Josef Bacik | 0e9cebe | 2015-03-20 10:50:37 -0400 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2014 Facebook. All rights reserved. |
| 3 | * |
| 4 | * This file is released under the GPL. |
| 5 | */ |
| 6 | |
| 7 | #include <linux/device-mapper.h> |
| 8 | |
| 9 | #include <linux/module.h> |
| 10 | #include <linux/init.h> |
| 11 | #include <linux/blkdev.h> |
| 12 | #include <linux/bio.h> |
| 13 | #include <linux/slab.h> |
| 14 | #include <linux/kthread.h> |
| 15 | #include <linux/freezer.h> |
| 16 | |
| 17 | #define DM_MSG_PREFIX "log-writes" |
| 18 | |
| 19 | /* |
| 20 | * This target will sequentially log all writes to the target device onto the |
| 21 | * log device. This is helpful for replaying writes to check for fs consistency |
| 22 | * at all times. This target provides a mechanism to mark specific events to |
| 23 | * check data at a later time. So for example you would: |
| 24 | * |
| 25 | * write data |
| 26 | * fsync |
| 27 | * dmsetup message /dev/whatever mark mymark |
| 28 | * unmount /mnt/test |
| 29 | * |
| 30 | * Then replay the log up to mymark and check the contents of the replay to |
| 31 | * verify it matches what was written. |
| 32 | * |
| 33 | * We log writes only after they have been flushed, this makes the log describe |
| 34 | * close to the order in which the data hits the actual disk, not its cache. So |
| 35 | * for example the following sequence (W means write, C means complete) |
| 36 | * |
| 37 | * Wa,Wb,Wc,Cc,Ca,FLUSH,FUAd,Cb,CFLUSH,CFUAd |
| 38 | * |
| 39 | * Would result in the log looking like this: |
| 40 | * |
| 41 | * c,a,flush,fuad,b,<other writes>,<next flush> |
| 42 | * |
| 43 | * This is meant to help expose problems where file systems do not properly wait |
| 44 | * on data being written before invoking a FLUSH. FUA bypasses cache so once it |
| 45 | * completes it is added to the log as it should be on disk. |
| 46 | * |
| 47 | * We treat DISCARDs as if they don't bypass cache so that they are logged in |
| 48 | * order of completion along with the normal writes. If we didn't do it this |
| 49 | * way we would process all the discards first and then write all the data, when |
| 50 | * in fact we want to do the data and the discard in the order that they |
| 51 | * completed. |
| 52 | */ |
| 53 | #define LOG_FLUSH_FLAG (1 << 0) |
| 54 | #define LOG_FUA_FLAG (1 << 1) |
| 55 | #define LOG_DISCARD_FLAG (1 << 2) |
| 56 | #define LOG_MARK_FLAG (1 << 3) |
| 57 | |
Geert Uytterhoeven | f4ad317 | 2015-04-19 00:07:30 +0200 | [diff] [blame] | 58 | #define WRITE_LOG_VERSION 1ULL |
| 59 | #define WRITE_LOG_MAGIC 0x6a736677736872ULL |
Josef Bacik | 0e9cebe | 2015-03-20 10:50:37 -0400 | [diff] [blame] | 60 | |
| 61 | /* |
| 62 | * The disk format for this is braindead simple. |
| 63 | * |
| 64 | * At byte 0 we have our super, followed by the following sequence for |
| 65 | * nr_entries: |
| 66 | * |
| 67 | * [ 1 sector ][ entry->nr_sectors ] |
| 68 | * [log_write_entry][ data written ] |
| 69 | * |
| 70 | * The log_write_entry takes up a full sector so we can have arbitrary length |
| 71 | * marks and it leaves us room for extra content in the future. |
| 72 | */ |
| 73 | |
| 74 | /* |
| 75 | * Basic info about the log for userspace. |
| 76 | */ |
| 77 | struct log_write_super { |
| 78 | __le64 magic; |
| 79 | __le64 version; |
| 80 | __le64 nr_entries; |
| 81 | __le32 sectorsize; |
| 82 | }; |
| 83 | |
| 84 | /* |
| 85 | * sector - the sector we wrote. |
| 86 | * nr_sectors - the number of sectors we wrote. |
| 87 | * flags - flags for this log entry. |
