drivers/md/raid1-10.c

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