frees, either here or in bio completion for last page */
2076 scrub_block_put(sblock);
2077 return 0;
2078}
2079
2080static void scrub_bio_end_io(struct bio *bio, int err)
2081{
2082 struct scrub_bio *sbio = bio->bi_private;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002083 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002084
2085 sbio->err = err;
2086 sbio->bio = bio;
2087
Qu Wenruo0339ef22014-02-28 10:46:17 +08002088 btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002089}
2090
2091static void scrub_bio_end_io_worker(struct btrfs_work *work)
2092{
2093 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002094 struct scrub_ctx *sctx = sbio->sctx;
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002095 int i;
2096
Stefan Behrensff023aa2012-11-06 11:43:11 +01002097 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002098 if (sbio->err) {
2099 for (i = 0; i < sbio->page_count; i++) {
2100 struct scrub_page *spage = sbio->pagev[i];
2101
2102 spage->io_error = 1;
2103 spage->sblock->no_io_error_seen = 0;
2104 }
2105 }
2106
2107 /* now complete the scrub_block items that have all pages completed */
2108 for (i = 0; i < sbio->page_count; i++) {
2109 struct scrub_page *spage = sbio->pagev[i];
2110 struct scrub_block *sblock = spage->sblock;
2111
2112 if (atomic_dec_and_test(&sblock->outstanding_pages))
2113 scrub_block_complete(sblock);
2114 scrub_block_put(sblock);
2115 }
2116
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002117 bio_put(sbio->bio);
2118 sbio->bio = NULL;
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002119 spin_lock(&sctx->list_lock);
2120 sbio->next_free = sctx->first_free;
2121 sctx->first_free = sbio->index;
2122 spin_unlock(&sctx->list_lock);
Stefan Behrensff023aa2012-11-06 11:43:11 +01002123
2124 if (sctx->is_dev_replace &&
2125 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2126 mutex_lock(&sctx->wr_ctx.wr_lock);
2127 scrub_wr_submit(sctx);
2128 mutex_unlock(&sctx->wr_ctx.wr_lock);
2129 }
2130
Stefan Behrensb6bfebc2012-11-02 16:44:58 +01002131 scrub_pending_bio_dec(sctx);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002132}
2133
2134static void scrub_block_complete(struct scrub_block *sblock)
2135{
Stefan Behrensff023aa2012-11-06 11:43:11 +01002136 if (!sblock->no_io_error_seen) {
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002137 scrub_handle_errored_block(sblock);
Stefan Behrensff023aa2012-11-06 11:43:11 +01002138 } else {
2139 /*
2140 * if has checksum error, write via repair mechanism in
2141 * dev replace case, otherwise write here in dev replace
2142 * case.
2143 */
2144 if (!scrub_checksum(sblock) && sblock->sctx->is_dev_replace)
2145 scrub_write_block_to_dev_replace(sblock);
2146 }
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002147}
2148
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002149static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
Arne Jansena2de7332011-03-08 14:14:00 +01002150 u8 *csum)
2151{
2152 struct btrfs_ordered_sum *sum = NULL;
Miao Xief51a4a12013-06-19 10:36:09 +08002153 unsigned long index;
Arne Jansena2de7332011-03-08 14:14:00 +01002154 unsigned long num_sectors;
Arne Jansena2de7332011-03-08 14:14:00 +01002155
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002156 while (!list_empty(&sctx->csum_list)) {
2157 sum = list_first_entry(&sctx->csum_list,
Arne Jansena2de7332011-03-08 14:14:00 +01002158 struct btrfs_ordered_sum, list);
2159 if (sum->bytenr > logical)
2160 return 0;
2161 if (sum->bytenr + sum->len > logical)
2162 break;
2163
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002164 ++sctx->stat.csum_discards;
Arne Jansena2de7332011-03-08 14:14:00 +01002165 list_del(&sum->list);
2166 kfree(sum);
2167 sum = NULL;
2168 }
2169 if (!sum)
2170 return 0;
2171
Miao Xief51a4a12013-06-19 10:36:09 +08002172 index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002173 num_sectors = sum->len / sctx->sectorsize;
Miao Xief51a4a12013-06-19 10:36:09 +08002174 memcpy(csum, sum->sums + index, sctx->csum_size);
2175 if (index == num_sectors - 1) {
Arne Jansena2de7332011-03-08 14:14:00 +01002176 list_del(&sum->list);
2177 kfree(sum);
2178 }
Miao Xief51a4a12013-06-19 10:36:09 +08002179 return 1;
Arne Jansena2de7332011-03-08 14:14:00 +01002180}
2181
2182/* scrub extent tries to collect up to 64 kB for each bio */
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002183static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002184 u64 physical, struct btrfs_device *dev, u64 flags,
Stefan Behrensff023aa2012-11-06 11:43:11 +01002185 u64 gen, int mirror_num, u64 physical_for_dev_replace)
Arne Jansena2de7332011-03-08 14:14:00 +01002186{
2187 int ret;
2188 u8 csum[BTRFS_CSUM_SIZE];
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002189 u32 blocksize;
2190
2191 if (flags & BTRFS_EXTENT_FLAG_DATA) {
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002192 blocksize = sctx->sectorsize;
2193 spin_lock(&sctx->stat_lock);
2194 sctx->stat.data_extents_scrubbed++;
2195 sctx->stat.data_bytes_scrubbed += len;
2196 spin_unlock(&sctx->stat_lock);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002197 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
Stefan Behrensff023aa2012-11-06 11:43:11 +01002198 WARN_ON(sctx->nodesize != sctx->leafsize);
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002199 blocksize = sctx->nodesize;
2200 spin_lock(&sctx->stat_lock);
2201 sctx->stat.tree_extents_scrubbed++;
2202 sctx->stat.tree_bytes_scrubbed += len;
2203 spin_unlock(&sctx->stat_lock);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002204 } else {
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002205 blocksize = sctx->sectorsize;
Stefan Behrensff023aa2012-11-06 11:43:11 +01002206 WARN_ON(1);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002207 }
Arne Jansena2de7332011-03-08 14:14:00 +01002208
2209 while (len) {
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002210 u64 l = min_t(u64, len, blocksize);
Arne Jansena2de7332011-03-08 14:14:00 +01002211 int have_csum = 0;
2212
2213 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2214 /* push csums to sbio */
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002215 have_csum = scrub_find_csum(sctx, logical, l, csum);
Arne Jansena2de7332011-03-08 14:14:00 +01002216 if (have_csum == 0)
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002217 ++sctx->stat.no_csum;
Stefan Behrensff023aa2012-11-06 11:43:11 +01002218 if (sctx->is_dev_replace && !have_csum) {
2219 ret = copy_nocow_pages(sctx, logical, l,
2220 mirror_num,
2221 physical_for_dev_replace);
2222 goto behind_scrub_pages;
2223 }
Arne Jansena2de7332011-03-08 14:14:00 +01002224 }
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002225 ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
Stefan Behrensff023aa2012-11-06 11:43:11 +01002226 mirror_num, have_csum ? csum : NULL, 0,
2227 physical_for_dev_replace);
2228behind_scrub_pages:
Arne Jansena2de7332011-03-08 14:14:00 +01002229 if (ret)
2230 return ret;
2231 len -= l;
2232 logical += l;
2233 physical += l;
Stefan Behrensff023aa2012-11-06 11:43:11 +01002234 physical_for_dev_replace += l;
Arne Jansena2de7332011-03-08 14:14:00 +01002235 }
2236 return 0;
2237}
2238
Wang Shilong3b080b22014-04-01 18:01:43 +08002239/*
2240 * Given a physical address, this will calculate it's
2241 * logical offset. if this is a parity stripe, it will return
2242 * the most left data stripe's logical offset.
2243 *
2244 * return 0 if it is a data stripe, 1 means parity stripe.
