David Sterba | c1d7c51 | 2018-04-03 19:23:33 +0200 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2 | /* |
| 3 | * Copyright (C) 2012 Fusion-io All rights reserved. |
| 4 | * Copyright (C) 2012 Intel Corp. All rights reserved. |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 5 | */ |
David Sterba | c1d7c51 | 2018-04-03 19:23:33 +0200 | [diff] [blame] | 6 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 7 | #include <linux/sched.h> |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 8 | #include <linux/bio.h> |
| 9 | #include <linux/slab.h> |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 10 | #include <linux/blkdev.h> |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 11 | #include <linux/raid/pq.h> |
| 12 | #include <linux/hash.h> |
| 13 | #include <linux/list_sort.h> |
| 14 | #include <linux/raid/xor.h> |
David Sterba | 818e010 | 2017-05-31 18:40:02 +0200 | [diff] [blame] | 15 | #include <linux/mm.h> |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 16 | #include "ctree.h" |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 17 | #include "disk-io.h" |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 18 | #include "volumes.h" |
| 19 | #include "raid56.h" |
| 20 | #include "async-thread.h" |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 21 | |
| 22 | /* set when additional merges to this rbio are not allowed */ |
| 23 | #define RBIO_RMW_LOCKED_BIT 1 |
| 24 | |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 25 | /* |
| 26 | * set when this rbio is sitting in the hash, but it is just a cache |
| 27 | * of past RMW |
| 28 | */ |
| 29 | #define RBIO_CACHE_BIT 2 |
| 30 | |
| 31 | /* |
| 32 | * set when it is safe to trust the stripe_pages for caching |
| 33 | */ |
| 34 | #define RBIO_CACHE_READY_BIT 3 |
| 35 | |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 36 | #define RBIO_CACHE_SIZE 1024 |
| 37 | |
Miao Xie | 1b94b55 | 2014-11-06 16:14:21 +0800 | [diff] [blame] | 38 | enum btrfs_rbio_ops { |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 39 | BTRFS_RBIO_WRITE, |
| 40 | BTRFS_RBIO_READ_REBUILD, |
| 41 | BTRFS_RBIO_PARITY_SCRUB, |
| 42 | BTRFS_RBIO_REBUILD_MISSING, |
Miao Xie | 1b94b55 | 2014-11-06 16:14:21 +0800 | [diff] [blame] | 43 | }; |
| 44 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 45 | struct btrfs_raid_bio { |
| 46 | struct btrfs_fs_info *fs_info; |
| 47 | struct btrfs_bio *bbio; |
| 48 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 49 | /* while we're doing rmw on a stripe |
| 50 | * we put it into a hash table so we can |
| 51 | * lock the stripe and merge more rbios |
| 52 | * into it. |
| 53 | */ |
| 54 | struct list_head hash_list; |
| 55 | |
| 56 | /* |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 57 | * LRU list for the stripe cache |
| 58 | */ |
| 59 | struct list_head stripe_cache; |
| 60 | |
| 61 | /* |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 62 | * for scheduling work in the helper threads |
| 63 | */ |
| 64 | struct btrfs_work work; |
| 65 | |
| 66 | /* |
| 67 | * bio list and bio_list_lock are used |
| 68 | * to add more bios into the stripe |
| 69 | * in hopes of avoiding the full rmw |
| 70 | */ |
| 71 | struct bio_list bio_list; |
| 72 | spinlock_t bio_list_lock; |
| 73 | |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 74 | /* also protected by the bio_list_lock, the |
| 75 | * plug list is used by the plugging code |
| 76 | * to collect partial bios while plugged. The |
| 77 | * stripe locking code also uses it to hand off |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 78 | * the stripe lock to the next pending IO |
| 79 | */ |
| 80 | struct list_head plug_list; |
| 81 | |
| 82 | /* |
| 83 | * flags that tell us if it is safe to |
| 84 | * merge with this bio |
| 85 | */ |
| 86 | unsigned long flags; |
| 87 | |
| 88 | /* size of each individual stripe on disk */ |
| 89 | int stripe_len; |
| 90 | |
| 91 | /* number of data stripes (no p/q) */ |
| 92 | int nr_data; |
| 93 | |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 94 | int real_stripes; |
| 95 | |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 96 | int stripe_npages; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 97 | /* |
| 98 | * set if we're doing a parity rebuild |
| 99 | * for a read from higher up, which is handled |
| 100 | * differently from a parity rebuild as part of |
| 101 | * rmw |
| 102 | */ |
Miao Xie | 1b94b55 | 2014-11-06 16:14:21 +0800 | [diff] [blame] | 103 | enum btrfs_rbio_ops operation; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 104 | |
| 105 | /* first bad stripe */ |
| 106 | int faila; |
| 107 | |
| 108 | /* second bad stripe (for raid6 use) */ |
| 109 | int failb; |
| 110 | |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 111 | int scrubp; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 112 | /* |
| 113 | * number of pages needed to represent the full |
| 114 | * stripe |
| 115 | */ |
| 116 | int nr_pages; |
| 117 | |
| 118 | /* |
| 119 | * size of all the bios in the bio_list. This |
| 120 | * helps us decide if the rbio maps to a full |
| 121 | * stripe or not |
| 122 | */ |
| 123 | int bio_list_bytes; |
| 124 | |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 125 | int generic_bio_cnt; |
| 126 | |
Elena Reshetova | dec9557 | 2017-03-03 10:55:26 +0200 | [diff] [blame] | 127 | refcount_t refs; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 128 | |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 129 | atomic_t stripes_pending; |
| 130 | |
| 131 | atomic_t error; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 132 | /* |
| 133 | * these are two arrays of pointers. We allocate the |
| 134 | * rbio big enough to hold them both and setup their |
| 135 | * locations when the rbio is allocated |
| 136 | */ |
| 137 | |
| 138 | /* pointers to pages that we allocated for |
| 139 | * reading/writing stripes directly from the disk (including P/Q) |
| 140 | */ |
| 141 | struct page **stripe_pages; |
| 142 | |
| 143 | /* |
| 144 | * pointers to the pages in the bio_list. Stored |
| 145 | * here for faster lookup |
| 146 | */ |
| 147 | struct page **bio_pages; |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 148 | |
| 149 | /* |
| 150 | * bitmap to record which horizontal stripe has data |
| 151 | */ |
| 152 | unsigned long *dbitmap; |
Kees Cook | 1389053 | 2018-05-29 16:44:59 -0700 | [diff] [blame] | 153 | |
| 154 | /* allocated with real_stripes-many pointers for finish_*() calls */ |
| 155 | void **finish_pointers; |
| 156 | |
| 157 | /* allocated with stripe_npages-many bits for finish_*() calls */ |
| 158 | unsigned long *finish_pbitmap; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 159 | }; |
| 160 | |
| 161 | static int __raid56_parity_recover(struct btrfs_raid_bio *rbio); |
| 162 | static noinline void finish_rmw(struct btrfs_raid_bio *rbio); |
| 163 | static void rmw_work(struct btrfs_work *work); |
| 164 | static void read_rebuild_work(struct btrfs_work *work); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 165 | static int fail_bio_stripe(struct btrfs_raid_bio *rbio, struct bio *bio); |
| 166 | static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed); |
| 167 | static void __free_raid_bio(struct btrfs_raid_bio *rbio); |
| 168 | static void index_rbio_pages(struct btrfs_raid_bio *rbio); |
| 169 | static int alloc_rbio_pages(struct btrfs_raid_bio *rbio); |
| 170 | |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 171 | static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, |
| 172 | int need_check); |
David Sterba | a81b747 | 2018-06-29 10:57:03 +0200 | [diff] [blame] | 173 | static void scrub_parity_work(struct btrfs_work *work); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 174 | |
David Sterba | ac63885 | 2018-06-29 10:56:56 +0200 | [diff] [blame] | 175 | static void start_async_work(struct btrfs_raid_bio *rbio, btrfs_func_t work_func) |
| 176 | { |
| 177 | btrfs_init_work(&rbio->work, btrfs_rmw_helper, work_func, NULL, NULL); |
| 178 | btrfs_queue_work(rbio->fs_info->rmw_workers, &rbio->work); |
| 179 | } |
| 180 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 181 | /* |
| 182 | * the stripe hash table is used for locking, and to collect |
| 183 | * bios in hopes of making a full stripe |
| 184 | */ |
| 185 | int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info) |
| 186 | { |
| 187 | struct btrfs_stripe_hash_table *table; |
| 188 | struct btrfs_stripe_hash_table *x; |
| 189 | struct btrfs_stripe_hash *cur; |
| 190 | struct btrfs_stripe_hash *h; |
| 191 | int num_entries = 1 << BTRFS_STRIPE_HASH_TABLE_BITS; |
| 192 | int i; |
David Sterba | 83c8266 | 2013-03-01 15:03:00 +0000 | [diff] [blame] | 193 | int table_size; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 194 | |
| 195 | if (info->stripe_hash_table) |
| 196 | return 0; |
| 197 | |
David Sterba | 83c8266 | 2013-03-01 15:03:00 +0000 | [diff] [blame] | 198 | /* |
| 199 | * The table is large, starting with order 4 and can go as high as |
| 200 | * order 7 in case lock debugging is turned on. |
| 201 | * |
| 202 | * Try harder to allocate and fallback to vmalloc to lower the chance |
| 203 | * of a failing mount. |
| 204 | */ |
| 205 | table_size = sizeof(*table) + sizeof(*h) * num_entries; |
David Sterba | 818e010 | 2017-05-31 18:40:02 +0200 | [diff] [blame] | 206 | table = kvzalloc(table_size, GFP_KERNEL); |
| 207 | if (!table) |
| 208 | return -ENOMEM; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 209 | |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 210 | spin_lock_init(&table->cache_lock); |
| 211 | INIT_LIST_HEAD(&table->stripe_cache); |
| 212 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 213 | h = table->table; |
| 214 | |
| 215 | for (i = 0; i < num_entries; i++) { |
| 216 | cur = h + i; |
| 217 | INIT_LIST_HEAD(&cur->hash_list); |
| 218 | spin_lock_init(&cur->lock); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 219 | } |
| 220 | |
| 221 | x = cmpxchg(&info->stripe_hash_table, NULL, table); |
Wang Shilong | f749303 | 2014-11-22 21:13:10 +0800 | [diff] [blame] | 222 | if (x) |
| 223 | kvfree(x); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 224 | return 0; |
| 225 | } |
| 226 | |
| 227 | /* |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 228 | * caching an rbio means to copy anything from the |
| 229 | * bio_pages array into the stripe_pages array. We |
| 230 | * use the page uptodate bit in the stripe cache array |
| 231 | * to indicate if it has valid data |
| 232 | * |
| 233 | * once the caching is done, we set the cache ready |
| 234 | * bit. |
| 235 | */ |
| 236 | static void cache_rbio_pages(struct btrfs_raid_bio *rbio) |
| 237 | { |
| 238 | int i; |
| 239 | char *s; |
| 240 | char *d; |
| 241 | int ret; |
| 242 | |
| 243 | ret = alloc_rbio_pages(rbio); |
| 244 | if (ret) |
| 245 | return; |
| 246 | |
| 247 | for (i = 0; i < rbio->nr_pages; i++) { |
| 248 | if (!rbio->bio_pages[i]) |
| 249 | continue; |
| 250 | |
| 251 | s = kmap(rbio->bio_pages[i]); |
| 252 | d = kmap(rbio->stripe_pages[i]); |
| 253 | |
David Sterba | 69d2480 | 2018-06-29 10:56:44 +0200 | [diff] [blame] | 254 | copy_page(d, s); |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 255 | |
| 256 | kunmap(rbio->bio_pages[i]); |
| 257 | kunmap(rbio->stripe_pages[i]); |
| 258 | SetPageUptodate(rbio->stripe_pages[i]); |
| 259 | } |
| 260 | set_bit(RBIO_CACHE_READY_BIT, &rbio->flags); |
| 261 | } |
| 262 | |
| 263 | /* |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 264 | * we hash on the first logical address of the stripe |
| 265 | */ |
| 266 | static int rbio_bucket(struct btrfs_raid_bio *rbio) |
| 267 | { |
Zhao Lei | 8e5cfb5 | 2015-01-20 15:11:33 +0800 | [diff] [blame] | 268 | u64 num = rbio->bbio->raid_map[0]; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 269 | |
| 270 | /* |
| 271 | * we shift down quite a bit. We're using byte |
| 272 | * addressing, and most of the lower bits are zeros. |
| 273 | * This tends to upset hash_64, and it consistently |
| 274 | * returns just one or two different values. |
| 275 | * |
| 276 | * shifting off the lower bits fixes things. |
| 277 | */ |
| 278 | return hash_64(num >> 16, BTRFS_STRIPE_HASH_TABLE_BITS); |
| 279 | } |
| 280 | |
| 281 | /* |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 282 | * stealing an rbio means taking all the uptodate pages from the stripe |
| 283 | * array in the source rbio and putting them into the destination rbio |
| 284 | */ |
| 285 | static void steal_rbio(struct btrfs_raid_bio *src, struct btrfs_raid_bio *dest) |
| 286 | { |
| 287 | int i; |
| 288 | struct page *s; |
| 289 | struct page *d; |
| 290 | |
| 291 | if (!test_bit(RBIO_CACHE_READY_BIT, &src->flags)) |
| 292 | return; |
| 293 | |
| 294 | for (i = 0; i < dest->nr_pages; i++) { |
| 295 | s = src->stripe_pages[i]; |
| 296 | if (!s || !PageUptodate(s)) { |
| 297 | continue; |
| 298 | } |
| 299 | |
| 300 | d = dest->stripe_pages[i]; |
| 301 | if (d) |
| 302 | __free_page(d); |
| 303 | |
| 304 | dest->stripe_pages[i] = s; |
| 305 | src->stripe_pages[i] = NULL; |
| 306 | } |
| 307 | } |
| 308 | |
| 309 | /* |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 310 | * merging means we take the bio_list from the victim and |
| 311 | * splice it into the destination. The victim should |
| 312 | * be discarded afterwards. |
| 313 | * |
| 314 | * must be called with dest->rbio_list_lock held |
| 315 | */ |
| 316 | static void merge_rbio(struct btrfs_raid_bio *dest, |
| 317 | struct btrfs_raid_bio *victim) |
| 318 | { |
| 319 | bio_list_merge(&dest->bio_list, &victim->bio_list); |
| 320 | dest->bio_list_bytes += victim->bio_list_bytes; |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 321 | dest->generic_bio_cnt += victim->generic_bio_cnt; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 322 | bio_list_init(&victim->bio_list); |
| 323 | } |
| 324 | |
| 325 | /* |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 326 | * used to prune items that are in the cache. The caller |
| 327 | * must hold the hash table lock. |
| 328 | */ |
| 329 | static void __remove_rbio_from_cache(struct btrfs_raid_bio *rbio) |
| 330 | { |
| 331 | int bucket = rbio_bucket(rbio); |
| 332 | struct btrfs_stripe_hash_table *table; |
| 333 | struct btrfs_stripe_hash *h; |
| 334 | int freeit = 0; |
| 335 | |
| 336 | /* |
| 337 | * check the bit again under the hash table lock. |
| 338 | */ |
| 339 | if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) |
| 340 | return; |
| 341 | |
| 342 | table = rbio->fs_info->stripe_hash_table; |
| 343 | h = table->table + bucket; |
| 344 | |
| 345 | /* hold the lock for the bucket because we may be |
| 346 | * removing it from the hash table |
| 347 | */ |
| 348 | spin_lock(&h->lock); |
| 349 | |
| 350 | /* |
| 351 | * hold the lock for the bio list because we need |
| 352 | * to make sure the bio list is empty |
| 353 | */ |
| 354 | spin_lock(&rbio->bio_list_lock); |
| 355 | |
| 356 | if (test_and_clear_bit(RBIO_CACHE_BIT, &rbio->flags)) { |
| 357 | list_del_init(&rbio->stripe_cache); |
| 358 | table->cache_size -= 1; |
| 359 | freeit = 1; |
| 360 | |
| 361 | /* if the bio list isn't empty, this rbio is |
| 362 | * still involved in an IO. We take it out |
| 363 | * of the cache list, and drop the ref that |
| 364 | * was held for the list. |
| 365 | * |
| 366 | * If the bio_list was empty, we also remove |
| 367 | * the rbio from the hash_table, and drop |
| 368 | * the corresponding ref |
| 369 | */ |
| 370 | if (bio_list_empty(&rbio->bio_list)) { |
| 371 | if (!list_empty(&rbio->hash_list)) { |
| 372 | list_del_init(&rbio->hash_list); |
