Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1// SPDX-License-Identifier: GPL-2.0
2/* Maximum size of each resync request */
3#define RESYNC_BLOCK_SIZE (64*1024)
4#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
5
6/* when we get a read error on a read-only array, we redirect to another
7 * device without failing the first device, or trying to over-write to
8 * correct the read error. To keep track of bad blocks on a per-bio
9 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
10 */
11#define IO_BLOCKED ((struct bio *)1)
12/* When we successfully write to a known bad-block, we need to remove the
13 * bad-block marking which must be done from process context. So we record
14 * the success by setting devs[n].bio to IO_MADE_GOOD
15 */
16#define IO_MADE_GOOD ((struct bio *)2)
17
18#define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
19#define MAX_PLUG_BIO 32
20
21/* for managing resync I/O pages */
22struct resync_pages {
23 void *raid_bio;
24 struct page *pages[RESYNC_PAGES];
25};
26
27struct raid1_plug_cb {
28 struct blk_plug_cb cb;
29 struct bio_list pending;
30 unsigned int count;
31};
32
33static void rbio_pool_free(void *rbio, void *data)
34{
35 kfree(rbio);
36}
37
38static inline int resync_alloc_pages(struct resync_pages *rp,
39 gfp_t gfp_flags)
40{
41 int i;
42
43 for (i = 0; i < RESYNC_PAGES; i++) {
44 rp->pages[i] = alloc_page(gfp_flags);
45 if (!rp->pages[i])
46 goto out_free;
47 }
48
49 return 0;
50
51out_free:
52 while (--i >= 0)
53 put_page(rp->pages[i]);
54 return -ENOMEM;
55}
56
57static inline void resync_free_pages(struct resync_pages *rp)
58{
59 int i;
60
61 for (i = 0; i < RESYNC_PAGES; i++)
62 put_page(rp->pages[i]);
63}
64
65static inline void resync_get_all_pages(struct resync_pages *rp)
66{
67 int i;
68
69 for (i = 0; i < RESYNC_PAGES; i++)
70 get_page(rp->pages[i]);
71}
72
73static inline struct page *resync_fetch_page(struct resync_pages *rp,
74 unsigned idx)
75{
76 if (WARN_ON_ONCE(idx >= RESYNC_PAGES))
77 return NULL;
78 return rp->pages[idx];
79}
80
81/*
82 * 'strct resync_pages' stores actual pages used for doing the resync
83 * IO, and it is per-bio, so make .bi_private points to it.
84 */
85static inline struct resync_pages *get_resync_pages(struct bio *bio)
86{
87 return bio->bi_private;
88}
89
90/* generally called after bio_reset() for reseting bvec */
91static void md_bio_reset_resync_pages(struct bio *bio, struct resync_pages *rp,
92 int size)
93{
94 int idx = 0;
95
96 /* initialize bvec table again */
97 do {
98 struct page *page = resync_fetch_page(rp, idx);
99 int len = min_t(int, size, PAGE_SIZE);
100
101 if (WARN_ON(!bio_add_page(bio, page, len, 0))) {
102 bio->bi_status = BLK_STS_RESOURCE;
103 bio_endio(bio);
104 return;
105 }
106
107 size -= len;
108 } while (idx++ < RESYNC_PAGES && size > 0);
109}
110
111
112static inline void raid1_submit_write(struct bio *bio)
113{
114 struct md_rdev *rdev = (void *)bio->bi_bdev;
115
116 bio->bi_next = NULL;
117 bio_set_dev(bio, rdev->bdev);
118 if (test_bit(Faulty, &rdev->flags))
119 bio_io_error(bio);
120 else if (unlikely(bio_op(bio) == REQ_OP_DISCARD &&
121 !bdev_max_discard_sectors(bio->bi_bdev)))
122 /* Just ignore it */
123 bio_endio(bio);
124 else
125 submit_bio_noacct(bio);
126}
127
128static inline bool raid1_add_bio_to_plug(struct mddev *mddev, struct bio *bio,
129 blk_plug_cb_fn unplug, int copies)
130{
131 struct raid1_plug_cb *plug = NULL;
132 struct blk_plug_cb *cb;
133
134 /*
135 * If bitmap is not enabled, it's safe to submit the io directly, and
136 * this can get optimal performance.
137 */
138 if (!md_bitmap_enabled(mddev, true)) {
139 raid1_submit_write(bio);
140 return true;
141 }
142
143 cb = blk_check_plugged(unplug, mddev, sizeof(*plug));
144 if (!cb)
145 return false;
146
147 plug = container_of(cb, struct raid1_plug_cb, cb);
148 bio_list_add(&plug->pending, bio);
149 if (++plug->count / MAX_PLUG_BIO >= copies) {
150 list_del(&cb->list);
151 cb->callback(cb, false);
152 }
153
154
155 return true;
156}
157
158/*
159 * current->bio_list will be set under submit_bio() context, in this case bitmap
160 * io will be added to the list and wait for current io submission to finish,
161 * while current io submission must wait for bitmap io to be done. In order to
162 * avoid such deadlock, submit bitmap io asynchronously.