2245 */
2246static int get_raid56_logic_offset(u64 physical, int num,
2247 struct map_lookup *map, u64 *offset)
2248{
2249 int i;
2250 int j = 0;
2251 u64 stripe_nr;
2252 u64 last_offset;
2253 int stripe_index;
2254 int rot;
2255
2256 last_offset = (physical - map->stripes[num].physical) *
2257 nr_data_stripes(map);
2258 *offset = last_offset;
2259 for (i = 0; i < nr_data_stripes(map); i++) {
2260 *offset = last_offset + i * map->stripe_len;
2261
2262 stripe_nr = *offset;
2263 do_div(stripe_nr, map->stripe_len);
2264 do_div(stripe_nr, nr_data_stripes(map));
2265
2266 /* Work out the disk rotation on this stripe-set */
2267 rot = do_div(stripe_nr, map->num_stripes);
2268 /* calculate which stripe this data locates */
2269 rot += i;
Wang Shilonge4fbaee2014-04-11 18:32:25 +08002270 stripe_index = rot % map->num_stripes;
Wang Shilong3b080b22014-04-01 18:01:43 +08002271 if (stripe_index == num)
2272 return 0;
2273 if (stripe_index < num)
2274 j++;
2275 }
2276 *offset = last_offset + j * map->stripe_len;
2277 return 1;
2278}
2279
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002280static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002281 struct map_lookup *map,
2282 struct btrfs_device *scrub_dev,
Stefan Behrensff023aa2012-11-06 11:43:11 +01002283 int num, u64 base, u64 length,
2284 int is_dev_replace)
Arne Jansena2de7332011-03-08 14:14:00 +01002285{
2286 struct btrfs_path *path;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002287 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
Arne Jansena2de7332011-03-08 14:14:00 +01002288 struct btrfs_root *root = fs_info->extent_root;
2289 struct btrfs_root *csum_root = fs_info->csum_root;
2290 struct btrfs_extent_item *extent;
Arne Jansene7786c32011-05-28 20:58:38 +00002291 struct blk_plug plug;
Arne Jansena2de7332011-03-08 14:14:00 +01002292 u64 flags;
2293 int ret;
2294 int slot;
Arne Jansena2de7332011-03-08 14:14:00 +01002295 u64 nstripes;
Arne Jansena2de7332011-03-08 14:14:00 +01002296 struct extent_buffer *l;
2297 struct btrfs_key key;
2298 u64 physical;
2299 u64 logical;
Liu Bo625f1c8d2013-04-27 02:56:57 +00002300 u64 logic_end;
Wang Shilong3b080b22014-04-01 18:01:43 +08002301 u64 physical_end;
Arne Jansena2de7332011-03-08 14:14:00 +01002302 u64 generation;
Jan Schmidte12fa9c2011-06-17 15:55:21 +02002303 int mirror_num;
Arne Jansen7a262852011-06-10 12:39:23 +02002304 struct reada_control *reada1;
2305 struct reada_control *reada2;
2306 struct btrfs_key key_start;
2307 struct btrfs_key key_end;
Arne Jansena2de7332011-03-08 14:14:00 +01002308 u64 increment = map->stripe_len;
2309 u64 offset;
Stefan Behrensff023aa2012-11-06 11:43:11 +01002310 u64 extent_logical;
2311 u64 extent_physical;
2312 u64 extent_len;
2313 struct btrfs_device *extent_dev;
2314 int extent_mirror_num;
Wang Shilong3b080b22014-04-01 18:01:43 +08002315 int stop_loop = 0;
David Woodhouse53b381b2013-01-29 18:40:14 -05002316
Arne Jansena2de7332011-03-08 14:14:00 +01002317 nstripes = length;
Wang Shilong3b080b22014-04-01 18:01:43 +08002318 physical = map->stripes[num].physical;
Arne Jansena2de7332011-03-08 14:14:00 +01002319 offset = 0;
2320 do_div(nstripes, map->stripe_len);
2321 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2322 offset = map->stripe_len * num;
2323 increment = map->stripe_len * map->num_stripes;
Jan Schmidt193ea742011-06-13 19:56:54 +02002324 mirror_num = 1;
Arne Jansena2de7332011-03-08 14:14:00 +01002325 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2326 int factor = map->num_stripes / map->sub_stripes;
2327 offset = map->stripe_len * (num / map->sub_stripes);
2328 increment = map->stripe_len * factor;
Jan Schmidt193ea742011-06-13 19:56:54 +02002329 mirror_num = num % map->sub_stripes + 1;
Arne Jansena2de7332011-03-08 14:14:00 +01002330 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2331 increment = map->stripe_len;
Jan Schmidt193ea742011-06-13 19:56:54 +02002332 mirror_num = num % map->num_stripes + 1;
Arne Jansena2de7332011-03-08 14:14:00 +01002333 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2334 increment = map->stripe_len;
Jan Schmidt193ea742011-06-13 19:56:54 +02002335 mirror_num = num % map->num_stripes + 1;
Wang Shilong3b080b22014-04-01 18:01:43 +08002336 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2337 BTRFS_BLOCK_GROUP_RAID6)) {
2338 get_raid56_logic_offset(physical, num, map, &offset);
2339 increment = map->stripe_len * nr_data_stripes(map);
2340 mirror_num = 1;
Arne Jansena2de7332011-03-08 14:14:00 +01002341 } else {
2342 increment = map->stripe_len;
Jan Schmidt193ea742011-06-13 19:56:54 +02002343 mirror_num = 1;
Arne Jansena2de7332011-03-08 14:14:00 +01002344 }
2345
2346 path = btrfs_alloc_path();
2347 if (!path)
2348 return -ENOMEM;
2349
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002350 /*
2351 * work on commit root. The related disk blocks are static as
2352 * long as COW is applied. This means, it is save to rewrite
2353 * them to repair disk errors without any race conditions
2354 */
Arne Jansena2de7332011-03-08 14:14:00 +01002355 path->search_commit_root = 1;
2356 path->skip_locking = 1;
2357
2358 /*
Arne Jansen7a262852011-06-10 12:39:23 +02002359 * trigger the readahead for extent tree csum tree and wait for
2360 * completion. During readahead, the scrub is officially paused
2361 * to not hold off transaction commits
Arne Jansena2de7332011-03-08 14:14:00 +01002362 */
2363 logical = base + offset;
Wang Shilong3b080b22014-04-01 18:01:43 +08002364 physical_end = physical + nstripes * map->stripe_len;
2365 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2366 BTRFS_BLOCK_GROUP_RAID6)) {
2367 get_raid56_logic_offset(physical_end, num,
2368 map, &logic_end);
2369 logic_end += base;
2370 } else {
2371 logic_end = logical + increment * nstripes;
2372 }
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002373 wait_event(sctx->list_wait,
Stefan Behrensb6bfebc2012-11-02 16:44:58 +01002374 atomic_read(&sctx->bios_in_flight) == 0);
Wang Shilongcb7ab022013-12-04 21:16:53 +08002375 scrub_blocked_if_needed(fs_info);
Arne Jansena2de7332011-03-08 14:14:00 +01002376
Arne Jansen7a262852011-06-10 12:39:23 +02002377 /* FIXME it might be better to start readahead at commit root */
2378 key_start.objectid = logical;
2379 key_start.type = BTRFS_EXTENT_ITEM_KEY;
2380 key_start.offset = (u64)0;
Wang Shilong3b080b22014-04-01 18:01:43 +08002381 key_end.objectid = logic_end;
Josef Bacik3173a182013-03-07 14:22:04 -05002382 key_end.type = BTRFS_METADATA_ITEM_KEY;
2383 key_end.offset = (u64)-1;
Arne Jansen7a262852011-06-10 12:39:23 +02002384 reada1 = btrfs_reada_add(root, &key_start, &key_end);
Arne Jansena2de7332011-03-08 14:14:00 +01002385
Arne Jansen7a262852011-06-10 12:39:23 +02002386 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
2387 key_start.type = BTRFS_EXTENT_CSUM_KEY;
2388 key_start.offset = logical;
2389 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
2390 key_end.type = BTRFS_EXTENT_CSUM_KEY;
Wang Shilong3b080b22014-04-01 18:01:43 +08002391 key_end.offset = logic_end;
Arne Jansen7a262852011-06-10 12:39:23 +02002392 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
Arne Jansena2de7332011-03-08 14:14:00 +01002393
Arne Jansen7a262852011-06-10 12:39:23 +02002394 if (!IS_ERR(reada1))
2395 btrfs_reada_wait(reada1);
2396 if (!IS_ERR(reada2))
2397 btrfs_reada_wait(reada2);
Arne Jansena2de7332011-03-08 14:14:00 +01002398
Arne Jansena2de7332011-03-08 14:14:00 +01002399
2400 /*
2401 * collect all data csums for the stripe to avoid seeking during
2402 * the scrub. This might currently (crc32) end up to be about 1MB
2403 */
Arne Jansene7786c32011-05-28 20:58:38 +00002404 blk_start_plug(&plug);
Arne Jansena2de7332011-03-08 14:14:00 +01002405
Arne Jansena2de7332011-03-08 14:14:00 +01002406 /*
2407 * now find all extents for each stripe and scrub them
2408 */
Arne Jansena2de7332011-03-08 14:14:00 +01002409 ret = 0;
Wang Shilong3b080b22014-04-01 18:01:43 +08002410 while (physical < physical_end) {
2411 /* for raid56, we skip parity stripe */
2412 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2413 BTRFS_BLOCK_GROUP_RAID6)) {
2414 ret = get_raid56_logic_offset(physical, num,
2415 map, &logical);
2416 logical += base;
2417 if (ret)
2418 goto skip;
2419 }
Arne Jansena2de7332011-03-08 14:14:00 +01002420 /*
2421 * canceled?