Elena Reshetova | dec9557 | 2017-03-03 10:55:26 +0200 | [diff] [blame] | 373 | refcount_dec(&rbio->refs); |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 374 | BUG_ON(!list_empty(&rbio->plug_list)); |
| 375 | } |
| 376 | } |
| 377 | } |
| 378 | |
| 379 | spin_unlock(&rbio->bio_list_lock); |
| 380 | spin_unlock(&h->lock); |
| 381 | |
| 382 | if (freeit) |
| 383 | __free_raid_bio(rbio); |
| 384 | } |
| 385 | |
| 386 | /* |
| 387 | * prune a given rbio from the cache |
| 388 | */ |
| 389 | static void remove_rbio_from_cache(struct btrfs_raid_bio *rbio) |
| 390 | { |
| 391 | struct btrfs_stripe_hash_table *table; |
| 392 | unsigned long flags; |
| 393 | |
| 394 | if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) |
| 395 | return; |
| 396 | |
| 397 | table = rbio->fs_info->stripe_hash_table; |
| 398 | |
| 399 | spin_lock_irqsave(&table->cache_lock, flags); |
| 400 | __remove_rbio_from_cache(rbio); |
| 401 | spin_unlock_irqrestore(&table->cache_lock, flags); |
| 402 | } |
| 403 | |
| 404 | /* |
| 405 | * remove everything in the cache |
| 406 | */ |
Eric Sandeen | 48a3b63 | 2013-04-25 20:41:01 +0000 | [diff] [blame] | 407 | static void btrfs_clear_rbio_cache(struct btrfs_fs_info *info) |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 408 | { |
| 409 | struct btrfs_stripe_hash_table *table; |
| 410 | unsigned long flags; |
| 411 | struct btrfs_raid_bio *rbio; |
| 412 | |
| 413 | table = info->stripe_hash_table; |
| 414 | |
| 415 | spin_lock_irqsave(&table->cache_lock, flags); |
| 416 | while (!list_empty(&table->stripe_cache)) { |
| 417 | rbio = list_entry(table->stripe_cache.next, |
| 418 | struct btrfs_raid_bio, |
| 419 | stripe_cache); |
| 420 | __remove_rbio_from_cache(rbio); |
| 421 | } |
| 422 | spin_unlock_irqrestore(&table->cache_lock, flags); |
| 423 | } |
| 424 | |
| 425 | /* |
| 426 | * remove all cached entries and free the hash table |
| 427 | * used by unmount |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 428 | */ |
| 429 | void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info) |
| 430 | { |
| 431 | if (!info->stripe_hash_table) |
| 432 | return; |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 433 | btrfs_clear_rbio_cache(info); |
Wang Shilong | f749303 | 2014-11-22 21:13:10 +0800 | [diff] [blame] | 434 | kvfree(info->stripe_hash_table); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 435 | info->stripe_hash_table = NULL; |
| 436 | } |
| 437 | |
| 438 | /* |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 439 | * insert an rbio into the stripe cache. It |
| 440 | * must have already been prepared by calling |
| 441 | * cache_rbio_pages |
| 442 | * |
| 443 | * If this rbio was already cached, it gets |
| 444 | * moved to the front of the lru. |
| 445 | * |
| 446 | * If the size of the rbio cache is too big, we |
| 447 | * prune an item. |
| 448 | */ |
| 449 | static void cache_rbio(struct btrfs_raid_bio *rbio) |
| 450 | { |
| 451 | struct btrfs_stripe_hash_table *table; |
| 452 | unsigned long flags; |
| 453 | |
| 454 | if (!test_bit(RBIO_CACHE_READY_BIT, &rbio->flags)) |
| 455 | return; |
| 456 | |
| 457 | table = rbio->fs_info->stripe_hash_table; |
| 458 | |
| 459 | spin_lock_irqsave(&table->cache_lock, flags); |
| 460 | spin_lock(&rbio->bio_list_lock); |
| 461 | |
| 462 | /* bump our ref if we were not in the list before */ |
| 463 | if (!test_and_set_bit(RBIO_CACHE_BIT, &rbio->flags)) |
Elena Reshetova | dec9557 | 2017-03-03 10:55:26 +0200 | [diff] [blame] | 464 | refcount_inc(&rbio->refs); |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 465 | |
| 466 | if (!list_empty(&rbio->stripe_cache)){ |
| 467 | list_move(&rbio->stripe_cache, &table->stripe_cache); |
| 468 | } else { |
| 469 | list_add(&rbio->stripe_cache, &table->stripe_cache); |
| 470 | table->cache_size += 1; |
| 471 | } |
| 472 | |
| 473 | spin_unlock(&rbio->bio_list_lock); |
| 474 | |
| 475 | if (table->cache_size > RBIO_CACHE_SIZE) { |
| 476 | struct btrfs_raid_bio *found; |
| 477 | |
| 478 | found = list_entry(table->stripe_cache.prev, |
| 479 | struct btrfs_raid_bio, |
| 480 | stripe_cache); |
| 481 | |
| 482 | if (found != rbio) |
| 483 | __remove_rbio_from_cache(found); |
| 484 | } |
| 485 | |
| 486 | spin_unlock_irqrestore(&table->cache_lock, flags); |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 487 | } |
| 488 | |
| 489 | /* |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 490 | * helper function to run the xor_blocks api. It is only |
| 491 | * able to do MAX_XOR_BLOCKS at a time, so we need to |
| 492 | * loop through. |
| 493 | */ |
| 494 | static void run_xor(void **pages, int src_cnt, ssize_t len) |
| 495 | { |
| 496 | int src_off = 0; |
| 497 | int xor_src_cnt = 0; |
| 498 | void *dest = pages[src_cnt]; |
| 499 | |
| 500 | while(src_cnt > 0) { |
| 501 | xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS); |
| 502 | xor_blocks(xor_src_cnt, len, dest, pages + src_off); |
| 503 | |
| 504 | src_cnt -= xor_src_cnt; |
| 505 | src_off += xor_src_cnt; |
| 506 | } |
| 507 | } |
| 508 | |
| 509 | /* |
David Sterba | 176571a | 2018-06-29 10:57:05 +0200 | [diff] [blame] | 510 | * Returns true if the bio list inside this rbio covers an entire stripe (no |
| 511 | * rmw required). |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 512 | */ |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 513 | static int rbio_is_full(struct btrfs_raid_bio *rbio) |
| 514 | { |
| 515 | unsigned long flags; |
David Sterba | 176571a | 2018-06-29 10:57:05 +0200 | [diff] [blame] | 516 | unsigned long size = rbio->bio_list_bytes; |
| 517 | int ret = 1; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 518 | |
| 519 | spin_lock_irqsave(&rbio->bio_list_lock, flags); |
David Sterba | 176571a | 2018-06-29 10:57:05 +0200 | [diff] [blame] | 520 | if (size != rbio->nr_data * rbio->stripe_len) |
| 521 | ret = 0; |
| 522 | BUG_ON(size > rbio->nr_data * rbio->stripe_len); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 523 | spin_unlock_irqrestore(&rbio->bio_list_lock, flags); |
David Sterba | 176571a | 2018-06-29 10:57:05 +0200 | [diff] [blame] | 524 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 525 | return ret; |
| 526 | } |
| 527 | |
| 528 | /* |
| 529 | * returns 1 if it is safe to merge two rbios together. |
| 530 | * The merging is safe if the two rbios correspond to |
| 531 | * the same stripe and if they are both going in the same |
| 532 | * direction (read vs write), and if neither one is |
| 533 | * locked for final IO |
| 534 | * |
| 535 | * The caller is responsible for locking such that |
| 536 | * rmw_locked is safe to test |
| 537 | */ |
| 538 | static int rbio_can_merge(struct btrfs_raid_bio *last, |
| 539 | struct btrfs_raid_bio *cur) |
| 540 | { |
| 541 | if (test_bit(RBIO_RMW_LOCKED_BIT, &last->flags) || |
| 542 | test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) |
| 543 | return 0; |
| 544 | |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 545 | /* |
| 546 | * we can't merge with cached rbios, since the |
| 547 | * idea is that when we merge the destination |
| 548 | * rbio is going to run our IO for us. We can |
Nicholas D Steeves | 0132761 | 2016-05-19 21:18:45 -0400 | [diff] [blame] | 549 | * steal from cached rbios though, other functions |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 550 | * handle that. |
| 551 | */ |
| 552 | if (test_bit(RBIO_CACHE_BIT, &last->flags) || |
| 553 | test_bit(RBIO_CACHE_BIT, &cur->flags)) |
| 554 | return 0; |
| 555 | |
Zhao Lei | 8e5cfb5 | 2015-01-20 15:11:33 +0800 | [diff] [blame] | 556 | if (last->bbio->raid_map[0] != |
| 557 | cur->bbio->raid_map[0]) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 558 | return 0; |
| 559 | |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 560 | /* we can't merge with different operations */ |
| 561 | if (last->operation != cur->operation) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 562 | return 0; |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 563 | /* |
| 564 | * We've need read the full stripe from the drive. |
| 565 | * check and repair the parity and write the new results. |
| 566 | * |
| 567 | * We're not allowed to add any new bios to the |
| 568 | * bio list here, anyone else that wants to |
| 569 | * change this stripe needs to do their own rmw. |
| 570 | */ |
Liu Bo | db34be1 | 2017-12-04 15:40:35 -0700 | [diff] [blame] | 571 | if (last->operation == BTRFS_RBIO_PARITY_SCRUB) |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 572 | return 0; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 573 | |
Liu Bo | db34be1 | 2017-12-04 15:40:35 -0700 | [diff] [blame] | 574 | if (last->operation == BTRFS_RBIO_REBUILD_MISSING) |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 575 | return 0; |
| 576 | |
Liu Bo | cc54ff6 | 2017-12-11 14:56:31 -0700 | [diff] [blame] | 577 | if (last->operation == BTRFS_RBIO_READ_REBUILD) { |
| 578 | int fa = last->faila; |
| 579 | int fb = last->failb; |
| 580 | int cur_fa = cur->faila; |
| 581 | int cur_fb = cur->failb; |
| 582 | |
| 583 | if (last->faila >= last->failb) { |
| 584 | fa = last->failb; |
| 585 | fb = last->faila; |
| 586 | } |
| 587 | |
| 588 | if (cur->faila >= cur->failb) { |
| 589 | cur_fa = cur->failb; |
| 590 | cur_fb = cur->faila; |
| 591 | } |
| 592 | |
| 593 | if (fa != cur_fa || fb != cur_fb) |
| 594 | return 0; |
| 595 | } |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 596 | return 1; |
| 597 | } |
| 598 | |
Zhao Lei | b7178a5 | 2015-03-03 20:38:46 +0800 | [diff] [blame] | 599 | static int rbio_stripe_page_index(struct btrfs_raid_bio *rbio, int stripe, |
| 600 | int index) |
| 601 | { |
| 602 | return stripe * rbio->stripe_npages + index; |
| 603 | } |
| 604 | |
| 605 | /* |
| 606 | * these are just the pages from the rbio array, not from anything |
| 607 | * the FS sent down to us |
| 608 | */ |
| 609 | static struct page *rbio_stripe_page(struct btrfs_raid_bio *rbio, int stripe, |
| 610 | int index) |
| 611 | { |
| 612 | return rbio->stripe_pages[rbio_stripe_page_index(rbio, stripe, index)]; |
| 613 | } |
| 614 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 615 | /* |
| 616 | * helper to index into the pstripe |
| 617 | */ |
| 618 | static struct page *rbio_pstripe_page(struct btrfs_raid_bio *rbio, int index) |
| 619 | { |
Zhao Lei | b7178a5 | 2015-03-03 20:38:46 +0800 | [diff] [blame] | 620 | return rbio_stripe_page(rbio, rbio->nr_data, index); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 621 | } |
| 622 | |
| 623 | /* |
| 624 | * helper to index into the qstripe, returns null |
| 625 | * if there is no qstripe |
| 626 | */ |
| 627 | static struct page *rbio_qstripe_page(struct btrfs_raid_bio *rbio, int index) |
| 628 | { |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 629 | if (rbio->nr_data + 1 == rbio->real_stripes) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 630 | return NULL; |
Zhao Lei | b7178a5 | 2015-03-03 20:38:46 +0800 | [diff] [blame] | 631 | return rbio_stripe_page(rbio, rbio->nr_data + 1, index); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 632 | } |
| 633 | |
| 634 | /* |
| 635 | * The first stripe in the table for a logical address |
| 636 | * has the lock. rbios are added in one of three ways: |
| 637 | * |
| 638 | * 1) Nobody has the stripe locked yet. The rbio is given |
| 639 | * the lock and 0 is returned. The caller must start the IO |
| 640 | * themselves. |
| 641 | * |
| 642 | * 2) Someone has the stripe locked, but we're able to merge |
| 643 | * with the lock owner. The rbio is freed and the IO will |
| 644 | * start automatically along with the existing rbio. 1 is returned. |
| 645 | * |
| 646 | * 3) Someone has the stripe locked, but we're not able to merge. |
| 647 | * The rbio is added to the lock owner's plug list, or merged into |
| 648 | * an rbio already on the plug list. When the lock owner unlocks, |
| 649 | * the next rbio on the list is run and the IO is started automatically. |
| 650 | * 1 is returned |
| 651 | * |
| 652 | * If we return 0, the caller still owns the rbio and must continue with |
| 653 | * IO submission. If we return 1, the caller must assume the rbio has |
| 654 | * already been freed. |
| 655 | */ |
| 656 | static noinline int lock_stripe_add(struct btrfs_raid_bio *rbio) |
| 657 | { |
| 658 | int bucket = rbio_bucket(rbio); |
| 659 | struct btrfs_stripe_hash *h = rbio->fs_info->stripe_hash_table->table + bucket; |
| 660 | struct btrfs_raid_bio *cur; |
| 661 | struct btrfs_raid_bio *pending; |
| 662 | unsigned long flags; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 663 | struct btrfs_raid_bio *freeit = NULL; |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 664 | struct btrfs_raid_bio *cache_drop = NULL; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 665 | int ret = 0; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 666 | |
| 667 | spin_lock_irqsave(&h->lock, flags); |
| 668 | list_for_each_entry(cur, &h->hash_list, hash_list) { |
Zhao Lei | 8e5cfb5 | 2015-01-20 15:11:33 +0800 | [diff] [blame] | 669 | if (cur->bbio->raid_map[0] == rbio->bbio->raid_map[0]) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 670 | spin_lock(&cur->bio_list_lock); |
| 671 | |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 672 | /* can we steal this cached rbio's pages? */ |
| 673 | if (bio_list_empty(&cur->bio_list) && |
| 674 | list_empty(&cur->plug_list) && |
| 675 | test_bit(RBIO_CACHE_BIT, &cur->flags) && |
| 676 | !test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) { |
| 677 | list_del_init(&cur->hash_list); |
Elena Reshetova | dec9557 | 2017-03-03 10:55:26 +0200 | [diff] [blame] | 678 | refcount_dec(&cur->refs); |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 679 | |
| 680 | steal_rbio(cur, rbio); |
| 681 | cache_drop = cur; |
| 682 | spin_unlock(&cur->bio_list_lock); |
| 683 | |
| 684 | goto lockit; |
| 685 | } |
| 686 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 687 | /* can we merge into the lock owner? */ |
| 688 | if (rbio_can_merge(cur, rbio)) { |
| 689 | merge_rbio(cur, rbio); |
| 690 | spin_unlock(&cur->bio_list_lock); |
| 691 | freeit = rbio; |
| 692 | ret = 1; |
| 693 | goto out; |
| 694 | } |
| 695 | |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 696 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 697 | /* |
| 698 | * we couldn't merge with the running |
| 699 | * rbio, see if we can merge with the |
| 700 | * pending ones. We don't have to |
| 701 | * check for rmw_locked because there |
| 702 | * is no way they are inside finish_rmw |
| 703 | * right now |
| 704 | */ |
| 705 | list_for_each_entry(pending, &cur->plug_list, |
| 706 | plug_list) { |
| 707 | if (rbio_can_merge(pending, rbio)) { |
| 708 | merge_rbio(pending, rbio); |
| 709 | spin_unlock(&cur->bio_list_lock); |
| 710 | freeit = rbio; |
| 711 | ret = 1; |
| 712 | goto out; |
| 713 | } |
| 714 | } |
| 715 | |
| 716 | /* no merging, put us on the tail of the plug list, |
| 717 | * our rbio will be started with the currently |
| 718 | * running rbio unlocks |
| 719 | */ |
| 720 | list_add_tail(&rbio->plug_list, &cur->plug_list); |
| 721 | spin_unlock(&cur->bio_list_lock); |
| 722 | ret = 1; |
| 723 | goto out; |
| 724 | } |
| 725 | } |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 726 | lockit: |
Elena Reshetova | dec9557 | 2017-03-03 10:55:26 +0200 | [diff] [blame] | 727 | refcount_inc(&rbio->refs); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 728 | list_add(&rbio->hash_list, &h->hash_list); |
| 729 | out: |
| 730 | spin_unlock_irqrestore(&h->lock, flags); |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 731 | if (cache_drop) |