163 */
164static inline void raid1_prepare_flush_writes(struct mddev *mddev)
165{
166 mddev->bitmap_ops->unplug(mddev, current->bio_list == NULL);
167}
168
169/*
170 * Used by fix_read_error() to decay the per rdev read_errors.
171 * We halve the read error count for every hour that has elapsed
172 * since the last recorded read error.
173 */
174static inline void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev)
175{
176 long cur_time_mon;
177 unsigned long hours_since_last;
178 unsigned int read_errors = atomic_read(&rdev->read_errors);
179
180 cur_time_mon = ktime_get_seconds();
181
182 if (rdev->last_read_error == 0) {
183 /* first time we've seen a read error */
184 rdev->last_read_error = cur_time_mon;
185 return;
186 }
187
188 hours_since_last = (long)(cur_time_mon -
189 rdev->last_read_error) / 3600;
190
191 rdev->last_read_error = cur_time_mon;
192
193 /*
194 * if hours_since_last is > the number of bits in read_errors
195 * just set read errors to 0. We do this to avoid
196 * overflowing the shift of read_errors by hours_since_last.
197 */
198 if (hours_since_last >= 8 * sizeof(read_errors))
199 atomic_set(&rdev->read_errors, 0);
200 else
201 atomic_set(&rdev->read_errors, read_errors >> hours_since_last);
202}
203
204static inline bool exceed_read_errors(struct mddev *mddev, struct md_rdev *rdev)
205{
206 int max_read_errors = atomic_read(&mddev->max_corr_read_errors);
207 int read_errors;
208
209 check_decay_read_errors(mddev, rdev);
210 read_errors = atomic_inc_return(&rdev->read_errors);
211 if (read_errors > max_read_errors) {
212 pr_notice("md/"RAID_1_10_NAME":%s: %pg: Raid device exceeded read_error threshold [cur %d:max %d]\n",
213 mdname(mddev), rdev->bdev, read_errors, max_read_errors);
214 pr_notice("md/"RAID_1_10_NAME":%s: %pg: Failing raid device\n",
215 mdname(mddev), rdev->bdev);
216 md_error(mddev, rdev);
217 return true;
218 }
219
220 return false;
221}
222
223/**
224 * raid1_check_read_range() - check a given read range for bad blocks,
225 * available read length is returned;
226 * @rdev: the rdev to read;
227 * @this_sector: read position;
228 * @len: read length;
229 *
230 * helper function for read_balance()
231 *
232 * 1) If there are no bad blocks in the range, @len is returned;
233 * 2) If the range are all bad blocks, 0 is returned;
234 * 3) If there are partial bad blocks:
235 * - If the bad block range starts after @this_sector, the length of first
236 * good region is returned;
237 * - If the bad block range starts before @this_sector, 0 is returned and
238 * the @len is updated to the offset into the region before we get to the
239 * good blocks;
240 */
241static inline int raid1_check_read_range(struct md_rdev *rdev,
242 sector_t this_sector, int *len)
243{
244 sector_t first_bad;
245 sector_t bad_sectors;
246
247 /* no bad block overlap */
248 if (!is_badblock(rdev, this_sector, *len, &first_bad, &bad_sectors))
249 return *len;
250
251 /*
252 * bad block range starts offset into our range so we can return the
253 * number of sectors before the bad blocks start.
254 */
255 if (first_bad > this_sector)
256 return first_bad - this_sector;
257
258 /* read range is fully consumed by bad blocks. */
259 if (this_sector + *len <= first_bad + bad_sectors)
260 return 0;
261
262 /*
263 * final case, bad block range starts before or at the start of our
264 * range but does not cover our entire range so we still return 0 but
265 * update the length with the number of sectors before we get to the
266 * good ones.
267 */
268 *len = first_bad + bad_sectors - this_sector;
269 return 0;
270}
271
272/*
273 * Check if read should choose the first rdev.
274 *
275 * Balance on the whole device if no resync is going on (recovery is ok) or
276 * below the resync window. Otherwise, take the first readable disk.
277 */
278static inline bool raid1_should_read_first(struct mddev *mddev,
279 sector_t this_sector, int len)
280{
281 if ((mddev->resync_offset < this_sector + len))
282 return true;
283
284 if (mddev_is_clustered(mddev) &&
285 mddev->cluster_ops->area_resyncing(mddev, READ, this_sector,
286 this_sector + len))
287 return true;
288
289 return false;
290}
291
292/*
293 * bio with REQ_RAHEAD or REQ_NOWAIT can fail at anytime, before such IO is
294 * submitted to the underlying disks, hence don't record badblocks or retry
295 * in this case.
296 *
297 * BLK_STS_INVAL means the bio was not valid for the underlying device. This
298 * is a user error, not a device failure, so retrying or recording bad blocks
299 * would be wrong.
300 */
301static inline bool raid1_should_handle_error(struct bio *bio)
302{
303 return !(bio->bi_opf & (REQ_RAHEAD | REQ_NOWAIT)) &&
304 bio->bi_status != BLK_STS_INVAL;
305}