2422 */
2423 if (atomic_read(&fs_info->scrub_cancel_req) ||
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002424 atomic_read(&sctx->cancel_req)) {
Arne Jansena2de7332011-03-08 14:14:00 +01002425 ret = -ECANCELED;
2426 goto out;
2427 }
2428 /*
2429 * check to see if we have to pause
2430 */
2431 if (atomic_read(&fs_info->scrub_pause_req)) {
2432 /* push queued extents */
Stefan Behrensff023aa2012-11-06 11:43:11 +01002433 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002434 scrub_submit(sctx);
Stefan Behrensff023aa2012-11-06 11:43:11 +01002435 mutex_lock(&sctx->wr_ctx.wr_lock);
2436 scrub_wr_submit(sctx);
2437 mutex_unlock(&sctx->wr_ctx.wr_lock);
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002438 wait_event(sctx->list_wait,
Stefan Behrensb6bfebc2012-11-02 16:44:58 +01002439 atomic_read(&sctx->bios_in_flight) == 0);
Stefan Behrensff023aa2012-11-06 11:43:11 +01002440 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
Wang Shilong3cb09292013-12-04 21:15:19 +08002441 scrub_blocked_if_needed(fs_info);
Arne Jansena2de7332011-03-08 14:14:00 +01002442 }
2443
Wang Shilong7c76edb2014-01-12 21:38:32 +08002444 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2445 key.type = BTRFS_METADATA_ITEM_KEY;
2446 else
2447 key.type = BTRFS_EXTENT_ITEM_KEY;
Arne Jansena2de7332011-03-08 14:14:00 +01002448 key.objectid = logical;
Liu Bo625f1c8d2013-04-27 02:56:57 +00002449 key.offset = (u64)-1;
Arne Jansena2de7332011-03-08 14:14:00 +01002450
2451 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2452 if (ret < 0)
2453 goto out;
Josef Bacik3173a182013-03-07 14:22:04 -05002454
Arne Jansen8c510322011-06-03 10:09:26 +02002455 if (ret > 0) {
Wang Shilongade2e0b2014-01-12 21:38:33 +08002456 ret = btrfs_previous_extent_item(root, path, 0);
Arne Jansena2de7332011-03-08 14:14:00 +01002457 if (ret < 0)
2458 goto out;
Arne Jansen8c510322011-06-03 10:09:26 +02002459 if (ret > 0) {
2460 /* there's no smaller item, so stick with the
2461 * larger one */
2462 btrfs_release_path(path);
2463 ret = btrfs_search_slot(NULL, root, &key,
2464 path, 0, 0);
2465 if (ret < 0)
2466 goto out;
2467 }
Arne Jansena2de7332011-03-08 14:14:00 +01002468 }
2469
Liu Bo625f1c8d2013-04-27 02:56:57 +00002470 stop_loop = 0;
Arne Jansena2de7332011-03-08 14:14:00 +01002471 while (1) {
Josef Bacik3173a182013-03-07 14:22:04 -05002472 u64 bytes;
2473
Arne Jansena2de7332011-03-08 14:14:00 +01002474 l = path->nodes[0];
2475 slot = path->slots[0];
2476 if (slot >= btrfs_header_nritems(l)) {
2477 ret = btrfs_next_leaf(root, path);
2478 if (ret == 0)
2479 continue;
2480 if (ret < 0)
2481 goto out;
2482
Liu Bo625f1c8d2013-04-27 02:56:57 +00002483 stop_loop = 1;
Arne Jansena2de7332011-03-08 14:14:00 +01002484 break;
2485 }
2486 btrfs_item_key_to_cpu(l, &key, slot);
2487
Josef Bacik3173a182013-03-07 14:22:04 -05002488 if (key.type == BTRFS_METADATA_ITEM_KEY)
2489 bytes = root->leafsize;
2490 else
2491 bytes = key.offset;
2492
2493 if (key.objectid + bytes <= logical)
Arne Jansena2de7332011-03-08 14:14:00 +01002494 goto next;
2495
Liu Bo625f1c8d2013-04-27 02:56:57 +00002496 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2497 key.type != BTRFS_METADATA_ITEM_KEY)
2498 goto next;
Arne Jansena2de7332011-03-08 14:14:00 +01002499
Liu Bo625f1c8d2013-04-27 02:56:57 +00002500 if (key.objectid >= logical + map->stripe_len) {
2501 /* out of this device extent */
2502 if (key.objectid >= logic_end)
2503 stop_loop = 1;
2504 break;
2505 }
Arne Jansena2de7332011-03-08 14:14:00 +01002506
2507 extent = btrfs_item_ptr(l, slot,
2508 struct btrfs_extent_item);
2509 flags = btrfs_extent_flags(l, extent);
2510 generation = btrfs_extent_generation(l, extent);
2511
2512 if (key.objectid < logical &&
2513 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
Frank Holtonefe120a2013-12-20 11:37:06 -05002514 btrfs_err(fs_info,
2515 "scrub: tree block %llu spanning "
2516 "stripes, ignored. logical=%llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02002517 key.objectid, logical);
Arne Jansena2de7332011-03-08 14:14:00 +01002518 goto next;
2519 }
2520
Liu Bo625f1c8d2013-04-27 02:56:57 +00002521again:
2522 extent_logical = key.objectid;
2523 extent_len = bytes;
2524
Arne Jansena2de7332011-03-08 14:14:00 +01002525 /*
2526 * trim extent to this stripe
2527 */
Liu Bo625f1c8d2013-04-27 02:56:57 +00002528 if (extent_logical < logical) {
2529 extent_len -= logical - extent_logical;
2530 extent_logical = logical;
Arne Jansena2de7332011-03-08 14:14:00 +01002531 }
Liu Bo625f1c8d2013-04-27 02:56:57 +00002532 if (extent_logical + extent_len >
Arne Jansena2de7332011-03-08 14:14:00 +01002533 logical + map->stripe_len) {
Liu Bo625f1c8d2013-04-27 02:56:57 +00002534 extent_len = logical + map->stripe_len -
2535 extent_logical;
Arne Jansena2de7332011-03-08 14:14:00 +01002536 }
2537
Liu Bo625f1c8d2013-04-27 02:56:57 +00002538 extent_physical = extent_logical - logical + physical;
Stefan Behrensff023aa2012-11-06 11:43:11 +01002539 extent_dev = scrub_dev;
2540 extent_mirror_num = mirror_num;
2541 if (is_dev_replace)
2542 scrub_remap_extent(fs_info, extent_logical,
2543 extent_len, &extent_physical,
2544 &extent_dev,
2545 &extent_mirror_num);
Liu Bo625f1c8d2013-04-27 02:56:57 +00002546
2547 ret = btrfs_lookup_csums_range(csum_root, logical,
2548 logical + map->stripe_len - 1,
2549 &sctx->csum_list, 1);
Arne Jansena2de7332011-03-08 14:14:00 +01002550 if (ret)
2551 goto out;
2552
Liu Bo625f1c8d2013-04-27 02:56:57 +00002553 ret = scrub_extent(sctx, extent_logical, extent_len,
2554 extent_physical, extent_dev, flags,
2555 generation, extent_mirror_num,
Stefan Behrens115930c2013-07-04 16:14:23 +02002556 extent_logical - logical + physical);
Liu Bo625f1c8d2013-04-27 02:56:57 +00002557 if (ret)
2558 goto out;
2559
Josef Bacikd88d46c2013-06-10 12:59:04 +00002560 scrub_free_csums(sctx);
Liu Bo625f1c8d2013-04-27 02:56:57 +00002561 if (extent_logical + extent_len <
2562 key.objectid + bytes) {
Wang Shilong3b080b22014-04-01 18:01:43 +08002563 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2564 BTRFS_BLOCK_GROUP_RAID6)) {
2565 /*
2566 * loop until we find next data stripe
2567 * or we have finished all stripes.