| 732 | remove_rbio_from_cache(cache_drop); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 733 | if (freeit) |
| 734 | __free_raid_bio(freeit); |
| 735 | return ret; |
| 736 | } |
| 737 | |
| 738 | /* |
| 739 | * called as rmw or parity rebuild is completed. If the plug list has more |
| 740 | * rbios waiting for this stripe, the next one on the list will be started |
| 741 | */ |
| 742 | static noinline void unlock_stripe(struct btrfs_raid_bio *rbio) |
| 743 | { |
| 744 | int bucket; |
| 745 | struct btrfs_stripe_hash *h; |
| 746 | unsigned long flags; |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 747 | int keep_cache = 0; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 748 | |
| 749 | bucket = rbio_bucket(rbio); |
| 750 | h = rbio->fs_info->stripe_hash_table->table + bucket; |
| 751 | |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 752 | if (list_empty(&rbio->plug_list)) |
| 753 | cache_rbio(rbio); |
| 754 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 755 | spin_lock_irqsave(&h->lock, flags); |
| 756 | spin_lock(&rbio->bio_list_lock); |
| 757 | |
| 758 | if (!list_empty(&rbio->hash_list)) { |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 759 | /* |
| 760 | * if we're still cached and there is no other IO |
| 761 | * to perform, just leave this rbio here for others |
| 762 | * to steal from later |
| 763 | */ |
| 764 | if (list_empty(&rbio->plug_list) && |
| 765 | test_bit(RBIO_CACHE_BIT, &rbio->flags)) { |
| 766 | keep_cache = 1; |
| 767 | clear_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); |
| 768 | BUG_ON(!bio_list_empty(&rbio->bio_list)); |
| 769 | goto done; |
| 770 | } |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 771 | |
| 772 | list_del_init(&rbio->hash_list); |
Elena Reshetova | dec9557 | 2017-03-03 10:55:26 +0200 | [diff] [blame] | 773 | refcount_dec(&rbio->refs); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 774 | |
| 775 | /* |
| 776 | * we use the plug list to hold all the rbios |
| 777 | * waiting for the chance to lock this stripe. |
| 778 | * hand the lock over to one of them. |
| 779 | */ |
| 780 | if (!list_empty(&rbio->plug_list)) { |
| 781 | struct btrfs_raid_bio *next; |
| 782 | struct list_head *head = rbio->plug_list.next; |
| 783 | |
| 784 | next = list_entry(head, struct btrfs_raid_bio, |
| 785 | plug_list); |
| 786 | |
| 787 | list_del_init(&rbio->plug_list); |
| 788 | |
| 789 | list_add(&next->hash_list, &h->hash_list); |
Elena Reshetova | dec9557 | 2017-03-03 10:55:26 +0200 | [diff] [blame] | 790 | refcount_inc(&next->refs); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 791 | spin_unlock(&rbio->bio_list_lock); |
| 792 | spin_unlock_irqrestore(&h->lock, flags); |
| 793 | |
Miao Xie | 1b94b55 | 2014-11-06 16:14:21 +0800 | [diff] [blame] | 794 | if (next->operation == BTRFS_RBIO_READ_REBUILD) |
David Sterba | e66d8d5 | 2018-06-29 10:57:00 +0200 | [diff] [blame] | 795 | start_async_work(next, read_rebuild_work); |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 796 | else if (next->operation == BTRFS_RBIO_REBUILD_MISSING) { |
| 797 | steal_rbio(rbio, next); |
David Sterba | e66d8d5 | 2018-06-29 10:57:00 +0200 | [diff] [blame] | 798 | start_async_work(next, read_rebuild_work); |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 799 | } else if (next->operation == BTRFS_RBIO_WRITE) { |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 800 | steal_rbio(rbio, next); |
David Sterba | cf6a4a7 | 2018-06-29 10:56:58 +0200 | [diff] [blame] | 801 | start_async_work(next, rmw_work); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 802 | } else if (next->operation == BTRFS_RBIO_PARITY_SCRUB) { |
| 803 | steal_rbio(rbio, next); |
David Sterba | a81b747 | 2018-06-29 10:57:03 +0200 | [diff] [blame] | 804 | start_async_work(next, scrub_parity_work); |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 805 | } |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 806 | |
| 807 | goto done_nolock; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 808 | } |
| 809 | } |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 810 | done: |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 811 | spin_unlock(&rbio->bio_list_lock); |
| 812 | spin_unlock_irqrestore(&h->lock, flags); |
| 813 | |
| 814 | done_nolock: |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 815 | if (!keep_cache) |
| 816 | remove_rbio_from_cache(rbio); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 817 | } |
| 818 | |
| 819 | static void __free_raid_bio(struct btrfs_raid_bio *rbio) |
| 820 | { |
| 821 | int i; |
| 822 | |
Elena Reshetova | dec9557 | 2017-03-03 10:55:26 +0200 | [diff] [blame] | 823 | if (!refcount_dec_and_test(&rbio->refs)) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 824 | return; |
| 825 | |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 826 | WARN_ON(!list_empty(&rbio->stripe_cache)); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 827 | WARN_ON(!list_empty(&rbio->hash_list)); |
| 828 | WARN_ON(!bio_list_empty(&rbio->bio_list)); |
| 829 | |
| 830 | for (i = 0; i < rbio->nr_pages; i++) { |
| 831 | if (rbio->stripe_pages[i]) { |
| 832 | __free_page(rbio->stripe_pages[i]); |
| 833 | rbio->stripe_pages[i] = NULL; |
| 834 | } |
| 835 | } |
Miao Xie | af8e2d1 | 2014-10-23 14:42:50 +0800 | [diff] [blame] | 836 | |
Zhao Lei | 6e9606d | 2015-01-20 15:11:34 +0800 | [diff] [blame] | 837 | btrfs_put_bbio(rbio->bbio); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 838 | kfree(rbio); |
| 839 | } |
| 840 | |
Liu Bo | 7583d8d | 2018-01-09 18:36:25 -0700 | [diff] [blame] | 841 | static void rbio_endio_bio_list(struct bio *cur, blk_status_t err) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 842 | { |
Liu Bo | 7583d8d | 2018-01-09 18:36:25 -0700 | [diff] [blame] | 843 | struct bio *next; |
| 844 | |
| 845 | while (cur) { |
| 846 | next = cur->bi_next; |
| 847 | cur->bi_next = NULL; |
| 848 | cur->bi_status = err; |
| 849 | bio_endio(cur); |
| 850 | cur = next; |
| 851 | } |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 852 | } |
| 853 | |
| 854 | /* |
| 855 | * this frees the rbio and runs through all the bios in the |
| 856 | * bio_list and calls end_io on them |
| 857 | */ |
Christoph Hellwig | 4e4cbee | 2017-06-03 09:38:06 +0200 | [diff] [blame] | 858 | static void rbio_orig_end_io(struct btrfs_raid_bio *rbio, blk_status_t err) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 859 | { |
| 860 | struct bio *cur = bio_list_get(&rbio->bio_list); |
Liu Bo | 7583d8d | 2018-01-09 18:36:25 -0700 | [diff] [blame] | 861 | struct bio *extra; |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 862 | |
| 863 | if (rbio->generic_bio_cnt) |
| 864 | btrfs_bio_counter_sub(rbio->fs_info, rbio->generic_bio_cnt); |
| 865 | |
Liu Bo | 7583d8d | 2018-01-09 18:36:25 -0700 | [diff] [blame] | 866 | /* |
| 867 | * At this moment, rbio->bio_list is empty, however since rbio does not |
| 868 | * always have RBIO_RMW_LOCKED_BIT set and rbio is still linked on the |
| 869 | * hash list, rbio may be merged with others so that rbio->bio_list |
| 870 | * becomes non-empty. |
| 871 | * Once unlock_stripe() is done, rbio->bio_list will not be updated any |
| 872 | * more and we can call bio_endio() on all queued bios. |
| 873 | */ |
| 874 | unlock_stripe(rbio); |
| 875 | extra = bio_list_get(&rbio->bio_list); |
| 876 | __free_raid_bio(rbio); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 877 | |
Liu Bo | 7583d8d | 2018-01-09 18:36:25 -0700 | [diff] [blame] | 878 | rbio_endio_bio_list(cur, err); |
| 879 | if (extra) |
| 880 | rbio_endio_bio_list(extra, err); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 881 | } |
| 882 | |
| 883 | /* |
| 884 | * end io function used by finish_rmw. When we finally |
| 885 | * get here, we've written a full stripe |
| 886 | */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 887 | static void raid_write_end_io(struct bio *bio) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 888 | { |
| 889 | struct btrfs_raid_bio *rbio = bio->bi_private; |
Christoph Hellwig | 4e4cbee | 2017-06-03 09:38:06 +0200 | [diff] [blame] | 890 | blk_status_t err = bio->bi_status; |
Zhao Lei | a6111d1 | 2016-01-12 17:52:13 +0800 | [diff] [blame] | 891 | int max_errors; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 892 | |
| 893 | if (err) |
| 894 | fail_bio_stripe(rbio, bio); |
| 895 | |
| 896 | bio_put(bio); |
| 897 | |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 898 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 899 | return; |
| 900 | |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 901 | err = BLK_STS_OK; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 902 | |
| 903 | /* OK, we have read all the stripes we need to. */ |
Zhao Lei | a6111d1 | 2016-01-12 17:52:13 +0800 | [diff] [blame] | 904 | max_errors = (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) ? |
| 905 | 0 : rbio->bbio->max_errors; |
| 906 | if (atomic_read(&rbio->error) > max_errors) |
Christoph Hellwig | 4e4cbee | 2017-06-03 09:38:06 +0200 | [diff] [blame] | 907 | err = BLK_STS_IOERR; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 908 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 909 | rbio_orig_end_io(rbio, err); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 910 | } |
| 911 | |
| 912 | /* |
| 913 | * the read/modify/write code wants to use the original bio for |
| 914 | * any pages it included, and then use the rbio for everything |
| 915 | * else. This function decides if a given index (stripe number) |
| 916 | * and page number in that stripe fall inside the original bio |
| 917 | * or the rbio. |
| 918 | * |
| 919 | * if you set bio_list_only, you'll get a NULL back for any ranges |
| 920 | * that are outside the bio_list |
| 921 | * |
| 922 | * This doesn't take any refs on anything, you get a bare page pointer |
| 923 | * and the caller must bump refs as required. |
| 924 | * |
| 925 | * You must call index_rbio_pages once before you can trust |
| 926 | * the answers from this function. |
| 927 | */ |
| 928 | static struct page *page_in_rbio(struct btrfs_raid_bio *rbio, |
| 929 | int index, int pagenr, int bio_list_only) |
| 930 | { |
| 931 | int chunk_page; |
| 932 | struct page *p = NULL; |
| 933 | |
| 934 | chunk_page = index * (rbio->stripe_len >> PAGE_SHIFT) + pagenr; |
| 935 | |
| 936 | spin_lock_irq(&rbio->bio_list_lock); |
| 937 | p = rbio->bio_pages[chunk_page]; |
| 938 | spin_unlock_irq(&rbio->bio_list_lock); |
| 939 | |
| 940 | if (p || bio_list_only) |
| 941 | return p; |
| 942 | |
| 943 | return rbio->stripe_pages[chunk_page]; |
| 944 | } |
| 945 | |
| 946 | /* |
| 947 | * number of pages we need for the entire stripe across all the |
| 948 | * drives |
| 949 | */ |
| 950 | static unsigned long rbio_nr_pages(unsigned long stripe_len, int nr_stripes) |
| 951 | { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 952 | return DIV_ROUND_UP(stripe_len, PAGE_SIZE) * nr_stripes; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 953 | } |
| 954 | |
| 955 | /* |
| 956 | * allocation and initial setup for the btrfs_raid_bio. Not |
| 957 | * this does not allocate any pages for rbio->pages. |
| 958 | */ |
Jeff Mahoney | 2ff7e61 | 2016-06-22 18:54:24 -0400 | [diff] [blame] | 959 | static struct btrfs_raid_bio *alloc_rbio(struct btrfs_fs_info *fs_info, |
| 960 | struct btrfs_bio *bbio, |
| 961 | u64 stripe_len) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 962 | { |
| 963 | struct btrfs_raid_bio *rbio; |
| 964 | int nr_data = 0; |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 965 | int real_stripes = bbio->num_stripes - bbio->num_tgtdevs; |
| 966 | int num_pages = rbio_nr_pages(stripe_len, real_stripes); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 967 | int stripe_npages = DIV_ROUND_UP(stripe_len, PAGE_SIZE); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 968 | void *p; |
| 969 | |
Kees Cook | 1389053 | 2018-05-29 16:44:59 -0700 | [diff] [blame] | 970 | rbio = kzalloc(sizeof(*rbio) + |
| 971 | sizeof(*rbio->stripe_pages) * num_pages + |
| 972 | sizeof(*rbio->bio_pages) * num_pages + |
| 973 | sizeof(*rbio->finish_pointers) * real_stripes + |
| 974 | sizeof(*rbio->dbitmap) * BITS_TO_LONGS(stripe_npages) + |
| 975 | sizeof(*rbio->finish_pbitmap) * |
| 976 | BITS_TO_LONGS(stripe_npages), |
| 977 | GFP_NOFS); |
Miao Xie | af8e2d1 | 2014-10-23 14:42:50 +0800 | [diff] [blame] | 978 | if (!rbio) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 979 | return ERR_PTR(-ENOMEM); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 980 | |
| 981 | bio_list_init(&rbio->bio_list); |
| 982 | INIT_LIST_HEAD(&rbio->plug_list); |
| 983 | spin_lock_init(&rbio->bio_list_lock); |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 984 | INIT_LIST_HEAD(&rbio->stripe_cache); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 985 | INIT_LIST_HEAD(&rbio->hash_list); |
| 986 | rbio->bbio = bbio; |
Jeff Mahoney | 2ff7e61 | 2016-06-22 18:54:24 -0400 | [diff] [blame] | 987 | rbio->fs_info = fs_info; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 988 | rbio->stripe_len = stripe_len; |
| 989 | rbio->nr_pages = num_pages; |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 990 | rbio->real_stripes = real_stripes; |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 991 | rbio->stripe_npages = stripe_npages; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 992 | rbio->faila = -1; |
| 993 | rbio->failb = -1; |
Elena Reshetova | dec9557 | 2017-03-03 10:55:26 +0200 | [diff] [blame] | 994 | refcount_set(&rbio->refs, 1); |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 995 | atomic_set(&rbio->error, 0); |
| 996 | atomic_set(&rbio->stripes_pending, 0); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 997 | |
| 998 | /* |
Kees Cook | 1389053 | 2018-05-29 16:44:59 -0700 | [diff] [blame] | 999 | * the stripe_pages, bio_pages, etc arrays point to the extra |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1000 | * memory we allocated past the end of the rbio |
| 1001 | */ |
| 1002 | p = rbio + 1; |
Kees Cook | 1389053 | 2018-05-29 16:44:59 -0700 | [diff] [blame] | 1003 | #define CONSUME_ALLOC(ptr, count) do { \ |
| 1004 | ptr = p; \ |
| 1005 | p = (unsigned char *)p + sizeof(*(ptr)) * (count); \ |
| 1006 | } while (0) |
| 1007 | CONSUME_ALLOC(rbio->stripe_pages, num_pages); |
| 1008 | CONSUME_ALLOC(rbio->bio_pages, num_pages); |
| 1009 | CONSUME_ALLOC(rbio->finish_pointers, real_stripes); |
| 1010 | CONSUME_ALLOC(rbio->dbitmap, BITS_TO_LONGS(stripe_npages)); |
| 1011 | CONSUME_ALLOC(rbio->finish_pbitmap, BITS_TO_LONGS(stripe_npages)); |
| 1012 | #undef CONSUME_ALLOC |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1013 | |
Zhao Lei | 10f1190 | 2015-01-20 15:11:43 +0800 | [diff] [blame] | 1014 | if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5) |
| 1015 | nr_data = real_stripes - 1; |
| 1016 | else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6) |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1017 | nr_data = real_stripes - 2; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1018 | else |
Zhao Lei | 10f1190 | 2015-01-20 15:11:43 +0800 | [diff] [blame] | 1019 | BUG(); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1020 | |
| 1021 | rbio->nr_data = nr_data; |
| 1022 | return rbio; |
| 1023 | } |
| 1024 | |
| 1025 | /* allocate pages for all the stripes in the bio, including parity */ |
| 1026 | static int alloc_rbio_pages(struct btrfs_raid_bio *rbio) |
| 1027 | { |
| 1028 | int i; |
| 1029 | struct page *page; |
| 1030 | |
| 1031 | for (i = 0; i < rbio->nr_pages; i++) { |
| 1032 | if (rbio->stripe_pages[i]) |