| 88 | * data_len - the size of the data in this log entry, this is for private log |
| 89 | * entry stuff, the MARK data provided by userspace for example. |
| 90 | */ |
| 91 | struct log_write_entry { |
| 92 | __le64 sector; |
| 93 | __le64 nr_sectors; |
| 94 | __le64 flags; |
| 95 | __le64 data_len; |
| 96 | }; |
| 97 | |
| 98 | struct log_writes_c { |
| 99 | struct dm_dev *dev; |
| 100 | struct dm_dev *logdev; |
| 101 | u64 logged_entries; |
| 102 | u32 sectorsize; |
| 103 | atomic_t io_blocks; |
| 104 | atomic_t pending_blocks; |
| 105 | sector_t next_sector; |
| 106 | sector_t end_sector; |
| 107 | bool logging_enabled; |
| 108 | bool device_supports_discard; |
| 109 | spinlock_t blocks_lock; |
| 110 | struct list_head unflushed_blocks; |
| 111 | struct list_head logging_blocks; |
| 112 | wait_queue_head_t wait; |
| 113 | struct task_struct *log_kthread; |
| 114 | }; |
| 115 | |
| 116 | struct pending_block { |
| 117 | int vec_cnt; |
| 118 | u64 flags; |
| 119 | sector_t sector; |
| 120 | sector_t nr_sectors; |
| 121 | char *data; |
| 122 | u32 datalen; |
| 123 | struct list_head list; |
| 124 | struct bio_vec vecs[0]; |
| 125 | }; |
| 126 | |
| 127 | struct per_bio_data { |
| 128 | struct pending_block *block; |
| 129 | }; |
| 130 | |
| 131 | static void put_pending_block(struct log_writes_c *lc) |
| 132 | { |
| 133 | if (atomic_dec_and_test(&lc->pending_blocks)) { |
| 134 | smp_mb__after_atomic(); |
| 135 | if (waitqueue_active(&lc->wait)) |
| 136 | wake_up(&lc->wait); |
| 137 | } |
| 138 | } |
| 139 | |
| 140 | static void put_io_block(struct log_writes_c *lc) |
| 141 | { |
| 142 | if (atomic_dec_and_test(&lc->io_blocks)) { |
| 143 | smp_mb__after_atomic(); |
| 144 | if (waitqueue_active(&lc->wait)) |
| 145 | wake_up(&lc->wait); |
| 146 | } |
| 147 | } |
| 148 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame^] | 149 | static void log_end_io(struct bio *bio) |
Josef Bacik | 0e9cebe | 2015-03-20 10:50:37 -0400 | [diff] [blame] | 150 | { |
| 151 | struct log_writes_c *lc = bio->bi_private; |
| 152 | struct bio_vec *bvec; |
| 153 | int i; |
| 154 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame^] | 155 | if (bio->bi_error) { |
Josef Bacik | 0e9cebe | 2015-03-20 10:50:37 -0400 | [diff] [blame] | 156 | unsigned long flags; |
| 157 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame^] | 158 | DMERR("Error writing log block, error=%d", bio->bi_error); |
Josef Bacik | 0e9cebe | 2015-03-20 10:50:37 -0400 | [diff] [blame] | 159 | spin_lock_irqsave(&lc->blocks_lock, flags); |
| 160 | lc->logging_enabled = false; |
| 161 | spin_unlock_irqrestore(&lc->blocks_lock, flags); |
| 162 | } |
| 163 | |
| 164 | bio_for_each_segment_all(bvec, bio, i) |
| 165 | __free_page(bvec->bv_page); |
| 166 | |
| 167 | put_io_block(lc); |
| 168 | bio_put(bio); |
| 169 | } |
| 170 | |
| 171 | /* |
| 172 | * Meant to be called if there is an error, it will free all the pages |
| 173 | * associated with the block. |
| 174 | */ |
| 175 | static void free_pending_block(struct log_writes_c *lc, |
| 176 | struct pending_block *block) |
| 177 | { |
| 178 | int i; |
| 179 | |
| 180 | for (i = 0; i < block->vec_cnt; i++) { |
| 181 | if (block->vecs[i].bv_page) |
| 182 | __free_page(block->vecs[i].bv_page); |
| 183 | } |
| 184 | kfree(block->data); |
| 185 | kfree(block); |
| 186 | put_pending_block(lc); |
| 187 | } |
| 188 | |
| 189 | static int write_metadata(struct log_writes_c *lc, void *entry, |
| 190 | size_t entrylen, void *data, size_t datalen, |
| 191 | sector_t sector) |
| 192 | { |
| 193 | struct bio *bio; |
| 194 | struct page *page; |