2568 */
2569 do {
2570 physical += map->stripe_len;
2571 ret = get_raid56_logic_offset(
2572 physical, num,
2573 map, &logical);
2574 logical += base;
2575 } while (physical < physical_end && ret);
2576 } else {
2577 physical += map->stripe_len;
2578 logical += increment;
2579 }
Liu Bo625f1c8d2013-04-27 02:56:57 +00002580 if (logical < key.objectid + bytes) {
2581 cond_resched();
2582 goto again;
2583 }
2584
Wang Shilong3b080b22014-04-01 18:01:43 +08002585 if (physical >= physical_end) {
Liu Bo625f1c8d2013-04-27 02:56:57 +00002586 stop_loop = 1;
2587 break;
2588 }
2589 }
Arne Jansena2de7332011-03-08 14:14:00 +01002590next:
2591 path->slots[0]++;
2592 }
Chris Mason71267332011-05-23 06:30:52 -04002593 btrfs_release_path(path);
Wang Shilong3b080b22014-04-01 18:01:43 +08002594skip:
Arne Jansena2de7332011-03-08 14:14:00 +01002595 logical += increment;
2596 physical += map->stripe_len;
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002597 spin_lock(&sctx->stat_lock);
Liu Bo625f1c8d2013-04-27 02:56:57 +00002598 if (stop_loop)
2599 sctx->stat.last_physical = map->stripes[num].physical +
2600 length;
2601 else
2602 sctx->stat.last_physical = physical;
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002603 spin_unlock(&sctx->stat_lock);
Liu Bo625f1c8d2013-04-27 02:56:57 +00002604 if (stop_loop)
2605 break;
Arne Jansena2de7332011-03-08 14:14:00 +01002606 }
Stefan Behrensff023aa2012-11-06 11:43:11 +01002607out:
Arne Jansena2de7332011-03-08 14:14:00 +01002608 /* push queued extents */
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002609 scrub_submit(sctx);
Stefan Behrensff023aa2012-11-06 11:43:11 +01002610 mutex_lock(&sctx->wr_ctx.wr_lock);
2611 scrub_wr_submit(sctx);
2612 mutex_unlock(&sctx->wr_ctx.wr_lock);
Arne Jansena2de7332011-03-08 14:14:00 +01002613
Arne Jansene7786c32011-05-28 20:58:38 +00002614 blk_finish_plug(&plug);
Arne Jansena2de7332011-03-08 14:14:00 +01002615 btrfs_free_path(path);
2616 return ret < 0 ? ret : 0;
2617}
2618
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002619static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002620 struct btrfs_device *scrub_dev,
2621 u64 chunk_tree, u64 chunk_objectid,
2622 u64 chunk_offset, u64 length,
Stefan Behrensff023aa2012-11-06 11:43:11 +01002623 u64 dev_offset, int is_dev_replace)
Arne Jansena2de7332011-03-08 14:14:00 +01002624{
2625 struct btrfs_mapping_tree *map_tree =
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002626 &sctx->dev_root->fs_info->mapping_tree;
Arne Jansena2de7332011-03-08 14:14:00 +01002627 struct map_lookup *map;
2628 struct extent_map *em;
2629 int i;
Stefan Behrensff023aa2012-11-06 11:43:11 +01002630 int ret = 0;
Arne Jansena2de7332011-03-08 14:14:00 +01002631
2632 read_lock(&map_tree->map_tree.lock);
2633 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2634 read_unlock(&map_tree->map_tree.lock);
2635
2636 if (!em)
2637 return -EINVAL;
2638
2639 map = (struct map_lookup *)em->bdev;
2640 if (em->start != chunk_offset)
2641 goto out;
2642
2643 if (em->len < length)
2644 goto out;
2645
2646 for (i = 0; i < map->num_stripes; ++i) {
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002647 if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
Arne Jansen859acaf2012-02-09 15:09:02 +01002648 map->stripes[i].physical == dev_offset) {
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002649 ret = scrub_stripe(sctx, map, scrub_dev, i,
Stefan Behrensff023aa2012-11-06 11:43:11 +01002650 chunk_offset, length,
2651 is_dev_replace);
Arne Jansena2de7332011-03-08 14:14:00 +01002652 if (ret)
2653 goto out;
2654 }
2655 }
2656out:
2657 free_extent_map(em);
2658
2659 return ret;
2660}
2661
2662static noinline_for_stack
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002663int scrub_enumerate_chunks(struct scrub_ctx *sctx,
Stefan Behrensff023aa2012-11-06 11:43:11 +01002664 struct btrfs_device *scrub_dev, u64 start, u64 end,
2665 int is_dev_replace)
Arne Jansena2de7332011-03-08 14:14:00 +01002666{
2667 struct btrfs_dev_extent *dev_extent = NULL;
2668 struct btrfs_path *path;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002669 struct btrfs_root *root = sctx->dev_root;
Arne Jansena2de7332011-03-08 14:14:00 +01002670 struct btrfs_fs_info *fs_info = root->fs_info;
2671 u64 length;
2672 u64 chunk_tree;
2673 u64 chunk_objectid;
2674 u64 chunk_offset;
2675 int ret;
2676 int slot;
2677 struct extent_buffer *l;
2678 struct btrfs_key key;
2679 struct btrfs_key found_key;
2680 struct btrfs_block_group_cache *cache;
Stefan Behrensff023aa2012-11-06 11:43:11 +01002681 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
Arne Jansena2de7332011-03-08 14:14:00 +01002682
2683 path = btrfs_alloc_path();
2684 if (!path)
2685 return -ENOMEM;
2686
2687 path->reada = 2;
2688 path->search_commit_root = 1;
2689 path->skip_locking = 1;
2690
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002691 key.objectid = scrub_dev->devid;
Arne Jansena2de7332011-03-08 14:14:00 +01002692 key.offset = 0ull;
2693 key.type = BTRFS_DEV_EXTENT_KEY;
2694
Arne Jansena2de7332011-03-08 14:14:00 +01002695 while (1) {
2696 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2697 if (ret < 0)
Arne Jansen8c510322011-06-03 10:09:26 +02002698 break;
2699 if (ret > 0) {
2700 if (path->slots[0] >=
2701 btrfs_header_nritems(path->nodes[0])) {
2702 ret = btrfs_next_leaf(root, path);
2703 if (ret)
2704 break;
2705 }
2706 }
Arne Jansena2de7332011-03-08 14:14:00 +01002707
2708 l = path->nodes[0];
2709 slot = path->slots[0];
2710
2711 btrfs_item_key_to_cpu(l, &found_key, slot);
2712
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002713 if (found_key.objectid != scrub_dev->devid)
Arne Jansena2de7332011-03-08 14:14:00 +01002714 break;
2715
Arne Jansen8c510322011-06-03 10:09:26 +02002716 if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY)
Arne Jansena2de7332011-03-08 14:14:00 +01002717 break;
2718
2719 if (found_key.offset >= end)
2720 break;
2721
2722 if (found_key.offset < key.offset)
2723 break;
2724
2725 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2726 length = btrfs_dev_extent_length(l, dev_extent);
2727
Qu Wenruoced96ed2014-06-19 10:42:51 +08002728 if (found_key.offset + length <= start)
2729 goto skip;
Arne Jansena2de7332011-03-08 14:14:00 +01002730
2731 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2732 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2733 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2734
2735 /*
2736 * get a reference on the corresponding block group to prevent
2737 * the chunk from going away while we scrub it
2738 */
2739 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
Qu Wenruoced96ed2014-06-19 10:42:51 +08002740
2741 /* some chunks are removed but not committed to disk yet,
2742 * continue scrubbing */
2743 if (!cache)
2744 goto skip;
2745
Stefan Behrensff023aa2012-11-06 11:43:11 +01002746 dev_replace->cursor_right = found_key.offset + length;
2747 dev_replace->cursor_left = found_key.offset;
2748 dev_replace->item_needs_writeback = 1;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002749 ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid,
Stefan Behrensff023aa2012-11-06 11:43:11 +01002750 chunk_offset, length, found_key.offset,
2751 is_dev_replace);
2752
2753 /*
2754 * flush, submit all pending read and write bios, afterwards
2755 * wait for them.
2756 * Note that in the dev replace case, a read request causes
2757 * write requests that are submitted in the read completion
2758 * worker. Therefore in the current situation, it is required
2759 * that all write requests are flushed, so that all read and
2760 * write requests are really completed when bios_in_flight
2761 * changes to 0.
2762 */
2763 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
2764 scrub_submit(sctx);
2765 mutex_lock(&sctx->wr_ctx.wr_lock);
2766 scrub_wr_submit(sctx);
2767 mutex_unlock(&sctx->wr_ctx.wr_lock);
2768
2769 wait_event(sctx->list_wait,
2770 atomic_read(&sctx->bios_in_flight) == 0);
Wang Shilong12cf9372014-02-19 19:24:17 +08002771 atomic_inc(&fs_info->scrubs_paused);
2772 wake_up(&fs_info->scrub_pause_wait);
2773
2774 /*
2775 * must be called before we decrease @scrub_paused.
2776 * make sure we don't block transaction commit while
2777 * we are waiting pending workers finished.
