drivers/md/dm-log-writes.c

   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
  58 #define WRITE_LOG_VERSION 1ULL
  59 #define WRITE_LOG_MAGIC 0x6a736677736872ULL
  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
 149 static void log_end_io(struct bio *bio)
 150 {
 151         struct log_writes_c *lc = bio->bi_private;
 152         struct bio_vec *bvec;
 153         int i;
 154
 155         if (bio->bi_error) {
 156                 unsigned long flags;
 157
 158                 DMERR("Error writing log block, error=%d", bio->bi_error);
 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;
 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         atomic_inc(&lc->io_blocks);
 262         bio = bio_alloc(GFP_KERNEL, min(block->vec_cnt, BIO_MAX_PAGES));
 263         if (!bio) {
 264                 DMERR("Couldn't alloc log bio");
 265                 goto error;
 266         }
 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;
 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, min(block->vec_cnt - i, BIO_MAX_PAGES));
 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;
 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         int ret;
 421
 422         as.argc = argc;
 423         as.argv = argv;
 424
 425         if (argc < 2) {
 426                 ti->error = "Invalid argument count";
 427                 return -EINVAL;
 428         }
 429
 430         lc = kzalloc(sizeof(struct log_writes_c), GFP_KERNEL);
 431         if (!lc) {
 432                 ti->error = "Cannot allocate context";
 433                 return -ENOMEM;
 434         }
 435         spin_lock_init(&lc->blocks_lock);
 436         INIT_LIST_HEAD(&lc->unflushed_blocks);
 437         INIT_LIST_HEAD(&lc->logging_blocks);
 438         init_waitqueue_head(&lc->wait);
 439         lc->sectorsize = 1 << SECTOR_SHIFT;
 440         atomic_set(&lc->io_blocks, 0);
 441         atomic_set(&lc->pending_blocks, 0);
 442
 443         devname = dm_shift_arg(&as);
 444         ret = dm_get_device(ti, devname, dm_table_get_mode(ti->table), &lc->dev);
 445         if (ret) {
 446                 ti->error = "Device lookup failed";
 447                 goto bad;
 448         }
 449
 450         logdevname = dm_shift_arg(&as);
 451         ret = dm_get_device(ti, logdevname, dm_table_get_mode(ti->table),
 452                             &lc->logdev);
 453         if (ret) {
 454                 ti->error = "Log device lookup failed";
 455                 dm_put_device(ti, lc->dev);
 456                 goto bad;
 457         }
 458
 459         lc->log_kthread = kthread_run(log_writes_kthread, lc, "log-write");
 460         if (IS_ERR(lc->log_kthread)) {
 461                 ret = PTR_ERR(lc->log_kthread);
 462                 ti->error = "Couldn't alloc kthread";
 463                 dm_put_device(ti, lc->dev);
 464                 dm_put_device(ti, lc->logdev);
 465                 goto bad;
 466         }
 467
 468         /* We put the super at sector 0, start logging at sector 1 */
 469         lc->next_sector = 1;
 470         lc->logging_enabled = true;
 471         lc->end_sector = logdev_last_sector(lc);
 472         lc->device_supports_discard = true;
 473
 474         ti->num_flush_bios = 1;
 475         ti->flush_supported = true;
 476         ti->num_discard_bios = 1;
 477         ti->discards_supported = true;
 478         ti->per_bio_data_size = sizeof(struct per_bio_data);
 479         ti->private = lc;
 480         return 0;
 481
 482 bad:
 483         kfree(lc);
 484         return ret;
 485 }
 486
 487 static int log_mark(struct log_writes_c *lc, char *data)
 488 {
 489         struct pending_block *block;
 490         size_t maxsize = lc->sectorsize - sizeof(struct log_write_entry);
 491
 492         block = kzalloc(sizeof(struct pending_block), GFP_KERNEL);
 493         if (!block) {
 494                 DMERR("Error allocating pending block");
 495                 return -ENOMEM;
 496         }
 497
 498         block->data = kstrndup(data, maxsize, GFP_KERNEL);
 499         if (!block->data) {
 500                 DMERR("Error copying mark data");
 501                 kfree(block);
 502                 return -ENOMEM;
 503         }
 504         atomic_inc(&lc->pending_blocks);
 505         block->datalen = strlen(block->data);
 506         block->flags |= LOG_MARK_FLAG;
 507         spin_lock_irq(&lc->blocks_lock);
 508         list_add_tail(&block->list, &lc->logging_blocks);
 509         spin_unlock_irq(&lc->blocks_lock);
 510         wake_up_process(lc->log_kthread);
 511         return 0;
 512 }
 513
 514 static void log_writes_dtr(struct dm_target *ti)
 515 {
 516         struct log_writes_c *lc = ti->private;
 517
 518         spin_lock_irq(&lc->blocks_lock);
 519         list_splice_init(&lc->unflushed_blocks, &lc->logging_blocks);
 520         spin_unlock_irq(&lc->blocks_lock);
 521
 522         /*
 523          * This is just nice to have since it'll update the super to include the
 524          * unflushed blocks, if it fails we don't really care.