| 195 | void *ptr; |
| 196 | size_t ret; |
| 197 | |
| 198 | bio = bio_alloc(GFP_KERNEL, 1); |
| 199 | if (!bio) { |
| 200 | DMERR("Couldn't alloc log bio"); |
| 201 | goto error; |
| 202 | } |
| 203 | bio->bi_iter.bi_size = 0; |
| 204 | bio->bi_iter.bi_sector = sector; |
| 205 | bio->bi_bdev = lc->logdev->bdev; |
| 206 | bio->bi_end_io = log_end_io; |
| 207 | bio->bi_private = lc; |
Josef Bacik | 0e9cebe | 2015-03-20 10:50:37 -0400 | [diff] [blame] | 208 | |
| 209 | page = alloc_page(GFP_KERNEL); |
| 210 | if (!page) { |
| 211 | DMERR("Couldn't alloc log page"); |
| 212 | bio_put(bio); |
| 213 | goto error; |
| 214 | } |
| 215 | |
| 216 | ptr = kmap_atomic(page); |
| 217 | memcpy(ptr, entry, entrylen); |
| 218 | if (datalen) |
| 219 | memcpy(ptr + entrylen, data, datalen); |
| 220 | memset(ptr + entrylen + datalen, 0, |
| 221 | lc->sectorsize - entrylen - datalen); |
| 222 | kunmap_atomic(ptr); |
| 223 | |
| 224 | ret = bio_add_page(bio, page, lc->sectorsize, 0); |
| 225 | if (ret != lc->sectorsize) { |
| 226 | DMERR("Couldn't add page to the log block"); |
| 227 | goto error_bio; |
| 228 | } |
| 229 | submit_bio(WRITE, bio); |
| 230 | return 0; |
| 231 | error_bio: |
| 232 | bio_put(bio); |
| 233 | __free_page(page); |
| 234 | error: |
| 235 | put_io_block(lc); |
| 236 | return -1; |
| 237 | } |
| 238 | |
| 239 | static int log_one_block(struct log_writes_c *lc, |
| 240 | struct pending_block *block, sector_t sector) |
| 241 | { |
| 242 | struct bio *bio; |
| 243 | struct log_write_entry entry; |
| 244 | size_t ret; |
| 245 | int i; |
| 246 | |
| 247 | entry.sector = cpu_to_le64(block->sector); |
| 248 | entry.nr_sectors = cpu_to_le64(block->nr_sectors); |
| 249 | entry.flags = cpu_to_le64(block->flags); |
| 250 | entry.data_len = cpu_to_le64(block->datalen); |
| 251 | if (write_metadata(lc, &entry, sizeof(entry), block->data, |
| 252 | block->datalen, sector)) { |
| 253 | free_pending_block(lc, block); |
| 254 | return -1; |
| 255 | } |
| 256 | |
| 257 | if (!block->vec_cnt) |
| 258 | goto out; |
| 259 | sector++; |
| 260 | |
| 261 | bio = bio_alloc(GFP_KERNEL, block->vec_cnt); |
| 262 | if (!bio) { |
| 263 | DMERR("Couldn't alloc log bio"); |
| 264 | goto error; |
| 265 | } |
| 266 | atomic_inc(&lc->io_blocks); |
| 267 | bio->bi_iter.bi_size = 0; |
| 268 | bio->bi_iter.bi_sector = sector; |
| 269 | bio->bi_bdev = lc->logdev->bdev; |
| 270 | bio->bi_end_io = log_end_io; |
| 271 | bio->bi_private = lc; |
Josef Bacik | 0e9cebe | 2015-03-20 10:50:37 -0400 | [diff] [blame] | 272 | |
| 273 | for (i = 0; i < block->vec_cnt; i++) { |
| 274 | /* |
| 275 | * The page offset is always 0 because we allocate a new page |
| 276 | * for every bvec in the original bio for simplicity sake. |
| 277 | */ |
| 278 | ret = bio_add_page(bio, block->vecs[i].bv_page, |
| 279 | block->vecs[i].bv_len, 0); |
| 280 | if (ret != block->vecs[i].bv_len) { |
| 281 | atomic_inc(&lc->io_blocks); |
| 282 | submit_bio(WRITE, bio); |
| 283 | bio = bio_alloc(GFP_KERNEL, block->vec_cnt - i); |
| 284 | if (!bio) { |
| 285 | DMERR("Couldn't alloc log bio"); |
| 286 | goto error; |
| 287 | } |
| 288 | bio->bi_iter.bi_size = 0; |
| 289 | bio->bi_iter.bi_sector = sector; |
| 290 | bio->bi_bdev = lc->logdev->bdev; |
| 291 | bio->bi_end_io = log_end_io; |
| 292 | bio->bi_private = lc; |
Josef Bacik | 0e9cebe | 2015-03-20 10:50:37 -0400 | [diff] [blame] | 293 | |
| 294 | ret = bio_add_page(bio, block->vecs[i].bv_page, |
| 295 | block->vecs[i].bv_len, 0); |
| 296 | if (ret != block->vecs[i].bv_len) { |
| 297 | DMERR("Couldn't add page on new bio?"); |