2778 */
Stefan Behrensff023aa2012-11-06 11:43:11 +01002779 wait_event(sctx->list_wait,
2780 atomic_read(&sctx->workers_pending) == 0);
Wang Shilong12cf9372014-02-19 19:24:17 +08002781 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
2782
2783 mutex_lock(&fs_info->scrub_lock);
2784 __scrub_blocked_if_needed(fs_info);
2785 atomic_dec(&fs_info->scrubs_paused);
2786 mutex_unlock(&fs_info->scrub_lock);
2787 wake_up(&fs_info->scrub_pause_wait);
Stefan Behrensff023aa2012-11-06 11:43:11 +01002788
Arne Jansena2de7332011-03-08 14:14:00 +01002789 btrfs_put_block_group(cache);
2790 if (ret)
2791 break;
Stefan Behrensaf1be4f2012-11-27 17:39:51 +00002792 if (is_dev_replace &&
2793 atomic64_read(&dev_replace->num_write_errors) > 0) {
Stefan Behrensff023aa2012-11-06 11:43:11 +01002794 ret = -EIO;
2795 break;
2796 }
2797 if (sctx->stat.malloc_errors > 0) {
2798 ret = -ENOMEM;
2799 break;
2800 }
Arne Jansena2de7332011-03-08 14:14:00 +01002801
Ilya Dryomov539f3582013-10-07 13:42:57 +03002802 dev_replace->cursor_left = dev_replace->cursor_right;
2803 dev_replace->item_needs_writeback = 1;
Qu Wenruoced96ed2014-06-19 10:42:51 +08002804skip:
Arne Jansena2de7332011-03-08 14:14:00 +01002805 key.offset = found_key.offset + length;
Chris Mason71267332011-05-23 06:30:52 -04002806 btrfs_release_path(path);
Arne Jansena2de7332011-03-08 14:14:00 +01002807 }
2808
Arne Jansena2de7332011-03-08 14:14:00 +01002809 btrfs_free_path(path);
Arne Jansen8c510322011-06-03 10:09:26 +02002810
2811 /*
2812 * ret can still be 1 from search_slot or next_leaf,
2813 * that's not an error
2814 */
2815 return ret < 0 ? ret : 0;
Arne Jansena2de7332011-03-08 14:14:00 +01002816}
2817
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002818static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
2819 struct btrfs_device *scrub_dev)
Arne Jansena2de7332011-03-08 14:14:00 +01002820{
2821 int i;
2822 u64 bytenr;
2823 u64 gen;
2824 int ret;
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002825 struct btrfs_root *root = sctx->dev_root;
Arne Jansena2de7332011-03-08 14:14:00 +01002826
Miao Xie87533c42013-01-29 10:14:48 +00002827 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002828 return -EIO;
2829
Arne Jansena2de7332011-03-08 14:14:00 +01002830 gen = root->fs_info->last_trans_committed;
2831
2832 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2833 bytenr = btrfs_sb_offset(i);
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002834 if (bytenr + BTRFS_SUPER_INFO_SIZE > scrub_dev->total_bytes)
Arne Jansena2de7332011-03-08 14:14:00 +01002835 break;
2836
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002837 ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
Stefan Behrensa36cf8b2012-11-02 13:26:57 +01002838 scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
Stefan Behrensff023aa2012-11-06 11:43:11 +01002839 NULL, 1, bytenr);
Arne Jansena2de7332011-03-08 14:14:00 +01002840 if (ret)
2841 return ret;
2842 }
Stefan Behrensb6bfebc2012-11-02 16:44:58 +01002843 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
Arne Jansena2de7332011-03-08 14:14:00 +01002844
2845 return 0;
2846}
2847
2848/*
2849 * get a reference count on fs_info->scrub_workers. start worker if necessary
2850 */
Stefan Behrensff023aa2012-11-06 11:43:11 +01002851static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
2852 int is_dev_replace)
Arne Jansena2de7332011-03-08 14:14:00 +01002853{
Josef Bacik0dc3b842011-11-18 14:37:27 -05002854 int ret = 0;
Qu Wenruo0339ef22014-02-28 10:46:17 +08002855 int flags = WQ_FREEZABLE | WQ_UNBOUND;
2856 int max_active = fs_info->thread_pool_size;
Arne Jansena2de7332011-03-08 14:14:00 +01002857
Arne Jansen632dd772011-06-10 12:07:07 +02002858 if (fs_info->scrub_workers_refcnt == 0) {
Stefan Behrensff023aa2012-11-06 11:43:11 +01002859 if (is_dev_replace)
Qu Wenruo0339ef22014-02-28 10:46:17 +08002860 fs_info->scrub_workers =
2861 btrfs_alloc_workqueue("btrfs-scrub", flags,
2862 1, 4);
Stefan Behrensff023aa2012-11-06 11:43:11 +01002863 else
Qu Wenruo0339ef22014-02-28 10:46:17 +08002864 fs_info->scrub_workers =
2865 btrfs_alloc_workqueue("btrfs-scrub", flags,
2866 max_active, 4);
2867 if (!fs_info->scrub_workers) {
2868 ret = -ENOMEM;
Josef Bacik0dc3b842011-11-18 14:37:27 -05002869 goto out;
Qu Wenruo0339ef22014-02-28 10:46:17 +08002870 }
2871 fs_info->scrub_wr_completion_workers =
2872 btrfs_alloc_workqueue("btrfs-scrubwrc", flags,
2873 max_active, 2);
2874 if (!fs_info->scrub_wr_completion_workers) {
2875 ret = -ENOMEM;
Stefan Behrensff023aa2012-11-06 11:43:11 +01002876 goto out;
Qu Wenruo0339ef22014-02-28 10:46:17 +08002877 }
2878 fs_info->scrub_nocow_workers =
2879 btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0);
2880 if (!fs_info->scrub_nocow_workers) {
2881 ret = -ENOMEM;
Stefan Behrensff023aa2012-11-06 11:43:11 +01002882 goto out;
Qu Wenruo0339ef22014-02-28 10:46:17 +08002883 }
Arne Jansen632dd772011-06-10 12:07:07 +02002884 }
Arne Jansena2de7332011-03-08 14:14:00 +01002885 ++fs_info->scrub_workers_refcnt;
Josef Bacik0dc3b842011-11-18 14:37:27 -05002886out:
Josef Bacik0dc3b842011-11-18 14:37:27 -05002887 return ret;
Arne Jansena2de7332011-03-08 14:14:00 +01002888}
2889
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002890static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
Arne Jansena2de7332011-03-08 14:14:00 +01002891{
Stefan Behrensff023aa2012-11-06 11:43:11 +01002892 if (--fs_info->scrub_workers_refcnt == 0) {
Qu Wenruo0339ef22014-02-28 10:46:17 +08002893 btrfs_destroy_workqueue(fs_info->scrub_workers);
2894 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
2895 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
Stefan Behrensff023aa2012-11-06 11:43:11 +01002896 }
Arne Jansena2de7332011-03-08 14:14:00 +01002897 WARN_ON(fs_info->scrub_workers_refcnt < 0);
Arne Jansena2de7332011-03-08 14:14:00 +01002898}
2899
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002900int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
2901 u64 end, struct btrfs_scrub_progress *progress,
Stefan Behrens63a212a2012-11-05 18:29:28 +01002902 int readonly, int is_dev_replace)
Arne Jansena2de7332011-03-08 14:14:00 +01002903{
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002904 struct scrub_ctx *sctx;
Arne Jansena2de7332011-03-08 14:14:00 +01002905 int ret;
2906 struct btrfs_device *dev;
2907
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002908 if (btrfs_fs_closing(fs_info))
Arne Jansena2de7332011-03-08 14:14:00 +01002909 return -EINVAL;
2910
2911 /*
2912 * check some assumptions
2913 */
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002914 if (fs_info->chunk_root->nodesize != fs_info->chunk_root->leafsize) {
Frank Holtonefe120a2013-12-20 11:37:06 -05002915 btrfs_err(fs_info,
2916 "scrub: size assumption nodesize == leafsize (%d == %d) fails",
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002917 fs_info->chunk_root->nodesize,
2918 fs_info->chunk_root->leafsize);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002919 return -EINVAL;
2920 }
2921
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002922 if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002923 /*
2924 * in this case scrub is unable to calculate the checksum
2925 * the way scrub is implemented. Do not handle this
2926 * situation at all because it won't ever happen.