| 1033 | continue; |
| 1034 | page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); |
| 1035 | if (!page) |
| 1036 | return -ENOMEM; |
| 1037 | rbio->stripe_pages[i] = page; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1038 | } |
| 1039 | return 0; |
| 1040 | } |
| 1041 | |
Zhao Lei | b7178a5 | 2015-03-03 20:38:46 +0800 | [diff] [blame] | 1042 | /* only allocate pages for p/q stripes */ |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1043 | static int alloc_rbio_parity_pages(struct btrfs_raid_bio *rbio) |
| 1044 | { |
| 1045 | int i; |
| 1046 | struct page *page; |
| 1047 | |
Zhao Lei | b7178a5 | 2015-03-03 20:38:46 +0800 | [diff] [blame] | 1048 | i = rbio_stripe_page_index(rbio, rbio->nr_data, 0); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1049 | |
| 1050 | for (; i < rbio->nr_pages; i++) { |
| 1051 | if (rbio->stripe_pages[i]) |
| 1052 | continue; |
| 1053 | page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); |
| 1054 | if (!page) |
| 1055 | return -ENOMEM; |
| 1056 | rbio->stripe_pages[i] = page; |
| 1057 | } |
| 1058 | return 0; |
| 1059 | } |
| 1060 | |
| 1061 | /* |
| 1062 | * add a single page from a specific stripe into our list of bios for IO |
| 1063 | * this will try to merge into existing bios if possible, and returns |
| 1064 | * zero if all went well. |
| 1065 | */ |
Eric Sandeen | 48a3b63 | 2013-04-25 20:41:01 +0000 | [diff] [blame] | 1066 | static int rbio_add_io_page(struct btrfs_raid_bio *rbio, |
| 1067 | struct bio_list *bio_list, |
| 1068 | struct page *page, |
| 1069 | int stripe_nr, |
| 1070 | unsigned long page_index, |
| 1071 | unsigned long bio_max_len) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1072 | { |
| 1073 | struct bio *last = bio_list->tail; |
| 1074 | u64 last_end = 0; |
| 1075 | int ret; |
| 1076 | struct bio *bio; |
| 1077 | struct btrfs_bio_stripe *stripe; |
| 1078 | u64 disk_start; |
| 1079 | |
| 1080 | stripe = &rbio->bbio->stripes[stripe_nr]; |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1081 | disk_start = stripe->physical + (page_index << PAGE_SHIFT); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1082 | |
| 1083 | /* if the device is missing, just fail this stripe */ |
| 1084 | if (!stripe->dev->bdev) |
| 1085 | return fail_rbio_index(rbio, stripe_nr); |
| 1086 | |
| 1087 | /* see if we can add this page onto our existing bio */ |
| 1088 | if (last) { |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1089 | last_end = (u64)last->bi_iter.bi_sector << 9; |
| 1090 | last_end += last->bi_iter.bi_size; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1091 | |
| 1092 | /* |
| 1093 | * we can't merge these if they are from different |
| 1094 | * devices or if they are not contiguous |
| 1095 | */ |
| 1096 | if (last_end == disk_start && stripe->dev->bdev && |
Christoph Hellwig | 4e4cbee | 2017-06-03 09:38:06 +0200 | [diff] [blame] | 1097 | !last->bi_status && |
Christoph Hellwig | 74d4699 | 2017-08-23 19:10:32 +0200 | [diff] [blame] | 1098 | last->bi_disk == stripe->dev->bdev->bd_disk && |
| 1099 | last->bi_partno == stripe->dev->bdev->bd_partno) { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1100 | ret = bio_add_page(last, page, PAGE_SIZE, 0); |
| 1101 | if (ret == PAGE_SIZE) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1102 | return 0; |
| 1103 | } |
| 1104 | } |
| 1105 | |
| 1106 | /* put a new bio on the list */ |
David Sterba | c5e4c3d | 2017-06-12 17:29:41 +0200 | [diff] [blame] | 1107 | bio = btrfs_io_bio_alloc(bio_max_len >> PAGE_SHIFT ?: 1); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1108 | bio->bi_iter.bi_size = 0; |
Christoph Hellwig | 74d4699 | 2017-08-23 19:10:32 +0200 | [diff] [blame] | 1109 | bio_set_dev(bio, stripe->dev->bdev); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1110 | bio->bi_iter.bi_sector = disk_start >> 9; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1111 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1112 | bio_add_page(bio, page, PAGE_SIZE, 0); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1113 | bio_list_add(bio_list, bio); |
| 1114 | return 0; |
| 1115 | } |
| 1116 | |
| 1117 | /* |
| 1118 | * while we're doing the read/modify/write cycle, we could |
| 1119 | * have errors in reading pages off the disk. This checks |
| 1120 | * for errors and if we're not able to read the page it'll |
| 1121 | * trigger parity reconstruction. The rmw will be finished |
| 1122 | * after we've reconstructed the failed stripes |
| 1123 | */ |
| 1124 | static void validate_rbio_for_rmw(struct btrfs_raid_bio *rbio) |
| 1125 | { |
| 1126 | if (rbio->faila >= 0 || rbio->failb >= 0) { |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1127 | BUG_ON(rbio->faila == rbio->real_stripes - 1); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1128 | __raid56_parity_recover(rbio); |
| 1129 | } else { |
| 1130 | finish_rmw(rbio); |
| 1131 | } |
| 1132 | } |
| 1133 | |
| 1134 | /* |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1135 | * helper function to walk our bio list and populate the bio_pages array with |
| 1136 | * the result. This seems expensive, but it is faster than constantly |
| 1137 | * searching through the bio list as we setup the IO in finish_rmw or stripe |
| 1138 | * reconstruction. |
| 1139 | * |
| 1140 | * This must be called before you trust the answers from page_in_rbio |
| 1141 | */ |
| 1142 | static void index_rbio_pages(struct btrfs_raid_bio *rbio) |
| 1143 | { |
| 1144 | struct bio *bio; |
| 1145 | u64 start; |
| 1146 | unsigned long stripe_offset; |
| 1147 | unsigned long page_index; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1148 | |
| 1149 | spin_lock_irq(&rbio->bio_list_lock); |
| 1150 | bio_list_for_each(bio, &rbio->bio_list) { |
Filipe Manana | 6592e58 | 2017-07-12 23:36:02 +0100 | [diff] [blame] | 1151 | struct bio_vec bvec; |
| 1152 | struct bvec_iter iter; |
| 1153 | int i = 0; |
| 1154 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1155 | start = (u64)bio->bi_iter.bi_sector << 9; |
Zhao Lei | 8e5cfb5 | 2015-01-20 15:11:33 +0800 | [diff] [blame] | 1156 | stripe_offset = start - rbio->bbio->raid_map[0]; |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1157 | page_index = stripe_offset >> PAGE_SHIFT; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1158 | |
Filipe Manana | 6592e58 | 2017-07-12 23:36:02 +0100 | [diff] [blame] | 1159 | if (bio_flagged(bio, BIO_CLONED)) |
| 1160 | bio->bi_iter = btrfs_io_bio(bio)->iter; |
| 1161 | |
| 1162 | bio_for_each_segment(bvec, bio, iter) { |
| 1163 | rbio->bio_pages[page_index + i] = bvec.bv_page; |
| 1164 | i++; |
| 1165 | } |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1166 | } |
| 1167 | spin_unlock_irq(&rbio->bio_list_lock); |
| 1168 | } |
| 1169 | |
| 1170 | /* |
| 1171 | * this is called from one of two situations. We either |
| 1172 | * have a full stripe from the higher layers, or we've read all |
| 1173 | * the missing bits off disk. |
| 1174 | * |
| 1175 | * This will calculate the parity and then send down any |
| 1176 | * changed blocks. |
| 1177 | */ |
| 1178 | static noinline void finish_rmw(struct btrfs_raid_bio *rbio) |
| 1179 | { |
| 1180 | struct btrfs_bio *bbio = rbio->bbio; |
Kees Cook | 1389053 | 2018-05-29 16:44:59 -0700 | [diff] [blame] | 1181 | void **pointers = rbio->finish_pointers; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1182 | int nr_data = rbio->nr_data; |
| 1183 | int stripe; |
| 1184 | int pagenr; |
| 1185 | int p_stripe = -1; |
| 1186 | int q_stripe = -1; |
| 1187 | struct bio_list bio_list; |
| 1188 | struct bio *bio; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1189 | int ret; |
| 1190 | |
| 1191 | bio_list_init(&bio_list); |
| 1192 | |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1193 | if (rbio->real_stripes - rbio->nr_data == 1) { |
| 1194 | p_stripe = rbio->real_stripes - 1; |
| 1195 | } else if (rbio->real_stripes - rbio->nr_data == 2) { |
| 1196 | p_stripe = rbio->real_stripes - 2; |
| 1197 | q_stripe = rbio->real_stripes - 1; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1198 | } else { |
| 1199 | BUG(); |
| 1200 | } |
| 1201 | |
| 1202 | /* at this point we either have a full stripe, |
| 1203 | * or we've read the full stripe from the drive. |
| 1204 | * recalculate the parity and write the new results. |
| 1205 | * |
| 1206 | * We're not allowed to add any new bios to the |
| 1207 | * bio list here, anyone else that wants to |
| 1208 | * change this stripe needs to do their own rmw. |
| 1209 | */ |
| 1210 | spin_lock_irq(&rbio->bio_list_lock); |
| 1211 | set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); |
| 1212 | spin_unlock_irq(&rbio->bio_list_lock); |
| 1213 | |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 1214 | atomic_set(&rbio->error, 0); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1215 | |
| 1216 | /* |
| 1217 | * now that we've set rmw_locked, run through the |
| 1218 | * bio list one last time and map the page pointers |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 1219 | * |
| 1220 | * We don't cache full rbios because we're assuming |
| 1221 | * the higher layers are unlikely to use this area of |
| 1222 | * the disk again soon. If they do use it again, |
| 1223 | * hopefully they will send another full bio. |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1224 | */ |
| 1225 | index_rbio_pages(rbio); |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 1226 | if (!rbio_is_full(rbio)) |
| 1227 | cache_rbio_pages(rbio); |
| 1228 | else |
| 1229 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1230 | |
Zhao Lei | 915e229 | 2015-03-03 20:42:48 +0800 | [diff] [blame] | 1231 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1232 | struct page *p; |
| 1233 | /* first collect one page from each data stripe */ |
| 1234 | for (stripe = 0; stripe < nr_data; stripe++) { |
| 1235 | p = page_in_rbio(rbio, stripe, pagenr, 0); |
| 1236 | pointers[stripe] = kmap(p); |
| 1237 | } |
| 1238 | |
| 1239 | /* then add the parity stripe */ |
| 1240 | p = rbio_pstripe_page(rbio, pagenr); |
| 1241 | SetPageUptodate(p); |
| 1242 | pointers[stripe++] = kmap(p); |
| 1243 | |
| 1244 | if (q_stripe != -1) { |
| 1245 | |
| 1246 | /* |
| 1247 | * raid6, add the qstripe and call the |
| 1248 | * library function to fill in our p/q |
| 1249 | */ |
| 1250 | p = rbio_qstripe_page(rbio, pagenr); |
| 1251 | SetPageUptodate(p); |
| 1252 | pointers[stripe++] = kmap(p); |
| 1253 | |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1254 | raid6_call.gen_syndrome(rbio->real_stripes, PAGE_SIZE, |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1255 | pointers); |
| 1256 | } else { |
| 1257 | /* raid5 */ |
David Sterba | 69d2480 | 2018-06-29 10:56:44 +0200 | [diff] [blame] | 1258 | copy_page(pointers[nr_data], pointers[0]); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1259 | run_xor(pointers + 1, nr_data - 1, PAGE_SIZE); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1260 | } |
| 1261 | |
| 1262 | |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1263 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1264 | kunmap(page_in_rbio(rbio, stripe, pagenr, 0)); |
| 1265 | } |
| 1266 | |
| 1267 | /* |
| 1268 | * time to start writing. Make bios for everything from the |
| 1269 | * higher layers (the bio_list in our rbio) and our p/q. Ignore |
| 1270 | * everything else. |
| 1271 | */ |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1272 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
Zhao Lei | 915e229 | 2015-03-03 20:42:48 +0800 | [diff] [blame] | 1273 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1274 | struct page *page; |
| 1275 | if (stripe < rbio->nr_data) { |
| 1276 | page = page_in_rbio(rbio, stripe, pagenr, 1); |
| 1277 | if (!page) |
| 1278 | continue; |
| 1279 | } else { |
| 1280 | page = rbio_stripe_page(rbio, stripe, pagenr); |
| 1281 | } |
| 1282 | |
| 1283 | ret = rbio_add_io_page(rbio, &bio_list, |
| 1284 | page, stripe, pagenr, rbio->stripe_len); |
| 1285 | if (ret) |
| 1286 | goto cleanup; |
| 1287 | } |
| 1288 | } |
| 1289 | |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1290 | if (likely(!bbio->num_tgtdevs)) |
| 1291 | goto write_data; |
| 1292 | |
| 1293 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
| 1294 | if (!bbio->tgtdev_map[stripe]) |
| 1295 | continue; |
| 1296 | |
Zhao Lei | 915e229 | 2015-03-03 20:42:48 +0800 | [diff] [blame] | 1297 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1298 | struct page *page; |
| 1299 | if (stripe < rbio->nr_data) { |
| 1300 | page = page_in_rbio(rbio, stripe, pagenr, 1); |
| 1301 | if (!page) |
| 1302 | continue; |
| 1303 | } else { |
| 1304 | page = rbio_stripe_page(rbio, stripe, pagenr); |
| 1305 | } |
| 1306 | |
| 1307 | ret = rbio_add_io_page(rbio, &bio_list, page, |
| 1308 | rbio->bbio->tgtdev_map[stripe], |
| 1309 | pagenr, rbio->stripe_len); |
| 1310 | if (ret) |
| 1311 | goto cleanup; |
| 1312 | } |
| 1313 | } |
| 1314 | |
| 1315 | write_data: |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 1316 | atomic_set(&rbio->stripes_pending, bio_list_size(&bio_list)); |
| 1317 | BUG_ON(atomic_read(&rbio->stripes_pending) == 0); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1318 | |
| 1319 | while (1) { |
| 1320 | bio = bio_list_pop(&bio_list); |
| 1321 | if (!bio) |
| 1322 | break; |
| 1323 | |
| 1324 | bio->bi_private = rbio; |
| 1325 | bio->bi_end_io = raid_write_end_io; |
David Sterba | ebcc326 | 2018-06-29 10:56:53 +0200 | [diff] [blame] | 1326 | bio->bi_opf = REQ_OP_WRITE; |
Mike Christie | 4e49ea4 | 2016-06-05 14:31:41 -0500 | [diff] [blame] | 1327 | |
| 1328 | submit_bio(bio); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1329 | } |
| 1330 | return; |
| 1331 | |
| 1332 | cleanup: |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 1333 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
Liu Bo | 785884f | 2017-09-22 12:11:18 -0600 | [diff] [blame] | 1334 | |
| 1335 | while ((bio = bio_list_pop(&bio_list))) |
| 1336 | bio_put(bio); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1337 | } |
| 1338 | |
| 1339 | /* |
| 1340 | * helper to find the stripe number for a given bio. Used to figure out which |
| 1341 | * stripe has failed. This expects the bio to correspond to a physical disk, |
| 1342 | * so it looks up based on physical sector numbers. |
| 1343 | */ |
| 1344 | static int find_bio_stripe(struct btrfs_raid_bio *rbio, |
| 1345 | struct bio *bio) |
| 1346 | { |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1347 | u64 physical = bio->bi_iter.bi_sector; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1348 | u64 stripe_start; |
| 1349 | int i; |
| 1350 | struct btrfs_bio_stripe *stripe; |
| 1351 | |
| 1352 | physical <<= 9; |
| 1353 | |
| 1354 | for (i = 0; i < rbio->bbio->num_stripes; i++) { |
| 1355 | stripe = &rbio->bbio->stripes[i]; |
| 1356 | stripe_start = stripe->physical; |
| 1357 | if (physical >= stripe_start && |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1358 | physical < stripe_start + rbio->stripe_len && |
Dmitriy Gorokh | 047fdea | 2018-02-16 19:51:38 +0000 | [diff] [blame] | 1359 | stripe->dev->bdev && |
Christoph Hellwig | 74d4699 | 2017-08-23 19:10:32 +0200 | [diff] [blame] | 1360 | bio->bi_disk == stripe->dev->bdev->bd_disk && |
| 1361 | bio->bi_partno == stripe->dev->bdev->bd_partno) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1362 | return i; |
| 1363 | } |
| 1364 | } |
| 1365 | return -1; |
| 1366 | } |
| 1367 | |
| 1368 | /* |
| 1369 | * helper to find the stripe number for a given |
| 1370 | * bio (before mapping). Used to figure out which stripe has |