| 298 | bio_put(bio); |
| 299 | goto error; |
| 300 | } |
| 301 | } |
| 302 | sector += block->vecs[i].bv_len >> SECTOR_SHIFT; |
| 303 | } |
| 304 | submit_bio(WRITE, bio); |
| 305 | out: |
| 306 | kfree(block->data); |
| 307 | kfree(block); |
| 308 | put_pending_block(lc); |
| 309 | return 0; |
| 310 | error: |
| 311 | free_pending_block(lc, block); |
| 312 | put_io_block(lc); |
| 313 | return -1; |
| 314 | } |
| 315 | |
| 316 | static int log_super(struct log_writes_c *lc) |
| 317 | { |
| 318 | struct log_write_super super; |
| 319 | |
| 320 | super.magic = cpu_to_le64(WRITE_LOG_MAGIC); |
| 321 | super.version = cpu_to_le64(WRITE_LOG_VERSION); |
| 322 | super.nr_entries = cpu_to_le64(lc->logged_entries); |
| 323 | super.sectorsize = cpu_to_le32(lc->sectorsize); |
| 324 | |
| 325 | if (write_metadata(lc, &super, sizeof(super), NULL, 0, 0)) { |
| 326 | DMERR("Couldn't write super"); |
| 327 | return -1; |
| 328 | } |
| 329 | |
| 330 | return 0; |
| 331 | } |
| 332 | |
| 333 | static inline sector_t logdev_last_sector(struct log_writes_c *lc) |
| 334 | { |
| 335 | return i_size_read(lc->logdev->bdev->bd_inode) >> SECTOR_SHIFT; |
| 336 | } |
| 337 | |
| 338 | static int log_writes_kthread(void *arg) |
| 339 | { |
| 340 | struct log_writes_c *lc = (struct log_writes_c *)arg; |
| 341 | sector_t sector = 0; |
| 342 | |
| 343 | while (!kthread_should_stop()) { |
| 344 | bool super = false; |
| 345 | bool logging_enabled; |
| 346 | struct pending_block *block = NULL; |
| 347 | int ret; |
| 348 | |
| 349 | spin_lock_irq(&lc->blocks_lock); |
| 350 | if (!list_empty(&lc->logging_blocks)) { |
| 351 | block = list_first_entry(&lc->logging_blocks, |
| 352 | struct pending_block, list); |
| 353 | list_del_init(&block->list); |
| 354 | if (!lc->logging_enabled) |
| 355 | goto next; |
| 356 | |
| 357 | sector = lc->next_sector; |
| 358 | if (block->flags & LOG_DISCARD_FLAG) |
| 359 | lc->next_sector++; |
| 360 | else |
| 361 | lc->next_sector += block->nr_sectors + 1; |
| 362 | |
| 363 | /* |
| 364 | * Apparently the size of the device may not be known |
| 365 | * right away, so handle this properly. |
| 366 | */ |
| 367 | if (!lc->end_sector) |
| 368 | lc->end_sector = logdev_last_sector(lc); |
| 369 | if (lc->end_sector && |
| 370 | lc->next_sector >= lc->end_sector) { |
| 371 | DMERR("Ran out of space on the logdev"); |
| 372 | lc->logging_enabled = false; |
| 373 | goto next; |
| 374 | } |
| 375 | lc->logged_entries++; |
| 376 | atomic_inc(&lc->io_blocks); |
| 377 | |
| 378 | super = (block->flags & (LOG_FUA_FLAG | LOG_MARK_FLAG)); |
| 379 | if (super) |
| 380 | atomic_inc(&lc->io_blocks); |
| 381 | } |
| 382 | next: |
| 383 | logging_enabled = lc->logging_enabled; |
| 384 | spin_unlock_irq(&lc->blocks_lock); |
| 385 | if (block) { |
| 386 | if (logging_enabled) { |
| 387 | ret = log_one_block(lc, block, sector); |
| 388 | if (!ret && super) |
| 389 | ret = log_super(lc); |
| 390 | if (ret) { |
| 391 | spin_lock_irq(&lc->blocks_lock); |
| 392 | lc->logging_enabled = false; |
| 393 | spin_unlock_irq(&lc->blocks_lock); |
| 394 | } |
| 395 | } else |
| 396 | free_pending_block(lc, block); |
| 397 | continue; |
| 398 | } |
| 399 | |
| 400 | if (!try_to_freeze()) { |
| 401 | set_current_state(TASK_INTERRUPTIBLE); |
| 402 | if (!kthread_should_stop() && |
| 403 | !atomic_read(&lc->pending_blocks)) |
| 404 | schedule(); |
| 405 | __set_current_state(TASK_RUNNING); |
| 406 | } |
| 407 | } |
| 408 | return 0; |
| 409 | } |
| 410 | |
| 411 | /* |
| 412 | * Construct a log-writes mapping: |
| 413 | * log-writes <dev_path> <log_dev_path> |