2927 */
Frank Holtonefe120a2013-12-20 11:37:06 -05002928 btrfs_err(fs_info,
2929 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002930 fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002931 return -EINVAL;
2932 }
2933
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002934 if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
Stefan Behrensb5d67f62012-03-27 14:21:27 -04002935 /* not supported for data w/o checksums */
Frank Holtonefe120a2013-12-20 11:37:06 -05002936 btrfs_err(fs_info,
2937 "scrub: size assumption sectorsize != PAGE_SIZE "
2938 "(%d != %lu) fails",
Geert Uytterhoeven27f9f022013-08-20 13:20:09 +02002939 fs_info->chunk_root->sectorsize, PAGE_SIZE);
Arne Jansena2de7332011-03-08 14:14:00 +01002940 return -EINVAL;
2941 }
2942
Stefan Behrens7a9e9982012-11-02 14:58:04 +01002943 if (fs_info->chunk_root->nodesize >
2944 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
2945 fs_info->chunk_root->sectorsize >
2946 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
2947 /*
2948 * would exhaust the array bounds of pagev member in
2949 * struct scrub_block
2950 */
Frank Holtonefe120a2013-12-20 11:37:06 -05002951 btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
2952 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
Stefan Behrens7a9e9982012-11-02 14:58:04 +01002953 fs_info->chunk_root->nodesize,
2954 SCRUB_MAX_PAGES_PER_BLOCK,
2955 fs_info->chunk_root->sectorsize,
2956 SCRUB_MAX_PAGES_PER_BLOCK);
2957 return -EINVAL;
2958 }
2959
Arne Jansena2de7332011-03-08 14:14:00 +01002960
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002961 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2962 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
Stefan Behrens63a212a2012-11-05 18:29:28 +01002963 if (!dev || (dev->missing && !is_dev_replace)) {
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002964 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
Arne Jansena2de7332011-03-08 14:14:00 +01002965 return -ENODEV;
2966 }
Arne Jansena2de7332011-03-08 14:14:00 +01002967
Wang Shilong3b7a0162013-10-12 02:11:12 +08002968 mutex_lock(&fs_info->scrub_lock);
Stefan Behrens63a212a2012-11-05 18:29:28 +01002969 if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
Arne Jansena2de7332011-03-08 14:14:00 +01002970 mutex_unlock(&fs_info->scrub_lock);
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002971 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002972 return -EIO;
Arne Jansena2de7332011-03-08 14:14:00 +01002973 }
2974
Stefan Behrens8dabb742012-11-06 13:15:27 +01002975 btrfs_dev_replace_lock(&fs_info->dev_replace);
2976 if (dev->scrub_device ||
2977 (!is_dev_replace &&
2978 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
2979 btrfs_dev_replace_unlock(&fs_info->dev_replace);
Arne Jansena2de7332011-03-08 14:14:00 +01002980 mutex_unlock(&fs_info->scrub_lock);
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002981 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
Arne Jansena2de7332011-03-08 14:14:00 +01002982 return -EINPROGRESS;
2983 }
Stefan Behrens8dabb742012-11-06 13:15:27 +01002984 btrfs_dev_replace_unlock(&fs_info->dev_replace);
Wang Shilong3b7a0162013-10-12 02:11:12 +08002985
2986 ret = scrub_workers_get(fs_info, is_dev_replace);
2987 if (ret) {
2988 mutex_unlock(&fs_info->scrub_lock);
2989 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2990 return ret;
2991 }
2992
Stefan Behrens63a212a2012-11-05 18:29:28 +01002993 sctx = scrub_setup_ctx(dev, is_dev_replace);
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002994 if (IS_ERR(sctx)) {
Arne Jansena2de7332011-03-08 14:14:00 +01002995 mutex_unlock(&fs_info->scrub_lock);
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01002996 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2997 scrub_workers_put(fs_info);
Stefan Behrensd9d181c2012-11-02 09:58:09 +01002998 return PTR_ERR(sctx);
Arne Jansena2de7332011-03-08 14:14:00 +01002999 }
Stefan Behrensd9d181c2012-11-02 09:58:09 +01003000 sctx->readonly = readonly;
3001 dev->scrub_device = sctx;
Wang Shilong3cb09292013-12-04 21:15:19 +08003002 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
Arne Jansena2de7332011-03-08 14:14:00 +01003003
Wang Shilong3cb09292013-12-04 21:15:19 +08003004 /*
3005 * checking @scrub_pause_req here, we can avoid
3006 * race between committing transaction and scrubbing.
3007 */
Wang Shilongcb7ab022013-12-04 21:16:53 +08003008 __scrub_blocked_if_needed(fs_info);
Arne Jansena2de7332011-03-08 14:14:00 +01003009 atomic_inc(&fs_info->scrubs_running);
3010 mutex_unlock(&fs_info->scrub_lock);
Arne Jansena2de7332011-03-08 14:14:00 +01003011
Stefan Behrensff023aa2012-11-06 11:43:11 +01003012 if (!is_dev_replace) {
Wang Shilong9b011ad2013-10-25 19:12:02 +08003013 /*
3014 * by holding device list mutex, we can
3015 * kick off writing super in log tree sync.
3016 */
Wang Shilong3cb09292013-12-04 21:15:19 +08003017 mutex_lock(&fs_info->fs_devices->device_list_mutex);
Stefan Behrensff023aa2012-11-06 11:43:11 +01003018 ret = scrub_supers(sctx, dev);
Wang Shilong3cb09292013-12-04 21:15:19 +08003019 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
Stefan Behrensff023aa2012-11-06 11:43:11 +01003020 }
Arne Jansena2de7332011-03-08 14:14:00 +01003021
3022 if (!ret)
Stefan Behrensff023aa2012-11-06 11:43:11 +01003023 ret = scrub_enumerate_chunks(sctx, dev, start, end,
3024 is_dev_replace);
Arne Jansena2de7332011-03-08 14:14:00 +01003025
Stefan Behrensb6bfebc2012-11-02 16:44:58 +01003026 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
Arne Jansena2de7332011-03-08 14:14:00 +01003027 atomic_dec(&fs_info->scrubs_running);
3028 wake_up(&fs_info->scrub_pause_wait);
3029
Stefan Behrensb6bfebc2012-11-02 16:44:58 +01003030 wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
Jan Schmidt0ef8e452011-06-13 20:04:15 +02003031
Arne Jansena2de7332011-03-08 14:14:00 +01003032 if (progress)
Stefan Behrensd9d181c2012-11-02 09:58:09 +01003033 memcpy(progress, &sctx->stat, sizeof(*progress));
Arne Jansena2de7332011-03-08 14:14:00 +01003034
3035 mutex_lock(&fs_info->scrub_lock);
3036 dev->scrub_device = NULL;
Wang Shilong3b7a0162013-10-12 02:11:12 +08003037 scrub_workers_put(fs_info);
Arne Jansena2de7332011-03-08 14:14:00 +01003038 mutex_unlock(&fs_info->scrub_lock);
3039
Stefan Behrensd9d181c2012-11-02 09:58:09 +01003040 scrub_free_ctx(sctx);
Arne Jansena2de7332011-03-08 14:14:00 +01003041
3042 return ret;
3043}
3044
Jeff Mahoney143bede2012-03-01 14:56:26 +01003045void btrfs_scrub_pause(struct btrfs_root *root)
Arne Jansena2de7332011-03-08 14:14:00 +01003046{
3047 struct btrfs_fs_info *fs_info = root->fs_info;
3048
3049 mutex_lock(&fs_info->scrub_lock);
3050 atomic_inc(&fs_info->scrub_pause_req);
3051 while (atomic_read(&fs_info->scrubs_paused) !=
3052 atomic_read(&fs_info->scrubs_running)) {
3053 mutex_unlock(&fs_info->scrub_lock);
3054 wait_event(fs_info->scrub_pause_wait,
3055 atomic_read(&fs_info->scrubs_paused) ==
3056 atomic_read(&fs_info->scrubs_running));
3057 mutex_lock(&fs_info->scrub_lock);
3058 }
3059 mutex_unlock(&fs_info->scrub_lock);
Arne Jansena2de7332011-03-08 14:14:00 +01003060}
3061
Jeff Mahoney143bede2012-03-01 14:56:26 +01003062void btrfs_scrub_continue(struct btrfs_root *root)
Arne Jansena2de7332011-03-08 14:14:00 +01003063{
3064 struct btrfs_fs_info *fs_info = root->fs_info;
3065
3066 atomic_dec(&fs_info->scrub_pause_req);
3067 wake_up(&fs_info->scrub_pause_wait);
Arne Jansena2de7332011-03-08 14:14:00 +01003068}
3069
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01003070int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
Arne Jansena2de7332011-03-08 14:14:00 +01003071{
Arne Jansena2de7332011-03-08 14:14:00 +01003072 mutex_lock(&fs_info->scrub_lock);
3073 if (!atomic_read(&fs_info->scrubs_running)) {
3074 mutex_unlock(&fs_info->scrub_lock);
3075 return -ENOTCONN;
3076 }
3077
3078 atomic_inc(&fs_info->scrub_cancel_req);