 525          */
 526         log_mark(lc, "dm-log-writes-end");
 527         wake_up_process(lc->log_kthread);
 528         wait_event(lc->wait, !atomic_read(&lc->io_blocks) &&
 529                    !atomic_read(&lc->pending_blocks));
 530         kthread_stop(lc->log_kthread);
 531
 532         WARN_ON(!list_empty(&lc->logging_blocks));
 533         WARN_ON(!list_empty(&lc->unflushed_blocks));
 534         dm_put_device(ti, lc->dev);
 535         dm_put_device(ti, lc->logdev);
 536         kfree(lc);
 537 }
 538
 539 static void normal_map_bio(struct dm_target *ti, struct bio *bio)
 540 {
 541         struct log_writes_c *lc = ti->private;
 542
 543         bio->bi_bdev = lc->dev->bdev;
 544 }
 545
 546 static int log_writes_map(struct dm_target *ti, struct bio *bio)
 547 {
 548         struct log_writes_c *lc = ti->private;
 549         struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data));
 550         struct pending_block *block;
 551         struct bvec_iter iter;
 552         struct bio_vec bv;
 553         size_t alloc_size;
 554         int i = 0;
 555         bool flush_bio = (bio->bi_rw & REQ_FLUSH);
 556         bool fua_bio = (bio->bi_rw & REQ_FUA);
 557         bool discard_bio = (bio->bi_rw & REQ_DISCARD);
 558
 559         pb->block = NULL;
 560
 561         /* Don't bother doing anything if logging has been disabled */
 562         if (!lc->logging_enabled)
 563                 goto map_bio;
 564
 565         /*
 566          * Map reads as normal.
 567          */
 568         if (bio_data_dir(bio) == READ)
 569                 goto map_bio;
 570
 571         /* No sectors and not a flush?  Don't care */
 572         if (!bio_sectors(bio) && !flush_bio)
 573                 goto map_bio;
 574
 575         /*
 576          * Discards will have bi_size set but there's no actual data, so just
 577          * allocate the size of the pending block.