| 1371 | * failed. This looks up based on logical block numbers. |
| 1372 | */ |
| 1373 | static int find_logical_bio_stripe(struct btrfs_raid_bio *rbio, |
| 1374 | struct bio *bio) |
| 1375 | { |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1376 | u64 logical = bio->bi_iter.bi_sector; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1377 | u64 stripe_start; |
| 1378 | int i; |
| 1379 | |
| 1380 | logical <<= 9; |
| 1381 | |
| 1382 | for (i = 0; i < rbio->nr_data; i++) { |
Zhao Lei | 8e5cfb5 | 2015-01-20 15:11:33 +0800 | [diff] [blame] | 1383 | stripe_start = rbio->bbio->raid_map[i]; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1384 | if (logical >= stripe_start && |
| 1385 | logical < stripe_start + rbio->stripe_len) { |
| 1386 | return i; |
| 1387 | } |
| 1388 | } |
| 1389 | return -1; |
| 1390 | } |
| 1391 | |
| 1392 | /* |
| 1393 | * returns -EIO if we had too many failures |
| 1394 | */ |
| 1395 | static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed) |
| 1396 | { |
| 1397 | unsigned long flags; |
| 1398 | int ret = 0; |
| 1399 | |
| 1400 | spin_lock_irqsave(&rbio->bio_list_lock, flags); |
| 1401 | |
| 1402 | /* we already know this stripe is bad, move on */ |
| 1403 | if (rbio->faila == failed || rbio->failb == failed) |
| 1404 | goto out; |
| 1405 | |
| 1406 | if (rbio->faila == -1) { |
| 1407 | /* first failure on this rbio */ |
| 1408 | rbio->faila = failed; |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 1409 | atomic_inc(&rbio->error); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1410 | } else if (rbio->failb == -1) { |
| 1411 | /* second failure on this rbio */ |
| 1412 | rbio->failb = failed; |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 1413 | atomic_inc(&rbio->error); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1414 | } else { |
| 1415 | ret = -EIO; |
| 1416 | } |
| 1417 | out: |
| 1418 | spin_unlock_irqrestore(&rbio->bio_list_lock, flags); |
| 1419 | |
| 1420 | return ret; |
| 1421 | } |
| 1422 | |
| 1423 | /* |
| 1424 | * helper to fail a stripe based on a physical disk |
| 1425 | * bio. |
| 1426 | */ |
| 1427 | static int fail_bio_stripe(struct btrfs_raid_bio *rbio, |
| 1428 | struct bio *bio) |
| 1429 | { |
| 1430 | int failed = find_bio_stripe(rbio, bio); |
| 1431 | |
| 1432 | if (failed < 0) |
| 1433 | return -EIO; |
| 1434 | |
| 1435 | return fail_rbio_index(rbio, failed); |
| 1436 | } |
| 1437 | |
| 1438 | /* |
| 1439 | * this sets each page in the bio uptodate. It should only be used on private |
| 1440 | * rbio pages, nothing that comes in from the higher layers |
| 1441 | */ |
| 1442 | static void set_bio_pages_uptodate(struct bio *bio) |
| 1443 | { |
Liu Bo | 0198e5b | 2018-01-12 18:07:01 -0700 | [diff] [blame] | 1444 | struct bio_vec *bvec; |
| 1445 | int i; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1446 | |
Liu Bo | 0198e5b | 2018-01-12 18:07:01 -0700 | [diff] [blame] | 1447 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
Filipe Manana | 6592e58 | 2017-07-12 23:36:02 +0100 | [diff] [blame] | 1448 | |
Liu Bo | 0198e5b | 2018-01-12 18:07:01 -0700 | [diff] [blame] | 1449 | bio_for_each_segment_all(bvec, bio, i) |
| 1450 | SetPageUptodate(bvec->bv_page); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1451 | } |
| 1452 | |
| 1453 | /* |
| 1454 | * end io for the read phase of the rmw cycle. All the bios here are physical |
| 1455 | * stripe bios we've read from the disk so we can recalculate the parity of the |
| 1456 | * stripe. |
| 1457 | * |
| 1458 | * This will usually kick off finish_rmw once all the bios are read in, but it |
| 1459 | * may trigger parity reconstruction if we had any errors along the way |
| 1460 | */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1461 | static void raid_rmw_end_io(struct bio *bio) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1462 | { |
| 1463 | struct btrfs_raid_bio *rbio = bio->bi_private; |
| 1464 | |
Christoph Hellwig | 4e4cbee | 2017-06-03 09:38:06 +0200 | [diff] [blame] | 1465 | if (bio->bi_status) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1466 | fail_bio_stripe(rbio, bio); |
| 1467 | else |
| 1468 | set_bio_pages_uptodate(bio); |
| 1469 | |
| 1470 | bio_put(bio); |
| 1471 | |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 1472 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1473 | return; |
| 1474 | |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 1475 | if (atomic_read(&rbio->error) > rbio->bbio->max_errors) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1476 | goto cleanup; |
| 1477 | |
| 1478 | /* |
| 1479 | * this will normally call finish_rmw to start our write |
| 1480 | * but if there are any failed stripes we'll reconstruct |
| 1481 | * from parity first |
| 1482 | */ |
| 1483 | validate_rbio_for_rmw(rbio); |
| 1484 | return; |
| 1485 | |
| 1486 | cleanup: |
| 1487 | |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 1488 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1489 | } |
| 1490 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1491 | /* |
| 1492 | * the stripe must be locked by the caller. It will |
| 1493 | * unlock after all the writes are done |
| 1494 | */ |
| 1495 | static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio) |
| 1496 | { |
| 1497 | int bios_to_read = 0; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1498 | struct bio_list bio_list; |
| 1499 | int ret; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1500 | int pagenr; |
| 1501 | int stripe; |
| 1502 | struct bio *bio; |
| 1503 | |
| 1504 | bio_list_init(&bio_list); |
| 1505 | |
| 1506 | ret = alloc_rbio_pages(rbio); |
| 1507 | if (ret) |
| 1508 | goto cleanup; |
| 1509 | |
| 1510 | index_rbio_pages(rbio); |
| 1511 | |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 1512 | atomic_set(&rbio->error, 0); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1513 | /* |
| 1514 | * build a list of bios to read all the missing parts of this |
| 1515 | * stripe |
| 1516 | */ |
| 1517 | for (stripe = 0; stripe < rbio->nr_data; stripe++) { |
Zhao Lei | 915e229 | 2015-03-03 20:42:48 +0800 | [diff] [blame] | 1518 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1519 | struct page *page; |
| 1520 | /* |
| 1521 | * we want to find all the pages missing from |
| 1522 | * the rbio and read them from the disk. If |
| 1523 | * page_in_rbio finds a page in the bio list |
| 1524 | * we don't need to read it off the stripe. |
| 1525 | */ |
| 1526 | page = page_in_rbio(rbio, stripe, pagenr, 1); |
| 1527 | if (page) |
| 1528 | continue; |
| 1529 | |
| 1530 | page = rbio_stripe_page(rbio, stripe, pagenr); |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 1531 | /* |
| 1532 | * the bio cache may have handed us an uptodate |
| 1533 | * page. If so, be happy and use it |
| 1534 | */ |
| 1535 | if (PageUptodate(page)) |
| 1536 | continue; |
| 1537 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1538 | ret = rbio_add_io_page(rbio, &bio_list, page, |
| 1539 | stripe, pagenr, rbio->stripe_len); |
| 1540 | if (ret) |
| 1541 | goto cleanup; |
| 1542 | } |
| 1543 | } |
| 1544 | |
| 1545 | bios_to_read = bio_list_size(&bio_list); |
| 1546 | if (!bios_to_read) { |
| 1547 | /* |
| 1548 | * this can happen if others have merged with |
| 1549 | * us, it means there is nothing left to read. |
| 1550 | * But if there are missing devices it may not be |
| 1551 | * safe to do the full stripe write yet. |
| 1552 | */ |
| 1553 | goto finish; |
| 1554 | } |
| 1555 | |
| 1556 | /* |
| 1557 | * the bbio may be freed once we submit the last bio. Make sure |
| 1558 | * not to touch it after that |
| 1559 | */ |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 1560 | atomic_set(&rbio->stripes_pending, bios_to_read); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1561 | while (1) { |
| 1562 | bio = bio_list_pop(&bio_list); |
| 1563 | if (!bio) |
| 1564 | break; |
| 1565 | |
| 1566 | bio->bi_private = rbio; |
| 1567 | bio->bi_end_io = raid_rmw_end_io; |
David Sterba | ebcc326 | 2018-06-29 10:56:53 +0200 | [diff] [blame] | 1568 | bio->bi_opf = REQ_OP_READ; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1569 | |
Jeff Mahoney | 0b246af | 2016-06-22 18:54:23 -0400 | [diff] [blame] | 1570 | btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1571 | |
Mike Christie | 4e49ea4 | 2016-06-05 14:31:41 -0500 | [diff] [blame] | 1572 | submit_bio(bio); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1573 | } |
| 1574 | /* the actual write will happen once the reads are done */ |
| 1575 | return 0; |
| 1576 | |
| 1577 | cleanup: |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 1578 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
Liu Bo | 785884f | 2017-09-22 12:11:18 -0600 | [diff] [blame] | 1579 | |
| 1580 | while ((bio = bio_list_pop(&bio_list))) |
| 1581 | bio_put(bio); |
| 1582 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1583 | return -EIO; |
| 1584 | |
| 1585 | finish: |
| 1586 | validate_rbio_for_rmw(rbio); |
| 1587 | return 0; |
| 1588 | } |
| 1589 | |
| 1590 | /* |
| 1591 | * if the upper layers pass in a full stripe, we thank them by only allocating |
| 1592 | * enough pages to hold the parity, and sending it all down quickly. |
| 1593 | */ |
| 1594 | static int full_stripe_write(struct btrfs_raid_bio *rbio) |
| 1595 | { |
| 1596 | int ret; |
| 1597 | |
| 1598 | ret = alloc_rbio_parity_pages(rbio); |
Miao Xie | 3cd846d | 2013-07-22 16:36:57 +0800 | [diff] [blame] | 1599 | if (ret) { |
| 1600 | __free_raid_bio(rbio); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1601 | return ret; |
Miao Xie | 3cd846d | 2013-07-22 16:36:57 +0800 | [diff] [blame] | 1602 | } |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1603 | |
| 1604 | ret = lock_stripe_add(rbio); |
| 1605 | if (ret == 0) |
| 1606 | finish_rmw(rbio); |
| 1607 | return 0; |
| 1608 | } |
| 1609 | |
| 1610 | /* |
| 1611 | * partial stripe writes get handed over to async helpers. |
| 1612 | * We're really hoping to merge a few more writes into this |
| 1613 | * rbio before calculating new parity |
| 1614 | */ |
| 1615 | static int partial_stripe_write(struct btrfs_raid_bio *rbio) |
| 1616 | { |
| 1617 | int ret; |
| 1618 | |
| 1619 | ret = lock_stripe_add(rbio); |
| 1620 | if (ret == 0) |
David Sterba | cf6a4a7 | 2018-06-29 10:56:58 +0200 | [diff] [blame] | 1621 | start_async_work(rbio, rmw_work); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1622 | return 0; |
| 1623 | } |
| 1624 | |
| 1625 | /* |
| 1626 | * sometimes while we were reading from the drive to |
| 1627 | * recalculate parity, enough new bios come into create |
| 1628 | * a full stripe. So we do a check here to see if we can |
| 1629 | * go directly to finish_rmw |
| 1630 | */ |
| 1631 | static int __raid56_parity_write(struct btrfs_raid_bio *rbio) |
| 1632 | { |
| 1633 | /* head off into rmw land if we don't have a full stripe */ |
| 1634 | if (!rbio_is_full(rbio)) |
| 1635 | return partial_stripe_write(rbio); |
| 1636 | return full_stripe_write(rbio); |
| 1637 | } |
| 1638 | |
| 1639 | /* |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1640 | * We use plugging call backs to collect full stripes. |
| 1641 | * Any time we get a partial stripe write while plugged |
| 1642 | * we collect it into a list. When the unplug comes down, |
| 1643 | * we sort the list by logical block number and merge |
| 1644 | * everything we can into the same rbios |
| 1645 | */ |
| 1646 | struct btrfs_plug_cb { |
| 1647 | struct blk_plug_cb cb; |
| 1648 | struct btrfs_fs_info *info; |
| 1649 | struct list_head rbio_list; |
| 1650 | struct btrfs_work work; |
| 1651 | }; |
| 1652 | |
| 1653 | /* |
| 1654 | * rbios on the plug list are sorted for easier merging. |
| 1655 | */ |
| 1656 | static int plug_cmp(void *priv, struct list_head *a, struct list_head *b) |
| 1657 | { |
| 1658 | struct btrfs_raid_bio *ra = container_of(a, struct btrfs_raid_bio, |
| 1659 | plug_list); |
| 1660 | struct btrfs_raid_bio *rb = container_of(b, struct btrfs_raid_bio, |
| 1661 | plug_list); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1662 | u64 a_sector = ra->bio_list.head->bi_iter.bi_sector; |
| 1663 | u64 b_sector = rb->bio_list.head->bi_iter.bi_sector; |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1664 | |
| 1665 | if (a_sector < b_sector) |
| 1666 | return -1; |
| 1667 | if (a_sector > b_sector) |
| 1668 | return 1; |
| 1669 | return 0; |
| 1670 | } |
| 1671 | |
| 1672 | static void run_plug(struct btrfs_plug_cb *plug) |
| 1673 | { |
| 1674 | struct btrfs_raid_bio *cur; |
| 1675 | struct btrfs_raid_bio *last = NULL; |
| 1676 | |
| 1677 | /* |
| 1678 | * sort our plug list then try to merge |
| 1679 | * everything we can in hopes of creating full |
| 1680 | * stripes. |
| 1681 | */ |
| 1682 | list_sort(NULL, &plug->rbio_list, plug_cmp); |
| 1683 | while (!list_empty(&plug->rbio_list)) { |
| 1684 | cur = list_entry(plug->rbio_list.next, |
| 1685 | struct btrfs_raid_bio, plug_list); |
| 1686 | list_del_init(&cur->plug_list); |
| 1687 | |
| 1688 | if (rbio_is_full(cur)) { |
David Sterba | c7b562c | 2018-06-29 10:57:10 +0200 | [diff] [blame] | 1689 | int ret; |
| 1690 | |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1691 | /* we have a full stripe, send it down */ |
David Sterba | c7b562c | 2018-06-29 10:57:10 +0200 | [diff] [blame] | 1692 | ret = full_stripe_write(cur); |
| 1693 | BUG_ON(ret); |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1694 | continue; |
| 1695 | } |
| 1696 | if (last) { |
| 1697 | if (rbio_can_merge(last, cur)) { |
| 1698 | merge_rbio(last, cur); |
| 1699 | __free_raid_bio(cur); |
| 1700 | continue; |
| 1701 | |
| 1702 | } |
| 1703 | __raid56_parity_write(last); |
| 1704 | } |
| 1705 | last = cur; |
| 1706 | } |
| 1707 | if (last) { |
| 1708 | __raid56_parity_write(last); |
| 1709 | } |
| 1710 | kfree(plug); |
| 1711 | } |
| 1712 | |
| 1713 | /* |
| 1714 | * if the unplug comes from schedule, we have to push the |
| 1715 | * work off to a helper thread |
| 1716 | */ |
| 1717 | static void unplug_work(struct btrfs_work *work) |
| 1718 | { |
| 1719 | struct btrfs_plug_cb *plug; |
| 1720 | plug = container_of(work, struct btrfs_plug_cb, work); |
| 1721 | run_plug(plug); |
| 1722 | } |
| 1723 | |
| 1724 | static void btrfs_raid_unplug(struct blk_plug_cb *cb, bool from_schedule) |
| 1725 | { |
| 1726 | struct btrfs_plug_cb *plug; |
| 1727 | plug = container_of(cb, struct btrfs_plug_cb, cb); |
| 1728 | |
| 1729 | if (from_schedule) { |
Liu Bo | 9e0af23 | 2014-08-15 23:36:53 +0800 | [diff] [blame] | 1730 | btrfs_init_work(&plug->work, btrfs_rmw_helper, |
| 1731 | unplug_work, NULL, NULL); |
Qu Wenruo | d05a33a | 2014-02-28 10:46:11 +0800 | [diff] [blame] | 1732 | btrfs_queue_work(plug->info->rmw_workers, |
| 1733 | &plug->work); |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1734 | return; |
| 1735 | } |
| 1736 | run_plug(plug); |
| 1737 | } |
| 1738 | |
| 1739 | /* |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1740 | * our main entry point for writes from the rest of the FS. |
| 1741 | */ |
Jeff Mahoney | 2ff7e61 | 2016-06-22 18:54:24 -0400 | [diff] [blame] | 1742 | int raid56_parity_write(struct btrfs_fs_info *fs_info, struct bio *bio, |
Zhao Lei | 8e5cfb5 | 2015-01-20 15:11:33 +0800 | [diff] [blame] | 1743 | struct btrfs_bio *bbio, u64 stripe_len) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1744 | { |
| 1745 | struct btrfs_raid_bio *rbio; |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1746 | struct btrfs_plug_cb *plug = NULL; |
| 1747 | struct blk_plug_cb *cb; |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 1748 | int ret; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1749 | |