3079 while (atomic_read(&fs_info->scrubs_running)) {
3080 mutex_unlock(&fs_info->scrub_lock);
3081 wait_event(fs_info->scrub_pause_wait,
3082 atomic_read(&fs_info->scrubs_running) == 0);
3083 mutex_lock(&fs_info->scrub_lock);
3084 }
3085 atomic_dec(&fs_info->scrub_cancel_req);
3086 mutex_unlock(&fs_info->scrub_lock);
3087
3088 return 0;
3089}
3090
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01003091int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
3092 struct btrfs_device *dev)
Jeff Mahoney49b25e02012-03-01 17:24:58 +01003093{
Stefan Behrensd9d181c2012-11-02 09:58:09 +01003094 struct scrub_ctx *sctx;
Arne Jansena2de7332011-03-08 14:14:00 +01003095
3096 mutex_lock(&fs_info->scrub_lock);
Stefan Behrensd9d181c2012-11-02 09:58:09 +01003097 sctx = dev->scrub_device;
3098 if (!sctx) {
Arne Jansena2de7332011-03-08 14:14:00 +01003099 mutex_unlock(&fs_info->scrub_lock);
3100 return -ENOTCONN;
3101 }
Stefan Behrensd9d181c2012-11-02 09:58:09 +01003102 atomic_inc(&sctx->cancel_req);
Arne Jansena2de7332011-03-08 14:14:00 +01003103 while (dev->scrub_device) {
3104 mutex_unlock(&fs_info->scrub_lock);
3105 wait_event(fs_info->scrub_pause_wait,
3106 dev->scrub_device == NULL);
3107 mutex_lock(&fs_info->scrub_lock);
3108 }
3109 mutex_unlock(&fs_info->scrub_lock);
3110
3111 return 0;
3112}
Stefan Behrens1623ede2012-03-27 14:21:26 -04003113
Arne Jansena2de7332011-03-08 14:14:00 +01003114int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
3115 struct btrfs_scrub_progress *progress)
3116{
3117 struct btrfs_device *dev;
Stefan Behrensd9d181c2012-11-02 09:58:09 +01003118 struct scrub_ctx *sctx = NULL;
Arne Jansena2de7332011-03-08 14:14:00 +01003119
3120 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
Stefan Behrensaa1b8cd2012-11-05 17:03:39 +01003121 dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
Arne Jansena2de7332011-03-08 14:14:00 +01003122 if (dev)
Stefan Behrensd9d181c2012-11-02 09:58:09 +01003123 sctx = dev->scrub_device;
3124 if (sctx)
3125 memcpy(progress, &sctx->stat, sizeof(*progress));
Arne Jansena2de7332011-03-08 14:14:00 +01003126 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3127
Stefan Behrensd9d181c2012-11-02 09:58:09 +01003128 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
Arne Jansena2de7332011-03-08 14:14:00 +01003129}
Stefan Behrensff023aa2012-11-06 11:43:11 +01003130
3131static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
3132 u64 extent_logical, u64 extent_len,
3133 u64 *extent_physical,
3134 struct btrfs_device **extent_dev,
3135 int *extent_mirror_num)
3136{
3137 u64 mapped_length;
3138 struct btrfs_bio *bbio = NULL;
3139 int ret;
3140
3141 mapped_length = extent_len;
3142 ret = btrfs_map_block(fs_info, READ, extent_logical,
3143 &mapped_length, &bbio, 0);
3144 if (ret || !bbio || mapped_length < extent_len ||
3145 !bbio->stripes[0].dev->bdev) {
3146 kfree(bbio);
3147 return;
3148 }
3149
3150 *extent_physical = bbio->stripes[0].physical;
3151 *extent_mirror_num = bbio->mirror_num;
3152 *extent_dev = bbio->stripes[0].dev;
3153 kfree(bbio);
3154}
3155
3156static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
3157 struct scrub_wr_ctx *wr_ctx,
3158 struct btrfs_fs_info *fs_info,
3159 struct btrfs_device *dev,
3160 int is_dev_replace)
3161{
3162 WARN_ON(wr_ctx->wr_curr_bio != NULL);
3163
3164 mutex_init(&wr_ctx->wr_lock);
3165 wr_ctx->wr_curr_bio = NULL;
3166 if (!is_dev_replace)
3167 return 0;
3168
3169 WARN_ON(!dev->bdev);
3170 wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO,
3171 bio_get_nr_vecs(dev->bdev));
3172 wr_ctx->tgtdev = dev;
3173 atomic_set(&wr_ctx->flush_all_writes, 0);
3174 return 0;
3175}
3176
3177static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx)
3178{
3179 mutex_lock(&wr_ctx->wr_lock);
3180 kfree(wr_ctx->wr_curr_bio);
3181 wr_ctx->wr_curr_bio = NULL;
3182 mutex_unlock(&wr_ctx->wr_lock);
3183}
3184
3185static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
3186 int mirror_num, u64 physical_for_dev_replace)
3187{
3188 struct scrub_copy_nocow_ctx *nocow_ctx;
3189 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
3190
3191 nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS);
3192 if (!nocow_ctx) {
3193 spin_lock(&sctx->stat_lock);
3194 sctx->stat.malloc_errors++;
3195 spin_unlock(&sctx->stat_lock);
3196 return -ENOMEM;
3197 }
3198
3199 scrub_pending_trans_workers_inc(sctx);
3200
3201 nocow_ctx->sctx = sctx;
3202 nocow_ctx->logical = logical;
3203 nocow_ctx->len = len;
3204 nocow_ctx->mirror_num = mirror_num;
3205 nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
Qu Wenruo0339ef22014-02-28 10:46:17 +08003206 btrfs_init_work(&nocow_ctx->work, copy_nocow_pages_worker, NULL, NULL);
Josef Bacik652f25a2013-09-12 16:58:28 -04003207 INIT_LIST_HEAD(&nocow_ctx->inodes);
Qu Wenruo0339ef22014-02-28 10:46:17 +08003208 btrfs_queue_work(fs_info->scrub_nocow_workers,
3209 &nocow_ctx->work);
Stefan Behrensff023aa2012-11-06 11:43:11 +01003210
3211 return 0;
3212}
3213
Josef Bacik652f25a2013-09-12 16:58:28 -04003214static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx)
3215{
3216 struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
3217 struct scrub_nocow_inode *nocow_inode;
3218
3219 nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS);
3220 if (!nocow_inode)
3221 return -ENOMEM;
3222 nocow_inode->inum = inum;
3223 nocow_inode->offset = offset;
3224 nocow_inode->root = root;
3225 list_add_tail(&nocow_inode->list, &nocow_ctx->inodes);
3226 return 0;
3227}
3228
3229#define COPY_COMPLETE 1
3230
Stefan Behrensff023aa2012-11-06 11:43:11 +01003231static void copy_nocow_pages_worker(struct btrfs_work *work)
3232{
3233 struct scrub_copy_nocow_ctx *nocow_ctx =
3234 container_of(work, struct scrub_copy_nocow_ctx, work);
3235 struct scrub_ctx *sctx = nocow_ctx->sctx;
3236 u64 logical = nocow_ctx->logical;
3237 u64 len = nocow_ctx->len;
3238 int mirror_num = nocow_ctx->mirror_num;
3239 u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
3240 int ret;
3241 struct btrfs_trans_handle *trans = NULL;
3242 struct btrfs_fs_info *fs_info;
3243 struct btrfs_path *path;
3244 struct btrfs_root *root;
3245 int not_written = 0;
3246
3247 fs_info = sctx->dev_root->fs_info;
3248 root = fs_info->extent_root;
3249
3250 path = btrfs_alloc_path();
3251 if (!path) {
3252 spin_lock(&sctx->stat_lock);
3253 sctx->stat.malloc_errors++;
3254 spin_unlock(&sctx->stat_lock);
3255 not_written = 1;
3256 goto out;
3257 }
3258
3259 trans = btrfs_join_transaction(root);
3260 if (IS_ERR(trans)) {
3261 not_written = 1;
3262 goto out;
3263 }
3264
3265 ret = iterate_inodes_from_logical(logical, fs_info, path,
Josef Bacik652f25a2013-09-12 16:58:28 -04003266 record_inode_for_nocow, nocow_ctx);
Stefan Behrensff023aa2012-11-06 11:43:11 +01003267 if (ret != 0 && ret != -ENOENT) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003268 btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
3269 "phys %llu, len %llu, mir %u, ret %d",
Geert Uytterhoeven118a0a22013-08-20 13:20:10 +02003270 logical, physical_for_dev_replace, len, mirror_num,
3271 ret);
Stefan Behrensff023aa2012-11-06 11:43:11 +01003272 not_written = 1;
3273 goto out;
3274 }
3275
Josef Bacik652f25a2013-09-12 16:58:28 -04003276 btrfs_end_transaction(trans, root);
3277 trans = NULL;
3278 while (!list_empty(&nocow_ctx->inodes)) {
3279 struct scrub_nocow_inode *entry;
3280 entry = list_first_entry(&nocow_ctx->inodes,
3281 struct scrub_nocow_inode,
3282 list);
3283 list_del_init(&entry->list);
3284 ret = copy_nocow_pages_for_inode(entry->inum, entry->offset,
3285 entry->root, nocow_ctx);
3286 kfree(entry);
3287 if (ret == COPY_COMPLETE) {
3288 ret = 0;
3289 break;
3290 } else if (ret) {
3291 break;
3292 }
3293 }
Stefan Behrensff023aa2012-11-06 11:43:11 +01003294out:
Josef Bacik652f25a2013-09-12 16:58:28 -04003295 while (!list_empty(&nocow_ctx->inodes)) {
3296 struct scrub_nocow_inode *entry;
3297 entry = list_first_entry(&nocow_ctx->inodes,