| 414 | */ |
| 415 | static int log_writes_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
| 416 | { |
| 417 | struct log_writes_c *lc; |
| 418 | struct dm_arg_set as; |
| 419 | const char *devname, *logdevname; |
| 420 | |
| 421 | as.argc = argc; |
| 422 | as.argv = argv; |
| 423 | |
| 424 | if (argc < 2) { |
| 425 | ti->error = "Invalid argument count"; |
| 426 | return -EINVAL; |
| 427 | } |
| 428 | |
| 429 | lc = kzalloc(sizeof(struct log_writes_c), GFP_KERNEL); |
| 430 | if (!lc) { |
| 431 | ti->error = "Cannot allocate context"; |
| 432 | return -ENOMEM; |
| 433 | } |
| 434 | spin_lock_init(&lc->blocks_lock); |
| 435 | INIT_LIST_HEAD(&lc->unflushed_blocks); |
| 436 | INIT_LIST_HEAD(&lc->logging_blocks); |
| 437 | init_waitqueue_head(&lc->wait); |
| 438 | lc->sectorsize = 1 << SECTOR_SHIFT; |
| 439 | atomic_set(&lc->io_blocks, 0); |
| 440 | atomic_set(&lc->pending_blocks, 0); |
| 441 | |
| 442 | devname = dm_shift_arg(&as); |
| 443 | if (dm_get_device(ti, devname, dm_table_get_mode(ti->table), &lc->dev)) { |
| 444 | ti->error = "Device lookup failed"; |
| 445 | goto bad; |
| 446 | } |
| 447 | |
| 448 | logdevname = dm_shift_arg(&as); |
| 449 | if (dm_get_device(ti, logdevname, dm_table_get_mode(ti->table), &lc->logdev)) { |
| 450 | ti->error = "Log device lookup failed"; |
| 451 | dm_put_device(ti, lc->dev); |
| 452 | goto bad; |
| 453 | } |
| 454 | |
| 455 | lc->log_kthread = kthread_run(log_writes_kthread, lc, "log-write"); |
| 456 | if (!lc->log_kthread) { |
| 457 | ti->error = "Couldn't alloc kthread"; |
| 458 | dm_put_device(ti, lc->dev); |
| 459 | dm_put_device(ti, lc->logdev); |
| 460 | goto bad; |
| 461 | } |
| 462 | |
| 463 | /* We put the super at sector 0, start logging at sector 1 */ |
| 464 | lc->next_sector = 1; |
| 465 | lc->logging_enabled = true; |
| 466 | lc->end_sector = logdev_last_sector(lc); |
| 467 | lc->device_supports_discard = true; |
| 468 | |
| 469 | ti->num_flush_bios = 1; |
| 470 | ti->flush_supported = true; |
| 471 | ti->num_discard_bios = 1; |
| 472 | ti->discards_supported = true; |
| 473 | ti->per_bio_data_size = sizeof(struct per_bio_data); |
| 474 | ti->private = lc; |
| 475 | return 0; |
| 476 | |
| 477 | bad: |
| 478 | kfree(lc); |
| 479 | return -EINVAL; |
| 480 | } |
| 481 | |
| 482 | static int log_mark(struct log_writes_c *lc, char *data) |
| 483 | { |
| 484 | struct pending_block *block; |
| 485 | size_t maxsize = lc->sectorsize - sizeof(struct log_write_entry); |
| 486 | |
| 487 | block = kzalloc(sizeof(struct pending_block), GFP_KERNEL); |
| 488 | if (!block) { |
| 489 | DMERR("Error allocating pending block"); |
| 490 | return -ENOMEM; |
| 491 | } |
| 492 | |
| 493 | block->data = kstrndup(data, maxsize, GFP_KERNEL); |
| 494 | if (!block->data) { |
| 495 | DMERR("Error copying mark data"); |
| 496 | kfree(block); |
| 497 | return -ENOMEM; |
| 498 | } |
| 499 | atomic_inc(&lc->pending_blocks); |
| 500 | block->datalen = strlen(block->data); |
| 501 | block->flags |= LOG_MARK_FLAG; |
| 502 | spin_lock_irq(&lc->blocks_lock); |
| 503 | list_add_tail(&block->list, &lc->logging_blocks); |
| 504 | spin_unlock_irq(&lc->blocks_lock); |
| 505 | wake_up_process(lc->log_kthread); |
| 506 | return 0; |
| 507 | } |
| 508 | |
| 509 | static void log_writes_dtr(struct dm_target *ti) |
| 510 | { |
| 511 | struct log_writes_c *lc = ti->private; |
| 512 | |
| 513 | spin_lock_irq(&lc->blocks_lock); |
| 514 | list_splice_init(&lc->unflushed_blocks, &lc->logging_blocks); |
| 515 | spin_unlock_irq(&lc->blocks_lock); |
| 516 | |
| 517 | /* |