Jeff Mahoney | 2ff7e61 | 2016-06-22 18:54:24 -0400 | [diff] [blame] | 1750 | rbio = alloc_rbio(fs_info, bbio, stripe_len); |
Miao Xie | af8e2d1 | 2014-10-23 14:42:50 +0800 | [diff] [blame] | 1751 | if (IS_ERR(rbio)) { |
Zhao Lei | 6e9606d | 2015-01-20 15:11:34 +0800 | [diff] [blame] | 1752 | btrfs_put_bbio(bbio); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1753 | return PTR_ERR(rbio); |
Miao Xie | af8e2d1 | 2014-10-23 14:42:50 +0800 | [diff] [blame] | 1754 | } |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1755 | bio_list_add(&rbio->bio_list, bio); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1756 | rbio->bio_list_bytes = bio->bi_iter.bi_size; |
Miao Xie | 1b94b55 | 2014-11-06 16:14:21 +0800 | [diff] [blame] | 1757 | rbio->operation = BTRFS_RBIO_WRITE; |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1758 | |
Jeff Mahoney | 0b246af | 2016-06-22 18:54:23 -0400 | [diff] [blame] | 1759 | btrfs_bio_counter_inc_noblocked(fs_info); |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 1760 | rbio->generic_bio_cnt = 1; |
| 1761 | |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1762 | /* |
| 1763 | * don't plug on full rbios, just get them out the door |
| 1764 | * as quickly as we can |
| 1765 | */ |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 1766 | if (rbio_is_full(rbio)) { |
| 1767 | ret = full_stripe_write(rbio); |
| 1768 | if (ret) |
Jeff Mahoney | 0b246af | 2016-06-22 18:54:23 -0400 | [diff] [blame] | 1769 | btrfs_bio_counter_dec(fs_info); |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 1770 | return ret; |
| 1771 | } |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1772 | |
Jeff Mahoney | 0b246af | 2016-06-22 18:54:23 -0400 | [diff] [blame] | 1773 | cb = blk_check_plugged(btrfs_raid_unplug, fs_info, sizeof(*plug)); |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1774 | if (cb) { |
| 1775 | plug = container_of(cb, struct btrfs_plug_cb, cb); |
| 1776 | if (!plug->info) { |
Jeff Mahoney | 0b246af | 2016-06-22 18:54:23 -0400 | [diff] [blame] | 1777 | plug->info = fs_info; |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1778 | INIT_LIST_HEAD(&plug->rbio_list); |
| 1779 | } |
| 1780 | list_add_tail(&rbio->plug_list, &plug->rbio_list); |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 1781 | ret = 0; |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1782 | } else { |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 1783 | ret = __raid56_parity_write(rbio); |
| 1784 | if (ret) |
Jeff Mahoney | 0b246af | 2016-06-22 18:54:23 -0400 | [diff] [blame] | 1785 | btrfs_bio_counter_dec(fs_info); |
Chris Mason | 6ac0f48 | 2013-01-31 14:42:28 -0500 | [diff] [blame] | 1786 | } |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 1787 | return ret; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1788 | } |
| 1789 | |
| 1790 | /* |
| 1791 | * all parity reconstruction happens here. We've read in everything |
| 1792 | * we can find from the drives and this does the heavy lifting of |
| 1793 | * sorting the good from the bad. |
| 1794 | */ |
| 1795 | static void __raid_recover_end_io(struct btrfs_raid_bio *rbio) |
| 1796 | { |
| 1797 | int pagenr, stripe; |
| 1798 | void **pointers; |
| 1799 | int faila = -1, failb = -1; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1800 | struct page *page; |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 1801 | blk_status_t err; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1802 | int i; |
| 1803 | |
David Sterba | 31e818f | 2015-02-20 18:00:26 +0100 | [diff] [blame] | 1804 | pointers = kcalloc(rbio->real_stripes, sizeof(void *), GFP_NOFS); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1805 | if (!pointers) { |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 1806 | err = BLK_STS_RESOURCE; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1807 | goto cleanup_io; |
| 1808 | } |
| 1809 | |
| 1810 | faila = rbio->faila; |
| 1811 | failb = rbio->failb; |
| 1812 | |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 1813 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
| 1814 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1815 | spin_lock_irq(&rbio->bio_list_lock); |
| 1816 | set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); |
| 1817 | spin_unlock_irq(&rbio->bio_list_lock); |
| 1818 | } |
| 1819 | |
| 1820 | index_rbio_pages(rbio); |
| 1821 | |
Zhao Lei | 915e229 | 2015-03-03 20:42:48 +0800 | [diff] [blame] | 1822 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 1823 | /* |
| 1824 | * Now we just use bitmap to mark the horizontal stripes in |
| 1825 | * which we have data when doing parity scrub. |
| 1826 | */ |
| 1827 | if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB && |
| 1828 | !test_bit(pagenr, rbio->dbitmap)) |
| 1829 | continue; |
| 1830 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1831 | /* setup our array of pointers with pages |
| 1832 | * from each stripe |
| 1833 | */ |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1834 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1835 | /* |
| 1836 | * if we're rebuilding a read, we have to use |
| 1837 | * pages from the bio list |
| 1838 | */ |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 1839 | if ((rbio->operation == BTRFS_RBIO_READ_REBUILD || |
| 1840 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) && |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1841 | (stripe == faila || stripe == failb)) { |
| 1842 | page = page_in_rbio(rbio, stripe, pagenr, 0); |
| 1843 | } else { |
| 1844 | page = rbio_stripe_page(rbio, stripe, pagenr); |
| 1845 | } |
| 1846 | pointers[stripe] = kmap(page); |
| 1847 | } |
| 1848 | |
| 1849 | /* all raid6 handling here */ |
Zhao Lei | 10f1190 | 2015-01-20 15:11:43 +0800 | [diff] [blame] | 1850 | if (rbio->bbio->map_type & BTRFS_BLOCK_GROUP_RAID6) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1851 | /* |
| 1852 | * single failure, rebuild from parity raid5 |
| 1853 | * style |
| 1854 | */ |
| 1855 | if (failb < 0) { |
| 1856 | if (faila == rbio->nr_data) { |
| 1857 | /* |
| 1858 | * Just the P stripe has failed, without |
| 1859 | * a bad data or Q stripe. |
| 1860 | * TODO, we should redo the xor here. |
| 1861 | */ |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 1862 | err = BLK_STS_IOERR; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1863 | goto cleanup; |
| 1864 | } |
| 1865 | /* |
| 1866 | * a single failure in raid6 is rebuilt |
| 1867 | * in the pstripe code below |
| 1868 | */ |
| 1869 | goto pstripe; |
| 1870 | } |
| 1871 | |
| 1872 | /* make sure our ps and qs are in order */ |
| 1873 | if (faila > failb) { |
| 1874 | int tmp = failb; |
| 1875 | failb = faila; |
| 1876 | faila = tmp; |
| 1877 | } |
| 1878 | |
| 1879 | /* if the q stripe is failed, do a pstripe reconstruction |
| 1880 | * from the xors. |
| 1881 | * If both the q stripe and the P stripe are failed, we're |
| 1882 | * here due to a crc mismatch and we can't give them the |
| 1883 | * data they want |
| 1884 | */ |
Zhao Lei | 8e5cfb5 | 2015-01-20 15:11:33 +0800 | [diff] [blame] | 1885 | if (rbio->bbio->raid_map[failb] == RAID6_Q_STRIPE) { |
| 1886 | if (rbio->bbio->raid_map[faila] == |
| 1887 | RAID5_P_STRIPE) { |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 1888 | err = BLK_STS_IOERR; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1889 | goto cleanup; |
| 1890 | } |
| 1891 | /* |
| 1892 | * otherwise we have one bad data stripe and |
| 1893 | * a good P stripe. raid5! |
| 1894 | */ |
| 1895 | goto pstripe; |
| 1896 | } |
| 1897 | |
Zhao Lei | 8e5cfb5 | 2015-01-20 15:11:33 +0800 | [diff] [blame] | 1898 | if (rbio->bbio->raid_map[failb] == RAID5_P_STRIPE) { |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1899 | raid6_datap_recov(rbio->real_stripes, |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1900 | PAGE_SIZE, faila, pointers); |
| 1901 | } else { |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1902 | raid6_2data_recov(rbio->real_stripes, |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1903 | PAGE_SIZE, faila, failb, |
| 1904 | pointers); |
| 1905 | } |
| 1906 | } else { |
| 1907 | void *p; |
| 1908 | |
| 1909 | /* rebuild from P stripe here (raid5 or raid6) */ |
| 1910 | BUG_ON(failb != -1); |
| 1911 | pstripe: |
| 1912 | /* Copy parity block into failed block to start with */ |
David Sterba | 69d2480 | 2018-06-29 10:56:44 +0200 | [diff] [blame] | 1913 | copy_page(pointers[faila], pointers[rbio->nr_data]); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1914 | |
| 1915 | /* rearrange the pointer array */ |
| 1916 | p = pointers[faila]; |
| 1917 | for (stripe = faila; stripe < rbio->nr_data - 1; stripe++) |
| 1918 | pointers[stripe] = pointers[stripe + 1]; |
| 1919 | pointers[rbio->nr_data - 1] = p; |
| 1920 | |
| 1921 | /* xor in the rest */ |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1922 | run_xor(pointers, rbio->nr_data - 1, PAGE_SIZE); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1923 | } |
| 1924 | /* if we're doing this rebuild as part of an rmw, go through |
| 1925 | * and set all of our private rbio pages in the |
| 1926 | * failed stripes as uptodate. This way finish_rmw will |
| 1927 | * know they can be trusted. If this was a read reconstruction, |
| 1928 | * other endio functions will fiddle the uptodate bits |
| 1929 | */ |
Miao Xie | 1b94b55 | 2014-11-06 16:14:21 +0800 | [diff] [blame] | 1930 | if (rbio->operation == BTRFS_RBIO_WRITE) { |
Zhao Lei | 915e229 | 2015-03-03 20:42:48 +0800 | [diff] [blame] | 1931 | for (i = 0; i < rbio->stripe_npages; i++) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1932 | if (faila != -1) { |
| 1933 | page = rbio_stripe_page(rbio, faila, i); |
| 1934 | SetPageUptodate(page); |
| 1935 | } |
| 1936 | if (failb != -1) { |
| 1937 | page = rbio_stripe_page(rbio, failb, i); |
| 1938 | SetPageUptodate(page); |
| 1939 | } |
| 1940 | } |
| 1941 | } |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 1942 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1943 | /* |
| 1944 | * if we're rebuilding a read, we have to use |
| 1945 | * pages from the bio list |
| 1946 | */ |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 1947 | if ((rbio->operation == BTRFS_RBIO_READ_REBUILD || |
| 1948 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) && |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1949 | (stripe == faila || stripe == failb)) { |
| 1950 | page = page_in_rbio(rbio, stripe, pagenr, 0); |
| 1951 | } else { |
| 1952 | page = rbio_stripe_page(rbio, stripe, pagenr); |
| 1953 | } |
| 1954 | kunmap(page); |
| 1955 | } |
| 1956 | } |
| 1957 | |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 1958 | err = BLK_STS_OK; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1959 | cleanup: |
| 1960 | kfree(pointers); |
| 1961 | |
| 1962 | cleanup_io: |
Liu Bo | 580c6ef | 2018-03-22 09:20:11 +0800 | [diff] [blame] | 1963 | /* |
| 1964 | * Similar to READ_REBUILD, REBUILD_MISSING at this point also has a |
| 1965 | * valid rbio which is consistent with ondisk content, thus such a |
| 1966 | * valid rbio can be cached to avoid further disk reads. |
| 1967 | */ |
| 1968 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
| 1969 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { |
Liu Bo | 44ac474 | 2018-01-12 18:07:02 -0700 | [diff] [blame] | 1970 | /* |
| 1971 | * - In case of two failures, where rbio->failb != -1: |
| 1972 | * |
| 1973 | * Do not cache this rbio since the above read reconstruction |
| 1974 | * (raid6_datap_recov() or raid6_2data_recov()) may have |
| 1975 | * changed some content of stripes which are not identical to |
| 1976 | * on-disk content any more, otherwise, a later write/recover |
| 1977 | * may steal stripe_pages from this rbio and end up with |
| 1978 | * corruptions or rebuild failures. |
| 1979 | * |
| 1980 | * - In case of single failure, where rbio->failb == -1: |
| 1981 | * |
| 1982 | * Cache this rbio iff the above read reconstruction is |
| 1983 | * excuted without problems. |
| 1984 | */ |
| 1985 | if (err == BLK_STS_OK && rbio->failb < 0) |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 1986 | cache_rbio_pages(rbio); |
| 1987 | else |
| 1988 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); |
| 1989 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1990 | rbio_orig_end_io(rbio, err); |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 1991 | } else if (err == BLK_STS_OK) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 1992 | rbio->faila = -1; |
| 1993 | rbio->failb = -1; |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 1994 | |
| 1995 | if (rbio->operation == BTRFS_RBIO_WRITE) |
| 1996 | finish_rmw(rbio); |
| 1997 | else if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) |
| 1998 | finish_parity_scrub(rbio, 0); |
| 1999 | else |
| 2000 | BUG(); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2001 | } else { |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 2002 | rbio_orig_end_io(rbio, err); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2003 | } |
| 2004 | } |
| 2005 | |
| 2006 | /* |
| 2007 | * This is called only for stripes we've read from disk to |
| 2008 | * reconstruct the parity. |
| 2009 | */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 2010 | static void raid_recover_end_io(struct bio *bio) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2011 | { |
| 2012 | struct btrfs_raid_bio *rbio = bio->bi_private; |
| 2013 | |
| 2014 | /* |
| 2015 | * we only read stripe pages off the disk, set them |
| 2016 | * up to date if there were no errors |
| 2017 | */ |
Christoph Hellwig | 4e4cbee | 2017-06-03 09:38:06 +0200 | [diff] [blame] | 2018 | if (bio->bi_status) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2019 | fail_bio_stripe(rbio, bio); |
| 2020 | else |
| 2021 | set_bio_pages_uptodate(bio); |
| 2022 | bio_put(bio); |
| 2023 | |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 2024 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2025 | return; |
| 2026 | |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 2027 | if (atomic_read(&rbio->error) > rbio->bbio->max_errors) |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 2028 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2029 | else |
| 2030 | __raid_recover_end_io(rbio); |
| 2031 | } |
| 2032 | |
| 2033 | /* |
| 2034 | * reads everything we need off the disk to reconstruct |
| 2035 | * the parity. endio handlers trigger final reconstruction |
| 2036 | * when the IO is done. |
| 2037 | * |
| 2038 | * This is used both for reads from the higher layers and for |
| 2039 | * parity construction required to finish a rmw cycle. |
| 2040 | */ |
| 2041 | static int __raid56_parity_recover(struct btrfs_raid_bio *rbio) |
| 2042 | { |
| 2043 | int bios_to_read = 0; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2044 | struct bio_list bio_list; |
| 2045 | int ret; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2046 | int pagenr; |
| 2047 | int stripe; |
| 2048 | struct bio *bio; |
| 2049 | |
| 2050 | bio_list_init(&bio_list); |
| 2051 | |
| 2052 | ret = alloc_rbio_pages(rbio); |
| 2053 | if (ret) |
| 2054 | goto cleanup; |
| 2055 | |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 2056 | atomic_set(&rbio->error, 0); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2057 | |
| 2058 | /* |
Chris Mason | 4ae10b3 | 2013-01-31 14:42:09 -0500 | [diff] [blame] | 2059 | * read everything that hasn't failed. Thanks to the |
| 2060 | * stripe cache, it is possible that some or all of these |