3298 struct scrub_nocow_inode,
3299 list);
3300 list_del_init(&entry->list);
3301 kfree(entry);
3302 }
Stefan Behrensff023aa2012-11-06 11:43:11 +01003303 if (trans && !IS_ERR(trans))
3304 btrfs_end_transaction(trans, root);
3305 if (not_written)
3306 btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
3307 num_uncorrectable_read_errors);
3308
3309 btrfs_free_path(path);
3310 kfree(nocow_ctx);
3311
3312 scrub_pending_trans_workers_dec(sctx);
3313}
3314
Josef Bacik652f25a2013-09-12 16:58:28 -04003315static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
3316 struct scrub_copy_nocow_ctx *nocow_ctx)
Stefan Behrensff023aa2012-11-06 11:43:11 +01003317{
Miao Xie826aa0a2013-06-27 18:50:59 +08003318 struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
Stefan Behrensff023aa2012-11-06 11:43:11 +01003319 struct btrfs_key key;
Miao Xie826aa0a2013-06-27 18:50:59 +08003320 struct inode *inode;
3321 struct page *page;
Stefan Behrensff023aa2012-11-06 11:43:11 +01003322 struct btrfs_root *local_root;
Josef Bacik652f25a2013-09-12 16:58:28 -04003323 struct btrfs_ordered_extent *ordered;
3324 struct extent_map *em;
3325 struct extent_state *cached_state = NULL;
3326 struct extent_io_tree *io_tree;
Stefan Behrensff023aa2012-11-06 11:43:11 +01003327 u64 physical_for_dev_replace;
Josef Bacik652f25a2013-09-12 16:58:28 -04003328 u64 len = nocow_ctx->len;
3329 u64 lockstart = offset, lockend = offset + len - 1;
Miao Xie826aa0a2013-06-27 18:50:59 +08003330 unsigned long index;
Liu Bo6f1c3602013-01-29 03:22:10 +00003331 int srcu_index;
Josef Bacik652f25a2013-09-12 16:58:28 -04003332 int ret = 0;
3333 int err = 0;
Stefan Behrensff023aa2012-11-06 11:43:11 +01003334
3335 key.objectid = root;
3336 key.type = BTRFS_ROOT_ITEM_KEY;
3337 key.offset = (u64)-1;
Liu Bo6f1c3602013-01-29 03:22:10 +00003338
3339 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
3340
Stefan Behrensff023aa2012-11-06 11:43:11 +01003341 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
Liu Bo6f1c3602013-01-29 03:22:10 +00003342 if (IS_ERR(local_root)) {
3343 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
Stefan Behrensff023aa2012-11-06 11:43:11 +01003344 return PTR_ERR(local_root);
Liu Bo6f1c3602013-01-29 03:22:10 +00003345 }
Stefan Behrensff023aa2012-11-06 11:43:11 +01003346
3347 key.type = BTRFS_INODE_ITEM_KEY;
3348 key.objectid = inum;
3349 key.offset = 0;
3350 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
Liu Bo6f1c3602013-01-29 03:22:10 +00003351 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
Stefan Behrensff023aa2012-11-06 11:43:11 +01003352 if (IS_ERR(inode))
3353 return PTR_ERR(inode);
3354
Miao Xieedd14002013-06-27 18:51:00 +08003355 /* Avoid truncate/dio/punch hole.. */
3356 mutex_lock(&inode->i_mutex);
3357 inode_dio_wait(inode);
3358
Stefan Behrensff023aa2012-11-06 11:43:11 +01003359 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
Josef Bacik652f25a2013-09-12 16:58:28 -04003360 io_tree = &BTRFS_I(inode)->io_tree;
3361
3362 lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state);
3363 ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
3364 if (ordered) {
3365 btrfs_put_ordered_extent(ordered);
3366 goto out_unlock;
3367 }
3368
3369 em = btrfs_get_extent(inode, NULL, 0, lockstart, len, 0);
3370 if (IS_ERR(em)) {
3371 ret = PTR_ERR(em);
3372 goto out_unlock;
3373 }
3374
3375 /*
3376 * This extent does not actually cover the logical extent anymore,
3377 * move on to the next inode.
3378 */
3379 if (em->block_start > nocow_ctx->logical ||
3380 em->block_start + em->block_len < nocow_ctx->logical + len) {
3381 free_extent_map(em);
3382 goto out_unlock;
3383 }
3384 free_extent_map(em);
3385
Stefan Behrensff023aa2012-11-06 11:43:11 +01003386 while (len >= PAGE_CACHE_SIZE) {
Stefan Behrensff023aa2012-11-06 11:43:11 +01003387 index = offset >> PAGE_CACHE_SHIFT;
Miao Xieedd14002013-06-27 18:51:00 +08003388again:
Stefan Behrensff023aa2012-11-06 11:43:11 +01003389 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
3390 if (!page) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003391 btrfs_err(fs_info, "find_or_create_page() failed");
Stefan Behrensff023aa2012-11-06 11:43:11 +01003392 ret = -ENOMEM;
Miao Xie826aa0a2013-06-27 18:50:59 +08003393 goto out;
Stefan Behrensff023aa2012-11-06 11:43:11 +01003394 }
3395
3396 if (PageUptodate(page)) {
3397 if (PageDirty(page))
3398 goto next_page;
3399 } else {
3400 ClearPageError(page);
Josef Bacik652f25a2013-09-12 16:58:28 -04003401 err = extent_read_full_page_nolock(io_tree, page,
3402 btrfs_get_extent,
3403 nocow_ctx->mirror_num);
Miao Xie826aa0a2013-06-27 18:50:59 +08003404 if (err) {
3405 ret = err;
Stefan Behrensff023aa2012-11-06 11:43:11 +01003406 goto next_page;
3407 }
Miao Xieedd14002013-06-27 18:51:00 +08003408
Miao Xie26b258912013-06-27 18:50:58 +08003409 lock_page(page);
Miao Xieedd14002013-06-27 18:51:00 +08003410 /*
3411 * If the page has been remove from the page cache,
3412 * the data on it is meaningless, because it may be
3413 * old one, the new data may be written into the new
3414 * page in the page cache.
3415 */
3416 if (page->mapping != inode->i_mapping) {
Josef Bacik652f25a2013-09-12 16:58:28 -04003417 unlock_page(page);
Miao Xieedd14002013-06-27 18:51:00 +08003418 page_cache_release(page);
3419 goto again;
3420 }
Stefan Behrensff023aa2012-11-06 11:43:11 +01003421 if (!PageUptodate(page)) {
3422 ret = -EIO;
3423 goto next_page;
3424 }
3425 }
Miao Xie826aa0a2013-06-27 18:50:59 +08003426 err = write_page_nocow(nocow_ctx->sctx,
3427 physical_for_dev_replace, page);
3428 if (err)
3429 ret = err;
Stefan Behrensff023aa2012-11-06 11:43:11 +01003430next_page:
Miao Xie826aa0a2013-06-27 18:50:59 +08003431 unlock_page(page);
3432 page_cache_release(page);
3433
3434 if (ret)
3435 break;
3436
Stefan Behrensff023aa2012-11-06 11:43:11 +01003437 offset += PAGE_CACHE_SIZE;
3438 physical_for_dev_replace += PAGE_CACHE_SIZE;
3439 len -= PAGE_CACHE_SIZE;
3440 }
Josef Bacik652f25a2013-09-12 16:58:28 -04003441 ret = COPY_COMPLETE;
3442out_unlock:
3443 unlock_extent_cached(io_tree, lockstart, lockend, &cached_state,
3444 GFP_NOFS);
Miao Xie826aa0a2013-06-27 18:50:59 +08003445out:
Miao Xieedd14002013-06-27 18:51:00 +08003446 mutex_unlock(&inode->i_mutex);
Miao Xie826aa0a2013-06-27 18:50:59 +08003447 iput(inode);
Stefan Behrensff023aa2012-11-06 11:43:11 +01003448 return ret;
3449}
3450
3451static int write_page_nocow(struct scrub_ctx *sctx,
3452 u64 physical_for_dev_replace, struct page *page)
3453{
3454 struct bio *bio;
3455 struct btrfs_device *dev;
3456 int ret;
Stefan Behrensff023aa2012-11-06 11:43:11 +01003457
3458 dev = sctx->wr_ctx.tgtdev;
3459 if (!dev)
3460 return -EIO;
3461 if (!dev->bdev) {
3462 printk_ratelimited(KERN_WARNING
Frank Holtonefe120a2013-12-20 11:37:06 -05003463 "BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n");
Stefan Behrensff023aa2012-11-06 11:43:11 +01003464 return -EIO;
3465 }
Chris Mason9be33952013-05-17 18:30:14 -04003466 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
Stefan Behrensff023aa2012-11-06 11:43:11 +01003467 if (!bio) {
3468 spin_lock(&sctx->stat_lock);
3469 sctx->stat.malloc_errors++;
3470 spin_unlock(&sctx->stat_lock);
3471 return -ENOMEM;
3472 }
Kent Overstreet4f024f32013-10-11 15:44:27 -07003473 bio->bi_iter.bi_size = 0;
3474 bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
Stefan Behrensff023aa2012-11-06 11:43:11 +01003475 bio->bi_bdev = dev->bdev;
3476 ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
3477 if (ret != PAGE_CACHE_SIZE) {
3478leave_with_eio:
3479 bio_put(bio);
3480 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
3481 return -EIO;
3482 }
Stefan Behrensff023aa2012-11-06 11:43:11 +01003483
Kent Overstreet33879d42013-11-23 22:33:32 -08003484 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
Stefan Behrensff023aa2012-11-06 11:43:11 +01003485 goto leave_with_eio;
3486
3487 bio_put(bio);
3488 return 0;
3489}