| 518 | * This is just nice to have since it'll update the super to include the |
| 519 | * unflushed blocks, if it fails we don't really care. |
| 520 | */ |
| 521 | log_mark(lc, "dm-log-writes-end"); |
| 522 | wake_up_process(lc->log_kthread); |
| 523 | wait_event(lc->wait, !atomic_read(&lc->io_blocks) && |
| 524 | !atomic_read(&lc->pending_blocks)); |
| 525 | kthread_stop(lc->log_kthread); |
| 526 | |
| 527 | WARN_ON(!list_empty(&lc->logging_blocks)); |
| 528 | WARN_ON(!list_empty(&lc->unflushed_blocks)); |
| 529 | dm_put_device(ti, lc->dev); |
| 530 | dm_put_device(ti, lc->logdev); |
| 531 | kfree(lc); |
| 532 | } |
| 533 | |
| 534 | static void normal_map_bio(struct dm_target *ti, struct bio *bio) |
| 535 | { |
| 536 | struct log_writes_c *lc = ti->private; |
| 537 | |
| 538 | bio->bi_bdev = lc->dev->bdev; |
| 539 | } |
| 540 | |
| 541 | static int log_writes_map(struct dm_target *ti, struct bio *bio) |
| 542 | { |
| 543 | struct log_writes_c *lc = ti->private; |
| 544 | struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data)); |
| 545 | struct pending_block *block; |
| 546 | struct bvec_iter iter; |
| 547 | struct bio_vec bv; |
| 548 | size_t alloc_size; |
| 549 | int i = 0; |
| 550 | bool flush_bio = (bio->bi_rw & REQ_FLUSH); |
| 551 | bool fua_bio = (bio->bi_rw & REQ_FUA); |
| 552 | bool discard_bio = (bio->bi_rw & REQ_DISCARD); |
| 553 | |
| 554 | pb->block = NULL; |
| 555 | |
| 556 | /* Don't bother doing anything if logging has been disabled */ |
| 557 | if (!lc->logging_enabled) |
| 558 | goto map_bio; |
| 559 | |
| 560 | /* |
| 561 | * Map reads as normal. |
| 562 | */ |
| 563 | if (bio_data_dir(bio) == READ) |
| 564 | goto map_bio; |
| 565 | |
| 566 | /* No sectors and not a flush? Don't care */ |
| 567 | if (!bio_sectors(bio) && !flush_bio) |
| 568 | goto map_bio; |
| 569 | |
| 570 | /* |
| 571 | * Discards will have bi_size set but there's no actual data, so just |
| 572 | * allocate the size of the pending block. |
| 573 | */ |
| 574 | if (discard_bio) |
| 575 | alloc_size = sizeof(struct pending_block); |
| 576 | else |
| 577 | alloc_size = sizeof(struct pending_block) + sizeof(struct bio_vec) * bio_segments(bio); |
| 578 | |
| 579 | block = kzalloc(alloc_size, GFP_NOIO); |
| 580 | if (!block) { |
| 581 | DMERR("Error allocating pending block"); |
| 582 | spin_lock_irq(&lc->blocks_lock); |
| 583 | lc->logging_enabled = false; |
| 584 | spin_unlock_irq(&lc->blocks_lock); |
| 585 | return -ENOMEM; |
| 586 | } |
| 587 | INIT_LIST_HEAD(&block->list); |
| 588 | pb->block = block; |
| 589 | atomic_inc(&lc->pending_blocks); |
| 590 | |
| 591 | if (flush_bio) |
| 592 | block->flags |= LOG_FLUSH_FLAG; |
| 593 | if (fua_bio) |
| 594 | block->flags |= LOG_FUA_FLAG; |
| 595 | if (discard_bio) |
| 596 | block->flags |= LOG_DISCARD_FLAG; |
| 597 | |
| 598 | block->sector = bio->bi_iter.bi_sector; |
| 599 | block->nr_sectors = bio_sectors(bio); |
| 600 | |
| 601 | /* We don't need the data, just submit */ |
| 602 | if (discard_bio) { |
| 603 | WARN_ON(flush_bio || fua_bio); |
| 604 | if (lc->device_supports_discard) |
| 605 | goto map_bio; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame^] | 606 | bio_endio(bio); |
Josef Bacik | 0e9cebe | 2015-03-20 10:50:37 -0400 | [diff] [blame] | 607 | return DM_MAPIO_SUBMITTED; |
| 608 | } |
| 609 | |
| 610 | /* Flush bio, splice the unflushed blocks onto this list and submit */ |
| 611 | if (flush_bio && !bio_sectors(bio)) { |
| 612 | spin_lock_irq(&lc->blocks_lock); |
| 613 | list_splice_init(&lc->unflushed_blocks, &block->list); |
| 614 | spin_unlock_irq(&lc->blocks_lock); |