| 2061 | * pages are going to be uptodate. |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2062 | */ |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 2063 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
Liu Bo | 5588383 | 2014-06-24 15:39:16 +0800 | [diff] [blame] | 2064 | if (rbio->faila == stripe || rbio->failb == stripe) { |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 2065 | atomic_inc(&rbio->error); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2066 | continue; |
Liu Bo | 5588383 | 2014-06-24 15:39:16 +0800 | [diff] [blame] | 2067 | } |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2068 | |
Zhao Lei | 915e229 | 2015-03-03 20:42:48 +0800 | [diff] [blame] | 2069 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2070 | struct page *p; |
| 2071 | |
| 2072 | /* |
| 2073 | * the rmw code may have already read this |
| 2074 | * page in |
| 2075 | */ |
| 2076 | p = rbio_stripe_page(rbio, stripe, pagenr); |
| 2077 | if (PageUptodate(p)) |
| 2078 | continue; |
| 2079 | |
| 2080 | ret = rbio_add_io_page(rbio, &bio_list, |
| 2081 | rbio_stripe_page(rbio, stripe, pagenr), |
| 2082 | stripe, pagenr, rbio->stripe_len); |
| 2083 | if (ret < 0) |
| 2084 | goto cleanup; |
| 2085 | } |
| 2086 | } |
| 2087 | |
| 2088 | bios_to_read = bio_list_size(&bio_list); |
| 2089 | if (!bios_to_read) { |
| 2090 | /* |
| 2091 | * we might have no bios to read just because the pages |
| 2092 | * were up to date, or we might have no bios to read because |
| 2093 | * the devices were gone. |
| 2094 | */ |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 2095 | if (atomic_read(&rbio->error) <= rbio->bbio->max_errors) { |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2096 | __raid_recover_end_io(rbio); |
| 2097 | goto out; |
| 2098 | } else { |
| 2099 | goto cleanup; |
| 2100 | } |
| 2101 | } |
| 2102 | |
| 2103 | /* |
| 2104 | * the bbio may be freed once we submit the last bio. Make sure |
| 2105 | * not to touch it after that |
| 2106 | */ |
Miao Xie | b89e1b0 | 2014-10-15 11:18:44 +0800 | [diff] [blame] | 2107 | atomic_set(&rbio->stripes_pending, bios_to_read); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2108 | while (1) { |
| 2109 | bio = bio_list_pop(&bio_list); |
| 2110 | if (!bio) |
| 2111 | break; |
| 2112 | |
| 2113 | bio->bi_private = rbio; |
| 2114 | bio->bi_end_io = raid_recover_end_io; |
David Sterba | ebcc326 | 2018-06-29 10:56:53 +0200 | [diff] [blame] | 2115 | bio->bi_opf = REQ_OP_READ; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2116 | |
Jeff Mahoney | 0b246af | 2016-06-22 18:54:23 -0400 | [diff] [blame] | 2117 | btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2118 | |
Mike Christie | 4e49ea4 | 2016-06-05 14:31:41 -0500 | [diff] [blame] | 2119 | submit_bio(bio); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2120 | } |
| 2121 | out: |
| 2122 | return 0; |
| 2123 | |
| 2124 | cleanup: |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 2125 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
| 2126 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 2127 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
Liu Bo | 785884f | 2017-09-22 12:11:18 -0600 | [diff] [blame] | 2128 | |
| 2129 | while ((bio = bio_list_pop(&bio_list))) |
| 2130 | bio_put(bio); |
| 2131 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2132 | return -EIO; |
| 2133 | } |
| 2134 | |
| 2135 | /* |
| 2136 | * the main entry point for reads from the higher layers. This |
| 2137 | * is really only called when the normal read path had a failure, |
| 2138 | * so we assume the bio they send down corresponds to a failed part |
| 2139 | * of the drive. |
| 2140 | */ |
Jeff Mahoney | 2ff7e61 | 2016-06-22 18:54:24 -0400 | [diff] [blame] | 2141 | int raid56_parity_recover(struct btrfs_fs_info *fs_info, struct bio *bio, |
Zhao Lei | 8e5cfb5 | 2015-01-20 15:11:33 +0800 | [diff] [blame] | 2142 | struct btrfs_bio *bbio, u64 stripe_len, |
| 2143 | int mirror_num, int generic_io) |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2144 | { |
| 2145 | struct btrfs_raid_bio *rbio; |
| 2146 | int ret; |
| 2147 | |
Liu Bo | abad60c | 2017-03-29 10:54:26 -0700 | [diff] [blame] | 2148 | if (generic_io) { |
| 2149 | ASSERT(bbio->mirror_num == mirror_num); |
| 2150 | btrfs_io_bio(bio)->mirror_num = mirror_num; |
| 2151 | } |
| 2152 | |
Jeff Mahoney | 2ff7e61 | 2016-06-22 18:54:24 -0400 | [diff] [blame] | 2153 | rbio = alloc_rbio(fs_info, bbio, stripe_len); |
Miao Xie | af8e2d1 | 2014-10-23 14:42:50 +0800 | [diff] [blame] | 2154 | if (IS_ERR(rbio)) { |
Zhao Lei | 6e9606d | 2015-01-20 15:11:34 +0800 | [diff] [blame] | 2155 | if (generic_io) |
| 2156 | btrfs_put_bbio(bbio); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2157 | return PTR_ERR(rbio); |
Miao Xie | af8e2d1 | 2014-10-23 14:42:50 +0800 | [diff] [blame] | 2158 | } |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2159 | |
Miao Xie | 1b94b55 | 2014-11-06 16:14:21 +0800 | [diff] [blame] | 2160 | rbio->operation = BTRFS_RBIO_READ_REBUILD; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2161 | bio_list_add(&rbio->bio_list, bio); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2162 | rbio->bio_list_bytes = bio->bi_iter.bi_size; |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2163 | |
| 2164 | rbio->faila = find_logical_bio_stripe(rbio, bio); |
| 2165 | if (rbio->faila == -1) { |
Jeff Mahoney | 0b246af | 2016-06-22 18:54:23 -0400 | [diff] [blame] | 2166 | btrfs_warn(fs_info, |
Liu Bo | e46a28c | 2016-07-29 10:57:55 -0700 | [diff] [blame] | 2167 | "%s could not find the bad stripe in raid56 so that we cannot recover any more (bio has logical %llu len %llu, bbio has map_type %llu)", |
| 2168 | __func__, (u64)bio->bi_iter.bi_sector << 9, |
| 2169 | (u64)bio->bi_iter.bi_size, bbio->map_type); |
Zhao Lei | 6e9606d | 2015-01-20 15:11:34 +0800 | [diff] [blame] | 2170 | if (generic_io) |
| 2171 | btrfs_put_bbio(bbio); |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2172 | kfree(rbio); |
| 2173 | return -EIO; |
| 2174 | } |
| 2175 | |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 2176 | if (generic_io) { |
Jeff Mahoney | 0b246af | 2016-06-22 18:54:23 -0400 | [diff] [blame] | 2177 | btrfs_bio_counter_inc_noblocked(fs_info); |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 2178 | rbio->generic_bio_cnt = 1; |
| 2179 | } else { |
Zhao Lei | 6e9606d | 2015-01-20 15:11:34 +0800 | [diff] [blame] | 2180 | btrfs_get_bbio(bbio); |
Miao Xie | 4245215 | 2014-11-25 16:39:28 +0800 | [diff] [blame] | 2181 | } |
| 2182 | |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2183 | /* |
Liu Bo | 8810f75 | 2018-01-02 13:36:41 -0700 | [diff] [blame] | 2184 | * Loop retry: |
| 2185 | * for 'mirror == 2', reconstruct from all other stripes. |
| 2186 | * for 'mirror_num > 2', select a stripe to fail on every retry. |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2187 | */ |
Liu Bo | 8810f75 | 2018-01-02 13:36:41 -0700 | [diff] [blame] | 2188 | if (mirror_num > 2) { |
| 2189 | /* |
| 2190 | * 'mirror == 3' is to fail the p stripe and |
| 2191 | * reconstruct from the q stripe. 'mirror > 3' is to |
| 2192 | * fail a data stripe and reconstruct from p+q stripe. |
| 2193 | */ |
| 2194 | rbio->failb = rbio->real_stripes - (mirror_num - 1); |
| 2195 | ASSERT(rbio->failb > 0); |
| 2196 | if (rbio->failb <= rbio->faila) |
| 2197 | rbio->failb--; |
| 2198 | } |
David Woodhouse | 53b381b | 2013-01-29 18:40:14 -0500 | [diff] [blame] | 2199 | |
| 2200 | ret = lock_stripe_add(rbio); |
| 2201 | |
| 2202 | /* |
| 2203 | * __raid56_parity_recover will end the bio with |
| 2204 | * any errors it hits. We don't want to return |
| 2205 | * its error value up the stack because our caller |
| 2206 | * will end up calling bio_endio with any nonzero |
| 2207 | * return |
| 2208 | */ |
| 2209 | if (ret == 0) |
| 2210 | __raid56_parity_recover(rbio); |
| 2211 | /* |
| 2212 | * our rbio has been added to the list of |
| 2213 | * rbios that will be handled after the |
| 2214 | * currently lock owner is done |
| 2215 | */ |
| 2216 | return 0; |
| 2217 | |
| 2218 | } |
| 2219 | |
| 2220 | static void rmw_work(struct btrfs_work *work) |
| 2221 | { |
| 2222 | struct btrfs_raid_bio *rbio; |
| 2223 | |
| 2224 | rbio = container_of(work, struct btrfs_raid_bio, work); |
| 2225 | raid56_rmw_stripe(rbio); |
| 2226 | } |
| 2227 | |
| 2228 | static void read_rebuild_work(struct btrfs_work *work) |
| 2229 | { |
| 2230 | struct btrfs_raid_bio *rbio; |
| 2231 | |
| 2232 | rbio = container_of(work, struct btrfs_raid_bio, work); |
| 2233 | __raid56_parity_recover(rbio); |
| 2234 | } |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2235 | |
| 2236 | /* |
| 2237 | * The following code is used to scrub/replace the parity stripe |
| 2238 | * |
Qu Wenruo | ae6529c | 2017-03-29 09:33:21 +0800 | [diff] [blame] | 2239 | * Caller must have already increased bio_counter for getting @bbio. |
| 2240 | * |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2241 | * Note: We need make sure all the pages that add into the scrub/replace |
| 2242 | * raid bio are correct and not be changed during the scrub/replace. That |
| 2243 | * is those pages just hold metadata or file data with checksum. |
| 2244 | */ |
| 2245 | |
| 2246 | struct btrfs_raid_bio * |
Jeff Mahoney | 2ff7e61 | 2016-06-22 18:54:24 -0400 | [diff] [blame] | 2247 | raid56_parity_alloc_scrub_rbio(struct btrfs_fs_info *fs_info, struct bio *bio, |
Zhao Lei | 8e5cfb5 | 2015-01-20 15:11:33 +0800 | [diff] [blame] | 2248 | struct btrfs_bio *bbio, u64 stripe_len, |
| 2249 | struct btrfs_device *scrub_dev, |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2250 | unsigned long *dbitmap, int stripe_nsectors) |
| 2251 | { |
| 2252 | struct btrfs_raid_bio *rbio; |
| 2253 | int i; |
| 2254 | |
Jeff Mahoney | 2ff7e61 | 2016-06-22 18:54:24 -0400 | [diff] [blame] | 2255 | rbio = alloc_rbio(fs_info, bbio, stripe_len); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2256 | if (IS_ERR(rbio)) |
| 2257 | return NULL; |
| 2258 | bio_list_add(&rbio->bio_list, bio); |
| 2259 | /* |
| 2260 | * This is a special bio which is used to hold the completion handler |
| 2261 | * and make the scrub rbio is similar to the other types |
| 2262 | */ |
| 2263 | ASSERT(!bio->bi_iter.bi_size); |
| 2264 | rbio->operation = BTRFS_RBIO_PARITY_SCRUB; |
| 2265 | |
Liu Bo | 9cd3a7e | 2017-08-03 13:53:31 -0600 | [diff] [blame] | 2266 | /* |
| 2267 | * After mapping bbio with BTRFS_MAP_WRITE, parities have been sorted |
| 2268 | * to the end position, so this search can start from the first parity |
| 2269 | * stripe. |
| 2270 | */ |
| 2271 | for (i = rbio->nr_data; i < rbio->real_stripes; i++) { |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2272 | if (bbio->stripes[i].dev == scrub_dev) { |
| 2273 | rbio->scrubp = i; |
| 2274 | break; |
| 2275 | } |
| 2276 | } |
Liu Bo | 9cd3a7e | 2017-08-03 13:53:31 -0600 | [diff] [blame] | 2277 | ASSERT(i < rbio->real_stripes); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2278 | |
| 2279 | /* Now we just support the sectorsize equals to page size */ |
Jeff Mahoney | 0b246af | 2016-06-22 18:54:23 -0400 | [diff] [blame] | 2280 | ASSERT(fs_info->sectorsize == PAGE_SIZE); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2281 | ASSERT(rbio->stripe_npages == stripe_nsectors); |
| 2282 | bitmap_copy(rbio->dbitmap, dbitmap, stripe_nsectors); |
| 2283 | |
Qu Wenruo | ae6529c | 2017-03-29 09:33:21 +0800 | [diff] [blame] | 2284 | /* |
| 2285 | * We have already increased bio_counter when getting bbio, record it |
| 2286 | * so we can free it at rbio_orig_end_io(). |
| 2287 | */ |
| 2288 | rbio->generic_bio_cnt = 1; |
| 2289 | |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2290 | return rbio; |
| 2291 | } |
| 2292 | |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 2293 | /* Used for both parity scrub and missing. */ |
| 2294 | void raid56_add_scrub_pages(struct btrfs_raid_bio *rbio, struct page *page, |
| 2295 | u64 logical) |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2296 | { |
| 2297 | int stripe_offset; |
| 2298 | int index; |
| 2299 | |
Zhao Lei | 8e5cfb5 | 2015-01-20 15:11:33 +0800 | [diff] [blame] | 2300 | ASSERT(logical >= rbio->bbio->raid_map[0]); |
| 2301 | ASSERT(logical + PAGE_SIZE <= rbio->bbio->raid_map[0] + |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2302 | rbio->stripe_len * rbio->nr_data); |
Zhao Lei | 8e5cfb5 | 2015-01-20 15:11:33 +0800 | [diff] [blame] | 2303 | stripe_offset = (int)(logical - rbio->bbio->raid_map[0]); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2304 | index = stripe_offset >> PAGE_SHIFT; |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2305 | rbio->bio_pages[index] = page; |
| 2306 | } |
| 2307 | |
| 2308 | /* |
| 2309 | * We just scrub the parity that we have correct data on the same horizontal, |
| 2310 | * so we needn't allocate all pages for all the stripes. |
| 2311 | */ |
| 2312 | static int alloc_rbio_essential_pages(struct btrfs_raid_bio *rbio) |
| 2313 | { |
| 2314 | int i; |
| 2315 | int bit; |
| 2316 | int index; |
| 2317 | struct page *page; |
| 2318 | |
| 2319 | for_each_set_bit(bit, rbio->dbitmap, rbio->stripe_npages) { |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 2320 | for (i = 0; i < rbio->real_stripes; i++) { |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2321 | index = i * rbio->stripe_npages + bit; |
| 2322 | if (rbio->stripe_pages[index]) |
| 2323 | continue; |
| 2324 | |
| 2325 | page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); |
| 2326 | if (!page) |
| 2327 | return -ENOMEM; |
| 2328 | rbio->stripe_pages[index] = page; |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2329 | } |
| 2330 | } |
| 2331 | return 0; |
| 2332 | } |
| 2333 | |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2334 | static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, |
| 2335 | int need_check) |
| 2336 | { |
Miao Xie | 7603597 | 2014-11-14 17:45:42 +0800 | [diff] [blame] | 2337 | struct btrfs_bio *bbio = rbio->bbio; |
Kees Cook | 1389053 | 2018-05-29 16:44:59 -0700 | [diff] [blame] | 2338 | void **pointers = rbio->finish_pointers; |
| 2339 | unsigned long *pbitmap = rbio->finish_pbitmap; |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2340 | int nr_data = rbio->nr_data; |
| 2341 | int stripe; |
| 2342 | int pagenr; |
| 2343 | int p_stripe = -1; |
| 2344 | int q_stripe = -1; |
| 2345 | struct page *p_page = NULL; |
| 2346 | struct page *q_page = NULL; |
| 2347 | struct bio_list bio_list; |
| 2348 | struct bio *bio; |
Miao Xie | 7603597 | 2014-11-14 17:45:42 +0800 | [diff] [blame] | 2349 | int is_replace = 0; |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2350 | int ret; |
| 2351 | |
| 2352 | bio_list_init(&bio_list); |
| 2353 | |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 2354 | if (rbio->real_stripes - rbio->nr_data == 1) { |
| 2355 | p_stripe = rbio->real_stripes - 1; |
| 2356 | } else if (rbio->real_stripes - rbio->nr_data == 2) { |
| 2357 | p_stripe = rbio->real_stripes - 2; |
| 2358 | q_stripe = rbio->real_stripes - 1; |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2359 | } else { |
| 2360 | BUG(); |
| 2361 | } |
| 2362 | |
Miao Xie | 7603597 | 2014-11-14 17:45:42 +0800 | [diff] [blame] | 2363 | if (bbio->num_tgtdevs && bbio->tgtdev_map[rbio->scrubp]) { |
| 2364 | is_replace = 1; |
| 2365 | bitmap_copy(pbitmap, rbio->dbitmap, rbio->stripe_npages); |
| 2366 | } |
| 2367 | |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2368 | /* |
| 2369 | * Because the higher layers(scrubber) are unlikely to |
| 2370 | * use this area of the disk again soon, so don't cache |
| 2371 | * it. |
| 2372 | */ |
| 2373 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); |