 578          */
 579         if (discard_bio)
 580                 alloc_size = sizeof(struct pending_block);
 581         else
 582                 alloc_size = sizeof(struct pending_block) + sizeof(struct bio_vec) * bio_segments(bio);
 583
 584         block = kzalloc(alloc_size, GFP_NOIO);
 585         if (!block) {
 586                 DMERR("Error allocating pending block");
 587                 spin_lock_irq(&lc->blocks_lock);
 588                 lc->logging_enabled = false;
 589                 spin_unlock_irq(&lc->blocks_lock);
 590                 return -ENOMEM;
 591         }
 592         INIT_LIST_HEAD(&block->list);
 593         pb->block = block;
 594         atomic_inc(&lc->pending_blocks);
 595
 596         if (flush_bio)
 597                 block->flags |= LOG_FLUSH_FLAG;
 598         if (fua_bio)
 599                 block->flags |= LOG_FUA_FLAG;
 600         if (discard_bio)
 601                 block->flags |= LOG_DISCARD_FLAG;
 602
 603         block->sector = bio->bi_iter.bi_sector;
 604         block->nr_sectors = bio_sectors(bio);
 605
 606         /* We don't need the data, just submit */
 607         if (discard_bio) {
 608                 WARN_ON(flush_bio || fua_bio);
 609                 if (lc->device_supports_discard)
 610                         goto map_bio;
 611                 bio_endio(bio);
 612                 return DM_MAPIO_SUBMITTED;
 613         }
 614
 615         /* Flush bio, splice the unflushed blocks onto this list and submit */
 616         if (flush_bio && !bio_sectors(bio)) {
 617                 spin_lock_irq(&lc->blocks_lock);
 618                 list_splice_init(&lc->unflushed_blocks, &block->list);
 619                 spin_unlock_irq(&lc->blocks_lock);
 620                 goto map_bio;
 621         }
 622
 623         /*
 624          * We will write this bio somewhere else way later so we need to copy
 625          * the actual contents into new pages so we know the data will always be
 626          * there.
 627          *
 628          * We do this because this could be a bio from O_DIRECT in which case we
 629          * can't just hold onto the page until some later point, we have to
 630          * manually copy the contents.
 631          */
 632         bio_for_each_segment(bv, bio, iter) {
 633                 struct page *page;
 634                 void *src, *dst;
 635
 636                 page = alloc_page(GFP_NOIO);
 637                 if (!page) {
 638                         DMERR("Error allocing page");
 639                         free_pending_block(lc, block);
 640                         spin_lock_irq(&lc->blocks_lock);
 641                         lc->logging_enabled = false;
 642                         spin_unlock_irq(&lc->blocks_lock);
 643                         return -ENOMEM;
 644                 }
 645
 646                 src = kmap_atomic(bv.bv_page);
 647                 dst = kmap_atomic(page);
 648                 memcpy(dst, src + bv.bv_offset, bv.bv_len);
 649                 kunmap_atomic(dst);
 650                 kunmap_atomic(src);
 651                 block->vecs[i].bv_page = page;
 652                 block->vecs[i].bv_len = bv.bv_len;
 653                 block->vec_cnt++;
 654                 i++;
 655         }
 656
 657         /* Had a flush with data in it, weird */
 658         if (flush_bio) {
 659                 spin_lock_irq(&lc->blocks_lock);
 660                 list_splice_init(&lc->unflushed_blocks, &block->list);
 661                 spin_unlock_irq(&lc->blocks_lock);
 662         }
 663 map_bio:
 664         normal_map_bio(ti, bio);
 665         return DM_MAPIO_REMAPPED;
 666 }
 667
 668 static int normal_end_io(struct dm_target *ti, struct bio *bio, int error)
 669 {
 670         struct log_writes_c *lc = ti->private;
 671         struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data));
 672
 673         if (bio_data_dir(bio) == WRITE && pb->block) {
 674                 struct pending_block *block = pb->block;
 675                 unsigned long flags;
 676
 677                 spin_lock_irqsave(&lc->blocks_lock, flags);
 678                 if (block->flags & LOG_FLUSH_FLAG) {
 679                         list_splice_tail_init(&block->list, &lc->logging_blocks);
 680                         list_add_tail(&block->list, &lc->logging_blocks);
 681                         wake_up_process(lc->log_kthread);
 682                 } else if (block->flags & LOG_FUA_FLAG) {
 683                         list_add_tail(&block->list, &lc->logging_blocks);
 684                         wake_up_process(lc->log_kthread);
 685                 } else
 686                         list_add_tail(&block->list, &lc->unflushed_blocks);
 687                 spin_unlock_irqrestore(&lc->blocks_lock, flags);
 688         }
 689
 690         return error;
 691 }
 692
 693 /*
 694  * INFO format: <logged entries> <highest allocated sector>
 695  */
 696 static void log_writes_status(struct dm_target *ti, status_type_t type,
 697                               unsigned status_flags, char *result,
 698                               unsigned maxlen)
 699 {
 700         unsigned sz = 0;
 701         struct log_writes_c *lc = ti->private;
 702
 703         switch (type) {
 704         case STATUSTYPE_INFO:
 705                 DMEMIT("%llu %llu", lc->logged_entries,
 706                        (unsigned long long)lc->next_sector - 1);
 707                 if (!lc->logging_enabled)
 708                         DMEMIT(" logging_disabled");
 709                 break;
 710
 711         case STATUSTYPE_TABLE:
 712                 DMEMIT("%s %s", lc->dev->name, lc->logdev->name);
 713                 break;
 714         }
 715 }
 716
 717 static int log_writes_prepare_ioctl(struct dm_target *ti,
 718                 struct block_device **bdev, fmode_t *mode)
 719 {
 720         struct log_writes_c *lc = ti->private;
 721         struct dm_dev *dev = lc->dev;
 722
 723         *bdev = dev->bdev;
 724         /*
 725          * Only pass ioctls through if the device sizes match exactly.