| 615 | goto map_bio; |
| 616 | } |
| 617 | |
| 618 | /* |
| 619 | * We will write this bio somewhere else way later so we need to copy |
| 620 | * the actual contents into new pages so we know the data will always be |
| 621 | * there. |
| 622 | * |
| 623 | * We do this because this could be a bio from O_DIRECT in which case we |
| 624 | * can't just hold onto the page until some later point, we have to |
| 625 | * manually copy the contents. |
| 626 | */ |
| 627 | bio_for_each_segment(bv, bio, iter) { |
| 628 | struct page *page; |
| 629 | void *src, *dst; |
| 630 | |
| 631 | page = alloc_page(GFP_NOIO); |
| 632 | if (!page) { |
| 633 | DMERR("Error allocing page"); |
| 634 | free_pending_block(lc, block); |
| 635 | spin_lock_irq(&lc->blocks_lock); |
| 636 | lc->logging_enabled = false; |
| 637 | spin_unlock_irq(&lc->blocks_lock); |
| 638 | return -ENOMEM; |
| 639 | } |
| 640 | |
| 641 | src = kmap_atomic(bv.bv_page); |
| 642 | dst = kmap_atomic(page); |
| 643 | memcpy(dst, src + bv.bv_offset, bv.bv_len); |
| 644 | kunmap_atomic(dst); |
| 645 | kunmap_atomic(src); |
| 646 | block->vecs[i].bv_page = page; |
| 647 | block->vecs[i].bv_len = bv.bv_len; |
| 648 | block->vec_cnt++; |
| 649 | i++; |
| 650 | } |
| 651 | |
| 652 | /* Had a flush with data in it, weird */ |
| 653 | if (flush_bio) { |
| 654 | spin_lock_irq(&lc->blocks_lock); |
| 655 | list_splice_init(&lc->unflushed_blocks, &block->list); |
| 656 | spin_unlock_irq(&lc->blocks_lock); |
| 657 | } |
| 658 | map_bio: |
| 659 | normal_map_bio(ti, bio); |
| 660 | return DM_MAPIO_REMAPPED; |
| 661 | } |
| 662 | |
| 663 | static int normal_end_io(struct dm_target *ti, struct bio *bio, int error) |
| 664 | { |
| 665 | struct log_writes_c *lc = ti->private; |
| 666 | struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data)); |
| 667 | |
| 668 | if (bio_data_dir(bio) == WRITE && pb->block) { |
| 669 | struct pending_block *block = pb->block; |
| 670 | unsigned long flags; |
| 671 | |
| 672 | spin_lock_irqsave(&lc->blocks_lock, flags); |
| 673 | if (block->flags & LOG_FLUSH_FLAG) { |
| 674 | list_splice_tail_init(&block->list, &lc->logging_blocks); |
| 675 | list_add_tail(&block->list, &lc->logging_blocks); |
| 676 | wake_up_process(lc->log_kthread); |
| 677 | } else if (block->flags & LOG_FUA_FLAG) { |
| 678 | list_add_tail(&block->list, &lc->logging_blocks); |
| 679 | wake_up_process(lc->log_kthread); |
| 680 | } else |
| 681 | list_add_tail(&block->list, &lc->unflushed_blocks); |
| 682 | spin_unlock_irqrestore(&lc->blocks_lock, flags); |
| 683 | } |
| 684 | |
| 685 | return error; |
| 686 | } |
| 687 | |
| 688 | /* |
| 689 | * INFO format: <logged entries> <highest allocated sector> |
| 690 | */ |
| 691 | static void log_writes_status(struct dm_target *ti, status_type_t type, |
| 692 | unsigned status_flags, char *result, |
| 693 | unsigned maxlen) |
| 694 | { |
| 695 | unsigned sz = 0; |
| 696 | struct log_writes_c *lc = ti->private; |
| 697 | |
| 698 | switch (type) { |
| 699 | case STATUSTYPE_INFO: |
| 700 | DMEMIT("%llu %llu", lc->logged_entries, |
| 701 | (unsigned long long)lc->next_sector - 1); |
| 702 | if (!lc->logging_enabled) |
| 703 | DMEMIT(" logging_disabled"); |
| 704 | break; |
| 705 | |
| 706 | case STATUSTYPE_TABLE: |
| 707 | DMEMIT("%s %s", lc->dev->name, lc->logdev->name); |
| 708 | break; |
| 709 | } |
| 710 | } |
| 711 | |
| 712 | static int log_writes_ioctl(struct dm_target *ti, unsigned int cmd, |
| 713 | unsigned long arg) |
| 714 | { |
| 715 | struct log_writes_c *lc = ti->private; |
| 716 | struct dm_dev *dev = lc->dev; |