| 2374 | |
| 2375 | if (!need_check) |
| 2376 | goto writeback; |
| 2377 | |
| 2378 | p_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); |
| 2379 | if (!p_page) |
| 2380 | goto cleanup; |
| 2381 | SetPageUptodate(p_page); |
| 2382 | |
| 2383 | if (q_stripe != -1) { |
| 2384 | q_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); |
| 2385 | if (!q_page) { |
| 2386 | __free_page(p_page); |
| 2387 | goto cleanup; |
| 2388 | } |
| 2389 | SetPageUptodate(q_page); |
| 2390 | } |
| 2391 | |
| 2392 | atomic_set(&rbio->error, 0); |
| 2393 | |
| 2394 | for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { |
| 2395 | struct page *p; |
| 2396 | void *parity; |
| 2397 | /* first collect one page from each data stripe */ |
| 2398 | for (stripe = 0; stripe < nr_data; stripe++) { |
| 2399 | p = page_in_rbio(rbio, stripe, pagenr, 0); |
| 2400 | pointers[stripe] = kmap(p); |
| 2401 | } |
| 2402 | |
| 2403 | /* then add the parity stripe */ |
| 2404 | pointers[stripe++] = kmap(p_page); |
| 2405 | |
| 2406 | if (q_stripe != -1) { |
| 2407 | |
| 2408 | /* |
| 2409 | * raid6, add the qstripe and call the |
| 2410 | * library function to fill in our p/q |
| 2411 | */ |
| 2412 | pointers[stripe++] = kmap(q_page); |
| 2413 | |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 2414 | raid6_call.gen_syndrome(rbio->real_stripes, PAGE_SIZE, |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2415 | pointers); |
| 2416 | } else { |
| 2417 | /* raid5 */ |
David Sterba | 69d2480 | 2018-06-29 10:56:44 +0200 | [diff] [blame] | 2418 | copy_page(pointers[nr_data], pointers[0]); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2419 | run_xor(pointers + 1, nr_data - 1, PAGE_SIZE); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2420 | } |
| 2421 | |
Nicholas D Steeves | 0132761 | 2016-05-19 21:18:45 -0400 | [diff] [blame] | 2422 | /* Check scrubbing parity and repair it */ |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2423 | p = rbio_stripe_page(rbio, rbio->scrubp, pagenr); |
| 2424 | parity = kmap(p); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2425 | if (memcmp(parity, pointers[rbio->scrubp], PAGE_SIZE)) |
David Sterba | 69d2480 | 2018-06-29 10:56:44 +0200 | [diff] [blame] | 2426 | copy_page(parity, pointers[rbio->scrubp]); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2427 | else |
| 2428 | /* Parity is right, needn't writeback */ |
| 2429 | bitmap_clear(rbio->dbitmap, pagenr, 1); |
| 2430 | kunmap(p); |
| 2431 | |
Andrea Righi | 1cf4ab0 | 2019-03-14 08:56:28 +0100 | [diff] [blame] | 2432 | for (stripe = 0; stripe < nr_data; stripe++) |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2433 | kunmap(page_in_rbio(rbio, stripe, pagenr, 0)); |
Andrea Righi | 1cf4ab0 | 2019-03-14 08:56:28 +0100 | [diff] [blame] | 2434 | kunmap(p_page); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2435 | } |
| 2436 | |
| 2437 | __free_page(p_page); |
| 2438 | if (q_page) |
| 2439 | __free_page(q_page); |
| 2440 | |
| 2441 | writeback: |
| 2442 | /* |
| 2443 | * time to start writing. Make bios for everything from the |
| 2444 | * higher layers (the bio_list in our rbio) and our p/q. Ignore |
| 2445 | * everything else. |
| 2446 | */ |
| 2447 | for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { |
| 2448 | struct page *page; |
| 2449 | |
| 2450 | page = rbio_stripe_page(rbio, rbio->scrubp, pagenr); |
| 2451 | ret = rbio_add_io_page(rbio, &bio_list, |
| 2452 | page, rbio->scrubp, pagenr, rbio->stripe_len); |
| 2453 | if (ret) |
| 2454 | goto cleanup; |
| 2455 | } |
| 2456 | |
Miao Xie | 7603597 | 2014-11-14 17:45:42 +0800 | [diff] [blame] | 2457 | if (!is_replace) |
| 2458 | goto submit_write; |
| 2459 | |
| 2460 | for_each_set_bit(pagenr, pbitmap, rbio->stripe_npages) { |
| 2461 | struct page *page; |
| 2462 | |
| 2463 | page = rbio_stripe_page(rbio, rbio->scrubp, pagenr); |
| 2464 | ret = rbio_add_io_page(rbio, &bio_list, page, |
| 2465 | bbio->tgtdev_map[rbio->scrubp], |
| 2466 | pagenr, rbio->stripe_len); |
| 2467 | if (ret) |
| 2468 | goto cleanup; |
| 2469 | } |
| 2470 | |
| 2471 | submit_write: |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2472 | nr_data = bio_list_size(&bio_list); |
| 2473 | if (!nr_data) { |
| 2474 | /* Every parity is right */ |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 2475 | rbio_orig_end_io(rbio, BLK_STS_OK); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2476 | return; |
| 2477 | } |
| 2478 | |
| 2479 | atomic_set(&rbio->stripes_pending, nr_data); |
| 2480 | |
| 2481 | while (1) { |
| 2482 | bio = bio_list_pop(&bio_list); |
| 2483 | if (!bio) |
| 2484 | break; |
| 2485 | |
| 2486 | bio->bi_private = rbio; |
Zhao Lei | a6111d1 | 2016-01-12 17:52:13 +0800 | [diff] [blame] | 2487 | bio->bi_end_io = raid_write_end_io; |
David Sterba | ebcc326 | 2018-06-29 10:56:53 +0200 | [diff] [blame] | 2488 | bio->bi_opf = REQ_OP_WRITE; |
Mike Christie | 4e49ea4 | 2016-06-05 14:31:41 -0500 | [diff] [blame] | 2489 | |
| 2490 | submit_bio(bio); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2491 | } |
| 2492 | return; |
| 2493 | |
| 2494 | cleanup: |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 2495 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
Liu Bo | 785884f | 2017-09-22 12:11:18 -0600 | [diff] [blame] | 2496 | |
| 2497 | while ((bio = bio_list_pop(&bio_list))) |
| 2498 | bio_put(bio); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2499 | } |
| 2500 | |
| 2501 | static inline int is_data_stripe(struct btrfs_raid_bio *rbio, int stripe) |
| 2502 | { |
| 2503 | if (stripe >= 0 && stripe < rbio->nr_data) |
| 2504 | return 1; |
| 2505 | return 0; |
| 2506 | } |
| 2507 | |
| 2508 | /* |
| 2509 | * While we're doing the parity check and repair, we could have errors |
| 2510 | * in reading pages off the disk. This checks for errors and if we're |
| 2511 | * not able to read the page it'll trigger parity reconstruction. The |
| 2512 | * parity scrub will be finished after we've reconstructed the failed |
| 2513 | * stripes |
| 2514 | */ |
| 2515 | static void validate_rbio_for_parity_scrub(struct btrfs_raid_bio *rbio) |
| 2516 | { |
| 2517 | if (atomic_read(&rbio->error) > rbio->bbio->max_errors) |
| 2518 | goto cleanup; |
| 2519 | |
| 2520 | if (rbio->faila >= 0 || rbio->failb >= 0) { |
| 2521 | int dfail = 0, failp = -1; |
| 2522 | |
| 2523 | if (is_data_stripe(rbio, rbio->faila)) |
| 2524 | dfail++; |
| 2525 | else if (is_parity_stripe(rbio->faila)) |
| 2526 | failp = rbio->faila; |
| 2527 | |
| 2528 | if (is_data_stripe(rbio, rbio->failb)) |
| 2529 | dfail++; |
| 2530 | else if (is_parity_stripe(rbio->failb)) |
| 2531 | failp = rbio->failb; |
| 2532 | |
| 2533 | /* |
| 2534 | * Because we can not use a scrubbing parity to repair |
| 2535 | * the data, so the capability of the repair is declined. |
| 2536 | * (In the case of RAID5, we can not repair anything) |
| 2537 | */ |
| 2538 | if (dfail > rbio->bbio->max_errors - 1) |
| 2539 | goto cleanup; |
| 2540 | |
| 2541 | /* |
| 2542 | * If all data is good, only parity is correctly, just |
| 2543 | * repair the parity. |
| 2544 | */ |
| 2545 | if (dfail == 0) { |
| 2546 | finish_parity_scrub(rbio, 0); |
| 2547 | return; |
| 2548 | } |
| 2549 | |
| 2550 | /* |
| 2551 | * Here means we got one corrupted data stripe and one |
| 2552 | * corrupted parity on RAID6, if the corrupted parity |
Nicholas D Steeves | 0132761 | 2016-05-19 21:18:45 -0400 | [diff] [blame] | 2553 | * is scrubbing parity, luckily, use the other one to repair |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2554 | * the data, or we can not repair the data stripe. |
| 2555 | */ |
| 2556 | if (failp != rbio->scrubp) |
| 2557 | goto cleanup; |
| 2558 | |
| 2559 | __raid_recover_end_io(rbio); |
| 2560 | } else { |
| 2561 | finish_parity_scrub(rbio, 1); |
| 2562 | } |
| 2563 | return; |
| 2564 | |
| 2565 | cleanup: |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 2566 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2567 | } |
| 2568 | |
| 2569 | /* |
| 2570 | * end io for the read phase of the rmw cycle. All the bios here are physical |
| 2571 | * stripe bios we've read from the disk so we can recalculate the parity of the |
| 2572 | * stripe. |
| 2573 | * |
| 2574 | * This will usually kick off finish_rmw once all the bios are read in, but it |
| 2575 | * may trigger parity reconstruction if we had any errors along the way |
| 2576 | */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 2577 | static void raid56_parity_scrub_end_io(struct bio *bio) |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2578 | { |
| 2579 | struct btrfs_raid_bio *rbio = bio->bi_private; |
| 2580 | |
Christoph Hellwig | 4e4cbee | 2017-06-03 09:38:06 +0200 | [diff] [blame] | 2581 | if (bio->bi_status) |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2582 | fail_bio_stripe(rbio, bio); |
| 2583 | else |
| 2584 | set_bio_pages_uptodate(bio); |
| 2585 | |
| 2586 | bio_put(bio); |
| 2587 | |
| 2588 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
| 2589 | return; |
| 2590 | |
| 2591 | /* |
| 2592 | * this will normally call finish_rmw to start our write |
| 2593 | * but if there are any failed stripes we'll reconstruct |
| 2594 | * from parity first |
| 2595 | */ |
| 2596 | validate_rbio_for_parity_scrub(rbio); |
| 2597 | } |
| 2598 | |
| 2599 | static void raid56_parity_scrub_stripe(struct btrfs_raid_bio *rbio) |
| 2600 | { |
| 2601 | int bios_to_read = 0; |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2602 | struct bio_list bio_list; |
| 2603 | int ret; |
| 2604 | int pagenr; |
| 2605 | int stripe; |
| 2606 | struct bio *bio; |
| 2607 | |
Liu Bo | 785884f | 2017-09-22 12:11:18 -0600 | [diff] [blame] | 2608 | bio_list_init(&bio_list); |
| 2609 | |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2610 | ret = alloc_rbio_essential_pages(rbio); |
| 2611 | if (ret) |
| 2612 | goto cleanup; |
| 2613 | |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2614 | atomic_set(&rbio->error, 0); |
| 2615 | /* |
| 2616 | * build a list of bios to read all the missing parts of this |
| 2617 | * stripe |
| 2618 | */ |
Miao Xie | 2c8cdd6 | 2014-11-14 16:06:25 +0800 | [diff] [blame] | 2619 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2620 | for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { |
| 2621 | struct page *page; |
| 2622 | /* |
| 2623 | * we want to find all the pages missing from |
| 2624 | * the rbio and read them from the disk. If |
| 2625 | * page_in_rbio finds a page in the bio list |
| 2626 | * we don't need to read it off the stripe. |
| 2627 | */ |
| 2628 | page = page_in_rbio(rbio, stripe, pagenr, 1); |
| 2629 | if (page) |
| 2630 | continue; |
| 2631 | |
| 2632 | page = rbio_stripe_page(rbio, stripe, pagenr); |
| 2633 | /* |
| 2634 | * the bio cache may have handed us an uptodate |
| 2635 | * page. If so, be happy and use it |
| 2636 | */ |
| 2637 | if (PageUptodate(page)) |
| 2638 | continue; |
| 2639 | |
| 2640 | ret = rbio_add_io_page(rbio, &bio_list, page, |
| 2641 | stripe, pagenr, rbio->stripe_len); |
| 2642 | if (ret) |
| 2643 | goto cleanup; |
| 2644 | } |
| 2645 | } |
| 2646 | |
| 2647 | bios_to_read = bio_list_size(&bio_list); |
| 2648 | if (!bios_to_read) { |
| 2649 | /* |
| 2650 | * this can happen if others have merged with |
| 2651 | * us, it means there is nothing left to read. |
| 2652 | * But if there are missing devices it may not be |
| 2653 | * safe to do the full stripe write yet. |
| 2654 | */ |
| 2655 | goto finish; |
| 2656 | } |
| 2657 | |
| 2658 | /* |
| 2659 | * the bbio may be freed once we submit the last bio. Make sure |
| 2660 | * not to touch it after that |
| 2661 | */ |
| 2662 | atomic_set(&rbio->stripes_pending, bios_to_read); |
| 2663 | while (1) { |
| 2664 | bio = bio_list_pop(&bio_list); |
| 2665 | if (!bio) |
| 2666 | break; |
| 2667 | |
| 2668 | bio->bi_private = rbio; |
| 2669 | bio->bi_end_io = raid56_parity_scrub_end_io; |
David Sterba | ebcc326 | 2018-06-29 10:56:53 +0200 | [diff] [blame] | 2670 | bio->bi_opf = REQ_OP_READ; |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2671 | |
Jeff Mahoney | 0b246af | 2016-06-22 18:54:23 -0400 | [diff] [blame] | 2672 | btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2673 | |
Mike Christie | 4e49ea4 | 2016-06-05 14:31:41 -0500 | [diff] [blame] | 2674 | submit_bio(bio); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2675 | } |
| 2676 | /* the actual write will happen once the reads are done */ |
| 2677 | return; |
| 2678 | |
| 2679 | cleanup: |
Omar Sandoval | 58efbc9 | 2017-08-22 23:45:59 -0700 | [diff] [blame] | 2680 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
Liu Bo | 785884f | 2017-09-22 12:11:18 -0600 | [diff] [blame] | 2681 | |
| 2682 | while ((bio = bio_list_pop(&bio_list))) |
| 2683 | bio_put(bio); |
| 2684 | |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2685 | return; |
| 2686 | |
| 2687 | finish: |
| 2688 | validate_rbio_for_parity_scrub(rbio); |
| 2689 | } |
| 2690 | |
| 2691 | static void scrub_parity_work(struct btrfs_work *work) |
| 2692 | { |
| 2693 | struct btrfs_raid_bio *rbio; |
| 2694 | |
| 2695 | rbio = container_of(work, struct btrfs_raid_bio, work); |
| 2696 | raid56_parity_scrub_stripe(rbio); |
| 2697 | } |
| 2698 | |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2699 | void raid56_parity_submit_scrub_rbio(struct btrfs_raid_bio *rbio) |
| 2700 | { |
| 2701 | if (!lock_stripe_add(rbio)) |
David Sterba | a81b747 | 2018-06-29 10:57:03 +0200 | [diff] [blame] | 2702 | start_async_work(rbio, scrub_parity_work); |
Miao Xie | 5a6ac9e | 2014-11-06 17:20:58 +0800 | [diff] [blame] | 2703 | } |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 2704 | |
| 2705 | /* The following code is used for dev replace of a missing RAID 5/6 device. */ |
| 2706 | |
| 2707 | struct btrfs_raid_bio * |
Jeff Mahoney | 2ff7e61 | 2016-06-22 18:54:24 -0400 | [diff] [blame] | 2708 | raid56_alloc_missing_rbio(struct btrfs_fs_info *fs_info, struct bio *bio, |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 2709 | struct btrfs_bio *bbio, u64 length) |
| 2710 | { |
| 2711 | struct btrfs_raid_bio *rbio; |
| 2712 | |
Jeff Mahoney | 2ff7e61 | 2016-06-22 18:54:24 -0400 | [diff] [blame] | 2713 | rbio = alloc_rbio(fs_info, bbio, length); |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 2714 | if (IS_ERR(rbio)) |
| 2715 | return NULL; |
| 2716 | |
| 2717 | rbio->operation = BTRFS_RBIO_REBUILD_MISSING; |
| 2718 | bio_list_add(&rbio->bio_list, bio); |
| 2719 | /* |
| 2720 | * This is a special bio which is used to hold the completion handler |
| 2721 | * and make the scrub rbio is similar to the other types |
| 2722 | */ |
| 2723 | ASSERT(!bio->bi_iter.bi_size); |
| 2724 | |
| 2725 | rbio->faila = find_logical_bio_stripe(rbio, bio); |
| 2726 | if (rbio->faila == -1) { |
| 2727 | BUG(); |
| 2728 | kfree(rbio); |
| 2729 | return NULL; |
| 2730 | } |
| 2731 | |
Qu Wenruo | ae6529c | 2017-03-29 09:33:21 +0800 | [diff] [blame] | 2732 | /* |
| 2733 | * When we get bbio, we have already increased bio_counter, record it |
| 2734 | * so we can free it at rbio_orig_end_io() |
| 2735 | */ |
| 2736 | rbio->generic_bio_cnt = 1; |
| 2737 | |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 2738 | return rbio; |
| 2739 | } |
| 2740 | |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 2741 | void raid56_submit_missing_rbio(struct btrfs_raid_bio *rbio) |
| 2742 | { |
| 2743 | if (!lock_stripe_add(rbio)) |
David Sterba | e66d8d5 | 2018-06-29 10:57:00 +0200 | [diff] [blame] | 2744 | start_async_work(rbio, read_rebuild_work); |
Omar Sandoval | b4ee178 | 2015-06-19 11:52:50 -0700 | [diff] [blame] | 2745 | } |