 726          */
 727         if (ti->len != i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT)
 728                 return 1;
 729         return 0;
 730 }
 731
 732 static int log_writes_iterate_devices(struct dm_target *ti,
 733                                       iterate_devices_callout_fn fn,
 734                                       void *data)
 735 {
 736         struct log_writes_c *lc = ti->private;
 737
 738         return fn(ti, lc->dev, 0, ti->len, data);
 739 }
 740
 741 /*
 742  * Messages supported:
 743  *   mark <mark data> - specify the marked data.
 744  */
 745 static int log_writes_message(struct dm_target *ti, unsigned argc, char **argv)
 746 {
 747         int r = -EINVAL;
 748         struct log_writes_c *lc = ti->private;
 749
 750         if (argc != 2) {
 751                 DMWARN("Invalid log-writes message arguments, expect 2 arguments, got %d", argc);
 752                 return r;
 753         }
 754
 755         if (!strcasecmp(argv[0], "mark"))
 756                 r = log_mark(lc, argv[1]);
 757         else
 758                 DMWARN("Unrecognised log writes target message received: %s", argv[0]);
 759
 760         return r;
 761 }
 762
 763 static void log_writes_io_hints(struct dm_target *ti, struct queue_limits *limits)
 764 {
 765         struct log_writes_c *lc = ti->private;
 766         struct request_queue *q = bdev_get_queue(lc->dev->bdev);
 767
 768         if (!q || !blk_queue_discard(q)) {
 769                 lc->device_supports_discard = false;
 770                 limits->discard_granularity = 1 << SECTOR_SHIFT;
 771                 limits->max_discard_sectors = (UINT_MAX >> SECTOR_SHIFT);
 772         }
 773 }
 774
 775 static struct target_type log_writes_target = {
 776         .name   = "log-writes",
 777         .version = {1, 0, 0},
 778         .module = THIS_MODULE,
 779         .ctr    = log_writes_ctr,
 780         .dtr    = log_writes_dtr,
 781         .map    = log_writes_map,
 782         .end_io = normal_end_io,
 783         .status = log_writes_status,
 784         .prepare_ioctl = log_writes_prepare_ioctl,
 785         .message = log_writes_message,
 786         .iterate_devices = log_writes_iterate_devices,
 787         .io_hints = log_writes_io_hints,
 788 };
 789
 790 static int __init dm_log_writes_init(void)
 791 {
 792         int r = dm_register_target(&log_writes_target);
 793
 794         if (r < 0)
 795                 DMERR("register failed %d", r);
 796
 797         return r;
 798 }
 799
 800 static void __exit dm_log_writes_exit(void)
 801 {
 802         dm_unregister_target(&log_writes_target);
 803 }
 804
 805 module_init(dm_log_writes_init);
 806 module_exit(dm_log_writes_exit);
 807
 808 MODULE_DESCRIPTION(DM_NAME " log writes target");
 809 MODULE_AUTHOR("Josef Bacik <jbacik@fb.com>");
 810 MODULE_LICENSE("GPL");