| 717 | int r = 0; |
| 718 | |
| 719 | /* |
| 720 | * Only pass ioctls through if the device sizes match exactly. |
| 721 | */ |
| 722 | if (ti->len != i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT) |
| 723 | r = scsi_verify_blk_ioctl(NULL, cmd); |
| 724 | |
| 725 | return r ? : __blkdev_driver_ioctl(dev->bdev, dev->mode, cmd, arg); |
| 726 | } |
| 727 | |
| 728 | static int log_writes_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
| 729 | struct bio_vec *biovec, int max_size) |
| 730 | { |
| 731 | struct log_writes_c *lc = ti->private; |
| 732 | struct request_queue *q = bdev_get_queue(lc->dev->bdev); |
| 733 | |
| 734 | if (!q->merge_bvec_fn) |
| 735 | return max_size; |
| 736 | |
| 737 | bvm->bi_bdev = lc->dev->bdev; |
| 738 | bvm->bi_sector = dm_target_offset(ti, bvm->bi_sector); |
| 739 | |
| 740 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); |
| 741 | } |
| 742 | |
| 743 | static int log_writes_iterate_devices(struct dm_target *ti, |
| 744 | iterate_devices_callout_fn fn, |
| 745 | void *data) |
| 746 | { |
| 747 | struct log_writes_c *lc = ti->private; |
| 748 | |
| 749 | return fn(ti, lc->dev, 0, ti->len, data); |
| 750 | } |
| 751 | |
| 752 | /* |
| 753 | * Messages supported: |
| 754 | * mark <mark data> - specify the marked data. |
| 755 | */ |
| 756 | static int log_writes_message(struct dm_target *ti, unsigned argc, char **argv) |
| 757 | { |
| 758 | int r = -EINVAL; |
| 759 | struct log_writes_c *lc = ti->private; |
| 760 | |
| 761 | if (argc != 2) { |
| 762 | DMWARN("Invalid log-writes message arguments, expect 2 arguments, got %d", argc); |
| 763 | return r; |
| 764 | } |
| 765 | |
| 766 | if (!strcasecmp(argv[0], "mark")) |
| 767 | r = log_mark(lc, argv[1]); |
| 768 | else |
| 769 | DMWARN("Unrecognised log writes target message received: %s", argv[0]); |
| 770 | |
| 771 | return r; |
| 772 | } |
| 773 | |
| 774 | static void log_writes_io_hints(struct dm_target *ti, struct queue_limits *limits) |
| 775 | { |
| 776 | struct log_writes_c *lc = ti->private; |
| 777 | struct request_queue *q = bdev_get_queue(lc->dev->bdev); |
| 778 | |
| 779 | if (!q || !blk_queue_discard(q)) { |
| 780 | lc->device_supports_discard = false; |
| 781 | limits->discard_granularity = 1 << SECTOR_SHIFT; |
| 782 | limits->max_discard_sectors = (UINT_MAX >> SECTOR_SHIFT); |
| 783 | } |
| 784 | } |
| 785 | |
| 786 | static struct target_type log_writes_target = { |
| 787 | .name = "log-writes", |
| 788 | .version = {1, 0, 0}, |
| 789 | .module = THIS_MODULE, |
| 790 | .ctr = log_writes_ctr, |
| 791 | .dtr = log_writes_dtr, |
| 792 | .map = log_writes_map, |
| 793 | .end_io = normal_end_io, |
| 794 | .status = log_writes_status, |
| 795 | .ioctl = log_writes_ioctl, |
| 796 | .merge = log_writes_merge, |
| 797 | .message = log_writes_message, |
| 798 | .iterate_devices = log_writes_iterate_devices, |
| 799 | .io_hints = log_writes_io_hints, |
| 800 | }; |
| 801 | |
| 802 | static int __init dm_log_writes_init(void) |
| 803 | { |
| 804 | int r = dm_register_target(&log_writes_target); |
| 805 | |
| 806 | if (r < 0) |
| 807 | DMERR("register failed %d", r); |
| 808 | |
| 809 | return r; |
| 810 | } |
| 811 | |
| 812 | static void __exit dm_log_writes_exit(void) |
| 813 | { |
| 814 | dm_unregister_target(&log_writes_target); |
| 815 | } |
| 816 | |
| 817 | module_init(dm_log_writes_init); |
| 818 | module_exit(dm_log_writes_exit); |
| 819 | |
| 820 | MODULE_DESCRIPTION(DM_NAME " log writes target"); |
| 821 | MODULE_AUTHOR("Josef Bacik <jbacik@fb.com>"); |
| 822 | MODULE_LICENSE("GPL"); |